JP2006209331A - Device diagnostic device and measuring device diagnostic device on network - Google Patents

Abstract

A device having an abnormal device characteristic parameter value and a measuring device having an abnormal measurement value are identified and the diagnosis result is displayed and output on a screen.
An optimization unit and a device characteristic equation or a stationary network simulator are provided, and a current measurement value and a specified value of a device characteristic parameter value are input to the optimization unit as input / output energy initial values of the target device, The current device characteristic parameter value is input to the equation, etc., and the calculated value for the measured value is obtained. The optimization unit uses the measured value and the calculated value based on the device characteristic equation, etc. to optimize the device characteristic parameter of all devices. The process of obtaining the value is repeated to output the optimum parameter value, the value calculated from the deviation between the optimum parameter value and the specified value and the weighting coefficient is compared with the set value to obtain the device group in which the device characteristic parameter value has changed. When the measurement value of the measuring instrument related to the device group is abnormal, the device having the abnormal device characteristic parameter value is identified from the device group by the optimization method.
[Selection] Figure 7

Description

  The present invention is directed to networks in which electricity, gas, water, steam, and the like are input and output, and among the devices constituting these networks, devices having abnormal device characteristic parameter values and measuring devices having abnormal measurement values This invention relates to a device diagnostic apparatus and a measuring instrument diagnostic apparatus on a network that can identify the result and display the diagnosis result on a screen.

When a device that constitutes a network through which a liquid or gas such as water (hot water / cold water) or steam for electricity, gas, or air conditioning flows is controlled by a control computer or a stand-alone computer, measurement was performed in relation to the device. Measurement values such as electrical output and heat output are input to a computer to calculate the controlled variable. At this time, it is necessary for the computer to grasp the characteristics of the devices constituting the network and execute control within the normal range of the characteristics and measured values.
For this reason, abnormality diagnosis of device characteristics and measurement values is extremely important in performing control.

Conventionally, as this kind of abnormality diagnosis apparatus, an apparatus for diagnosing abnormality of a measurement value obtained from a network is known.
For example, in Patent Document 1, the distortion or sharpness of a measurement value and the number of Geary are obtained, and when these values exceed a preset threshold, it is determined as abnormal.
Moreover, in patent document 2, abnormality is diagnosed by calculating | requiring the power spectrum of the time series data of a measured value.

JP-A-5-142000 (paragraphs [0006] to [0017], FIG. 1 etc.) JP-A-7-244522 (paragraphs [0012] to [0040], FIG. 1 etc.)

In the above-mentioned conventional technology, abnormality diagnosis is performed using only the measurement values of the measuring instrument alone, but the characteristics of the devices that make up the network have changed over time, and the measured values are improperly adjusted. Depending on the cause, it may deviate from the normal value range. Further, when viewed individually, there may be a case where a device failure or the like has occurred even though it is within a normal range.
Under such circumstances, it is necessary to identify a device with abnormal characteristics or a measuring instrument with abnormal measurement values, and perform repair or adjustment as necessary. Furthermore, it is desirable that normal device characteristic values and measurement values can be acquired and control can be continued even after an abnormality has occurred.

Therefore, the present invention considers the relationship between a plurality of actual data on the network while ensuring the same time property of the measurement values, and uses the network-related equations to make various non-linear device characteristics and operational rules that cannot be formulated. Using a stationary network simulator or the like as another program that considers all of the above, and using optimization methods such as enumeration, genetic algorithm or its improved method, tab search or its improved method, Particle Swarm Optimization or its improved method , To provide a device / measurement device diagnostic device that identifies devices with abnormal device characteristic parameter values and measuring devices with abnormal measurement values, and displays the diagnosis results on the screen. Is.
In addition, by displaying and displaying the above diagnostic results on the screen, it is possible to realize Condition-based Maintenance (CBM) for various devices and measuring instruments and their associated equipment, and normal device characteristic parameter values and measured values. It is intended to provide a device / measuring device diagnostic apparatus that can continuously control a network using only the network.

In order to solve the above-described problem, the device diagnostic apparatus according to claim 1 is an error between a measured value and a calculated value of energy such as electricity, gas, water, and steam inputted to and outputted from each device constituting the network. And an optimization unit that obtains and outputs each device characteristic parameter value as a state variable by an optimization method using an objective function that minimizes an error from a specified value of each device characteristic parameter value;
It has a device model, device operation rules, etc. of each device that constitutes the network inside, and based on each device characteristic parameter value input from the optimization unit, it calculates the output energy relative to the input energy of the device. The device characteristic equation that is output as a calculated value for the device, or the device model of each device that constitutes the network, the device operation rule, the connection state of each device, and the equation relating to the energy flow of the network, etc. Under each device characteristic parameter value input from the unit, a steady network simulator that obtains a state quantity of each unit in the network in a time section with a steady state and outputs it as a calculated value for the measured value, and
Input the current measured value as the initial value of the input / output energy of the target device to the optimization unit, input the specified value of the device characteristic parameter value, and input the current device characteristic parameter value to the device characteristic equation or steady network simulator To obtain the calculated value for the measured value,
The optimization unit repeats the process of obtaining the device characteristic parameter values of all devices by the optimization method using the input measurement value and the device characteristic equation or the calculated value obtained by the steady network simulator. The parameter value when the set number of times is reached is output as the optimum parameter value for each device characteristic equation.
Further, a device group in which the device characteristic parameter value is changed is obtained by comparing a value calculated based on a deviation between the optimum parameter value and the specified value and a predetermined weighting factor with a set value, and measuring instruments related to these device groups. When the measured value is abnormal, the device having the abnormal device characteristic parameter value is identified from the device group using the optimization method.

  A device diagnosis apparatus according to a second aspect uses the enumeration method as an optimization method for identifying a device having an abnormal device characteristic parameter value in the first aspect.

  A device diagnosis apparatus according to claim 3 uses a genetic algorithm or an improved method thereof as an optimization method for identifying a device having an abnormal device characteristic parameter value.

  A device diagnostic apparatus according to a fourth aspect uses a tab search or an improved method thereof as an optimization method for identifying a device whose device characteristic parameter value is abnormal in the first aspect.

  The device diagnostic apparatus according to claim 5 uses Particle Swarm Optimization or an improved method thereof as an optimization method for identifying a device having an abnormal device characteristic parameter value.

  The measuring instrument diagnostic device according to claim 6 has an abnormal measurement value for a measuring instrument related to the device specified by the device diagnostic device according to any one of claims 1 to 5. A measuring instrument is specified using an optimization method.

  The device diagnostic device and the measuring device diagnostic device according to claim 7 are information related to the device specified by the device diagnostic device according to any one of claims 1 to 5 and the measuring device diagnosis according to claim 6. Information relating to a measuring instrument identified as an abnormal measurement value by the apparatus and a measurement correction value estimated to be most likely for the abnormal measurement value are displayed on the screen.

According to the present invention, among the devices constituting the network in which electricity, gas, water, steam and the like are input and output, the device having the abnormal device characteristic parameter value and the measuring device having the abnormal measurement value are identified. The diagnosis result can be displayed on the screen. Moreover, it is also possible to implement Condition-based Maintenance (CBM) for equipment, measuring instruments, and facilities associated therewith based on the diagnosis results.
Furthermore, by replacing the measurement value determined to be abnormal with the measurement correction value estimated to be most likely, the plant or the like can be operated continuously while using this measurement correction value together with the normal measurement value.

Hereinafter, embodiments of the present invention will be described with reference to the drawings.
1. First, a stationary network simulator used in the present invention will be described.
The stationary network simulator calculates the steady state of the entire network in a certain time section by modeling the network and its constituent devices and solving equations relating to the energy flow of each device and the network.

This type of network simulator is an existing technology described as a tidal current calculation technology in the “Power System Analysis Theory” (Yoshiji Sekine, Denki Shoin, January 1971), for example, regarding power energy. The flow and water supply is an existing technology described as a pipeline network analysis technique in "Introduction to Computer Hydraulics" (CA Brevia / AJ Ferrant, translated by Masahiko Isobe, Science, 1983). Furthermore, in the case of water for air conditioning (hot water / cold water), the heat balance equation is obtained.
The energy flow equation is formulated as a linear or non-linear simultaneous equation, and by using the Newton method or the like, a solution, that is, a steady state of the network in a certain time section can be obtained.

(1) A device model used for a stationary network simulator is expressed as a branch, and its input / output relationship is expressed by a function (equipment characteristic equation) of Equation 1.

  Here, the operation rule of the device is also expressed in the function f. The input state quantity can be calculated from the input energy for the branch, and the output energy output from the branch can also be calculated using the output state quantity. In addition, since power transmission lines, water pipes, gas pipes and the like generally include a resistance component, a branch in which energy loss is caused by this resistance component is also expressed by Equation 1.

(2) Next, a network model used for a stationary network simulator will be described. A network model is formed by connecting a device model as a branch by a large number of nodes. When the input energy and the output energy are balanced in each node, it can be expressed by the following function.

(3) The steady network simulation result is obtained by solving the equation by designating the input energy and output energy for the network for the nonlinear simultaneous equation of Equation 2. Each device in the network can express the input / output relationship by designating the parameter of the characteristic function as shown in Equation 1, and in the present invention, this parameter value a _j (called device characteristic parameter value) is finally given. ).
The stationary network simulator has equations relating to the device model and the energy flow of the network inside, and outputs state quantities of each part in the network in a certain time section by designating input energy and output energy.

2. Next, a first algorithm for specifying a device having an abnormal device characteristic parameter value in claim 1 will be described.
(1) Estimation of device characteristic parameter value of each device First, for example, the device characteristic parameter value is estimated by the method described in Japanese Patent Application No. 2003-284292 which is a prior application.
First, the state variable is set as a parameter value in the characteristic equation of the target device. In the formulation, it is assumed that the input / output state quantities of the target device can be measured (measured).

  As a formulation for estimating the optimum parameter, as shown in Equation 3, the objective function is to minimize the norm of the output measurement value and the calculated value at the measurement point and the norm of the deviation from the specified value of the parameter. . In addition, Formula 4 shows a constraint condition.

Here, W _mi1 is determined by the method described later. a _refl1 is a parameter value of a characteristic formula obtained from experience or a parameter value of a characteristic formula obtained from a catalog value. _Further , values such as w _pj1 , W ₁ , W ₂ , p ₁ , _{q 1} , a _{11, min} , a _{11, max} are set in advance.
Optimization methods include device characteristic parameter values so that the measured values of all measurement point combinations to be calculated match the calculated values at the measurement points, and so that each parameter value does not deviate as much as possible from the specified value. Will be adjusted.

Next, the basic concept of the optimum parameter estimation method will be described.
The state equation h _i is calculated from the characteristic formula of the target device. As described above, this characteristic equation includes device characteristics, device operation rules, and the like, and output energy is calculated when input energy of the device is input. Using this output energy, it is possible to evaluate the measured value and the calculated value by Equation 3.
Here, the optimization problem that minimizes the objective function of Equation 3 in consideration of the constraints of Equation 4 is expressed as a metaheuristic optimization method (genetic algorithm or its improved method, tab search or its improved method, Particle Swarm Optimization Alternatively, it is solved using the improved method.

First, a method for determining the weighting factor w _mi1 in Equation 3 will be described below.
w _mi1 represents the relative weight of each measurement value with respect to the certainty of the measurement value. This weight can be evaluated by the following two items.
a. Measurement sensor accuracy (Sensor _PRE )
b. Measurement sensor failure rate (Sensor _FAU )
The accuracy of the measurement sensor is expressed in a% error range such as ± 5%. The failure rate is expressed as the number of failures per year (number of failures / year). Using these values, the weighting factor w _mi1 for each measurement value is determined as in Equation 5.

The weighting factor w _mi1 can be expressed by Equation 6, for example. Note that a and b in Equation 6 are specified in advance.

Next, the device characteristic parameter value is obtained by the following algorithm.
Step 1: The combination of the device characteristic equation information of the target device, the specified value for each parameter of the device characteristic equation, and the target measurement value is input to the optimization unit.
Step 2: The current state variable value (each device characteristic parameter value) is input to a device characteristic formula including device characteristics, device operation rules, etc., and a calculated value for the output measurement value is obtained.
Step 3: The optimization unit obtains an evaluation value using the output measurement value input in Step 1 and the calculated value for the output measurement value calculated in Step 2 and Equations 3 to 5.
Step 4: Using the metaheuristic optimization method (genetic algorithm or its improved method, tab search or its improved method, or PSO or its improved method), the current state variable value is calculated based on Equations 3 and 4. change.
Step 5: If the number of searches set in advance is reached, proceed to Step 6; otherwise, return to Step 2.
Step 6: Output the current parameter value as the optimum parameter value.

  Here, FIG. 1 is a diagram showing a calculation concept of the optimum parameter value in the present embodiment. In FIG. 1, reference numeral 10 denotes a current state for the purpose of minimizing an error between a measured value and a calculated value and minimizing an error between a state variable (equipment characteristic parameter value) and a parameter specified value by a metaheuristic optimization method. An optimization unit 20 that changes a variable value (solves Equations 3 and 4), and 20 is an input that is input in consideration of device characteristics, operation rules, etc., using state variables (device characteristic parameter values) from the optimization unit 10 The device characteristic formula for calculating the output energy relative to the energy and outputting it as a calculated value is shown.

(2) Identification of device whose device characteristic parameter value has changed The following Formula 7 is calculated to obtain the f ₁ value.

Here, a _{ref, i} is a parameter value obtained from experience, a parameter value obtained from input / output data of a device, and a parameter value obtained from a catalog value.
When the f ₁ value calculated by Equation 7 is larger than the preset value D _evj for each device, the device is identified as a device whose device characteristic parameter value has changed.

(3) Identification of a device having an abnormal device characteristic parameter value Using a plurality of devices whose device characteristic parameter values obtained in the above (2) have changed, the parameter value is determined by various optimization methods described later. Identify the device that is malfunctioning.
If there is no device whose device characteristic parameter value has changed in (2) above, the process returns to the parameter value estimation processing in (1).

3. Next, a second algorithm for identifying a device whose device characteristic parameter value is abnormal will be described.
(1) Estimation of device characteristic parameter value of each device The parameter value is estimated by the same method as the first algorithm.
(2) Identification of device whose device characteristic parameter value has changed The following Formula 8 is calculated to obtain the f ₂ value.

Here, a _{ref, i} is a parameter value obtained from experience, a parameter value obtained from input / output data of a device, and a parameter value obtained from a catalog value.
When the f ₁ value calculated by Equation 7 is larger than the preset value D _evj for each device, the device is identified as a device whose device characteristic parameter value has changed.

(3) Identification of a device whose device characteristic parameter value is abnormal Using the device (s) whose device characteristic parameter value obtained in (2) above has changed, the parameter value is determined by various optimization methods described below. Identify the device that is abnormal.
If there is no device whose device characteristic parameter value has changed in (2) above, the process returns to the parameter value estimation processing in (1).

4). Next, a method for identifying a device having an abnormal device characteristic parameter value using an enumeration method will be described.
In the first and second algorithms described above, (3) an embodiment in which an enumeration method is used as an optimization method will be described as a method for identifying a device whose device characteristic parameter value is abnormal.
In this embodiment, based on the assumption that the device characteristic parameter value related to the device in which the measured value is abnormal is most changed (becomes abnormal), the parameter values for each device are set as a set. When the parameter value of the device is changed again to the one before the change, the order in ascending order is obtained from the combination in which the measured value and the calculated value are best matched. At this time, the device changed to the parameter value before the change Is identified as a device having an abnormal device characteristic parameter value.

(1) Expression of state variables An array having the number of elements corresponding to the number of target devices is prepared with the parameter values for each device as a set. Each element of the array represents a response to the parameter value for each corresponding numbered target device, and represents the element with a binary value of 1, 0. Here, 1 indicates that the current parameter value before correction is used, and 0 indicates that the calculated new parameter value is used.

(2) Solution algorithm step 1: Generation of parameter value combinations 1 or 0 elements are generated for each element of the array, and 2 ^NumEquip values (in this case, target device) A combination of parameter values is generated by fixing the parameter value as a state variable to the current parameter value for those using the current parameter value under the designation.

Step 2: Calculation of Measurement Correction Values for Each Combination Measurement correction values for all combinations generated in Step 1 are measured by the method described in Japanese Patent Application No. 2004-218788 “Measurement Value Scrutinization Device in Network”, which is a prior application, for example. Calculate
In addition, the above-mentioned "measurement value scrutinizing device in the network" minimizes the error between the measured value and the calculated value of energy such as electricity, gas, water, and steam that are input to and output from each device constituting the network. Using an objective function, the optimization unit that calculates and outputs the network input / output energy as a state variable using the metaheuristic optimization method, the device model of each device that configures the network, device operation rules, and the connection state of each device And a linear or non-linear equation related to the energy flow of the network, etc., which is stationary under the operation / stop state of each component device input from the optimization unit or input / output energy The amount of state of each part in the network in a time section with a state is measured by sequentially calculating from the device on the input side toward the end. A steady-state network simulator that outputs as a calculated value for, inputs the current measured value as the initial value of the network input / output energy to the optimization unit, and inputs the current input / output energy to the steady-state network simulator The optimization unit repeatedly calculates the input and output energy using the metaheuristic optimization method using the input measurement value and the calculation value obtained by the steady network simulator, and the number of times is calculated in advance. The calculated value for the current measurement value obtained by the stationary network simulator when the set number of times is reached is output as the most probable measurement correction value. As a metaheuristic optimization method by the optimization unit, genetic Algorithm or its improved method, tab search It is intended to use any of the improvements techniques, Particle Swarm Optimization or improved techniques.

Step 3: Determination of the combination with the smallest calculation error for each combination Combination with the smallest error (evaluation value) between the measurement correction value and the measured value in Step 2 above for each combination (the smallest objective function value) Store in order.
• Step 4: Identifying devices with abnormal device characteristic parameter values In order from the combination with the smallest evaluation value, the device combination that has been changed to the parameter value before the change (current parameter value) is determined to be abnormal. Identifies and outputs as a device.

5. Next, a method for identifying a device having an abnormal device characteristic parameter value using a genetic algorithm (GA) as an optimization method or an improved method thereof will be described.
Even in this embodiment, based on the assumption that the device characteristic parameter value related to the device in which the measured value is abnormal is most changed (becomes abnormal), several parameter values for each device are used as a set. When the parameter value of the device is re-changed to the one before the change, the ascending order is obtained from the combination in which the measured value and the calculated value are best matched. At this time, the device value changed to the parameter value before the change is obtained. The combination is identified as a device whose device characteristic parameter value is abnormal.
Here, the genetic algorithm or its improved method is the Simple Genetic Algorithm (hereinafter referred to as SGA) described in DE Goldberg, Genetic Algorithms in Search, Optimization, and Machine Learning, Addison-Wesley, 1989, and its algorithm. Refers to improved methods. In the following, an optimal parameter estimation method will be described using this SGA as an example.

(1) Expression of state variables Number the target device characteristic parameter values. Prepare genes with as many loci as there are target parameters. Each gene locus represents a response to a parameter for each corresponding numbered target device, and represents an element with a binary value of 1, 0. Here, 1 indicates that the current parameter value before correction is used, and 0 indicates that the calculated new parameter value is used.

(2) Solving algorithm step 1: Steps for setting preconditions and initial conditions 1) Input each measured value.
2) Set the number of strings, crossing probability, mutation probability, and maximum number of generations.
Step 2: Generation of initial values 1) A binary value (1, 0) is randomly generated for each string locus.
2) The current generation number is 1.
3) The best evaluation value (Obj ^* [0]) is set to infinity, and the current minimum value number is set to Optnum = 0. Here, Obj ^* [] is an array for storing the best value.

-Step 3: Evaluation and selection of each string 1) Specify whether to use the current parameter value or the modified parameter for each device according to the 1, 0 value of the locus of the string, and under the designation, For those using the current parameter value, the parameter value to be the state variable is fixed to the current parameter value, and the measurement correction value and the measurement correction value are used by using the method described in Japanese Patent Application No. 2004-218788 described above. Calculate the objective function (evaluation function) value for the hour.
2) A string is selected by roulette wheel selection using the reciprocal of the evaluation function value (evaluation value) as a Fitness function value.
3) If the minimum value of the evaluation values of the selected string is smaller than the current Obj ^* [Optnum], Optnum = Optnum + 1 is set, and the evaluation value is set to Obj ^* [Optnum]. The value of the state variable at that time is X ^* [Optnum].

Step 4: String manipulation Crossing and mutation are performed on the string set using crossing probability and mutation probability.
Step 5: Determination of the end of search by GA When the generation reaches a predetermined value, the process proceeds to Step 6, and when the generation does not reach the predetermined value, 1 is added to the number of generations and Step 3 Return to.
-Step 6: Identifying the device that is abnormal
In X ^* [Optnum] to X ^* [1], in this order, the device that uses the current parameter value is output as a device whose device characteristic parameter value is abnormal, and the process ends.

6). Next, a method for identifying a device having an abnormal device characteristic parameter value using a tab search as an optimization method or an improved method thereof will be described.
Even in this embodiment, based on the assumption that the device characteristic parameter value related to the device in which the measured value is abnormal is most changed (becomes abnormal), several parameter values for each device are used as a set. When the parameter value of the device is re-changed to the one before the change, the ascending order is obtained from the combination in which the measured value and the calculated value are best matched. At this time, the device value changed to the parameter value before the change is obtained. The combination is identified as a device whose device characteristic parameter value is abnormal.
At that time, a tab search or an improved technique thereof is used for a portion for obtaining a combination that best matches the measured value and the calculated value.
Here, tab search or its improved method is tabu search (hereinafter referred to as TS) described in F. Glover, "Tabu Search Part I", ORSA Journal of Computing, Vol. 1, NO. 3, Summer 1989. And its improved method. Hereinafter, a method for identifying a device having an abnormal device characteristic parameter value will be described using this TS as an example.

(1) Expression of state variables Number the parameters of the target device characteristic equation. Prepare an array with as many elements as the target number of parameters. Each element of the array represents a response to a parameter for each corresponding numbered target device, and represents an element with a binary value of 1, 0. Here, 1 indicates that the current parameter value before correction is used, and 0 indicates that the calculated new parameter value is used.

(2) Solution Algorithm Step 1: Precondition and Initial Condition Setting 1) Input each measured value.
2) Set the taboo length and maximum number of searches.

Step 2: Generation of initial value 1) A binary value (1, 0) is randomly generated for each element of the state expression array.
2) Put the current state into the taboo list.
3) The current number of searches is 1.
4) Specifying whether to use the current parameter value or the modified parameter value for each device by using the 1, 0 value of the array element in the current state, and using the current parameter value under the specification , The parameter value that becomes the state variable is fixed to the current parameter value, and the measurement correction value and the objective function (evaluation function) when the measurement correction value is used using the method described in Japanese Patent Application No. 2004-218788 described above. Calculate the value.
5) The evaluation function value (evaluation value) for the current state is the best value Obj ^* . The current minimum number is Optnum = 1. Here, Obj ^* [] is an array indicating the best value.

-Step 3: Generation of adjacent state and determination of next state 1) Generate a binary value different from the current value for each array element in the current state, and sequentially generate array values with the number of other elements as the current value This is the adjacent state.
For example, if the number of measurement values is 3, the number of elements in the array is 3, but if the value of this array is (1, 1, 0), only the first element is a binary value different from the current one, only the second element Since the binary value different from the current and the binary value different from the current only for the third element are individually created and the adjacent state is created, the adjacent states are (0, 1, 0), (1, 0, 0), ( 1, 1, 1). In this method, adjacent states are always generated for the number of target devices.
2) Specify whether to use the current parameter value or the modified parameter value for each device by the 1, 0 value of the array element of each state of the adjacent state. Under the specification, the current parameter value is Regarding the parameters to be used, the parameter value to be the state variable is fixed to the current parameter value, and the measurement correction value and the objective function when the measurement correction value is used using the method described in Japanese Patent Application No. 2004-218788 described above ( (Evaluation function) Calculate the value.
3) Among the adjacent states, the one that is not taboo and has the best evaluation is set as the next state. If the evaluation value of the next state is smaller than the current Obj ^* [Optnum], Optnum = Optnum + 1 is set, and the evaluation value is set to Obj ^* [Optnum]. The value of the state variable at that time is X ^* [Optnum].
4) Let the current state be the next state.

Step 4: Termination determination by tab search When the number of searches reaches a preset value, the process proceeds to the next step 5. If the number of searches does not reach the preset value, the current number of searches is set to 1. Return to step 3.
・ Step 5 :: Identifying the device that is abnormal
In X ^* [Optnum] to X ^* [1], in this order, the device that uses the current parameter value is output as a device whose device characteristic parameter value is abnormal, and the process ends.

7). Next, a method of identifying a device having an abnormal device characteristic parameter value using Particle Swarm Optimization as an optimization method or its improved method will be described.
Even in this embodiment, based on the assumption that the device characteristic parameter value related to the device in which the measured value is abnormal is most changed (becomes abnormal), several parameter values for each device are used as a set. When the parameter value of the device is re-changed to the one before the change, the ascending order is obtained from the combination in which the measured value and the calculated value are best matched. At this time, the device value changed to the parameter value before the change is obtained. The combination is identified as a device whose device characteristic parameter value is abnormal.
At that time, Particle Swarm Optimization or an improved method thereof is used for a part for obtaining a combination in which the measured value and the calculated value are best matched.
Here, Particle Swarm Optimization (hereinafter referred to as PSO) or its improvement method was developed by Eberhart and others as described in J. Kennedy and R. Eberhart, Swarm Intelligence, Morgan Kaufmann Publishers, 2001. It refers to a method that includes various variations of Particle Swarm Optimization such as Gbest model, Lbest model of Particle Swarm Optimization based on swarm theory, or Hybrid Particle Swarm Optimization developed by Angeline. Hereinafter, a method for identifying a device whose device characteristic parameter value is abnormal will be described using a binary model and Hybrid PSO as examples.

A. Binary model (1) First, the background of PSO research will be described.
For example, it has been found that the movement of a group of natural creatures such as a school of fish or a school of birds can be described by simple rules. In other words, even if the movement of an individual (agent) is described by simple rules, it is possible to describe a complex movement of a group.
Boid (Reynolds, CW, “Flocks, Herds, and Schools: A Distributed Behavioral Model in Computer Graphics” developed by Reynolds, Proceedings of SIGGRAPH '87, pp. 25-34, 1987.) Only rules are used.
1) Proceed in a direction to leave the adjacent agent.
2) Proceed in the direction the whole group is heading.
3) Proceed toward the center of the flock.

In other words, the movement of individual agents in the group can be considered by simple vector composition.
Boyd and Richerson investigated the human decision-making process and developed the concept of personal learning and cultural communication (R. Boyd and P. Richerson, Culture and the Evolutionary Process, University of Chicago Press, 1985). According to their research, humans use two important pieces of information in decision making. The first is my own experience. In other words, I know what the situation was good and how good it was because of my choices. The second is the experience of other people. In other words, we know how other people acted and what the results were. A person makes a decision using such two pieces of information, that is, information on a situation that is good based on his own experience and the experience of others.

  As described above, the fact that the individuals of the group are moving using simple vectors, and that human decision-making is performed using information on the situation that was good in terms of their own experience and the experience of others, 2 One viewpoint is the background of PSO, which is a new optimization method.

(2) PSO search method using gbest model
The PSO using the gbest model was developed as follows from the above viewpoint. The position (state quantity) of each agent is expressed by XY coordinates, and the speed is expressed by v _x (speed in the X direction) and v _y (speed in the Y direction). The position of each agent at the next time point can be updated from this position and speed information. Based on this concept, the following optimization can be considered if the whole flock of birds takes action that optimizes some objective function.
Consider a search where multiple agents work together to find a good solution. At this time, each agent remembers the best value (personal best: pbest) of the objective function in the past search and its XY position (state quantity). This pbest corresponds to the personal experience presented by Boyd and Richerson et al. Further, each agent shares the best value in the group among pbest, that is, information on the best value (group best: gbest) of the objective function in the past search of the group.
This gbest corresponds to the experience of others shown by Boyd and Richerson et al. Each agent present XY value and v _x, v _y values, and tries to change the position in that direction according to the distance between the pest and gbest. The action to be changed is expressed by speed. Using the current speed, pbest, and gbest, the speed of each agent is modified by Equation 9 below.

In other words, the complicated search direction for realizing the optimization can be expressed by vector synthesis as in Boyd. In addition, the second term on the right side of Equation 9 is a direction that converges to pbest, and the third term is a direction that converges to gbest as well. In other words, this search direction is a direction in which the experiences of individuals and the experiences of others are fused. In addition, by calculating the search direction according to Equation 9, the PSO performs a global search (first term on the right side of Equation 9) that maintains the previous speed, and a local search that attempts to approach it using pbest and gbest (Equation 9). This is a search method having a mechanism that balances the second and third terms on the right side of the above.
By using Equation 9 above, a speed that probabilistically approaches the best solution of each agent and the best solution of the group is obtained, and the current position (search point) of each agent is calculated by the following equation. 10 to correct.

[Equation 10]
s _i ^{k + 1} = s _i ^k + v _i ^{k + 1}

The concept of search point correction described above is shown in FIG. In FIG. 2, the current search point s _k (velocity v _orig ) of the agent is corrected to a new search point s _{k + 1} according to Equations 9 and 10 by the velocity v _pbest based on _pbest and the velocity v _gbest based on _gbest. .
PSO is a multi-point search having a plurality of search points like GA, etc., and the global optimal solution (by using each search point's pbest and collective gbest to change probabilistically ( This is a method for obtaining the best solution. In addition, the PSO needs to evaluate the objective function value at each step, but there is an advantage that the number of evaluations need only be the number of agents regardless of the scale of the problem. Therefore, it can be easily applied to large-scale problems.
FIG. 3 shows the concept of solution search by PSO in a multidimensional space. This figure shows a case where an n-dimensional problem is searched by m agents.

(3) Parameters of PSO The important parameters in PSO are the number of trials, the number of agents, the number of iterations, and the weighting factors of Equation 9. These can all be changed as set values.
1) Number of trials Since PSO is a probabilistic optimization method, the optimization is usually performed not only by one optimization but also by changing the seed of random numbers and performing optimization several times, and the best result is obtained. Let the solution be the optimal solution. The number of times that the optimization is executed is the number of trials.
2) Number of agents and number of iterations The number of agents is a numerical value of how many simultaneous searches are performed. The number of iterations is a numerical value indicating how many times each search point (agent) updates the position. That is, in the PSO algorithm, the objective function value of each agent is evaluated for the number of iterations. Therefore, the calculation time required for evaluating the objective function value × the number of agents × the number of iterations is required for one optimization trial. It is necessary to determine these parameters in consideration of this calculation time.

3) Weight coefficient (w, c ₁ , c ₂ in Equation 9)
It has been empirically known that the weighting coefficients w, c ₁ and c ₂ are empirically set as in the following formulas 11 and 12.

[Equation 12]
c ₁ = c ₂ = 2.0

Here, w _max = 0.9, w _min = 0.4, k _max : maximum number of iterations, k: current number of iterations.

  As shown in FIG. 4, it can be seen that the weight w determined by Equation 11 decreases as the search proceeds (the number of iterations increases). In Equation 9, the term with the weight w is a term corresponding to the global search as described above, and the specific gravity for the global search is relatively large at the start of the search, and the specific gravity is increased as the search proceeds. This is because the characteristics are such that the specific gravity of the local search is increased.

(4) Optimization algorithm by PSO FIG. 5 shows an outline of the algorithm in one optimization trial.
That is, a plurality of initial states corresponding to the number of agents are randomly generated (S1), and an objective function value for each agent is calculated by Equation 7 (S2).
Next, the objective function is evaluated in the search process of each agent, and the state where the best solution is obtained is stored as p _best (S3). Thereafter, the p _{bests of} all agents are compared, and the best solution obtained is stored as g _best (S4).
Next, it is determined whether or not the maximum number of iterations has been reached. If it has been reached, the current state of g _best (the agent that has obtained the best solution) is output as the optimum solution (S5yes, S7).
If the maximum number of iterations has not been reached, the speed of each agent is calculated within a range that satisfies the constraint conditions according to equations 9 and 10, the search point is moved, and the process returns to step S2 (S5no, S6).

(5) Binary PSO
Binary PSO when the state variable is expressed as 1, 0 was developed by Everhart et al. (J. Kennedy and R. Eberhart, "A discrete binary version of the particle swarm optimization algorithm", Proc. Of the 1997 conference on Systems, Man and Cybernetics (SMC'97), pp.4104-4109, 1997.).

The gbest model has been developed for nonlinear optimization problems using continuous variables.
However, many real problems can be formulated as combinatorial optimization problems. Everhart et al. Proposed a model like Formula 13 in which the agent's yes or no, right or wrong, or the probability P of making other decisions is expressed as a function of individuals and social factors.

[Equation 13]
P (s _i ^{k + 1} = 1) = f (s _i ^k , v _i ^k , pbest _i , gbest)

  In other words, the probability that the next state is 1 can be expressed as a function using the one used in the original PSO (state, speed, pbest, gbest). Also, the parameter v, which is one or another decision tendency in the agent, determines the threshold value of the probability value. If v is large, the agent selects 1, and if v is small, selects 0. Such a threshold is determined in the range [0, 1]. One function for realizing such characteristics is a sigmoid function shown in Equation 14. This sigmoid function is also used to express the non-linear characteristic of the output function of the neural network.

  An agent's trend must be corrected for the success of that group of agents. Thus, each velocity equation is a function of the difference between the agent's current state and the agent and group's most evaluated state so far. That is, like the PSO for the original continuous quantity, the binary PSO can be described as the following formulas 15 and 16.

As a result, only the expressions 15 and 16 are used instead of the expressions 9 and 10 described above, and the same algorithm as the original PSO can be used thereafter. The speed v needs to be determined so that sig (vi ^k ) does not suddenly approach 0 or 1.
In other words, the speed must always be changed from 1 to 0 or from 0 to 1. The maximum value of the speed needs to be determined in advance, and is set to [−4.0, +4.0], for example.

B. Method for identifying a device having an abnormal device characteristic parameter value using a binary model model (1) State variable Numbering the device characteristic parameter value of the target device. The agent state is expressed by an array with elements equal to the number of target devices. Each element of the array expresses a response to the parameter for each corresponding numbered target device, and expresses the element with a binary value of 1, 0. Here, 1 indicates that the current parameter value before correction is used, and 0 indicates that the calculated new parameter value is used.

(2) Solving algorithm step 1: Steps for setting preconditions and initial conditions 1) Input each measured value.
2) Set the number of agents, each optimization parameter value, and the maximum number of searches.

Step 2: Generation of initial value 1) A binary value (1, 0) is randomly set for each element of the array of state expressions of each agent.
2) Initial setting of pbest and gbest The pbest value and gbest [0] value for each agent are set to large values set in advance. Also, gbestnum = 0.

Step 3: Evaluation of each agent 1) Calculation of evaluation value for each agent Based on the 1, 0 value of the array element in the current state, the current parameter value is used for each device, or a new parameter value that has been modified The method described in the above-mentioned Japanese Patent Application No. 2004-218788 is used to specify whether to use the current parameter value under the designation, and to fix the parameter value as a state variable to the current parameter value. Is used to calculate the measurement correction value and the objective function (evaluation function) value when the measurement correction value is used.
2) Update of pbest value and gbest value If the evaluation value for each agent calculated in 1) above is better than the current pbest value for each agent, the current value is changed to the pbest value.
If the best value of the pbest values calculated above is better than the current gbest [gbestnum], then gbestnum = gbestnum + 1 is set and the value is stored in gbest [gbestnum].

Step 4: Correction of search point of each agent The state variable value is corrected using Formulas 15 and 16 for each agent.
Step 5: Checking the search termination condition by PSO When the number of searches reaches the maximum number of searches that has been input, the procedure ends.
If the number of searches does not reach the maximum number of searches, the process returns to step 3.
-Step 6: Identifying the device that is abnormal
In gbest [gbestnum] to gbest [1], the devices that use the current parameter values in this order are output as devices having abnormal device characteristic parameter values, and the process ends.

C. Hybrid PSO (HPSO)
HPSO is a hybrid method that combines the mechanism of PSO and the concept of natural selection used in GA (P. Angeline, “Using Selection to Improve Particle Swarm Optimization”, Proc. Of IEEE International Conference on Evolutionary Computation (ICEC). ), Anchorage, May 1998.).
Many of the metaheuristic methods (MH methods) incorporate the concept of selection corresponding to natural selection. The selection has a function of redirecting the search direction to a search point that is relatively effective as compared to a point recently searched by another search point in the search space. The PSO also uses search points (pbest, gbest) that have been effective in the previous search, but the search points thereafter are limited depending on the points themselves.

On the other hand, HPSO realizes a function that uses an effective point in the same way as a normal MH method, but gradually becomes less dependent on that point. Specifically, the evaluation values of the current search points are arranged in a good order, and the evaluation values are set in order from the search point and speed of the agent having the worst evaluation value with respect to the current search points by the preset ratio. Replace with the best agent value.
At this time, the information of the search point (pbest) that has been searched so far for each agent so far is left. By such a method, it is possible to realize functions such as concentration of search in an effective area, redirection of a search direction to an effective area that has been searched so far, and weak dependence on the direction.

A general algorithm of HPSO is as shown in FIG.
That is, a plurality of initial states corresponding to the number of agents are randomly generated (S11), and an objective function value for each agent is calculated by Equation 7 (S12).
Next, the objective function is evaluated in the search process of each agent, and the state where the best solution is obtained is stored as pbest (S13). Thereafter, the pbest of all agents are compared, and the best solution obtained is saved as gbest (S14).
Further, by changing the search point and speed of the poorly evaluated agent to the search point and speed of the highly evaluated agent, selection which is a main part of HPSO is executed (S15).
Next, it is determined whether or not the maximum number of iterations has been reached. If it has been reached, the current state of g _best (the agent that has obtained the best solution) is output as the optimum solution (S16yes, S18).
If the maximum number of iterations has not been reached, the speed of each agent is calculated within the range satisfying the constraint conditions according to equations 15 and 16, the search point is moved, and the process returns to step S2 (S16no, S17).

  With this algorithm, HPSO is a mixture of PSO and MH method selection dynamics. Although there is little difference between the mechanisms of HPSO and conventional PSO, by adding this selection method, a mechanism for intensively searching an area with high evaluation is added, and a search mechanism that can find a better solution is obtained.

D. A method for identifying a device having an abnormal device characteristic parameter value using HPSO The formulation and the method of taking state variables are the same as the method using PSO in the previous section. The solution algorithm using HPSO is as follows.
・ Step 1: Data input 1) Input each measured value.
2) Set the number of agents, each optimization parameter value, and the maximum number of searches.

Step 2: Generation of initial value 1) A binary value (1, 0) is randomly set for each element of the array of state expressions of each agent.
2) Initial setting of pbest and gbest The pbest value and gbest [0] value for each agent are set to large values set in advance. Also, gbestnum = 0.

Step 3: Evaluation of each agent 1) Calculation of evaluation value for each agent Based on the 1, 0 value of the array element in the current state, the current parameter value is used for each device, or a new parameter value that has been modified The method described in the above-mentioned Japanese Patent Application No. 2004-218788 is used to specify whether to use the current parameter value under the designation, and to fix the parameter value as a state variable to the current parameter value. Is used to calculate the measurement correction value and the objective function (evaluation function) value when the measurement correction value is used.
2) Update of pbest value and gbest value If the evaluation value for each agent calculated in 1) above is better than the current pbest value for each agent, the current value is changed to the pbest value.
If the best value of the pbest values calculated above is better than the current gbest [gbestnum], then gbestnum = gbestnum + 1 is set and the value is stored in gbest [gbestnum].

Step 4: Execution of selection 1) Sort the current evaluation values of each agent in good order.
2) Using the ratio (Sr) set in advance, the number of individuals to be changed (Ns) is calculated by the following formula.
[Equation 17]
Ns ＝ Round (Agentnum × Sr)
3) Ns search points and speeds from the worst-rated agent are changed to Ns search points and speeds from the best-rated agent. At this time, the pbest information of the changed agent is left as it is.

Step 5: Correction of search point of each agent 1) For each agent, the state variable value is corrected using Formulas 15 and 16.
Step 6: Checking the search end condition by PSO When the number of searches reaches the input maximum number of searches, the process ends.
If the number of searches does not reach the maximum number of searches, the process returns to step 3.
-Step 7: Identifying the device that is abnormal
In gbest [gbestnum] to gbest [1], the devices that use the current parameter values in this order are output as devices having abnormal device characteristic parameter values, and the process ends.

8). Method for identifying measuring instrument with abnormal measured value Next, select the measuring instrument related to this instrument for the instrument whose parameter value is abnormal by the various methods described above, and select the selected measurement. Identify the measuring instrument whose measured value is abnormal.
Specifically, as shown in claim 5 of the above-mentioned Japanese Patent Application No. 2004-218788 “Measured Value Examination Device in Network”, electricity, gas, water, steam, etc. input / output to / from each device constituting the network An objective function that specifies normal values and abnormal values from the state of measured values of the energy, minimizes the error between the measured value and the calculated value for only the measured value of energy specified as the normal value, and the measured value A method that outputs the measured value specified as an abnormal value as an abnormal value in the event with the smallest evaluation value based on the overall objective function, using an overall objective function that combines the objective function that takes into account the event probability when the is normal Thus, the measuring instrument in which the measured value is abnormal is specified. Here, the normal values and abnormal values are specified, and the measurement abnormal values output by the evaluation using the overall objective function include enumeration method, genetic algorithm or its improved method, tab search or its improved method, Particle Swarm Optimization or The improved technique is used.
In addition, when specifying a measuring instrument having an abnormal measurement value, a measurement correction value obtained incidentally is output.
The algorithm is as follows.

・ Step 1
By the above-described method, a device whose device characteristic parameter value is abnormal is specified.
・ Step 2
A measuring instrument whose measurement value is abnormal by the method described in the above-mentioned Japanese Patent Application No. 2004-218788 “Scrutinized Device for Measuring Values in a Network” for only the measurement value of the measuring instrument related to the specified device. Is identified.
Step 3: The most probable measurement correction value obtained in the process of Step 2 is output for the measuring instrument in which the measurement value is abnormal and the measurement value.

9. A device / measurement device diagnostic apparatus according to claim 7 will be described below with reference to FIG.
In this device / measuring instrument diagnostic device, a measurement value ensuring simultaneity is input, and the device characteristic parameter value is abnormal according to any one of claims 1 to 6 and the measurement value is abnormal. Information output device 100 for identifying the measuring instrument, the device characteristic parameter value, the device (device name), the measuring device (measurement device name) in which the measured value is abnormal, and the measurement correction value are displayed on the screen. Display device 200.

In FIG. 7, synchronized measurement values are input to the information output apparatus 100 from each measuring instrument installed in the network. For this synchronization, for example, using GPS or the like, measurement data with a time tag is input to the device at the same time as the measurement value at a predetermined period (for example, in units of 10 minutes), and the device also adjusts to GPS. A method is conceivable in which, among the measurement data of each measuring device, the data closest to the measurement time on the apparatus side is used as the measurement value at that time.
This method does not use GPS, but connects each measuring instrument via a network, synchronizes the absolute time with an NTP server, etc., and inputs the time tag and measurement value of this absolute measurement time to the device. Another possible method is to view the measurement values for the same measurement time as synchronized measurement values.

The measurement value group input in this way is input to the machining apparatus 110, and the machining apparatus 110 obtains an average value of the measurement values for each preset period and sends the measurement value group for the same period to the identification apparatus 120. (Fig. 7a). Here, the specifying device 120 includes the above-described optimal parameter calculation device 121 and the device characteristic equation or steady network simulator 122.
The specifying device 120 uses the input measurement value group to specify a device whose device characteristic parameter value is abnormal according to any one of claims 1 to 5 for each period set in advance. At this time, if the parameter value does not become abnormal, each parameter value is stored in the parameter value storage database 130 (b in FIG. 7). Further, when specifying a device whose device characteristic parameter value is abnormal, the current device characteristic parameter value is input from the database 130 (c in FIG. 7).
The device information identified as having an abnormal parameter value is sent to the measuring device identifying device 140 having an abnormal measurement value (FIG. 7D), and the identifying device 140 has an abnormal measurement value based on the device information. The right instrument. At this time, incidentally, the most probable measurement value for the abnormal measurement value is obtained by calculation (measurement correction value).
The device / measurement device information generation device 150 has an abnormal device characteristic parameter value based on the data input from the specific device 140, the device (device name), and a measurement device (measurement) in which the measurement value is abnormal. Device name) and measurement correction value data are generated (e in FIG. 7), and these are displayed on the screen by the display device 200.

Next, examples of the present invention will be described.
1. Calculation Conditions FIG. 8 shows a cogeneration system to which this embodiment is applied. In the figure, 301 is a gas turbine, 302 is a boiler, 303 is a steam turbine, 304 is a heat load, 305 is an electric load, and 10 measurement points m _{1 to} m ₁₀ are assumed as data measurement points. .

(1) Equipment model 1) Gas turbine 301
Here, the exhaust gas boiler is modeled as a gas turbine 301. The gas turbine 301 is a model that outputs electricity and steam by inputting fuel. The input / output relationship of the gas turbine 301 is as shown in Equations 18 and 19, and the input / output amount is expressed in linear form as shown in Equations 20 and 21, respectively.
[Equation 18]
y ₁ = f ₁ (x ₁ )
[Equation 19]
y ₂ = f ₂ (x ₁ )
[Equation 20]
y ₁ = a ₁ x ₁ + b ₁
[Equation 21]
y ₂ = a ₂ x ₁ + b ₂
Where x ₁ : fuel input to the gas turbine
y ₁ : Electric output amount of gas turbine
y ₂ : Steam output amount of gas turbine
a ₁ , b ₁ , a ₂ , b ₂ : equipment characteristic parameters of the gas turbine

2) Steam turbine 303
The steam turbine 303 is a model that outputs electricity and heat (steam or hot water) by inputting steam. The input / output relationship of the steam turbine 303 is as shown in Formulas 22 and 23, and the input / output amounts are expressed in linear form as shown in Formulas 24 and 25, respectively.
[Equation 22]
y ₃ = f ₃ (x ₂ )
[Equation 23]
y ₄ = f ₄ (x ₂ )
[Equation 24]
y ₃ = a ₃ x ₂ + b ₃
[Equation 25]
y ₄ = a ₄ x ₂ + b ₄
Here, x ₂ : Steam input amount to the steam turbine (= y ₂ )
y ₃ : Electric power output of steam turbine
y ₄ : Heat output of steam turbine
a ₃ , b ₃ , a ₄ , b ₄ : equipment characteristic parameters of the steam turbine

3) Boiler 302
The boiler 302 is a model that outputs heat (steam or hot water) by inputting fuel. The input / output relationship of the boiler 302 is as shown in Expression 26, and the input / output amount is expressed in a linear form as shown in Expression 27.
[Equation 26]
y ₅ = f ₅ (x ₃ )
[Equation 27]
y ₅ = a ₅ x ₃ + b ₅
Where x ₃ : fuel input to the boiler
y ₅ : heat output of steam turbine
a _{5, b 5:} Equipment characteristics of the boiler parameters

4) Total electricity output (electric load 305)
The total electrical output amount is a total electrical output value generated by the gas turbine 301 and the steam turbine 303 as shown in Equation 28.
[Equation 28]
y ₇ = f ₇ (y ₁ , y ₃ , y ₆ ) = y ₁ + y ₃ + y ₆
Where y ₆ : Electricity purchase
y _7: electrical total output quantity

5) Total heat output (heat load 304)
The total heat output amount is the heat output from the steam turbine 303 and the boiler 302, and is expressed by Equation 29.
[Equation 29]
y ₈ = f ₈ (y ₄ , y ₅ ) = y ₄ + y ₅
Where y ₈ : total heat output

In the above embodiment, a _{1 to} a ₅ and b _{1 to} b ₅ are parameters for determining whether or not the device characteristic parameter value is abnormal. The target devices that determine whether or not the parameter value is abnormal are the gas turbine 301, the steam turbine 303, and the boiler 302.
Here, for simplification, equipment such as a cooling tower, a condenser, and a refrigerator is ignored.

(2) Steady Network Simulator In the steady network simulator, the electric output amounts y ₁ and y ₃ from the gas turbine 301 and the steam turbine 303 are obtained from the fuel input amount x ₁ to the gas turbine 301, and these and the electric purchase amount y by summing the ₆ to simulate a condition of obtaining an electrical total output quantity y ₇ to the electrical load 305. Further, a state in which the heat output amount y ₄ from the steam turbine 303 and the heat output amount y ₅ from the boiler 302 are summed to obtain a total heat output amount y ₈ for the heat load 304 is simulated.
In this example, since all are linear models, no convergence calculation is required, and the output can be obtained from the input in the following steps.

Step 1: The gas turbine output amounts y ₁ and y ₂ are calculated from the fuel input amount x ₁ to the gas turbine 301 and the equations 20 and 21.
Step 2: The steam turbine output amounts y ₃ and y ₄ are calculated from the gas turbine output amount y ₂ calculated in Step 1 and the formulas 24 and 25.
Step 3: The boiler output amount y ₅ is calculated from the fuel input amount x ₃ to the boiler 302 and Equation 27.
Step 4: The electric power output y ₁ of the gas turbine obtained in Steps 1 and 2, the electric power output y ₃ of the steam turbine, the electric purchase amount y ₆ , and Equation 28 are used to calculate the total electric power output y ₇ . calculate.
Step 5: Calculate the total heat output y ₈ from the heat output y ₄ of the steam turbine obtained in Step 2 and the heat output y ₅ of the boiler obtained in Step 3 and Equation 29.

(3) Measurement Point, Measurement Value, and Calculation Value Measurement values at measurement points m _{1 to} m ₁₀ in FIG. 10 are m _{1meas to} m _10meas, and m _imeas means the i-th measurement value. In addition, the calculated values at the respective measurement points m _{1 to} m ₁₀ are m _{1calc to} m _10calc, and _micalc means the i-th calculated value.
Each measured value can be calculated as in Expressions 30 to 39.
[Equation 30]
m _1calc = x ₁
[Equation 31]
m _2calc = x ₃
[Formula 32]
m _3calc = y ₆
[Equation 33]
m _4calc = y ₁
[Formula 34]
m _5calc = y ₂
[Equation 35]
m _6calc = y ₅
[Equation 36]
m _7calc = y ₃
[Equation 37]
m _8calc = y ₇
[Equation 38]
m _9calc = y ₄
[Equation 39]
m _10calc = y ₈

2. Execution of calculation Method for Identifying Device with Device Characteristic Parameter Value Abnormal Using the above-described calculation conditions, a device with an abnormal device characteristic parameter value can be identified by the inventions of claims 1 to 5. Here, as an example, a case where the method 2 in the embodiment of claim 1 and the enumeration method according to claim 2 are used will be described.
The optimum parameter value is calculated for each cycle determined in advance, and it is determined whether the parameter value of the characteristic equation has changed using Equation 8 described above. At this time, if it is determined that the parameter value has changed, a combination of parameter values is created for the gas turbine, steam turbine, and boiler. This combination is as follows.
(Combination 1): (Gas turbine)
(Combination 2): (Steam turbine)
(Combination 3): (Boiler)
(Combination 4): (Gas turbine, steam turbine)
(Combination 5): (Gas turbine, boiler)
(Combination 6): (Steam turbine, boiler)
(Combination 7): (Gas turbine, steam turbine, boiler)

  When it is determined that the device characteristic parameter value has changed, a new calculated value is obtained as the device characteristic parameter value. For each of the above combinations, the current parameter value of the target device is input from the parameter value storage database 130 of FIG. 7, and the parameter value for only the target device is calculated from the new calculated value currently obtained. Change to the current parameter value. For example, for the combination 1, only the gas turbine parameter value is changed to the current parameter value input from the database 130.

By using the parameter value group of the gas turbine, steam turbine, and boiler for each combination obtained in this way, measurement is performed by the method described in the above-mentioned Japanese Patent Application No. 2004-218788 “Measurement Value Examination Device in Network”. Find the correction value. Among them, objective functions, that is, devices with a high probability that the device characteristic parameter value is abnormal according to the order of this combination, in order from the combination that minimizes the error between the measured value and the calculated value. As specified. For example, the top three of such combinations are (combination 1): (gas turbine)
(Combination 4): (Gas turbine, steam turbine)
(Combination 5): (Gas turbine, boiler)
In this case, the devices whose device characteristic parameter values are abnormal are identified in the following order.
(1) Gas turbine (2) Gas turbine and steam turbine (3) Gas turbine and boiler

B. Identification of a measuring instrument having an abnormal measurement value For example, assume that a device having an abnormal parameter value is specified as a gas turbine (combination 1). Measurements associated with this gas turbine are m ₁ , m ₄ , m ₅ .
Therefore, when obtaining a measuring instrument having an abnormal measurement value by the method of claim 5 or later of the above-mentioned Japanese Patent Application No. 2004-218788 “Measurement apparatus for measured values in a network”, the target measuring instrument is changed to these three measurement values. Only corresponding measuring instruments. Here, the case where the method using the enumeration method described in claim 6 of the above-mentioned Japanese Patent Application No. 2004-218788 “measurement value scrutinizing device in a network” will be described.

First, a combination of measurement values that are excluded from a target that minimizes an error between a measurement value and a calculation value is generated.
This combination is as follows.
(Combination 1): (m ₁ )
(Combination 2): (m ₄ )
(Combination 3): (m ₅ )
(Combination 4): (m ₁ , m ₄ )
(Combination 5): (m ₁ , m ₅ )
(Combination 6): (m ₄ , m ₅ )
(Combination 7): (m ₁ , m ₄ , m ₅ )
For these combinations, a measurement correction value that minimizes the error between the measurement value and the calculation value is calculated, and arranged in order from the smallest error. For example, those most error becomes smaller when a combination 1, the instrument corresponding to m _1, the probability that the measurement value becomes abnormal is identified as the highest instrument. In addition, the most probable measurement value is simultaneously output for the measurement value of the measuring instrument.

The above calculation results are displayed and output on the screen of the display device 200 as follows, for example.
“・ Devices with abnormal device characteristic parameter values have a high probability in the following order.
(1) Boiler (2) Gas turbine (3) Steam turbine ・ A measuring instrument with an abnormal measurement value has a high probability in the following order.
(1) OO with respect to the measured value m1 of the boiler. The most probable measurement is XX.
(2) ΔΔ with respect to boiler measured value m4. The most probable measurement is ●●. "

The operator who sees the display as described above can implement Condition-based Maintenance (CBM) in which the measuring instrument of the boiler 302 and the communication equipment associated therewith are first inspected, and then the gas turbine 301 is inspected. it can.
Moreover, it becomes possible to continue the operation of the target plant by using the most probable measurement correction values for the operation operation and operation control.

It is a figure which shows the calculation concept of the optimal parameter in embodiment of this invention. It is a figure which shows the concept of the correction of the search point in PSO. It is a figure which shows the concept of the solution search in PSO. It is explanatory drawing of the weighting coefficient in Numerical formula 11. It is a figure which shows the outline of the algorithm in one optimization trial of PSO. It is a figure which shows the outline of the algorithm of HPSO. It is a block diagram of the apparatus and measuring device diagnostic apparatus in embodiment of this invention. It is a figure which shows the cogeneration system with which the Example of this invention is applied.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10: Optimization part 20: Equipment characteristic formula 100: Information output apparatus 110: Measurement information processing apparatus 120: Equipment specification apparatus with an abnormal equipment characteristic parameter value 121: Optimization section 122: Equipment characteristic formula or stationary network simulator 130 : Parameter value storage database 140: Device for identifying measuring instrument with abnormal measurement value 150: Device / measuring device information generating device 200: Display device 301: Gas turbine 302: Boiler 303: Steam turbine 304: Thermal load 305: Electric load

Claims (7)

Minimize errors between measured and calculated values of energy such as electricity, gas, water, and steam that are input to and output from each device that makes up the network, and errors from specified values of each device characteristic parameter value. An optimization unit that obtains and outputs each device characteristic parameter value as a state variable by an optimization method using an objective function,
It has a device model, device operation rules, etc. of each device that constitutes the network inside, and based on each device characteristic parameter value input from the optimization unit, it calculates the output energy relative to the input energy of the device. The device characteristic equation that is output as a calculated value for the device, or the device model of each device that constitutes the network, the device operation rule, the connection state of each device, and the equation relating to the energy flow of the network, etc. Under each device characteristic parameter value input from the unit, a steady network simulator that obtains a state quantity of each unit in the network in a time section with a steady state and outputs it as a calculated value for the measured value, and
Input the current measured value as the initial value of the input / output energy of the target device to the optimization unit, input the specified value of the device characteristic parameter value, and input the current device characteristic parameter value to the device characteristic equation or steady network simulator To obtain the calculated value for the measured value,
The optimization unit repeats the process of obtaining the device characteristic parameter values of all devices by the optimization method using the input measurement value and the device characteristic equation or the calculated value obtained by the steady network simulator. The parameter value when the set number of times is reached is output as the optimum parameter value for each device characteristic equation.
Further, a device group in which the device characteristic parameter value is changed is obtained by comparing a value calculated based on a deviation between the optimum parameter value and the specified value and a predetermined weighting factor with a set value, and measuring instruments related to these device groups. A device diagnostic apparatus on a network, characterized in that, when the measured value is abnormal, a device having an abnormal device characteristic parameter value is identified from the device group using an optimization method. In the apparatus diagnostic apparatus according to claim 1,
An apparatus diagnosis apparatus using an enumeration method as an optimization technique for identifying an apparatus having an abnormal apparatus characteristic parameter value. In the apparatus diagnostic apparatus according to claim 1,
A device diagnostic apparatus characterized by using a genetic algorithm or an improved method thereof as an optimization method for identifying a device having an abnormal device characteristic parameter value. In the apparatus diagnostic apparatus according to claim 1,
A device diagnostic apparatus characterized by using a tab search or an improved method thereof as an optimization method for identifying a device having an abnormal device characteristic parameter value. In the apparatus diagnostic apparatus according to claim 1,
A device diagnostic apparatus using Particle Swarm Optimization or an improved method thereof as an optimization method for identifying a device having an abnormal device characteristic parameter value.   Identifying a measuring instrument having an abnormal measurement value using an optimization method for a measuring instrument related to the device specified by the device diagnostic apparatus according to any one of claims 1 to 5. A measuring instrument diagnostic device on the network.   Information related to the device specified by the device diagnostic device according to any one of claims 1 to 5, and information related to a measuring device specified by the measurement device diagnostic device according to claim 6 that the measurement value is abnormal And an apparatus diagnostic apparatus and a measuring instrument diagnostic apparatus on a network, characterized in that a measurement correction value estimated to be most probable for an abnormal measurement value is displayed on a screen.

Images