JP3336140B2 - Generator load distribution device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to power system operation,
In particular, the present invention relates to an apparatus for determining load sharing between generators so as to minimize fuel cost when using a generator to meet power demand.

[0002]

2. Description of the Related Art At present, basic power demand is covered by nuclear power generation, and fluctuations above it are handled by thermal power generation. To respond to fluctuations in power demand using multiple thermal power generators that are geographically dispersed,
It is necessary to distribute the load among multiple thermal power generators. A generator load distribution device is used to determine this load sharing.

[0003] In a conventional generator load distribution device, an equal-incremental (equal-λ) fuel cost method described in Yasuji Sekine: “Electric Power System Engineering” (Electrical Publishing) pp. 103-112 is applied. The equal incremental fuel cost method assumes that the fuel cost of each generator is expressed by a quadratic function of the generator output, and minimizes the total fuel cost using the quadratic function of each generator. In this manner, the load distribution between the generators is determined. Lagrange's undetermined constant method is used to determine the load distribution. Under the assumption that the fuel cost of each generator is represented by a quadratic function of the generator output, it is mathematically guaranteed that the equal incremental fuel cost method gives an optimal solution.

[0004]

As described above, the equal incremental fuel cost method is a method on the premise that the fuel cost of each generator is represented by a quadratic function of the generator output. However, actually, the boiler and control system of the generator have dynamic characteristics,
The relationship between fuel cost and generator output cannot be expressed by a quadratic function. This will be briefly described based on actual measurement data.

FIG. 11 shows the correlation between the generator output and the fuel flow rate. This figure plots the relationship between the output P at each point of time of a certain generator and the corresponding fuel flow rate Q measured for 10 hours in a 3-minute cycle. From this figure, the points (P,
It can be seen that Q) draws a trajectory like a vortex. This means that there is a dynamic characteristic between P and Q.

As a result, in the electric power economic load distribution apparatus to which the equal incremental fuel cost method is applied, an optimum solution cannot be actually obtained.

[0007] The problem to be solved by the present invention is to provide a power economical load distribution device which provides a more optimal load distribution.

[0008]

As described above, the relationship between the generator output and the fuel cost has a dynamic characteristic. Therefore, in order to express this relationship, it is more accurate to apply the dynamic characteristic expression than the static characteristic expression such as a quadratic function.

A generator load distribution device according to the present invention is a generator load distribution device for distributing a given load to a plurality of thermal power generators, wherein for each generator, the relationship between the generator output and the fuel consumption is determined. A dynamic characteristic model identification unit expressed by an autoregressive moving average model, and an objective function representing a total amount of fuel consumption of all generators is created using the autoregressive moving average model, and under a predetermined constraint condition. And a load distribution calculation unit for obtaining each generator output such that the value of the objective function is minimized based on the expected demand value of the generator output.

The generator load distribution device according to the present invention uses an autoregressive moving average model as a dynamic characteristic relational expression. The essence of this model is a constant-coefficient linear discrete-time filter, which is the most typical model expressing the dynamic characteristics of the system. When this model is applied, the generator load distribution problem can be formulated as a linear programming problem, and can be solved by a normal linear programming method such as the simplex method.

Regarding the autoregressive moving average model, Takayoshi Nakamizo: “Signal Analysis and System Identification” (Corona) pp.
121-148, and the details of linear programming are described in Nishikawa et al .: "Optimization" (Iwanami Shoten) pp. 9-41.

[0012]

According to the present invention, the relationship between fuel cost and generator output is accurately represented by an autoregressive moving average model, which is a dynamic characteristic model, and the total fuel cost of a plurality of generators is minimized based on the model. Since each generator output is determined, the optimum generator output that minimizes fuel cost can be determined more strictly than in the past. That is, when the present invention is applied, it is possible to save fuel cost more than before.

In particular, by using the autoregressive moving average model, it is possible to obtain each generator output command value at the present time corresponding to the demand prediction up to a plurality of control time points (for example, at least 30 minutes). Therefore, it is possible to control the fuel consumption more realistically.

[0014]

Embodiments of the present invention will be described below in detail with reference to the drawings.

FIG. 1 is a functional block diagram of a generator load distribution device according to the present invention. The generator load distribution device 3 includes a dynamic characteristic model identification unit 1 and a load distribution calculation unit 2. The dynamic characteristic model identification unit 1 calculates the measured generator output (MW) and the fuel consumption (ton) included in the measured generator output / fuel cost data 4.
, An autoregressive moving average model 7 that best represents these relationships is created using the least squares method. The load distribution calculation unit 2 includes the autoregressive moving average model 7, the predicted demand value 8 predicted by the predicted demand unit 5, and various constraints stored in the generator constraint data 6 (supply / demand balance constraint, output upper / lower limit constraint). And output change speed constraint)
Is used to formulate a generator output command value 9. According to the generator output command value 9, the generator 10 supplies the generated power to the power system 11. Although only one generator 10 is shown in FIG. 1, there are actually a plurality of generators 10, each of which independently includes a generator output command value 9 from the load distribution calculation unit 22.
Is given.

As described above, by providing the dynamic characteristic model identifying unit 1 and the load distribution calculating unit 2 inside the generator load distribution device 3, the generator output command value is determined in consideration of the dynamic characteristics of the generator. It becomes possible.

As one of the system configurations, a configuration shown in FIG. 2 in which the dynamic characteristic model identification unit 1 is included in the offline system and the load distribution calculation unit 2 is included in the online system is considered. This configuration makes the present invention relatively easy when the equipment that can capture the measured data of the generator online is not available, or when the generator characteristics do not change much over time and it is considered unnecessary to update the model online. It is a system configuration that can be realized in

As another system configuration, a configuration shown in FIG. 3 in which both the dynamic characteristic model identification unit 1 and the load distribution calculation unit 2 are included in the online system can be considered. However, in order to adopt this configuration, the generator output / fuel cost actual measurement data 4 to the dynamic characteristic model identification unit 2 must be supplied online from the power plant 32 via the communication line 31 for online data acquisition. It is. Therefore, the configuration of FIG. 3 is a system configuration for realizing the present invention, which is suitable when the preconditions are satisfied and the generator characteristics change over time and it is necessary to update the model online. is there. This configuration has the advantage that the model can be updated using the latest data.

FIG. 12 shows each of these systems 20, 21 and 30.
A specific hardware configuration for constructing is shown. Each system has a central processing unit (CP) including an operation unit and a control unit.
U) 111, an internal storage device (memory) 114 for storing a program for performing the processing of the present invention, an external storage device 115 such as a hard disk for storing various parameters and data described later, a keyboard for inputting various data and the like. , An input device 112 for displaying various data, a communication device 116 for performing communication with the power station, and the like. This can be configured by a normal workstation or the like.

Hereinafter, a method of realizing the dynamic characteristic model identification unit 1 will be described, and an example of a model creation result will be shown.

The dynamic characteristic model identification unit 1 creates an autoregressive moving average model using the generator output and fuel consumption as input and output for each of a plurality of generators to be controlled. The autoregressive moving average model of each generator is described by the following equation.

[0022]

(Equation 1)

Here, αj, βj, and γ are model parameters, m is the model order, and Pj and Qj are the generator output and fuel consumption at the j-th time, respectively. In particular, the simplest first-order (m = 1) autoregressive moving average model is described by the following equation.

[0024]

(Equation 2)

Here, the upper expression is for one generator, and a and b are α ₀ ,
corresponding to α _1. Also, c corresponds to β ₁ and d corresponds to γ. The lower equation is for the i-th generator of the plurality of generators. That is, ai, bi, ci,
di is the model parameter of the i-th generator, Pi, t,
Qi, t is the generator output and the fuel consumption at the time point t of the i-th generator, respectively. As a method of determining the value of the model parameter, a least square method as described in Takayoshi Nakamizo: “Signal Analysis and System Identification” (Corona) pp. 121-148 is applied. As described above, as input data to the dynamic characteristic model identification unit 1, actual measurement data measured offline or online data obtained via a communication line may be used. However, considering the model creation calculation time, in the former case, it is better to apply the offline least squares method that uses all the measured data at once, and in the latter case, the model is updated every time new measured data is obtained It is appropriate to apply a recursive least squares method.

As a method for determining the model order m, Nakajima, Inada, Harada and Tsurugai: "Economic load distribution considering boiler dynamic characteristics (I)-Modeling" Proceedings of the Institute of Electrical Engineers of Japan.
Apply the method described on p. That is, as will be described later with reference to FIG. 5, models of various orders are created using certain measured data, and these models are applied to different (different time range) measured data of the same generator, and the error is reduced. Let the minimum order be the optimal order.

Since the load distribution calculation unit 2 at the subsequent stage aims to determine the generator output command value with the minimum fuel cost, the dynamic characteristic model identification unit 1 at the front stage determines the fuel consumption (to
It is desirable to convert n) to fuel cost (\) to create a model, or to convert fuel consumption (ton) to fuel cost (\) after model creation.

Hereinafter, an example in which a first-order autoregressive moving average model is created using offline measured data of a generator output and fuel consumption of a certain generator will be described. The actual measurement data used are the generator output (MW) and fuel consumption (ton) for 10 hours (that is, 200 points) measured in a 3-minute cycle. The model parameter values are determined using an off-line least squares method. The calculation formula of the off-line least squares method is as follows.

[0029]

(Equation 3)

Where

[0031]

(Equation 4)

[0032]

(Equation 5)

Table 1 shows model parameter values for a certain generator, obtained for the actually measured data used. Here, the dimension of the constant c is dimensionless.

[0034]

[Table 1]

FIG. 4 shows the time characteristics of a conventional model using a quadratic function (upper graph) and the autoregressive moving average model (lower graph) of the present invention in comparison with measured data. As can be seen from this graph, the use of the autoregressive moving average model can improve the model accuracy over the conventional quadratic function.

Next, an example of determining the model order m using this example will be described. FIG. 5 is a graph showing model errors for models of each order. In FIG. 5, the white dots represent the model error of the model with respect to the data targeted for modeling.
The black square points represent the model error of the model with respect to other data of the same generator (data at different times). As can be seen by observing FIG. 5, as for the data to be modeled, the higher the degree, the smaller the model error. However, in such a high-order model, the model error may be rather large for other data. The data in the black square in the figure indicates this.
The minimum error point of the black square data has a model order in the range of the first to fifth order. In this range, the model order of the data to be modeled substantially matches, so the model order to be obtained is in the first to fifth order range. Is considered appropriate. As described above, in this embodiment, m = 1 is adopted as the model order. Even if m = 2, 3, etc., the calculation of variables F and G described later becomes complicated, but there is no particular problem.

Hereinafter, a method of realizing the load distribution calculation section 2 will be described, and an example of a generator output command value formulation result will be shown.

The load distribution calculation unit 2 formulates a generator output command value 9 using the autoregressive moving average model 7, the predicted demand value 8, and the generator constraint condition data 6. As the predicted demand value 8, the predicted demand up to the consideration target time (for example, two hours ahead) is used. As a method of realizing the demand prediction unit 5,
Nakamura / Higasaki: "Several-Hour Advance Power Load Prediction Using Linear Regression Model Updated by Kalman Filter", the method described in IEEJ Transactions on Electronics, B103, vol. 4, pp. 251-257 may be used.

As the generator constraint condition data 6, a supply-demand balance constraint, an output upper / lower limit constraint, and an output change speed constraint are considered. Each constraint can be described as follows.

[0040]

(Equation 6)

[0041]

(Equation 7)

[0042]

(Equation 8)

Here, N is the number of target generators, T is the number of control points corresponding to the time to be considered, Dt is the predicted demand at the t-th point, Li, Hi and δi are the output lower limit and output of the i-th generator, respectively. The upper limit and the output change speed. The supply-demand balance constraint of Equation 6 means that the sum of the outputs Pt of the N generators at the time point t must be equal to the total demand Dt at that time. The output upper and lower limit constraints in Equation 7 define the upper and lower limits of the output according to the characteristics of each generator. Similarly, the output change speed constraint of Expression 8 defines an upper limit and a lower limit of the output change amount from one control time point to the next control time point of each generator.

The objective function to be minimized is defined by the sum of the fuel costs for the time considered. In this case, by using the autoregressive moving average model, the objective function can be described by a linear function of each generator output at each time point. The reason is that the fuel cost of each generator at each point in time can be expressed using an autoregressive moving average model (Equation 1), and the left side of (Equation 1) is successively substituted into the right side of (Equation 1). This is because the term of the fuel cost Q can be deleted from the left side of Expression 1). For example, when a first-order model is applied as an autoregressive moving average model, an objective function J, which is the sum of fuel costs at the time to be considered, is described by the following equation.

[0045]

(Equation 9)

Fi, t, Gi, t can be described as follows using model parameters.

[0047]

(Equation 10)

[0048]

[Equation 11]

Here, t = 1 is the output of each generator (distribution ratio).
Is the control point in time when t is determined, and t = 0 is 1
Indicates before the time point. Qi, 0 in Equation 10 represents the actual data one time before.

Such a modification can be similarly performed with a second-order or higher-order autoregressive moving average model.

As described above, the problem to be solved by the load distribution calculation unit 2 is reduced to an optimization problem in which all objective functions and constraints are linear, that is, a linear programming problem. As a method for solving this problem, the aforementioned simplex method described in Nishikawa et al .: “Optimization” (Iwanami Shoten) pp. 9-41 may be applied. The specific solution of the linear programming problem is well known and will not be described in detail here. That is, once the objective function, the demand forecast value, and various constraint conditions are determined, a solution is uniquely obtained thereafter by a known method. In this way, each generator output at each time point at which the objective function J is minimized under various constraint conditions (Equations 6 to 8) is obtained.

Hereinafter, calculation results of the load distribution calculation unit 2 for the three-machine system will be described. Table 2 shows parameter values of a conventional quadratic function model of each generator, and Table 3 shows parameter values of an autoregressive moving average model.

[0053]

[Table 2]

[0054]

[Table 3]

Table 4 shows numerical examples of the generator constraint conditions.

[0056]

[Table 4]

FIG. 6 shows the target loads. Hereinafter, it is assumed that the demand prediction unit 5 can predict the target load in FIG. 6 without a prediction error. FIG. 7 shows the distribution result of the equal λ method using the conventional quadratic function model, and FIG.
Shown in In the steady state at the center of the graphs of FIGS. 6 to 8, for convenience of illustration, the load is shown to have no change at all, but there is actually a change. As can be seen from the figure, in the present invention, the use ratio of the inefficient generator A is suppressed, and accordingly, the efficient generator C is used. As a result, according to the present invention, a smaller fuel consumption is required for the plurality of generators as a whole to obtain the same total output.

When the fuel cost was evaluated using the autoregressive moving average model, the fuel cost of the equal λ method was 100,
In the method of the present invention, a fuel cost saving effect of 99.875 and 0.125% was obtained. Although this percentage may seem small, it can be expected to save the order of 100 million yen a year when applied to the actual fuel consumption of a power company.

As described above, by applying this method, economic load distribution based on the dynamic characteristics between the generator output and the fuel cost can be realized, so that a more economical load distribution result than the conventional method based on static characteristics can be obtained. Obtainable.

FIG. 9 shows an overall conceptual processing flow of the generator load distribution device according to the present invention.

First, for each generator, the time series data of the generator output and fuel consumption are actually measured (91), and based on this data, an autoregressive moving average model of each generator output and fuel consumption is created. (92). Using the autoregressive moving average model obtained in this way, an objective function is created (93). Also, a constraint condition is created using the demand forecast value and the generator constraint condition data (94). The process of step 94 may be performed before the next step 95,
The order of the steps 91 to 93 does not matter. Therefore,
A linear programming problem defined by the above objective function and constraints is solved using the simplex method (95). The solution obtained in this manner is used as each generator output command.

Next, FIG. 10 shows a more practical processing flow of the generator load distribution device of the present invention.

First, the parameters a, b, c, and d are determined by the least squares method using the equation (3) so that the model of the equation (2) best represents the actual data based on the actual fuel consumption data of the generator. (101). The generator model parameters thus obtained, that is, the autoregressive moving average model 7
Based on the above, an objective function J of Expression 9 is created (102). At this time, the actually measured value of the fuel flow rate (Qi, 0) one time before is used.
On the other hand, based on the demand forecast value 8 and the generator constraint condition 6, formulas of constraint conditions of Formulas 6, 7, and 8 are created (10
3). Therefore, the output command value of each generator that minimizes the value of the objective function J within a range that satisfies the constraint condition is obtained using the simplex method (104). Steps 102 and 10
4 is repeated every control cycle (for example, 3 minutes) (10
5).

[0064]

According to the present invention, the relationship between fuel cost and generator output is accurately represented by a dynamic characteristic model, and the generator output with the minimum fuel cost is determined based on the model. The machine output can be determined exactly. That is, when the present invention is applied, fuel costs can be reduced more than before.

[Brief description of the drawings]

FIG. 1 is a functional configuration diagram of a generator load distribution device of the present invention.

FIG. 2 is a system configuration diagram of the present invention which is used in combination of offline and online.

FIG. 3 is an online system configuration diagram of the present invention.

FIG. 4 is a graph showing time characteristics of an autoregressive moving average model used in the present invention.

FIG. 5 is a graph showing a relationship between a model order of an autoregressive moving average model and a model error.

FIG. 6 is a graph showing target loads for load distribution.

FIG. 7 is a graph showing a distribution result by an equal λ method which is a conventional method.

FIG. 8 is a graph showing a distribution result according to the method of the present invention.

FIG. 9 is a flowchart showing a conceptual processing flow of the generator load distribution device of the present invention.

FIG. 10 is a flowchart of a practical processing flow of the generator load distribution device of the present invention.

FIG. 11 is a correlation diagram between the output of the generator and the fuel flow rate.

FIG. 12 is a block diagram showing a hardware configuration of the system of the present invention.

[Explanation of symbols]

 Reference Signs List 1 Dynamic characteristic model identification unit 2 Load distribution calculation unit 3 Generator load distribution device 4 Generator output / fuel cost actual measurement data 5 Demand prediction unit 6 Generator constraint condition data 7 Autoregressive moving average model 8 Predicted demand value 9 Generator output Command value 10 Generator 11 Power system 20 Offline model identification system 21 Online load distribution calculation system 30 Online model identification / load distribution calculation system 31 Online data acquisition communication line 32 Power plant

──────────────────────────────────────────────────続 き Continuing on the front page (72) Hiroshi Nakajima 3-3-22 Nakanoshima, Kita-ku, Osaka City, Osaka Prefecture Inside Kansai Electric Power Company (72) Inventor Tsuyoshi Inada 3-3-1 Nakanoshima, Kita-ku, Osaka City, Osaka Prefecture 22 Kansai Electric Power Co., Inc. (56) References JP-A-58-140806 (JP, A) JP-A-59-1006827 (JP, A) Gonsuke Noda, Power Control and Information Series, Volume 1 Electric power system control, Japan, Denki Shoin Co., Ltd., April 25, 1986, 1st edition, 33-41 (58) Fields investigated (Int. Cl. ⁷ , DB name) H02J 3/00-5/00 H02P 9/04

Claims (9)

(57) [Claims] 1. A generator load distribution device for distributing a given load to a plurality of thermal power generators, wherein for each generator, the relationship between the generator output and fuel consumption is represented by an autoregressive moving average model. Using the dynamic characteristic model identification unit and the autoregressive moving average model, create an objective function representing the total amount of fuel consumption of all generators, and under predetermined constraints, the expected demand value of the generator output. Bei example and a load allocation calculating unit for obtaining the respective generator output as a value of the objective function is minimized based on said dynamic characteristic model identification section, .alpha.j, .beta.j, model para to γ
Meter, m is the model order, and Pj and Qj are the j
The generator output and fuel consumption at the point are described by the following equation (1).
Generator load distribution apparatus having a self-regressive moving average model . (Equation 1) 2. The generator load distribution device according to claim 1, wherein said predetermined constraint condition is a demand-supply balance constraint condition that a sum of generator outputs must be equal to said expected demand value at any time. . 3. The predetermined constraint condition includes, in addition to the supply and demand balance constraint condition, an output upper and lower limit constraint condition that defines an upper and lower limit of the output of each generator, and an upper and lower limit constraint condition of an output change speed of each generator. 3. The generator load distribution device according to claim 2, further comprising at least one of an output change speed constraint condition defining a lower limit. 4. The load distribution calculating unit, as the objective function, based on expected demand values from a current time to a plurality of control time points, and outputs all the generators within a time period up to the plurality of control time points. The generator load distribution device according to any one of claims 1 to 3, wherein an objective function representing a total amount of fuel consumption is created, and each generator output at the present time is obtained using the objective function. 5. The dynamic characteristic model identification unit according to claim 1 , wherein the model order is
Autoregressive moving average model that is one of the first to fifth order
5. The method according to claim 1, wherein a Dell is created.
A generator load distribution device according to any one of the above. 6. The dynamic characteristic model identification unit described in the following equation (2).
The autoregressive moving plane of the first-order model order (m = 1)
2. The method according to claim 1, further comprising:
Electric load distribution device. (Equation 2) Here, the upper equation is for one generator.
And a and b respectively correspond to α ₀ and α ₁ in Equation ₁ .
Where c corresponds to β ₁ and d corresponds to γ. Lower formula
Indicates the i-th generator of the plurality of generators
Where ai, bi, ci, and di are the model parameters of the i-th generator.
Where Pi, t and Qi, t are i-th generators, respectively.
The generator output and fuel consumption at time t. 7. The method according to claim 6, wherein the predetermined constraint condition is a time t
, The sum of the outputs Pt of the N generators at time t
Means that it must be equal to the forecast demand value Dt
Supply-demand balance constraint, and T
The corresponding number of control points can be described by the following Equation 6.
The generator load distribution device according to claim 6, characterized in that: (Equation 6) 8. The method according to claim 1, wherein the predetermined constraint condition is
In addition to the supply balance constraints, the upper and lower limits of the output of each generator
Output upper and lower limit constraints and the change in output of each generator
Output change speed constraint that sets upper and lower speed limits
When N is the number of target generators and T is considered
The control time points Li, Ui and δi corresponding to the
Output lower limit, output upper limit and output change speed
Therefore, the output upper / lower limit condition is expressed by the following equation (7).
The degree constraint is described by Equation 8 respectively.
8. The generator load distribution device according to claim 7, wherein:
Place. (Equation 7) (Equation 8) 9. The load distribution calculation unit according to claim 6 , wherein the objective function is
Therefore, using Equation 9, the schedule from the present to a plurality of control time points
Based on the expected demand value,
The objective function representing the total fuel consumption of all generators within
And use this objective function to calculate the current output of each generator.
A generator load distribution according to any one of claims 6 to 8 to be determined.
apparatus. (Equation 9) Here, Fi, t, Gi, t use model parameters, and
It is described as follows. (Equation 10) [Equation 11] Here, t = 1 means that each generator output (distribution ratio) is to be obtained.
Control point in time, and t = 0 represents one point before that point.
You. Qi, 0 in Equation 10 represents actual data one time before.
You.