JP7692803B2 - Demand management system, power quality management system, and method - Google Patents

Description

The present invention relates to a demand management system, a power quality management system, and a method for estimating and utilizing power demand from information on consumers, including new power consumers.

In recent years, the environment surrounding the energy business has changed significantly, with changes in medium- to long-term electricity demand, such as an increase in renewable energy and a decrease in electricity demand due to population decline, changes in the relationship between new businesses and existing electricity infrastructure operators, and the establishment of an electricity trading market. Electricity infrastructure operators are being asked to respond to these changes in the environment. As a result, there are concerns about overinvestment in electricity distribution facilities due to the uneven distribution of consumers in the distribution system, and there is a growing demand to curb capital investment. For this reason, it is necessary to switch to a new electricity network in order to improve stability of supply, environmental compatibility, and efficiency while making the most of the existing system. Against this background, there is a need to properly grasp and guide the electricity demand of electricity consumers. Furthermore, there is a growing need to provide new services to electricity consumers who have been grasped in this way.

Patent Document 1 is a background technology in this technical field. Patent Document 1 discloses a method of generating multiple power consumption models from power consumption information that stores various information on the power consumption of each consumer who has a smart meter installed, and then identifying a power consumption model that satisfies a similarity criterion from among the multiple models for second power consumption information that stores various information on each consumer who has a power meter installed but not a smart meter installed, and estimating the amount of power consumption.

Patent document 2 also discloses a power distribution system state estimation system that uses equipment information on the power distribution line and measurement information from measuring instruments at some nodes to estimate the state (active power, reactive power, voltage, etc.) of loads and power sources installed at nodes where no measuring instruments are installed, and determines the PQ amount at each point in the power distribution system.

Furthermore, Patent Document 3 discloses that a representative load pattern can be found from the accumulated long-term power usage data by collecting the actual power usage data of each power consumer for a specified period of time, calculating the feature parameters that best represent the shapes of all load patterns for the load patterns of multiple consumers for the number of days during that period, and creating clusters based on the similarity of the feature parameters.

Furthermore, Patent Document 4 discloses that a voltage monitoring and control system, a voltage monitoring and control device, a measuring device, and a voltage monitoring and control method maintain voltage by following voltage fluctuations in a power distribution system without increasing the communication load, and by issuing an appropriate voltage range command to a transformer-type voltage control device, the load is estimated from the statistical values of the smart data, and the voltage range is determined.

JP 2015-106952 A JP 2010-263754 A JP 2008-15921 A WO14/207849

By using the method disclosed in Patent Document 1, it is possible to provide an electricity demand matching means that includes a consumer extraction means for extracting consumers that are similar to other existing consumer information, an existing consumer load pattern clustering unit for classifying consumers according to their load patterns of electricity usage, a means for estimating consumers' electricity usage, and a consumer load pattern matching unit for matching the estimated electricity usage with similar load patterns from among the set of consumers extracted by the consumer extraction means.

However, the method described in Patent Document 1 does not disclose that it has a PQ amount calculation unit that calculates the PQ amount at each point in the power distribution system based on the matching results, and a power demand sensitivity analysis unit that modifies the load patterns of consumers based on the calculated PQ amount to determine the power distribution facility plan.

The method disclosed in Patent Document 2 has a PQ amount calculation unit that calculates the PQ amount at each point of the power distribution system based on the matching results, and a power demand sensitivity analysis unit that modifies the load patterns of consumers based on the calculated PQ amount and determines the power distribution facility plan. However, the method described in Patent Document 2 does not disclose any means based on consumer information.

The method disclosed in Patent Document 3 can provide power demand matching by a consumer load pattern matching unit that matches the estimated power usage with a similar load pattern among a set of consumers extracted by the consumer extraction unit. However, the method described in Patent Document 3 does not disclose that it has a consumer extraction unit that extracts consumers that are similar to other existing consumer information, an existing consumer load pattern clustering unit that classifies consumers by their load patterns of power usage, a PQ amount calculation unit that calculates the PQ amount of each point in the power distribution system based on the matching results with the means for estimating the consumer's power usage, and a power demand sensitivity analysis unit that determines a power distribution facility plan by modifying the consumer's load pattern based on the calculated PQ amount.

By using the method disclosed in Patent Document 4, it is possible to maintain voltage by following voltage fluctuations in the power distribution system without increasing the communication load while utilizing smart voltage data, and by issuing commands to transformer-type voltage control devices to set an appropriate voltage range, centralized voltage management becomes possible.

However, there is little accumulated smart data on new customers, making it difficult to estimate their load, and no method has been disclosed for accurately estimating the load of new customers.

The fact is that the aforementioned Patent Documents 1, 2, 3, and 4 do not disclose a specific method for extracting similar consumers, nor a means or system that simultaneously satisfies the need to grasp both the characteristics of consumers' power demand with high accuracy and the response to information provided to consumers under even more uncertain conditions, including a means for correcting consumer load patterns.

The present invention was made in consideration of the above-mentioned problems, and aims to provide a demand management system, a power quality management system, and a method that estimate and utilize power demand from information on consumers, including new power consumers.

In view of the above, the present invention provides a demand management system characterized by having "consumer information complementation means for complementing information on unknown items in consumer information composed of multiple items, existing consumer clustering means for extracting consumers having information similar to that of existing consumers from information on existing consumers composed of multiple items and classifying the existing consumers, demand pattern generation means for generating a demand pattern for new consumers from the demand pattern in a collection of consumers having similar information, and a demand amount calculation unit for each point for calculating the load amount at each point of the power distribution system as demand forecast information based on the demand pattern of new consumers and the demand pattern of existing consumers."

The present invention is also described as "a power quality management system that manages the power quality of a power distribution system using demand forecast information, which is the load amount at each point of the power distribution system obtained in a demand management system, the power quality management system including an SVC control device that is arranged in the power distribution system and adjusts the secondary tap position according to a setting value, and characterized in that the demand forecast information, which is the load amount at each point of the power distribution system, is used to calculate the setting value."

The present invention is also a power quality control system that performs power demand forecasting for a power distribution system using demand forecast information, which is the load amount at each point of the power distribution system obtained in a demand management system, and is characterized in that the power quality control system uses the demand forecast information, which is the load amount at each point of the power distribution system, when performing power demand forecasting for the power distribution system.

The present invention also provides a demand management method characterized by "complementing information on unknown items of consumer information composed of multiple items, extracting consumers having information similar to that of existing consumers from information on existing consumers composed of multiple items, classifying the existing consumers, generating a demand pattern for new consumers from the demand pattern of a set of consumers having similar information, and calculating the load amount at each point of the power distribution system as demand forecast information based on the demand pattern of the new consumers and the demand pattern of the existing consumers."

The present invention also provides a power quality management method for managing the power quality of a power distribution system using demand forecast information, which is the load amount at each point of the power distribution system obtained by a demand management method, characterized in that in controlling an SVC that is arranged in the power distribution system and adjusts the secondary tap position according to a setting value, the demand forecast information, which is the load amount at each point of the power distribution system, is used to calculate the setting value.

The present invention also provides a power quality management method for forecasting the power demand of a power distribution system using demand forecast information, which is the load at each point of the power distribution system obtained in a demand management method, characterized in that the power quality management method uses the demand forecast information, which is the load at each point of the power distribution system, when forecasting the power demand of the power distribution system.

According to a representative embodiment of the present invention, it is possible to grasp the effective power demand of consumers and obtain response characteristics to the provision of information, which can result in the achievement of effects such as improved business viability by more appropriately guiding consumers.

1 is a diagram showing an example of the overall configuration of a power demand management system 10 according to an embodiment of the present invention. FIG. 2 is a diagram showing an example of the data structure of a new customer information database DB1. FIG. 4 is a diagram showing the processing flow in the new customer information complementation unit 30. 13 shows a processing flow in the existing customer clustering unit 40. 13 shows a processing flow in a demand pattern generating unit 50. 11 shows a processing flow in the demand calculation unit 60 for each point. FIG. 2 is a diagram showing an example of the data structure of a demand forecast information database DB2. FIG. 1 is a diagram showing an example of application of a power quality control system to an SVR. FIG. 1 shows general process examples for various applications of the power quality control system. FIG. 1 is a diagram showing an example of an application of the power quality control system to a power distribution line demand forecasting system that forecasts demand on a power distribution line. FIG. 13 is a diagram showing the relationship between demand (forecast) and induction (forecast) taking nonlinearity into account.

The following describes an embodiment of the present invention with reference to the drawings.

In the first embodiment, a power demand management system is described. FIG. 1 shows an example of the overall configuration of a power demand management system 10 according to the first embodiment of the present invention.

The electricity demand management system 10 formed using a computer includes a new consumer information database DB1 that holds new consumer information D1 in advance, and a demand forecast information database DB2 that holds demand forecast information D2 after the demand management plan is implemented. In terms of the processing content of the calculation unit in the computer, it is configured to include at least the functions of a new consumer information complementation unit 30, an existing consumer clustering unit 40, a demand pattern generation unit 50, and a demand amount calculation unit 60 for each point.

With the configuration of Example 1 illustrated in FIG. 1, the electricity demand management system 10 adds information D1 about new consumers to information about existing consumers to obtain demand forecast information D2 that estimates electricity demand in a new electricity system regionally and over time. In particular, the information D1 about new consumers, which tends to be insufficient in terms of available information, is supplemented and enriched to improve the estimation accuracy of the demand forecast information D2.

Figure 2 shows an example of the data structure of the new consumer information database DB1. The new consumer information database DB1 is composed of new consumer information D1, including consumer ID (D11), address D12, contract information D13, contract capacity D14, business type D15, number of residents D16, stored appliances D17, and daytime at-home rate D18.

This information is publicly obtained by the power company from the consumer based on the timing when the consumer and the power company concluded a contract, or based on questionnaires given to the consumer, and some values are determined for each electricity consumer with whom the power company has a contract. However, information such as the number of residents D16, stored electrical appliances D17, and the rate of people at home during the day D18 cannot always be obtained accurately from the consumer.

As a result, if necessary, the daytime at-home rate D18 is determined by dividing it between the same contracted capacity D14 and the number of residents D16 based on the value (%) of the hourly daytime at-home rate from the "National Life Statistics" published by NHK as one of the embodiments. If the number of residents D18 is not available, it may be necessary to take measures such as determining it from the contracted capacity D14 alone. However, these measures will be carried out manually as appropriate.

As such, not all items in the new customer information database DB1 in Figure 2 are filled, and the current situation is that items can only be filled manually.

Figure 3 shows the processing flow in the new consumer information complementation unit 30. First, in processing step S301, a questionnaire is conducted on new consumers and the answers to the questionnaire questions are quantified. In this case, the questionnaire questions are determined so as to obtain data items in the new consumer information database DB1 in Figure 2, but in reality, it is possible that answers corresponding to the items are not obtained, resulting in unknown information.

In processing step S302, the Euclidean distance of each consumer in an N-dimensional space for the number of questions N is calculated. Next, in processing step S303, groups are searched for that reduce the Euclidean distance between consumers. For example, when the respondent is person A, other new consumers who show similar response tendencies to person A are extracted as group A. This allows, for example, single people, young married couples, and families with children to be identified as groups that show similar response tendencies.

Furthermore, in processing step S304, the average of the responses from each group is calculated, and the center of gravity vector is calculated. After the consumer information is extracted from the consumer data, the consumer information is normalized to the range of 0 to 1, and statistical information such as the average and variance for each data item in the other consumer information is calculated. This is expected to reveal distributions within groups that reflect differences in work style, income, etc., even within the same group.

In a subsequent processing step S305, a feature value is calculated for a specific consumer information by dividing the deviation from the average by the variance. Then, a feature vector of the consumer information is obtained for each consumer, and a classification tree is generated based on the feature vector. If the consumer information data for an individual consumer group classified based on the classification tree contains unknown information, the consumer information may be supplemented with information such as the average value and confidence interval from the vector components of the same classified consumer.

For example, if a feature in the same classification tree indicates a high probability that the household is a "two-person full-time dual-income household," it is possible to supplement the data by adding "2 people" or "0%" to the previously unknown "number of residents" and "percentage of people at home during the day."

Figure 4 shows the processing of the existing consumer clustering unit 40. First, in processing step S401, the feature vectors of each existing consumer are calculated. Note that, although not shown in Figure 1, data of the same configuration as that of Figure 2 is also held for existing consumers. Furthermore, the calculation of the feature vectors may include not only information on existing consumers, but also information on new consumers.

In processing step S402, the consumer's feature vectors are clustered. Examples of clustering methods include the Kmeans method, the Ward method, and random forests.

In processing step S403, combination patterns with a reliability of, for example, 90% are extracted, and in processing step S404, the consumer feature vectors are classified into multiple consumer groups. If classification is not complete after confirmation processing in processing step S405 that classification is complete, the process proceeds to processing step S407, where the values of the consumer feature vectors are updated, and processing returns to processing step S404, and is repeated until classification is complete.

Figure 5 shows the processing of the demand pattern generation unit 50. In the processing of the demand pattern generation unit 50, the address D12, contract information D13, contract capacity D14, business type D15, number of residents D16, and daytime at-home rate D18 of the consumer information supplemented by the new consumer information supplementation unit 30 are linked to the new consumer information and meteorological information, and clustering processing based on the load pattern is performed.

From the consumer groups based on the load patterns obtained, a decision tree is created based on the supplemented consumer information: address D12, contract information D13, contract capacity D14, business type D15, number of residents D16, and rate of being at home during the day D18. This allows consumers with similar load patterns to be further reclassified based on their characteristics.

For example, if, according to the supplemented consumer information, there is a high probability that Mr. A's household is a "full-time, two-income household," and similar load patterns in other "full-time, two-income households" have low demand on weekdays from 8:00 a.m. to around 8:00 p.m., and high load demand from 8:00 p.m. to 2:00 a.m., then the load pattern of Mr. A's household can be estimated to be similar to such similar load patterns. Load patterns can be formed appropriately by day, week, month, season, weekday, holiday, etc.

Figure 6 shows the processing of the demand calculation unit 60 for each point. In the flow of Figure 6, first, in processing step S601, an area is determined. The area corresponds to an administrative district or a map mesh. In processing step S602, consumers in the corresponding area are selected based on their addresses. In processing step S603, for the set of selected consumers, the load is calculated as the sum of the estimated demand and actual demand at each time slice. Note that there may be cases where information representing consumers is given as a point. In this case, the point may be a latitude and longitude, or the sum of the load within an area obtained by dividing a map into a 10 km x 10 km mesh.

In processing step S604, the load power factor in the area is set. In processing step S605, the load power factor is used to calculate the active power P and reactive power Q from the demand for each time section. The difference between the total load and the maximum allowable load is the load margin at each point. If the power factor is given, the load margin can be calculated as active power P and reactive power Q. It is possible to grasp the active and reactive power of power generation by combining the equipment capacity data of distributed power sources such as solar power generation connected to the distribution line in the area with weather data. In addition, even if the power generation and load are behind the smart data, if weather data is available, it is also possible to grasp the actual load by estimating the amount of power generation from the weather data and adding it to the smart data.

Figure 7 shows an example of the data structure of the demand forecast information database DB2. Demand forecast information D2 is composed of area ID D21, active power by time D22, reactive power D23, connected distribution lines D24, etc.

According to the demand forecast information D2 obtained by the process of FIG. 6 and stored in the demand forecast information database DB2 of FIG. 7, it is possible to grasp the geographical (area ID, D21) distribution of the load (active power D22, reactive power D23) in the new power system configuration, including existing and new consumers. Furthermore, this load distribution can be grasped not only geographically, but also including time-series (vertical axis of active power D22 and reactive power D23) fluctuations in demand, such as by day, week, or month. Furthermore, since the geographical load distribution grasped here includes address information, it can be grasped in relation to the positions of equipment such as switches and disconnectors on the distribution line in the power system (connected distribution line D24) and information on the branching of the distribution line.

According to the electricity demand management system 10 illustrated in the first embodiment, the accuracy of the consumer data is improved by supplementing the missing information about new consumers, and as a result, the accuracy of the finally obtained demand forecast information D2 can be improved.

In Example 2, a power quality control system is described in which the demand forecast information D2 estimated by the power demand control system in Example 1 is reflected in the monitoring and control of the power system. By using the power quality control system, the demand forecast information D2 for each distribution line can be accurately estimated, making it possible to apply the power quality control system.

Figure 8 shows an example of application of a power quality control system to an SVR (Step Voltage Regulator). The left side of Figure 8 shows the power demand control system 10 (load control unit) and the demand forecast information D2 estimated here. The right side of Figure 8 shows the realization of the function of the setting unit 30 that determines the setting value of the SVR, as an example of a power quality control system. DB3 is an existing consumer information database that accumulates existing consumer information D3.

To briefly explain the necessity of installing SVRs, first, the voltage of a distribution line fluctuates depending on the supply point and time of day due to line voltage drop caused by load power. High-voltage voltage regulators are installed midway along the line to adjust the voltage of high-voltage distribution lines so that it falls within a specified voltage range, and SVRs are generally used as a typical example. SVRs use autotransformers, and voltage control is performed by automatically switching the tap. By setting the reference voltage and LDC operating conditions, the SVR's sending voltage is controlled to a value based on the reference voltage plus an LDC compensation value (the line voltage drop value from the SVR installation point to the subsequent load center point) proportional to the magnitude of the load current. LDC (Line Drop Compensation) is a method of keeping the voltage at a certain point constant.

The voltage calculation unit 31 in the settling unit 30 in FIG. 8 performs voltage calculations for the power distribution system using power flow calculations based on the network topology and line impedance for the power distribution line stored in the power distribution system information, based on the demand forecast information D2 obtained by the power demand management system 10 (load management unit). This calculation makes it possible to calculate the current and voltage values at each node (point) of the power distribution line for each time section.

Based on this result, the voltage regulator setting unit 32 in the setting unit 30 calculates setting values 33 (Vref, R, X) for estimating the load center voltage of the distribution line. This calculation method is called LDC. The SVC installed on the distribution line performs tap control according to the setting values 33, and controls the voltage of the distribution line.

These setting values 33 are used for a period determined for each distribution line. The period may be common for several years, such as one year, or may be used for each non-weekend. It is also possible to use different values for different seasons, or for daytime and nighttime.

These setpoint values can be calculated by closely understanding the behavior of consumers' electricity usage using smart data, etc. The electricity demand management system 10 can accurately predict the behavior of new consumers, making it possible to calculate setpoints at various time intervals with even greater precision.

Figure 9 shows an example of general processing content for more diverse applications of the power quality management system, and shows the process flow when using demand forecast information D2 estimated by the power demand management system for general power analysis, not just voltage. The flow in Figure 9 shows an example of analysis processing such as voltage estimation performed by the voltage calculation unit 31.

The upper part of the speech bubble in Figure 9 shows an example of a power distribution system that is typically monitored. The power distribution lines are appropriately branched out from the circuit breaker FCB at the substation exit via sensor-equipped switches DS, manual switches DSh, etc. Quality control of such a power distribution system requires obtaining voltage information at each location, and currently available data include measurement data from the sensor-equipped switches DS, SMAME (measurement data: amount of power used (30-minute value)) data, and customer load data.

However, none of these methods are widely used in sufficient numbers, cannot be obtained over the time span required for quality control, and there are problems with the accuracy of the data. In terms of data accuracy, customer load data is the most suitable, but there is also an issue of how widely used they are, and it is unavoidable that there will be unmeasured sections on the distribution system.

In response to these current measurement circumstances, the demand forecast information D2 from the power demand management system 10 can accurately forecast demand data for each distribution line regardless of location, making it possible to apply this to planning support. Conventionally, simple monitoring was based on measurements from a limited number of sensor-equipped switches DS, but the use of the demand forecast information D2 from the power demand management system 10 will enable even more advanced monitoring technology in the future.

On the other hand, there are the following needs for the advancement of monitoring technology for power distribution systems, depending on operational objectives. First, with regard to overload, it is necessary to grasp the maximum values and the exact time of their occurrence to ensure that the equipment (distribution lines and switches) does not deviate from the allowable values. With regard to equipment utilization rate, it is necessary to manage the amount of electricity used in a day in detail and use this information when considering equipment capacity during equipment planning. With regard to voltage distribution, it is necessary to grasp in detail when and where voltage increases occur due to the interconnection of renewable energy sources such as PV.

To meet the above needs, current power distribution systems are so large that sensors are installed in only a limited number of locations, making it difficult to accurately grasp the current and voltage conditions in non-measured branch sections and low-voltage systems.

Therefore, in the second embodiment, a method using clustering is applied to extract features that will require strengthened monitoring for each operational purpose using the demand forecast information D2. Clustering is a method for extracting features by grouping individuals with similar evaluation values into the same group for individuals with multiple evaluation values.

In the field of power distribution systems, attention is focused on overload, capacity utilization rate, and voltage distribution. For this reason, in FIG. 9, which illustrates the processing in the voltage calculation unit 31, in the first processing step S801, a load model is generated according to the management items. These are to generate a model for the maximum value and occurrence time of overload in power flow management (processing step S801a), to generate a model for monthly power usage in the capacity utilization rate in power flow management (processing step S801b), and to generate a model for determining the hourly voltage distribution in voltage management (processing step S801c).

For these models, since the power distribution system shown in the upper part of the balloon in Figure 9 has sections where sensors are installed and sections where measurements are not made, in processing step S802, a load model for the section where sensors are installed is estimated, and in processing step S803, a load model for the section where measurements are not made is estimated, resulting in a corrected load model.

For this estimation, it is advisable to use the demand forecast information D2 from the power demand management system 10. For example, in the power distribution system shown in the upper part of the balloon in Figure 9, when a combination of two sensor-installed sections and two non-measured sections is used, if the sensor-installed section can be estimated with high accuracy from the sensor detection value, the measurement data is used, but when the sensor-installed section has a problem with low sensor accuracy or when the section is non-measured, it is advisable to use the demand forecast information D2 from the power demand management system 10.

By geographically connecting these sensor-installed sections and non-measured sections, the load state at each point from the substation exit to the end of the distribution line can be determined as a geographical distribution. Also, as shown in the lower part of the balloon in Figure 9, it can be determined as, for example, a time-series change in the amount of power at a specific point on the distribution line, and in this example, this is determined by comparing the result obtained using the measurement sensor with the result obtained using the demand forecast information D2 from the power demand management system 10. These processes are performed in processing step S804, and the corrected load model is used to determine, for example, the voltage and current over the entire section.

In processing step S805, the analysis results are output and displayed on the display. The display format at this time should be as shown in the balloon in Figure 9.

According to the above-mentioned second embodiment, the voltage can be estimated by grasping the load distribution for each section with high accuracy, and the deviation from the true value can be reduced.

In the second embodiment, a power quality control system is described in which the demand forecast information D2 estimated by the power demand control system in the first embodiment is reflected in demand management in the power grid's VPP (Virtual Power Plant) and DR (Demand Response).

Note that a VPP is when the owner or a third party of a consumer-side energy resource, power generation equipment directly connected to the power grid, or storage equipment controls the energy resources to provide functions equivalent to those of a power plant, and a DR is when the owner or a third party of a consumer-side energy resource controls the energy resource to change the electricity demand pattern.

Generally, bulk power systems make detailed demand forecasts using various information such as temperature and solar radiation. On the other hand, distribution systems make simple predictions using only information from sensors installed in the distribution system. In the future, more advanced demand forecasts for distribution lines will be necessary to achieve even more efficient facility planning.

Unlike trunk systems, the demand for distribution systems varies depending on the region, distribution line and section, customer contract type and capacity, and lifestyle (whether or not the customer is at home). For this reason, distribution line forecasting methods require forecasting methods that utilize the results of detailed smart data and data analysis of temperature, etc. It is possible to collect loads from multiple distribution lines, analyze the demand curve, and apply big data analysis (clustering technology, regression analysis, etc.) to the factors that cause fluctuations in the smart data.

Figure 10 shows an example of a power quality management system applied to a power distribution line demand forecasting system that forecasts demand on power distribution lines. The power distribution line demand forecasting system includes a power distribution line information database DB5 and an estimated parameter database DB6. The existing power distribution line information database DB5 includes smart data, sensor measurements, average temperatures, etc., but in the third embodiment of the present invention, it is preferable to further hold demand forecast information D2 estimated by the power demand management system of the first embodiment.

Specifically, the distribution line demand forecasting system processes the smart data of general lighting consumers, analyzes the demand curve, and clarifies factors of fluctuation such as region, weather information, and consumer contract type and capacity. Finally, methods for predicting future demand curves using the following parameters (day of the week, time of day, temperature, etc.) are considered and evaluated. Based on these predictions, a plan is drawn up for whether or not to implement demand response.

In the example of Figure 10, the distribution line demand forecast is composed of a part that is started every day, performs statistical processing (cluster analysis and regression analysis) of the previous day's smart data, and learns the parameters for the demand forecast by machine learning, and a part that is started when a demand forecast is required and performs the distribution line demand forecast.

In processing step S101, the information in the distribution line information database DB5 is used to perform an analysis (cluster analysis) of the demand curve by time division of the previous day. Similarly, in processing step S102, the information in the distribution line information database DB5 is used to perform an analysis (cluster analysis) of the demand curve by day of the week division of the previous day. As a result, for example, characteristic trends for each day of the week are measured for the amount of power consumed every 30 minutes from midnight to midnight, around 8:00 a.m. and around 8:00 p.m., and these are each classified as new events.

Furthermore, in processing step S103, the correlation between the average temperature of the previous day and the amount of electricity used is obtained by regression analysis using information from the power distribution line information database DB5. This makes the relationship between temperature and electricity clear, making it possible to estimate the amount of electricity used with the influence of temperature as a separate factor. In processing step S104, taking these clustering and comparisons into consideration, estimated parameters are obtained as feature quantities and stored in the estimated parameter database DB6.

Next, in the demand forecast process, when demand forecasting is started in processing step S105, the distribution line to be the subject of the demand forecast is determined in processing step S106, and demand forecasting is performed and output using the parameters stored in the estimated parameter database DB6 in processing step S107. Note that conventional distribution line demand forecasting methods perform demand forecasting from the growth rate of past sensor measurement values.

In this invention, distribution line demand is forecast based on the results of an analysis of the factors that cause fluctuations in the demand curve obtained from the SmartMe data and demand forecast information D2 estimated by the power demand management system. By using SmartMe data, time segments, day segments, contracted kW, and temperature as input data, it becomes possible to improve the distribution line demand forecast by capturing the individual characteristics of each consumer.

Figure 11 shows the relationship between demand (forecast) and induced amount (forecast) taking nonlinearity into account. The axes in Figure 11 are time, temperature, and power amount. Here, the nonlinear function is expressed as a polynomial with a maximum degree of 5 in the time direction and a maximum degree of 2 in the temperature direction over a one-day range, making it possible to simulate the behavior of daily power demand at various temperatures. Then, partial differential coefficients are calculated for each time and temperature cross section. These differential coefficients are regarded as the sensitivity to temperature and time and are stored in a table, from which demand and induced amounts are calculated.

To model human behavior such as whether or not to use electricity, a sigmoid curve, which is one type of cumulative distribution function, is used. The sigmoid curve in equation (1) is a horizontally stretched S-shaped function that takes on a value of 0 at the limit of -∞ and a value of 1 at the limit of +∞. In physics, it is known that the change in electron spin in the up and down directions when an external magnetic field is applied to a magnetic substance, as well as growth curves in economics and mortality curves in biology, can also be approximated by a sigmoid curve.

In addition, this sigmoid curve can be expressed as a function of the original sigmoid curve by differentiation using equation (2), which has the advantage that it is easy to calculate without the need for numerical calculations for differentiation operations in computer simulations.

The sigmoid curve can be translated in the x-axis direction, or the magnitude of the slope at the inflection point can be changed by changing the value of a in the formula. Mathematically, the cumulative distribution function is differentiated to obtain a probability density function. In this invention, the sigmoid curve (cumulative distribution function) can be obtained by assigning the partial differential coefficient described above to a in the formula, and the probability density function can be obtained by differentiating the sigmoid curve. This makes it possible to express the response probability to demand response.

10: Demand management system DB1: New consumer information database 30: New consumer information complementation unit 40: Existing consumer clustering unit 50: Demand pattern generation unit 60: Demand amount calculation unit for each point DB2: Demand forecast information database

Claims (12)

The power distribution system includes a new consumer information complementing unit that complements information of unknown items in information of new consumers composed of multiple items , an existing consumer clustering unit that extracts consumers having information similar to that of existing consumers from information of existing consumers composed of multiple items and classifies the existing consumers , a demand pattern generating unit that generates a demand pattern of new consumers from a demand pattern in a set of consumers having the similar information, and a demand amount calculation unit for each point that calculates a load amount at each point of the power distribution system as demand forecast information based on the demand pattern of new consumers and the demand pattern of existing consumers ,
The new consumer information complementation unit quantifies the answers to each question in a questionnaire administered to a plurality of new consumers, calculates the Euclidean distance between each new consumer in a three-dimensional space with the same number of questions as the number of questions in the questionnaire, searches for groups in which the Euclidean distance between each new consumer is small, calculates a center of gravity vector by averaging the questionnaire answers in each of the groups, extracts characteristics of each of the groups based on the center of gravity vector, and, if there is information on an unknown item in the information of any of the new consumers classified into each of the groups according to the characteristics, complements the information on the unknown item from information on other new consumers in each of the groups . The demand management system according to claim 1,
The demand management system is characterized in that the existing consumer clustering unit uses a K-means method. A power quality management system that manages power quality of a power distribution system by using the demand forecast information, which is a load amount at each point of the power distribution system obtained in the demand management system according to claim 1 or 2 ,
The power quality control system includes an SVC control device that is placed in a power distribution system and adjusts a secondary tap position in accordance with a setting value, and is characterized in that the demand forecast information, which is the load at each point in the power distribution system, is used to calculate the setting value. A power quality management system that performs power demand forecasting for a power distribution system using the demand forecast information, which is a load amount at each point of the power distribution system obtained in the demand management system according to claim 1 or 2 ,
The power quality control system is characterized in that, when predicting power demand in a power distribution system, the demand forecast information, which is the load at each point in the power distribution system, is used. 5. The power quality management system according to claim 4 ,
The power quality control system is characterized in that , when forecasting power demand in a power distribution system, it has: means for approximating factors of power demand with a nonlinear function and performing multidimensional sensitivity analysis using partial differential coefficients at each coordinate of the nonlinear function; and means for determining a probability density function based on the partial differential coefficients and calculating a consumer response probability. 6. The power quality management system according to claim 5 ,
The power quality management system is characterized in that the nonlinear function has a product of terms with a maximum order of 5 in the time direction and a maximum order of 2 in the temperature direction , and a sigmoid function is used as a probability density function projected from the partial differential coefficient of the nonlinear function. For information on new consumers consisting of multiple items, information on unknown items is supplemented ; for information on existing consumers consisting of multiple items, consumers having information similar to that of existing consumers are extracted, the existing consumers are classified , a demand pattern for new consumers is generated from a demand pattern in the set of consumers having the similar information, and a load amount at each point of the power distribution system is calculated as demand forecast information based on the demand pattern for new consumers and the demand pattern for existing consumers ;
In complementing the information on the unknown items, the demand management method includes quantifying the responses to each question in a questionnaire administered to a plurality of new consumers, calculating the Euclidean distance between each new consumer in a dimensional space with the same number of questions as the number of questions in the questionnaire, searching for groups in which the Euclidean distance between each new consumer is small, calculating a center of gravity vector by averaging the questionnaire responses of each of the groups, extracting characteristics of each of the groups based on the center of gravity vector, and if the information on any of the new consumers classified into each of the groups according to the characteristics contains information on unknown items, the information on the unknown items is complemented from information on other new consumers in each of the groups . 8. A demand management method according to claim 7 , comprising:
A demand management method characterized in that the K-means method is used to classify existing customers. A power quality management method for managing power quality of a power distribution system by using the demand forecast information, which is a load amount at each point of the power distribution system obtained in the demand management method according to claim 7 or 8,
A power quality management method, comprising the steps of: controlling an SVC that is placed in a power distribution system and adjusts a secondary tap position in accordance with a setting value; and using the demand forecast information, which is the load at each point in the power distribution system, to calculate the setting value. A power quality management method for forecasting power demand in a power distribution system using the demand forecast information, which is a load amount at each point in the power distribution system obtained in the demand management method according to claim 7 or 8 , comprising:
A power quality management method comprising the steps of: predicting power demand in a power distribution system; and using the demand forecast information, which is the load at each point in the power distribution system. A power quality control method according to claim 10 , comprising:
A power quality management method characterized by the steps of : approximating factors of power demand using a nonlinear function; performing a multidimensional sensitivity analysis using partial differential coefficients at each coordinate of the nonlinear function ; determining a probability density function based on the partial differential coefficients ; and calculating a consumer response probability when forecasting power demand in a power distribution system. The power quality control method according to claim 11 ,
A power quality management method characterized in that the nonlinear function has a product of terms with a maximum order of 5 in the time direction and a maximum order of 2 in the temperature direction , and a sigmoid function is used as a probability density function projected from the partial differential coefficient of the nonlinear function.

Images