Prediction system, prediction method, and system management system

Abstract

To update a determined prediction model by determining a prediction model to be predicted in which response characteristics are changed with a low load.SOLUTION: A prediction system generates update data. The update data include sample data for each observation period for each of one or multiple prediction objects. For each prediction object, sample data include at least one observation value obtained during an observation period for the prediction object, and an observation value obtained during the observation period for each of one or multiple factors. The prediction system calculates dissimilarity between a first sample group and a second sample group for each prediction object. For each prediction object, the first sample group is one or multiple sample data used for identifying a prediction model corresponding to the prediction object, and the second sample group is one or multiple sample data not used yet for identifying a prediction model corresponding to the prediction object. The prediction system updates a prediction model corresponding to a prediction object having dissimilarity equal to or more than a threshold.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] The present invention generally relates to data prediction.

Background Art

[0002] The prediction of future data is carried out in various fields such as weather prediction, energy demand prediction, and prediction of various economic indicators. For example, in the power field, there is an example of reducing transmission losses by controlling grid equipment based on the prediction of power demand every several tens to several hours for tens to thousands of substations.

[0003] The object to be predicted (prediction target) has a property (response characteristic) of showing a specific response under the influence of one or more factors. For example, the power demand of a substation (an example of a prediction target) changes under the influence of fluctuations in temperature (an example of a factor). When the temperature is high, the operation amount of air conditioning equipment increases, and the power demand also tends to increase. In prediction, a prediction model showing the response characteristic of the prediction target with respect to the factors estimated from past observation data is identified. <^

[0004] However, the response characteristic changes over time. For example, in power demand, the response characteristic changes over time due to seasonal changes, changes in people's lifestyles, or changes in the operating conditions of factories. In order for the response characteristic of the prediction target to follow the change over time, it is desirable to continuously update the prediction model. Furthermore, when prediction and control are repeated in a short cycle (for example, when control based on demand prediction for each substation is performed), it is desirable to continuously update the prediction model in a short cycle.

[0005] As technologies for updating prediction models, Japanese Unexamined Patent Application Publication No. 2000-276460 (Patent Document 1), Japanese Unexamined Patent Application Publication No. 2004-86896 (Patent Document 2), and Japanese Unexamined Patent Application Publication No. 2022-28338 (Patent Document 3) are known.

[0006] Patent Document 1 discloses a technique for predicting electricity demand by training a prediction model and using the trained prediction model as the best prediction model only when the error in the output data of the trained prediction model decreases. Patent Document 2 discloses a technique for updating a prediction model when the error in the predicted value output from the prediction model exceeds a set tolerance value. Patent Document 3 discloses a technique for updating a prediction model when an outlier is detected in the observed state values ​​of the training dataset. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Japanese Patent Publication No. 2000-276460 [Patent Document 2] Japanese Patent Publication No. 2004-86896 [Patent Document 3] Japanese Patent Publication No. 2022-28338 [Overview of the project] [Problems that the invention aims to solve]

[0008] In Patent Document 1, model training is required to determine whether to update the prediction model, resulting in a heavy processing load for determining whether to update all prediction models, making it difficult to complete updates within a predetermined time. In Patent Document 2, the prediction model is not updated until the error in the predicted value exceeds an acceptable value, and the timing of model updates is delayed in response to changes in the response characteristics of the target being predicted, resulting in errors occurring over a long period. In Patent Document 3, the state values ​​of the training data are factor data used to identify the prediction model, and even if outliers occur in the observed values, it does not necessarily mean that the response characteristics have changed, so there is a risk of unnecessary model updates (model updates that lead to a decrease in prediction accuracy).

[0009] This invention was made in consideration of the above points, and aims to determine the prediction model of a target whose response characteristics have changed with low load and to update the determined prediction model. [Means for solving the problem]

[0010] The prediction system generates update data from the prediction target observation data and factor observation data. The prediction target observation data includes time series data of past observations obtained for each of the one or more prediction targets. The factor observation data includes time series data of past observations obtained for each of the one or more factors that may influence the one or more prediction targets. The update data includes sample data for each of the one or more prediction targets for each observation period. For each prediction target, for each observation period, the sample data includes one or more observations obtained for the prediction target during that observation period, and observations obtained for each of the one or more factors during that observation period. The prediction system calculates the degree of difference between the first sample group and the second sample group for each prediction target. For each prediction target, the first sample group is one or more sample data used to identify the prediction model corresponding to that prediction target, and the second sample group is one or more sample data that has not yet been used to identify the prediction model corresponding to that prediction target. For each prediction target, the prediction model takes factor prediction values, which are data containing predicted values ​​for future periods for each of one or more factors, as input, and outputs predicted values ​​of observed values ​​obtained for the prediction target during those future periods. The prediction system performs an update necessity determination for each prediction target, which is a determination of whether the degree of difference is above a threshold, and outputs the result of the update necessity determination. A prediction model for which the update necessity determination result is true is a prediction model that needs to be updated. A prediction model for which the update necessity determination result is false is a prediction model that does not need to be updated. The prediction system updates the prediction models that have been determined to need updating from among the one or more prediction models corresponding to each of the one or more prediction targets. [Effects of the Invention]

[0011] According to the present invention, it is possible to determine the prediction model of a target whose response characteristics have changed with low load and update the determined prediction model.

[0012] Problems, configurations, and effects other than those described above will be clarified by the following description of the embodiments.

Brief Description of the Drawings

[0013] [Figure 1] It is an overall configuration diagram of the system management system. [Figure 2] It is a conceptual diagram showing the data flow of the system management system. [Figure 3] It is a functional configuration diagram of each device constituting the power demand prediction system. [Figure 4] It is a functional block diagram of the data extraction unit. [Figure 5] It is a functional block diagram of the model update control unit. [Figure 6] It is a functional block diagram of the model management unit. [Figure 7] It is a configuration diagram of the prediction target data for update. [Figure 8] It is a configuration diagram of the factor data for update. [Figure 9] It is a configuration diagram of the factor data for prediction. [Figure 10] It is a configuration diagram of the classified data for update. [Figure 11] It is a configuration diagram of the model management data. [Figure 12] It is a configuration diagram of the model update control data. [Figure 13] It is a configuration diagram of the model data. [Figure 14] It is a flowchart showing the power demand prediction process. [Figure 15A] It is a diagram showing the concept of determining whether update is necessary based on the degree of difference. [Figure 15B] It is a diagram showing the concept of determining whether update is necessary based on the degree of difference. [Figure 16A] It is a diagram showing the model update timing and update processing load according to the comparative example. [Figure 16B] It is a diagram showing the model update timing and update processing load according to the embodiment. [Figure 17A]This figure shows a specific example of a case where a model update is necessary. [Figure 17B] This figure shows a specific example of a case where a model update is not necessary. [Figure 17C] This figure shows a specific example of a case where a model update is not necessary. [Figure 18] This is a functional block diagram of the data extraction unit in one modified example. [Modes for carrying out the invention]

[0014] In the following explanation, "interface device" may refer to one or more interface devices. These one or more interface devices may be at least one of the following: • One or more I / O (Input / Output) interface devices. An I / O (Input / Output) interface device is an interface device to at least one of the following: an I / O device and a remote display computer. The I / O interface device to the display computer may be a communication interface device. The at least one I / O device may be either a user interface device, such as an input device like a keyboard and a pointing device, or an output device like a display device. • One or more communication interface devices. One or more communication interface devices may be one or more identical communication interface devices (e.g., one or more NICs (Network Interface Cards)) or two or more different communication interface devices (e.g., a NIC and an HBA (Host Bus Adapter)).

[0015] Furthermore, in the following explanation, "memory" refers to one or more memory devices, which are typically main memory devices. At least one memory device in memory may be a volatile memory device or a non-volatile memory device.

[0016] Furthermore, in the following explanation, "persistent storage device" refers to one or more persistent storage devices. Persistent storage devices are typically non-volatile storage devices (e.g., auxiliary storage devices), specifically, for example, HDDs (Hard Disk Drives) or SSDs (Solid State Drives).

[0017] Furthermore, in the following explanation, "storage device" may refer to at least memory, including both memory and persistent storage.

[0018] Furthermore, in the following explanation, "processor" refers to one or more processor devices. At least one processor device is typically a microprocessor device such as a CPU (Central Processing Unit), but may be other types of processor devices such as a GPU (Graphics Processing Unit). At least one processor device may be single-core or multi-core. At least one processor device may be a processor core. At least one processor device may be a broader processor device such as a hardware circuit that performs some or all of the processing (e.g., an FPGA (Field-Programmable Gate Array) or ASIC (Application Specific Integrated Circuit)).

[0019] Furthermore, in the following explanation, functions may be described using the expression "yyy section," but a function may be the computer program itself, implemented by one or more computer programs executed by a processor, implemented by one or more hardware circuits (e.g., FPGA or ASIC), or implemented by a combination thereof. When a function is implemented by a program executed by a processor, the defined processing is carried out using memory and / or interface devices as appropriate, so the function may be at least a part of the processor. Processing described with a function as the subject may be processing performed by the processor or a device having that processor. Programs may be installed from program source. Program source may be, for example, a program distribution computer or a computer-readable recording medium (e.g., a non-temporary recording medium). The description of each function is an example, and multiple functions may be combined into one function, or one function may be divided into multiple functions.

[0020] One embodiment of the present invention will be described in detail below with reference to the drawings.

[0021] In this embodiment, for the sake of explanation, we will use the power demand at the substation level in a power grid as an example of the target of prediction. (1) Overall structure

[0022] Figure 1 shows the overall configuration of the system management system 1 in this embodiment.

[0023] The power system management system 1 is a system for calculating predicted power demand values ​​for each substation at a predetermined date and time based on factor data including observed data (and predicted data) of factors and observed power demand data for each substation, and for controlling power system equipment 41 based on the predicted power demand values ​​for each substation. The power system management system 1 consists of various devices owned by the power operator 2, the power system 4, and the external attribute information provider 5. Furthermore, each of the communication paths 60, 61, and 62 is a communication path (typically a communication network) that connects various devices such as a WAN (Wide Area Network) or LAN (Local Area Network) so that they can communicate with each other.

[0024] Factor data is created from attribute data and includes the values ​​of one or more factors. Attribute data is data that includes information representing elements that can be factors, and includes grid attribute data held by the power operator 2 and external attribute data provided by the external attribute information provider 5. A factor is an element that can influence the prediction target. Factor data may include one or more values ​​(one or more factor values) for each of one or more factors. For example, if the factor data is for one day, the factor data may include hourly factor values ​​for each factor. One or more factors may be at least one of the following, for example. Weather conditions such as temperature, humidity, solar radiation, wind speed, and atmospheric pressure. • Energy supply from renewable energy sources such as solar power. • Trading volume and price of fuels such as crude oil and natural gas. • Subject to transmission capacity and other aspects related to power transmission lines. • Generator operating status, such as the generator's operation or maintenance schedule. • Calendar date including year, month, and day, day of the week, and a flag value indicating the type of day (as arbitrarily set). • Whether or not sudden events such as typhoons or other unforeseen circumstances occur. • Economic conditions such as the number of energy consumers, industry trends, and business sentiment indices. • Information on the movement of people and vehicles, such as the occupancy rate of express trains, the number of passengers, the number of reserved seats, or road traffic conditions. • The number of communication terminals connected to a communication base station.

[0025] Power Operator 2 is the operator that controls the grid equipment 41. Power Operator 2 is equipped with a demand forecasting data management device 30 that manages data for calculating predicted power demand values, such as grid data and external attribute data; a power demand forecasting device 31 that calculates predicted demand values; a grid control calculation device 21 that calculates the control content of the grid equipment 41; a grid equipment management device 22 that manages and controls the grid equipment 41; and an information input / output terminal 20 for exchanging data with these devices. Grid data includes grid observation data and grid attribute data. Grid observation data includes measurement data from each point in the power system 4 measured by the measuring device 40 of the power system 4 (for example, measured values ​​(observed values) such as voltage, current, power, and frequency). Data on the power demand (target of forecasting) of each substation is included in the grid observation data. System attribute data includes, for example, the functional specifications of the system equipment 41 of power system 4, system configuration information showing the connection relationships of the system equipment 41, location information showing the geographical location of the system equipment 41, planning information showing the planned future settings and control values ​​of the system equipment 41, and planning information showing the planned future system configuration. Control of the system equipment 41 refers to operations such as changing the winding ratio of a transformer by changing the tap position of the transformer to adjust the output voltage.

[0026] Power System 4 is a system that transmits electricity generated from power generation facilities to load facilities via power distribution facilities. Power System 4 consists of system equipment 41 that supplies, distributes, and consumes electricity, and measuring devices 40 that measure the state of each point in Power System 4. System equipment 41 includes at least power generation facilities, power distribution facilities, and load facilities. Power generation facilities are facilities that supply electricity and include, for example, power generation devices that generate electricity through thermal power generation, hydroelectric power generation, nuclear power generation, solar power generation, wind power generation, etc., as well as power generation control devices that control power generation devices, and storage batteries that store energy and supply electricity as needed. Power distribution facilities include various facilities related to power transmission and distribution and include, for example, overhead transmission lines, underground transmission lines, transformers, circuit breakers, disconnectors, and phase adjustment equipment. Load facilities are facilities that consume electricity and include, for example, household electrical appliances, industrial electrical appliances, and storage batteries. The measuring device 40 is a device that measures measurement data at each point in the power system 4, and includes, for example, a voltage measuring device, a current measuring device, a power measuring device, and a frequency measuring device.

[0027] External attribute information provider 5 is a business that provides past observation data and future prediction data of external attribute data. External attribute information provider 5 is equipped with an external attribute data distribution device 50 for distributing external attribute data. External attribute data includes first information that indicates elements having a certain periodicity that is the same as or similar to the periodicity of electricity demand, and second information that indicates elements that have a small periodicity but may have a correlation with electricity demand. The first information may include, for example, meteorological information such as temperature, humidity, cloud cover, and solar radiation, and calendar information such as holidays. The second information may include event information such as sports competitions, mobile information such as GPS of cars and mobile terminals, consumer information such as factory operating status, and social activity information such as industrial dynamics. The factor data created from the attribute data includes data created from the first information (for example, factor data of factors having a certain periodicity) and data created from the second information (for example, factor data of factors with a periodicity smaller than a certain periodicity).

[0028] Next, we will explain the system management system 1 with reference to Figure 2.

[0029] The power system management system 1 consists of a measuring device 40, a power system equipment management device 22, an external attribute data distribution device 50, a power demand forecasting system 3, a power system control calculation device 21, and power system equipment 41.

[0030] The measuring device 40 measures measurement data at each point in the power system 4 and transmits the measurement data from each point to the system equipment management device 22.

[0031] The grid equipment management device 22 transmits grid data 3006A to the power demand forecasting system 3. The grid data 3006A includes grid observation data, which includes measurement data from each point in the power system 4 received from the measuring device 40, and grid attribute data held by the grid equipment management device 22.

[0032] The external attribute data distribution device 50 transmits the external attribute data to the power demand forecasting system 3. The external attribute data includes external attribute observation data 3007A, which is observation data of the external attribute data held by the external attribute data distribution device 50, and external attribute forecast data 3008A, which is forecast data of the external attribute data.

[0033] The power demand forecasting system 3 calculates a predicted power demand for each substation based on the grid data 3006A received from the grid equipment management device 22 and the external attribute data received from the external attribute data distribution device 50, and transmits the predicted power demand for each substation to the grid control calculation device 21.

[0034] The system control calculation device 21 generates control data for the system equipment 41 based on the predicted power demand values ​​for each substation received from the power demand forecasting device 31, and transmits the control data to the system equipment management device 22.

[0035] The system equipment management device 22 controls the system equipment 41 based on the control data for the system equipment 41 received from the system control calculation device 21.

[0036] In this embodiment, the power demand forecasting system 3 is a physical computer system (one or more physical computers), but it may also be a logical computer system based on a physical computer system (for example, a cloud computing system). The power demand forecasting system 3 consists of a demand forecasting data management device 30 and a power demand forecasting device 31.

[0037] The demand forecasting data management device 30 stores the grid data 3006A received from the grid equipment management device 22 and the external attribute observation data 3007A and external attribute forecast data 3008A included in the external attribute data received from the external attribute data distribution device 50, and transmits these data 3006A, 3007A, and 3008A to the power demand forecasting device 31.

[0038] The power demand forecasting device 31 calculates predicted power demand values ​​for each substation based on data 3006A, 3007A, and 3008A received from the demand forecasting data management device 30, and transmits these predicted power demand values ​​for each substation to the system control calculation device 21.

[0039] Furthermore, in the power demand forecasting device 31, the data extraction unit 3106 receives data 3006A, 3007A, and 3008A, extracts observed values ​​of the power demand for each forecast target included in the grid data 3006A, and outputs the observed values ​​of the power demand for each forecast target used to update the forecasting model as update forecast target data 3110A. The data extraction unit 3106 also extracts factor data of predetermined factors from the external attribute observation data 3007A and the external attribute forecast data 3008A, and outputs the factor data used to update the forecasting model as update factor data 3111A and the factor data used to calculate the forecast values ​​as forecast factor data 3112A. The update factor data 3111A includes observed values ​​of one or more factors correlated with the update forecast target data 3110A. The forecast factor data 3112A includes forecast values ​​of one or more factors for a predetermined date and time. The model update control unit 3107 receives the update target prediction data 3110A and update factor data 3111A from the data extraction unit 3106, and the model management data 3115A from the model management unit 3108. The model update control unit 3107 determines whether each prediction model (specifically, the prediction model for each substation) needs to be updated and outputs model update control data 3114A that records the determination result for each prediction model. The model management unit 3108 receives the model update control data 3114A, the update target prediction data 3110A, and the update factor data 3111A. The model management unit 3108 updates the model data 3116A, which is the data for each prediction model used to calculate the prediction value, and the model management data 3115A, which records the update history information for each prediction model, based on the received data 3114A, 3110A, and 3111A, and outputs the updated model management data 3115A and model data 3116A. The prediction value calculation unit 3109 receives the model data 3116A output by the model management unit 3108 and the prediction factor data 3112A output by the data extraction unit 3106, calculates the predicted value of the power demand for each prediction target at a predetermined date and time based on the received data 3115A and 3116A, and transmits the predicted value of the power demand to the grid control calculation device 21. (2) Internal configuration

[0040] Figure 3 shows the functional configuration of each device that makes up the power demand forecasting system 3.

[0041] The demand forecasting data management device 30 is composed of information processing equipment such as a personal computer or a server computer, and includes a CPU 3001, an input device 3002, an output device 3003, a communication device 3004, and a storage device 3005 that comprehensively control the operation of the demand forecasting data management device 30. The CPU 3001 is connected to the input device 3002, the output device 3003, the communication device 3004, and the storage device 3005.

[0042] The input device 3002 consists of a keyboard or mouse, and the output device 3003 consists of a display or printer. The communication device 3004 is equipped with a NIC for connecting to a wireless LAN or a wired LAN. The storage device 3005 may include memory such as RAM (Random Access Memory) and ROM (Read Only Memory).

[0043] The storage device 3005 includes a system data storage area 3006, an external attribute observation data storage area 3007, and an external attribute prediction data storage area 3008.

[0044] The system data storage area 3006 stores system data 3006A. System data 3006A is data (e.g., a database) that includes system observation data, which includes data such as observed power demand values ​​for each substation, and system attribute data, which includes attribute data such as system configuration information indicating the connection relationships of the system equipment 41.

[0045] The external attribute observation data storage area 3007 stores the external attribute observation data 3007A. The external attribute observation data 3007A is data (e.g., a database) that includes observation data such as meteorological observation data among external attribute data such as meteorological data.

[0046] The external attribute prediction data storage area 3008 stores the external attribute prediction data 3008A. The external attribute prediction data 3008A is data (e.g., a database) that includes prediction data such as weather forecast data, which is part of the external attribute data such as weather data.

[0047] The power demand forecasting device 31 is composed of information processing equipment such as a personal computer or a server computer, and includes a CPU 3101 that comprehensively controls the operation of the power demand forecasting device 31, an input device 3102, an output device 3103, a communication device 3104, and a storage device 3105. The CPU 3101 is connected to the input device 3102, the output device 3103, the communication device 3104, and the storage device 3105. The CPU 3101 (and 3001) is an example of a processor. The communication device 3104 (and 3004) is an example of an interface device.

[0048] The input device 3102, output device 3103, communication device 3104, and storage device 3105 may have the same configuration as the input device 3002, output device 3003, communication device 3004, and storage device 3005.

[0049] The storage device 3105 stores computer programs such as the data extraction unit 3106, the model update control unit 3107, the model management unit 3108, and the predicted value calculation unit 3109. These computer programs are executed by the CPU 3101.

[0050] The data extraction unit 3106 extracts observed power demand values ​​for a given substation over a given past period and a given factor for a given substation from the system data 3006A received from the system equipment management device 22 and the external attribute data received from the external attribute data distribution device 50. The data extraction unit 3106 generates update prediction target data 3110A, update factor data 3111A, and prediction factor data 3112A for calculating prediction values, which are used to update the prediction model.

[0051] The model update control unit 3107 generates classified update data 3113A from the update prediction target data 3110A and update factor data 3111A generated by the data extraction unit 3106, and the model management data 3115A, which is the update history information for each prediction model. The classified update data 3113A is data in which each data (sample data) in the update data (data that combines the update prediction target data 3110A and update factor data 3111A into one) is assigned a classification to either the first sample group or the second sample group. Based on the classified update data 3113A, the model update control unit 3107 determines whether each prediction model needs to be updated and generates model update control data 3114A that records the determination result.

[0052] The model management unit 3108 updates the model data 3116A and model management data 3115A using the model update control data 3114A generated by the model update control unit 3107 and the update prediction target data 3110A and update factor data 3111A generated by the data extraction unit 3106 (generating the updated model data 3116A and model management data 3115A).

[0053] The prediction value calculation unit 3109 calculates the predicted power demand for each substation at a predetermined date and time from the model data 3116A generated by the model management unit 3108 and the prediction factor data 3112A generated by the data extraction unit 3106, and transmits the predicted power demand for each substation to the system control calculation device 21.

[0054] The storage device 3105 also stores the update prediction target data storage area 3110, the update factor data storage area 3111, the prediction factor data storage area 3112, the classification update data storage area 3113, the model update control data storage area 3114, the model management data storage area 3115, and the model data storage area 3116.

[0055] The update prediction target data storage area 3110 stores the update prediction target data 3110A. The update prediction target data 3110A includes prediction target sample data for each substation (for each prediction model) used for updating the prediction model. For each substation (prediction model), the prediction target sample data is data obtained by dividing the observed power demand data of the substation for a predetermined arbitrary time range into predetermined time units, and includes the identifier of the prediction model corresponding to the substation and a value representing the observation date and time (e.g., a database). For example, the update prediction target data 3110A is data obtained by dividing the power demand data for the past year of each substation into daily units, and assigning an identifier to identify the substation and the observation date and time to each data unit.

[0056] The update factor data storage area 3111 stores the update factor data 3111A. The update factor data 3111A is data (e.g., a database) containing the factor values ​​of one or more factors corresponding to the update prediction target data 3110A (each prediction target sample data used for updating the prediction model).

[0057] The prediction factor data storage area 3112 stores the prediction factor data 3112A. The prediction factor data 3112A is data (e.g., a database) containing one or more factors for any predetermined date and time, used to calculate the predicted value of the power demand for each substation.

[0058] The classified update data storage area 3113 stores the classified update data 3113A. The classified update data 3113A is data (for example, a database) in which each sample data of the update data (data that combines the corresponding data of the update prediction target data 3110A and the update factor data 3111A) is assigned a classification (a classification of the sample, which is either the classification of the first sample group or the second sample group) based on the observation date and time of the sample represented by the sample data.

[0059] The model update control data storage area 3114 stores model update control data 3114A. Model update control data 3114A is data (e.g., a database) that includes update control values ​​indicating whether each prediction model needs to be updated.

[0060] The model management data storage area 3115 stores model management data 3115A. Model management data 3115A is data (e.g., a database) that includes information about the samples used for updates, such as the last update date and time of each prediction model and the observation date and time range of the samples.

[0061] The model data storage area 3116 stores model data 3116A. Model data 3116A is data (e.g., a database) that includes data (e.g., formulas and parameters) for each prediction model obtained through the identification of prediction models. (3) Details of each processing unit (program)

[0062] Hereafter, the details of each processing unit of the power demand forecasting device 31 will be explained with reference to the diagrams. (3-1) Data extraction unit 3106

[0063] Figure 4 shows the configuration of the data extraction unit 3106.

[0064] The data extraction unit 3106 consists of an update data extraction unit 31061 and a prediction data extraction unit 31062.

[0065] The update data extraction unit 31061 extracts observed power demand values ​​for each substation within a predetermined time range (e.g., one day) from the system data 3006A, and divides the extracted observed values ​​into predetermined time granularities (e.g., one hour). For each substation (prediction model), the update data extraction unit 31061 generates prediction target sample data, which is data in which the identifier of the prediction model and the observation date and time are added to the divided data (e.g., data including hourly observed power demand values ​​for one day). The update data extraction unit 31061 generates update prediction target data 3110A, which is data containing one or more prediction target sample data. The update data extraction unit 31061 may also use values ​​obtained by processing the observed power demand values ​​extracted from the system data 3006A, such as standardization or normalization, to generate the update prediction target data 3110A.

[0066] Figure 7 shows the structure of the update target data 3110A. Specifically, for example, the update target data 3110A is a table, with rows (entries) for each pair of prediction model and observation date and time. Each row contains data such as model ID 3110A1, which represents the identifier of the prediction model (model ID), sample observation date and time 3110A2, which represents the observation date and time of the target sample, and power demand observation value 3110A3, which represents the observed power demand value for each point in time within a predetermined time range. Therefore, in the case of Figure 7, each row contains one target sample data for prediction, which is divided into daily time granularity values ​​for the observed power demand. The data contained in the first row is the target sample data for prediction used to update the prediction model of model ID "M0001", and the observed power demand values ​​for the substation corresponding to model ID "M0001" on December 10, 2020 are "0.9" at 0:00 and "0.8" at 0:05. In the example in Figure 7, for simplicity, the unit of the observation date and time is shown as a date, but it is not limited to a date; for example, the unit of the observation date and time may include more detailed units such as hours, minutes, and seconds.

[0067] Next, the update data extraction unit 31061 extracts observed values ​​of one or more predetermined factors within a predetermined time range from the systematic data 3006A and the external attribute observation data 3007A, and divides them at a predetermined time granularity to correspond to each prediction target sample of the update prediction target data 3110A. The update data extraction unit 31061 generates factor sample data, which is data to which the identifier of the prediction model and the observation date and time are the same as those of each prediction target sample data, and generates update factor data 3111A, which is data containing one or more factor sample data. The update data extraction unit 31061 may also use values ​​obtained by processing the observed values ​​of the factors extracted from the systematic data 3006A and the external attribute observation data 3007A, such as standardization or normalization, to generate the update factor data 3111A.

[0068] Figure 8 shows the structure of the update factor data 3111A. Specifically, for example, the update factor data 3111A is a table, with a row (entry) for each pair of prediction model and observation date and time. Each row contains data such as model ID 3111A1 representing the identifier of the prediction model, sample observation date and time 3111A2 representing the observation date and time of the factor sample, and factor observation value 3111A3 containing the observed value of each factor. Therefore, in the case of Figure 8, each row contains one factor sample data divided into daily time granularity for the observed value of each factor. The data in the first row is a factor sample with model ID "M0001" and sample observation date and time "2020-12-10", and is the factor sample data corresponding to the prediction target sample data shown in the first row of Figure 7. The observed values ​​for each factor are "0.4" for factor A and "0.1" for factor B. In the example in Figure 8, for simplicity, the unit of the observation date and time of the factor sample is shown as a date, but it is not limited to a date. For example, the unit of the observation date and time of the factor sample may include more detailed units such as hours, minutes, and seconds.

[0069] Next, the prediction data extraction unit 31062 extracts predicted values ​​for one or more predetermined factors for a predetermined target date from the lineage data 3006A and the external attribute prediction data 3008A for each prediction model. The prediction data extraction unit 31062 generates prediction factor sample data by adding the prediction model identifier and the target date and time to the predicted values ​​of the extracted factors, and generates prediction factor data 3112A which is data containing one or more prediction factor samples. The prediction data extraction unit 31062 may also use values ​​obtained by processing the predicted values ​​of the factors extracted from the lineage data 3006A and the external attribute prediction data 3008A, such as standardization or normalization, to generate the prediction factor data 3112A.

[0070] Figure 9 shows a conceptual diagram of the prediction factor data 3112A. Specifically, for example, the prediction factor data 3112A is a table, with rows (entries) for each pair of prediction model and observation date and time. Each row contains data such as Model ID 3112A1, which represents the identifier of the prediction model (Model ID), Prediction target date and time 3112A2, which represents the prediction target date and time, and Factor prediction value 3112A3, which contains the predicted value of each factor. Therefore, in the case of Figure 9, each row is the prediction factor sample data for a predetermined prediction target date and time for each prediction model. The first row is the prediction factor sample data used to calculate the predicted value by the prediction model with Model ID "M0001," and the predicted values ​​for each factor on the prediction target date, December 11, 2020, are "0.3" for Factor A and "0.2" for Factor B. In the example of Figure 9, for simplicity, the unit of the prediction target date and time in the prediction factor sample is a date, but it is not limited to a date; for example, the unit of the prediction target date and time in the prediction factor sample may include more detailed units such as hours, minutes, and seconds. (3-2) Model update control unit 3107

[0071] Figure 5 shows the configuration of the model update control unit 3107.

[0072] The model update control unit 3107 consists of an update data classification unit 31071, a difference degree summation unit 31072, and a model update determination unit 31073.

[0073] First, the update data classification unit 31071 generates update data by combining the data of each factor sample in the corresponding update factor data 3111A for each prediction target sample in the update prediction target data 3110A. Next, the update data classification unit 31071 generates classified update data 3113A by assigning a classification to each sample data in the update data as either the first sample group or the second sample group, based on the sample information used for the previous identification of each prediction model in the model management data 3115A and the observation date and time information of each sample in the update data. The "first sample group" means the samples in the sample group used for the previous identification of the prediction model, and the "second sample group" means the samples in the sample group newly acquired after the date and time of the previous identification of the prediction model.

[0074] Figure 10 shows the structure of the classified update data 3113A. Specifically, for example, the classified update data 3113A is a table, with rows (entries) for each pair of prediction model and observation date and time. Each row contains data such as model ID 3113A1 representing the identifier of the prediction model, classification 3113A2 representing the classification of the sample data, sample observation date and time 3113A3 representing the observation date and time of the sample, factor observation value 3113A4 containing the observed values ​​of each factor, and power demand observation value 3113A5 containing the observed values ​​of power demand at each time. Note that a value of "1" in classification 3113A2 means the first sample group, and a value of "2" in classification 3113A2 means the second sample group. Factor observation value 3113A4 may be the same as the factor observation value 3111A3 mentioned above, and power demand observation value 3113A5 may be the same as the power demand observation value 3110A3 mentioned above.

[0075] Figure 11 shows the structure of the model management data 3115A. Specifically, for example, the model management data 3115A is a table with a row (entry) for each prediction model. Each row contains data such as the model ID 3115A1, which represents the identifier of the prediction model; the identification date and time 3115A2, which represents the date and time the prediction model was identified; and the latest sample used date and time 3115A3, which represents the observation date and time of the latest sample used for identification.

[0076] Therefore, in the example in Figure 10, each row is a sample formed by combining each prediction target sample of the update prediction target data 3110A and each factor sample of the update factor data 3111A. The first row of the update data in Figure 10 contains sample data used to update the prediction model of model ID "M0001", and the sample observation date and time is "2020-12-10". The update data classification unit 31071 refers to model ID 3115A1 "M0001" and the latest used sample date and time 3115A3 "2020-12-9" in the first row of the model management data 3115A in Figure 11. The sample observation date and time 3113A3 in the first row of the update data in Figure 10 is after the date and time represented by the latest used sample date and time 3115A3. Therefore, the update data classification unit 31071 classifies the sample data in the first row of update data 3113A in Figure 10 into the second sample group, and sets the value of classification 3113A2 for that row to "2". Also, in the second row of update data in Figure 10, the sample observation date and time 3113A3 for model ID "M0001" is "2020-12-09", which is not after the latest used sample date and time 3115A3 "2020-12-09" for model ID "M0001". Therefore, the update data classification unit 31071 classifies the sample data in the second row of update data 3113A in Figure 10 into the first sample group, and sets the value of classification 3113A2 for that row to "1".

[0077] The difference sum calculation unit 31072 calculates the difference degree, which is an index value indicating the difference between the samples classified into the first sample group and the samples classified into the second sample group for each prediction model of the classified update data 3113A. The difference degree is an index value indicating the difference between the two sample groups. In addition, representative values ​​of the index values ​​indicating the differences between individual samples in the first and second sample groups may be used to calculate the difference degree. Representative values ​​of the index values ​​indicating the differences between individual samples include the mean, maximum, and minimum values ​​of the index values ​​indicating the differences between individual samples, the mean, maximum, and minimum values ​​of the index values ​​indicating the differences between the representative sample and individual samples, and the index values ​​indicating the differences between representative samples and representative samples. A representative sample is, for example, the mean sample obtained by averaging the values ​​of each observation of all samples that make up each sample group, or the latest sample with the most recent sample observation date and time. In addition, known methods such as distance calculation methods and similarity calculation methods may be applied to calculate the index values ​​indicating the differences between individual samples. Publicly known methods include, for example, methods for calculating distances in sample space such as the Euclidean distance and the Manhattan distance, and methods for calculating the similarity between samples such as the inner product.

[0078] Using the classified update data 3113A shown in Figure 10, we will explain an example of calculating the degree of difference using the Euclidean distance between the mean sample of the first sample group and the latest sample of the second sample group. For simplicity, we will only calculate the degree of difference for model ID "M0001", and assume that the second sample group consists only of the sample in the first row, and the first sample group consists only of the samples in the second and third rows. Furthermore, we assume that the factor observations 3113A4 consist only of the observations of "Factor A", "Factor B", and "Factor C", and that the power demand observations 3113A5 consist only of the observations at "0:00", "0:05", and "0:10".

[0079] The mean sample of the first sample group is the average of the values ​​in each column of the factor observations 3113A4 and the power demand observations 3113A5 of the samples that make up the first sample group. For example, the mean value of the column for factor A is "(0.4 + 0.3) / 2 = 0.35", so the value of factor A in the mean sample of the first sample group is "0.35". Similarly, by calculating the mean of each column, the values ​​in each column of the mean sample are obtained as "0.35, 0.15, 0.75, 0.95, 0.9, 0.9" (the composition is "Factor A", "Factor B", "Factor C", "0:00", "0:05", "0:10"). The Euclidean distance between the mean sample and the sample of the second sample group is the square of the mean of the squared differences in the values ​​of each column of the two samples. The difference between the values ​​in each column of the mean sample "0.35, 0.15, 0.75, 0.95, 0.9, 0.9" and the second sample group sample "0.4, 0.1, 0.9, 0.9, 0.8, 0.7" is "-0.05, 0.05, -0.15, 0.05, 0.1, 0.2", and the square of the mean of the squared differences in the values ​​in each column is "0.12". As described above, the degree of difference between the first sample group, which consists of the samples in the second and third rows of Figure 10, and the second sample group, which consists of the samples in the first row of Figure 10, is "0.12".

[0080] The model update determination unit 31073 determines whether an update is necessary based on the comparison result between the degree of difference of each prediction model and a predetermined threshold, and generates model update control data 3114A indicating the determination result. The determination of whether an update is necessary only requires determining whether the degree of difference is greater than or equal to the threshold. Prediction models with a degree of difference greater than or equal to the threshold are determined to require updating, and prediction models with a degree of difference less than the threshold are determined not to require updating. The threshold values ​​for each prediction model are predetermined in advance, such as through input from the information input / output terminal 20.

[0081] Figure 12 shows the structure of the model update control data 3114A. Specifically, for example, the model update control data 3114A is a table with a row (entry) for each prediction model. Each row contains data such as the model ID 3114A1 representing the identifier of the prediction model, the difference degree 3114A2 representing the degree of difference, the threshold 3114A3 representing the threshold for the degree of difference, and the judgment 3114A4 representing the result of the determination of whether or not the prediction model needs to be updated. In the case of Figure 12, according to the first row, for the prediction model with model ID "M0001", the difference degree is "0.12", which is greater than or equal to the threshold "0.1". Therefore, it is determined that the prediction model needs to be updated, and the value of judgment 3114A4 is "necessary". Also, according to the second row, for the prediction model with model ID "M0002", the difference degree is "0.3", which is less than the threshold "0.6". Therefore, the prediction model is determined to not require updating, and the value for determination 3114A4 becomes "not required". (3-3) Model Management Department 3108

[0082] Figure 6 shows the configuration of the model management unit 3108.

[0083] The model management unit 3108 consists of a model identification unit 31081 and a model storage unit 31082.

[0084] First, the model identification unit 31081 identifies the prediction models that have been determined to require model updating based on the model update control data 3114A, using the sample of each prediction model included in the update prediction target data 3110A and the update factor data 3111A.

[0085] Therefore, according to the model update control data 3114A illustrated in Figure 12, the prediction model for model ID "M0001" and the prediction model for model ID "M0003" are identified because judgment 3114A4 is "required". The prediction model for model ID "M0002" is not identified because judgment 3114A4 is "unnecessary". Furthermore, for example, the prediction model for model ID "M0001" is identified using the prediction target sample data for model ID "M0001" in the update prediction target data 3110A in Figure 8 and the factor sample data for model ID "M0001" in the update factor data 3111A in Figure 9.

[0086] Next, the model storage unit 31082 receives data for each prediction model identified by the model identification unit 31081 and replaces (updates) the existing model data 3116A with model data 3116A, such as the parameters of the identified prediction model. Furthermore, based on the update prediction target data 3110A and update factor data 3111A, the model storage unit 31082 replaces (updates) the existing model management data 3115A with model management data 3115A, such as the observation date and time of the sample used to identify the identified prediction model and the identification date and time of the prediction model.

[0087] Figure 13 shows the structure of model data 3116A. Specifically, for example, model data 3116A is a table with a row (entry) for each prediction model. Each row contains data such as model ID 3116A1 representing the identifier of the prediction model, formula 3116A2 representing the formula of the prediction model (an example of the model configuration), and parameter 3116A3 containing the parameter values ​​for each parameter item of the formula. Therefore, in the example in Figure 10, the model storage unit 31082 receives data for the prediction models with model ID "M0001" and model ID "M0003," and rewrites the values ​​of formula 3116A2 and parameter 3116A3 for model ID "M0001" and model ID "M0003" in model data 3116A in Figure 13. Furthermore, the model storage unit 31082 rewrites the identification date and time 3115A2 and the latest used sample date and time 3115A3 values ​​for model ID "M0001" and model ID "M0003" in the model management data 3115A shown in Figure 11. (3-4) Prediction value calculation unit 3109

[0088] The prediction value calculation unit 3109 receives the model data 3116A and prediction factor data 3112A updated by the model management unit 3108, inputs the factor prediction values ​​3112A3 from the prediction factor data 3112A into the prediction model represented by the model data 3116A, calculates the predicted power demand for a given date and time for the substation corresponding to that prediction model, and transmits the predicted power demand for each substation to the system control calculation device 21.

[0089] With this, the power demand forecasting process of the power demand forecasting device 31 is completed.

[0090] Next, the system control calculation device 21 generates optimal voltage settings at each point in the power system at a predetermined time in the future, based on the predicted power demand values ​​received from each substation. It then generates control instructions for each system equipment 41 to achieve these optimal voltage settings and transmits control data, including the control instructions for each system equipment 41, to the system equipment management device 22. The optimal voltage setting is, for example, the voltage that maximizes the reduction in transmission losses across the entire system, or a voltage that suppresses failures of each piece of equipment in the system equipment 41. The control instructions for each system equipment 41 are, for example, tap switching of transformers or increasing or suppressing the reactive power output of phase-shifting equipment.

[0091] Next, the system equipment management device 22 operates the system equipment 41 based on the control data received from the system equipment 41, for example, by changing the tap of a transformer from tap 1 to tap 2. The system equipment 41 changes its operation as a result of the operation, and physical changes occur, such as the output voltage of the transformer becoming higher than before the change. (4) Flowchart

[0092] Figure 14 shows the processing procedure for power demand forecasting. This process begins when the power demand forecasting device 31 receives an input operation from the power operator 2, and the power demand forecasting device 31 executes the processes from S101 to S106.

[0093] In practice, processing is performed based on various computer programs stored in the CPU 3101 and storage device 3105 of the power demand forecasting device 31. For the sake of explanation, the processing will be described as being handled by various computer programs of the demand type management device.

[0094] First, the data extraction unit 3106 extracts observed values ​​of power demand for a predetermined substation from the system data 3006A and generates update prediction target data 3110A. Furthermore, the data extraction unit 3106 extracts observed and predicted values ​​of predetermined factors from the system data 3006A, external attribute observation data 3007A, and external attribute prediction data 3008A and generates update factor data 3111A and prediction factor data 3112A (S101).

[0095] Next, the model update control unit 3107 determines whether or not to update each prediction model that predicts the power demand of each substation based on the update target prediction data 3110A, update factor data 3111A, and model management data 3115A, and generates model update control data 3114A showing the determination result (S102).

[0096] Next, the model management unit 3108 determines whether there are any prediction models to be updated based on the determination 3114A4 for each prediction model in the model update control data 3114A (S103). If the determination result in S103 is true (S103:Yes), the model management unit 3108 identifies each prediction model using the update prediction target data 3110A and update factor data 3111A for the prediction models to be updated, and updates the model management data 3115A and model data 3116A (S104).

[0097] Furthermore, the model management unit 3108 transmits the model data 3116A to the prediction value calculation unit 3109 (S105).

[0098] Next, the prediction value calculation unit 3109 inputs the predicted values ​​of the factors in the prediction factor data 3112A into the data of each prediction model in the model data 3116A, and calculates the predicted value of the power demand for each substation (S106).

[0099] With the above processing complete, the power demand forecasting process in this embodiment is finished. (5) Effects of this embodiment

[0100] Figures 15A and 15B illustrate the concept of determining whether an update is necessary based on the degree of difference. Figure 16A shows the model update timing and update processing load for a comparative example. Figure 16B shows the model update timing and update processing load according to this embodiment.

[0101] In Figures 15A and 15B, the plots in the graphs with factors on the horizontal axis and electricity demand on the vertical axis represent sample data constituting either the first or second sample group of a prediction model. In this paragraph, "sample data" refers to data containing the factor observation value 3113A4 and the electricity demand observation value 3113A5 in a single entry. The determination of whether or not a model update is necessary differs depending on the relative positions of the sample data in the first and second sample groups. Here, we assume that there is only one sample data set constituting the second sample group.

[0102] According to Figure 15A, the relative positions of the first and second sample groups on the graph are close, meaning the degree of difference is small. This indicates that there is little change in the response characteristics between the first sample group, which was used in the previous update of the prediction model, and the second sample group, which was obtained after the previous update of the prediction model. Therefore, the result is that an update of the prediction model is unnecessary.

[0103] On the other hand, as shown in Figure 15B, the relative positions of the first and second sample groups on the graph are far apart, meaning there is a large degree of difference. This indicates that there is a significant change in the response characteristics between the first sample group, which was used in the previous update of the prediction model, and the second sample group, which was obtained after the previous update of the prediction model. Therefore, the result indicates that an update of the prediction model is necessary to keep up with the change in response characteristics.

[0104] In Figures 16A and 16B, the horizontal axis of the graph indicates the timing of the prediction model update, the vertical axis indicates the processing load of the prediction model update process, and the dotted line indicates the upper limit of the processing load that can be handled within a given time.

[0105] As shown in Figure 16A, all prediction models are updated at each update timing. Because all prediction models are constantly being updated, the processing load exceeds the processing load that can be handled within the given time, regardless of the update timing, and the update of all prediction models, including those with changed response characteristics, cannot be completed within the given time.

[0106] On the other hand, as shown in Figure 16B, prediction models are updated only for those that detect changes in the response characteristics of factors and demand (prediction models where the degree of difference is above a threshold). As a result, it becomes possible to complete the updating of prediction models with changed response characteristics within a given time.

[0107] Furthermore, we will explain why the model update method according to this embodiment can suppress the degradation of prediction accuracy.

[0108] A prediction model that needs updating is one that has encountered sample data showing a different trend within the normal range, rather than samples showing abnormal values; in other words, a prediction model that indicates a possible change in response characteristics. As shown in Figure 17A, if the second sample group contains sample data showing a change in the trend of response characteristics within the normal range, updating prediction model 1700a to the new prediction model 1700b allows the model to track the response characteristics, thus improving prediction accuracy.

[0109] On the other hand, a prediction model that does not require updating is a model for which sample data showing a similar trend to the first sample group, which is the sample data observed before the update, or sample data showing outliers, has been obtained. If the former sample data (sample data showing a similar trend to the first sample group, which is the sample observed before the update) is obtained, as shown in Figure 17B, updating prediction model 1700a to the new prediction model 1701b does not substantially change the prediction model, and therefore does not improve prediction accuracy. If the latter sample data (sample data showing outliers) is obtained, as shown in Figure 17C, updating prediction model 1700a to the new prediction model 1702b causes the prediction model to follow the outliers, and therefore the prediction accuracy deteriorates.

[0110] As described above, prediction models that are likely to improve prediction accuracy are determined to require updating (i.e., these prediction models are updated), while prediction models that are likely to remain unchanged or deteriorate in prediction accuracy are determined not to require updating (i.e., these prediction models are not updated). As a result, the deterioration of prediction accuracy is suppressed.

[0111] Furthermore, by completing the update of prediction models whose response characteristics have changed within a time frame and suppressing the deterioration of prediction accuracy of numerous prediction models, it is possible to calculate highly accurate prediction values ​​for the power demand of numerous substations. This enables the control of a wide range of system equipment 41 in the power system 4 based on the prediction values ​​for the power demand of numerous substations, thereby enabling efficient operation of system equipment 41, such as reducing transmission losses in the power system 4.

[0112] The above explanation can be summarized as follows, for example.

[0113] The forecasting system (for example, power demand forecasting system 3) comprises an interface device (for example, communication device 3004) that accepts input of forecast target observation data and factor observation data, a storage device (for example, storage device 3005) that stores the forecast target observation data and factor observation data, and a processor (for example, CPU 3101) connected to the interface device and the storage device that updates one or more forecasting models corresponding to one or more forecast targets, which are determined to require updating.

[0114] For each prediction target, the prediction model (for example, the model represented by equation 3116A2) takes factor prediction values ​​(for example, factor prediction value 3112A3), which are data containing predicted values ​​for future periods for each of one or more factors that may influence one or more prediction targets, as input, and outputs predicted values ​​of observed values ​​to be obtained for the prediction target in that future period. Factor prediction values ​​3112A3 may be data extracted from external attribute prediction data 3008A.

[0115] The prediction target observation data includes time-series data of past observations obtained for each of one or more prediction targets. The prediction target observation data may be, for example, the update prediction target data 3110A, or data extracted from the external attribute observation data 3007A.

[0116] Factor observation data includes time-series data of past observations obtained for each of one or more factors. Factor observation data may be, for example, update factor data 3111A, or data extracted from external attribute observation data 3007A.

[0117] The processor generates update data from the prediction target observation data and factor observation data. The update data includes sample data for each of the one or more prediction targets for each observation period. For each prediction target, the sample data for each observation period includes one or more observations obtained for that prediction target during that observation period (e.g., power demand observation value 3113A5) and observations obtained for each of the one or more factors during that observation period (e.g., factor observation value 3113A4).

[0118] The processor calculates the degree of difference between the first and second sample sets for each prediction target. For each prediction target, the first sample set consists of one or more sample data used to identify the prediction model corresponding to that target (for example, sample data where the time represented by sample observation date and time 3113A3 is the same as or earlier than the time represented by the most recent sample used date and time 3115A3). The second sample set consists of one or more sample data that have not yet been used to identify the prediction model corresponding to that target (for example, sample data where the time represented by sample observation date and time 3113A3 is later than the time represented by the most recent sample used date and time 3115A3).

[0119] The processor performs an update requirement determination for each prediction target, which is to determine whether the degree of difference is above a threshold, and outputs the result of the update requirement determination. Prediction models for which the update requirement determination result is true are prediction models that need to be updated. Prediction models for which the update requirement determination result is false are prediction models that do not need to be updated.

[0120] Prediction targets whose degree of difference exceeds a threshold are those whose response characteristics have changed. For such targets, training of the prediction model is unnecessary, as they can be identified from the relationship between the degree of difference between the first and second sample groups and the threshold. In other words, the prediction model for targets with changed response characteristics can be determined with low overhead.

[0121] Furthermore, if there are multiple prediction targets, the prediction models to be updated will be narrowed down to those that have been determined to require updating (the prediction models corresponding to the prediction targets whose response characteristics have changed), thus increasing the likelihood of completing the update process within the given time.

[0122] Furthermore, even if the second sample group contains sample data with outliers, the degree of difference between the first and second sample groups does not necessarily exceed a threshold (for example, in the second sample group, there may be more sample data without outliers than sample data with outliers). This reduces the possibility of unnecessarily updating the prediction model.

[0123] Furthermore, for each prediction target, the determination of whether an update is necessary may also be made by determining whether the degree of difference is greater than or equal to the first threshold but less than the second threshold (second threshold > first threshold).

[0124] Furthermore, for at least one of the one or more prediction targets, the processor may calculate a representative value of an index that shows the difference between each sample data from the first sample group and each sample data from the second sample group, and calculate the degree of difference based on the representative value. This is expected to result in a more appropriate degree of difference.

[0125] Furthermore, for at least one of the one or more prediction targets, the processor calculates the degree of difference based on the distance or similarity between the first and second sample groups. This is expected to result in a more appropriate degree of difference.

[0126] Furthermore, in the above-described embodiment, one or more factors include at least one of weather and energy supply from renewable energy sources, and each of the one or more forecast targets is the power demand of a substation. The power demand of a substation has a relationship specific to the power sector with weather (and / or renewable energy such as solar power generation) such as temperature and solar radiation, and the sample data includes observed values ​​(observed values ​​of power demand and observed values ​​of factors) that have such a relationship, so it is possible to identify forecast targets whose response characteristics have changed and update the forecast model corresponding to those forecast targets, that is, to realize model updates that follow changes in response characteristics. (6) Modifications (other embodiments) (6-1) Variations of the predicted subject

[0127] The prediction target may be any of the following, instead of electricity demand (e.g., electricity consumption): demand (e.g., consumption) of energy other than electricity (e.g., gas or water), energy output (e.g., power generation) such as solar or wind power generation, trading volume and price of energy traded on wholesale exchanges, communication volume measured by communication base stations, or location history of mobile objects such as automobiles. Time series data may include energy demand such as gas and heat, commodity demand, industrial data such as factory production volume, and meteorological data such as temperature and solar radiation. Since there can be numerous targets for time series data other than electricity demand values, such as factories and regions, this technology is applicable and makes it possible to make predictions while suppressing the degradation of prediction accuracy of multiple prediction models. (6-2) Variations of the predicted subject

[0128] The processor may calculate the similarity between prediction targets based on the observed data of the prediction targets, classify multiple prediction targets into a predetermined number of groups based on the similarity so that prediction targets with high similarity belong to the same group, and integrate the sample data of all prediction targets belonging to each group. In this case, for each of the one or more prediction targets in the update data, the sample data for each observation period may be the integrated sample data obtained for the group to which the prediction target belongs. Specifically, for example, it may be as follows.

[0129] The target of the prediction may be the combined power demand of multiple substations. Combination refers to the process of converting multiple data series into a single data series through statistical processing such as summing or averaging. One method for selecting the substations to be combined is to select substations that show similar tendencies. For example, one method for selecting substations that show similar tendencies is to cluster substations using features such as the average daily demand curve and the maximum daily power value for each substation, grouping substations with high similarity, and selecting substations classified into the same group as the target for integration. The clustering method may be a known method, such as hierarchical clustering such as Ward's method or non-hierarchical clustering such as k-means. This reduces the processing load by decreasing the number of prediction targets and mitigates the degradation of prediction accuracy due to the law of large numbers. (6-3) Modified form of data extraction unit 3106

[0130] The processor may generate factor prediction values ​​for each prediction target based on factor prediction data (for example, data extracted from external attribute prediction data 3008A) containing the predicted values ​​for one or more factors, and factor observation data. Specifically, for example, it may be done as follows:

[0131] The prediction data extraction unit 31062 of the data extraction unit 3106 may generate prediction factor data 3112A based on observed values ​​of predetermined factors extracted from grid data 3006A and external attribute observation data 3007A. Here, predetermined factors include the setting value of grid equipment 41 several hours ago, the power demand value several hours ago, the temperature several hours ago, etc. That is, the prediction data extraction unit 31062 extracts predicted values ​​of one or more predetermined factors for a predetermined target date for each prediction model from grid data 3006A, external attribute observation data 3007A, and external attribute prediction data 3008A, and forms a prediction factor sample by adding a prediction model identifier and the target date and time for prediction to the predicted values ​​of the extracted factors, thereby generating prediction factor data 3112A, which is data containing one or more prediction factor samples. Factors may include factors that have a time delay in affecting the target of prediction, such as the temperature several hours ago affecting power demand. When making predictions using factors that affect the target subject with a time delay, the observed values ​​of the factors are available at the time the predicted value of the target subject is calculated, and the predicted value of the target subject can be calculated based on these observed values. This reduces the impact of prediction errors on the predicted value of the factors, thereby improving prediction accuracy. (6-4) Modified form of data extraction unit 3106

[0132] For at least one of the one or more prediction targets, the processor may detect non-conforming sample data, which are sample data with abnormal values, from the update data, and perform corrective processing on the non-conforming sample data (for example, processing to remove the non-conforming sample data from the update data). Specifically, for example, this may be done as follows:

[0133] As shown in Figure 18, the data extraction unit 3106 may further include a non-conforming sample processing unit 31063. The non-conforming sample processing unit 31063 receives the update prediction target data 3110A and update factor data 3111A output by the update data extraction unit 31061 and detects non-conforming samples within the update prediction target data 3110A and update factor data 3111A. The non-conforming sample processing unit 31063 performs corrective processing, such as deleting the detected non-conforming samples within the update prediction target data 3110A and update factor data 3111A, and outputs the updated prediction target data 3110A and update factor data 3111A after the corrective processing has been applied. A non-conforming sample is a prediction target sample and factor sample that has an abnormal value compared to other samples. For example, it is a prediction target sample that has a power demand value that is N times (where N is an integer of 2 or more) larger than the average power demand value of other prediction target samples with the same prediction model identifier. One method for detecting non-compliant samples is to compare each sample with a predetermined threshold and identify samples with values ​​above the threshold as non-compliant. This allows for the avoidance of unnecessary model updates due to temporary anomalies in observational data and improves prediction accuracy. (6-5) Modified model of the model update determination unit 31073

[0134] For at least one of the one or more prediction targets, the processor may identify a threshold at which a performance metric value, which is a value based on at least one of the processing load for updating the prediction model corresponding to that prediction target and the prediction accuracy of that prediction model, is best performed, and set the identified threshold as the difference threshold. Specifically, for example, it may be as follows:

[0135] The model update determination unit 31073 may pre-calculate a threshold for the degree of difference using a parameter calculation method based on predetermined performance index values, and determine whether each prediction model needs to be updated based on the calculated threshold. The parameter calculation method may include a grid search method, where the results of several threshold performance index values ​​are calculated in advance, and the threshold that yields the best performance index value is calculated. Alternatively, known methods such as Bayesian optimization, which sequentially and probabilistically searches for and determines the threshold that is most likely to yield the best performance index value, may be used. Performance index values ​​are indicators that show the processing load of updating the prediction model and the prediction accuracy of the prediction model. Examples include the sum of the processing times required for updating each prediction model, or the average prediction error, which is the difference between the predicted and observed power demand values ​​calculated by each prediction model. This allows for accurate determination of whether the prediction model needs to be updated, thereby suppressing the deterioration of prediction accuracy. (6-6) Modified model of the model update determination unit 31073

[0136] For at least one of the one or more prediction targets, the processor may set a threshold for the degree of difference as a value obtained by multiplying the standard deviation of the distance between sample data in the first sample group by a predetermined multiple. Specifically, for example, it may be as follows:

[0137] The model update determination unit 31073 may determine whether each prediction model needs to be updated by calculating the standard deviation of the inter-sample distance within the first sample group of each prediction model and using a value obtained by multiplying the standard deviation by a predetermined multiple as a threshold. The standard deviation of the inter-sample distance within the first sample group is calculated by focusing on each sample in the first sample group, removing that sample from the first sample group to form the other sample group, calculating the distance between that sample and the other sample group for each sample, and then calculating the standard deviation of the data series of distances calculated for each sample. The predetermined multiple may be a value determined in advance by input from the information input / output terminal 20 or a value determined in advance by a parameter search method. As for the parameter calculation method, known methods such as a grid search method in which the results of performance index values ​​for several thresholds are calculated in advance and the threshold that yields the best performance index value among the calculated values ​​is calculated, or Bayesian optimization which sequentially and probabilistically searches for and determines a threshold that is likely to yield the best performance index value, may be used. The performance index value may be the same as the performance index value described in (6-5). This allows for more accurate determination of whether or not the prediction model needs to be updated, thereby suppressing the deterioration of prediction accuracy. (6-7) Modified model of the model update determination unit 31073

[0138] The processor may determine whether the re-evaluation start condition is met, which is that the number of predictive models determined to need updating exceeds the upper limit of models, which is the number of predictive models that can be updated within a predetermined time. If the re-evaluation start condition is met, the processor may output data representing the predictive models that have a relatively high degree of difference among the predictive models determined to need updating, which are less than or equal to the upper limit of models. Specifically, for example, it may be as follows.

[0139] The model update determination unit 31073 generates a first determination result based on the comparison of the degree of difference between the models and a threshold, and if the first determination result satisfies a predetermined re-determination start condition, it generates a second determination result by re-determining whether each model needs to be updated based on the first determination result and other update determination criteria, and generates model update control data 3114A showing the new second determination result. The re-determination start condition is, for example, the condition that the number of prediction models determined to need updating in the first determination result exceeds the number of prediction models that can be updated within the time. The number of prediction models that can be updated within the time is predetermined based on the processing load that can be processed within the time. Other update determination criteria are, for example, the criteria for determining that the number of prediction models that can be updated within the time, in order of the degree of difference among the prediction models determined to need updating in the first determination result, in descending order of degree of difference, need updating, and the remaining prediction models and prediction models that were determined not to need updating in the first determination result do not need updating. This makes it possible to avoid a significant degradation in prediction accuracy by prioritizing updates to prediction models that particularly need updating, even when the response characteristics of numerous prediction targets change. (6-8) Modified model of the model update determination unit 31073

[0140] The processor may determine, based on at least one of the factor prediction values ​​and factor prediction data containing the prediction values ​​of one or more factors, whether there are any unsuitable prediction targets, which are prediction targets for which the prediction values ​​output using the prediction model are expected not to meet the conditions, and whether the re-evaluation start condition, which indicates the presence of unsuitable prediction targets, is met. If the re-evaluation start condition is met, the processor may output data representing the prediction models, excluding the prediction models corresponding to the unsuitable prediction targets from the prediction models that have been determined to need updating, as the result of the update necessity determination. Specifically, for example, it may be as follows.

[0141] The model update determination unit 31073 generates a first determination result based on a comparison of the degree of model difference with a threshold, and may calculate a prediction result for determination to be used to determine a predetermined re-determination start condition, which will be described later. Based on the prediction result for determination, the model update determination unit 31073 determines whether the predetermined re-determination start condition is met, and if the re-determination start condition is met, it generates a second determination result by re-determining whether each model needs to be updated based on the first determination result and other update determination criteria, and generates model update control data 3114A showing the new second determination result. The prediction result for determination is, for example, a prediction result made by the model update determination unit 31073 based on the predicted values ​​of factors such as solar radiation included in the prediction factor data 3112A, which predicts whether the power demand of each substation will deviate from a predetermined range of values ​​due to the influence of the amount of solar power generation at a predetermined prediction target date and time. Specifically, for example, if the solar radiation value included in the prediction factor data 3112A is less than a predetermined value, it is predicted that the electricity demand for a predetermined target date and time will not deviate from the predetermined range; if it is greater than or equal to the predetermined value, it is predicted that the electricity demand for a predetermined target date and time will deviate from the predetermined range. The re-determination start condition is, for example, the condition that a judgment prediction result has been obtained indicating that there are substations where the electricity demand will not deviate from the predetermined range due to the influence of solar power generation, etc., at a predetermined target date and time. Other update judgment criteria include, for example, among the prediction models that were determined to require updating in the first judgment result, substations where, by referring to the judgment prediction result, it is predicted that the electricity demand will not deviate from the predetermined range due to the influence of solar power generation, etc., at a predetermined target date and time are deemed not to require updating, and the remaining prediction models and those determined to require updating in the first judgment result are deemed not to require updating. need The prediction model that was determined to be updated need This is the criterion for making such a determination. This makes it possible to avoid unnecessary updates to the prediction model, which would not lead to a deterioration in the control accuracy of the system equipment 41 even without updating, and thus reduce the processing load.

[0142] For at least one of the one or more prediction targets, the processor may, based on the degree of difference calculated for that prediction target, assign a sample weight to each sample data for that prediction target, which is the relative importance of that sample data compared to other sample data, and then identify a corresponding prediction model for that prediction target based on the assigned sample weight. Specifically, for example, this may be done as follows:

[0143] The model identification unit 31081 may set a sample weighting value for each sample when identifying a prediction model and perform the identification based on the sample weighting value. The sample weighting value is a value that indicates the relative importance of a sample compared to other samples. Identification based on the sample weighting value means identifying a prediction model that is particularly well-fitting to samples with large sample weighting values. Methods for setting the sample weighting value include a time-weighting method, which sets a larger sample weighting value for samples with more recent observation dates and times based on the observation date and time of the sample, and a difference-weighting method, which sets a larger sample weighting value for samples with more recent observation dates and times based on an index value indicating the degree of difference in the model update control data 3114A, if the index value indicating the degree of difference is greater than or equal to a predetermined value. Note that both the time-weighting method and the difference-weighting method may be used in combination to set the sample weighting value. Identification based on the sample weighting value set by the time-weighting method makes it possible to update the prediction model to one that follows the latest response characteristics, thereby suppressing the deterioration of prediction accuracy. Identification based on sample weights set using a difference-weighting method allows for updating the prediction model to one that more closely follows the latest response characteristics when there are significant changes in response characteristics, thereby suppressing the degradation of prediction accuracy. (6-10) Modified form of the prediction value calculation unit 3109

[0144] For at least one of the one or more prediction targets, the processor may calculate a predicted value for that target using an identified prediction model corresponding to that target. If certain conditions regarding the degree of difference are met, the processor may apply a correction process to the calculated predicted value and calculate a new predicted value. Specifically, for example, this may be done as follows:

[0145] The prediction value calculation unit 3109 calculates a predicted value of the power demand for a predetermined date and time for each prediction target based on each prediction model and factor data of the prediction factor data 3112A. If conditions are met, correction processing may be performed on the calculated predicted value of power demand. The conditions are based on an index value indicating the degree of difference, for example, the index value indicating the degree of difference of each prediction model in the model update control data 3114A is greater than or equal to a predetermined value. That is, the prediction value calculation unit 3109 receives the model management data 3115A, model data 3116A, and prediction factor data 3112A, inputs predetermined factor data corresponding to each prediction model in the prediction factor data 3112A into each prediction model, and calculates a predicted value of the power demand for a predetermined date and time for each prediction target. Next, the prediction value calculation unit 3109 performs correction processing on the calculated predicted value of power demand for the prediction models that meet the conditions to calculate a new predicted value, and transmits the new predicted value to the grid control calculation device 21. The correction process adjusts the predicted values ​​to track the current observed electricity demand. For example, it calculates the difference between the average value of predicted electricity demand within a predetermined range and the current observed electricity demand, and then adds this difference to the predicted electricity demand to calculate a new predicted value. This makes it possible to suppress the deterioration of prediction accuracy by fitting the predicted electricity demand to the current observed values, even in situations where the prediction accuracy of the prediction model may decrease due to rapid changes in response characteristics.

[0146] As a correction process, it is also possible to extract observed electricity demand data that show a similar trend to the most recent observed electricity demand data, ensuring that the time range of the predicted electricity demand data is included. Then, a new predicted value may be calculated by weighting the predicted value and the extracted observed electricity demand data showing a similar trend. This makes it possible to suppress the deterioration of prediction accuracy by fitting the predicted electricity demand data to past observed data showing similar trends, even in situations where the prediction accuracy of the prediction model may decrease due to abrupt changes in response characteristics. (6-11) Other Variations

[0147] Furthermore, in the power demand forecasting system 3 of this embodiment, the display unit is omitted for the sake of simplicity of explanation, but the calculation results of each processing unit and the intermediate results of each processing unit may be calculated as appropriate through an output device such as a display or printer.

[0148] Although one embodiment of the present invention and several modifications have been described above, these are merely illustrative examples for the purpose of explaining the present invention and are not intended to limit the scope of the present invention to these embodiments and modifications. The present invention can be carried out in various other forms.

[0149] For example, it is possible to combine any two or more of the above-mentioned variations.

[0150] Furthermore, the method for constructing the predictive model can be any of the following, for example: Linear models such as ridge regression, lasso regression, and elastic nets. • Tree models such as regression trees, random forests, and boosting trees. • Kernel methods such as support vector regression, kernel ridge regression, and Gaussian process regression. Nonlinear models such as neural networks including recurrent networks and Long Short-Term Memory. [Explanation of symbols]

[0151] 1 System Management System 3. Electricity Demand Forecasting System

Claims

1. An interface device that accepts input of prediction target observation data and factor observation data, A storage device in which the prediction target observation data and the factor observation data are stored, A processor connected to the interface device and the storage device, which updates one or more prediction models that are determined to require updating, among the one or more prediction models corresponding to one or more prediction targets. Equipped with, For each prediction target, the prediction model takes factor prediction values, which are data containing predicted values ​​for future periods for each of the one or more factors that may influence the one or more prediction targets, as input, and outputs predicted values ​​of observed values ​​obtained for the prediction target during that future period. The aforementioned prediction target observation data includes, for each of the one or more prediction targets, time-series data of past observation values ​​obtained for that prediction target. The factor observation data includes, for each of the one or more factors, time-series data of past observation values ​​obtained for that factor. The processor generates update data from the prediction target observation data and the factor observation data. The update data includes, for each of the one or more prediction targets, sample data for each observation period. For each prediction target, the sample data for each observation period includes one or more observed values ​​obtained for that prediction target during that observation period, and the observed values ​​obtained for each of the one or more factors during that observation period. The processor calculates the degree of difference between the first sample group and the second sample group for each prediction target. For each prediction target, The first sample group is one or more sample data used to identify the prediction model corresponding to the target of prediction. The second sample group is one or more sample data that have not yet been used to identify the corresponding prediction model for the target of prediction. The processor performs an update necessity determination for each prediction target, which is a determination of whether the degree of difference is above a threshold, and outputs a first determination result as a result of the update necessity determination. A predictive model in which the above update necessity determination result is true is a predictive model that indicates an update is necessary. A predictive model for which the above update necessity determination result is false is a predictive model that does not require updating. The aforementioned processor, As a prediction result for determining the re-determination start condition, the system outputs a prediction result indicating whether the predicted value output from the current prediction model, which is input to factor prediction values ​​that include predicted values ​​for future periods for each of the one or more factors, deviates from a predetermined range of values. A determination is made, based on the prediction results for determination, to determine whether the re-determination start condition is met, which is that there are prediction targets that do not require updating and whose output prediction values ​​do not deviate from the predetermined range. If the re-determination start condition is met, a second determination result is generated as a result of performing the update necessity determination on each current prediction model again using the first determination result and other update determination criteria, and data representing the prediction models, excluding the current prediction models corresponding to prediction targets that do not require updating from the current prediction models that were determined to require updating because the degree of difference is greater than or equal to the threshold, is output as the result of the update necessity determination. The aforementioned other update determination criteria determine that, among the prediction models determined to require updating in the first determination result, prediction models whose predicted values ​​output for a predetermined target date and time do not deviate from the predetermined range of values, when referring to the prediction results for determination, do not require updating, while the remaining prediction models and prediction models determined to require updating in the first determination result require updating. Prediction system.

2. With respect to at least one of the one or more prediction targets, the processor: We calculate a representative value of the index that shows the difference between each sample data from the first sample group and each sample data from the second sample group. Based on the aforementioned representative values, the degree of difference is calculated. The prediction system according to claim 1.

3. For at least one of the one or more prediction targets, the processor identifies a threshold at which a performance indicator value, which is a value based on at least one of the processing load for updating the prediction model corresponding to that prediction target and the prediction accuracy of that prediction model, is best, and sets the identified threshold as the difference threshold. The prediction system according to claim 1.

4. For at least one of the one or more prediction targets, the processor calculates the degree of difference based on the distance or similarity between the first sample group and the second sample group. The prediction system according to claim 1.

5. For at least one of the one or more prediction targets, the processor excludes unsuitable sample data, which are sample data with abnormal values, from the update data. For at least one of the prediction targets, the abnormal value is either a value that is N times (where N is an integer of 2 or more) larger than the mean value of the sample values ​​of other prediction targets for that prediction target, or a value that is greater than or equal to a predetermined threshold, with respect to the prediction model of that prediction target. The prediction system according to claim 1.

6. The processor generates the factor prediction values ​​for each prediction target based on factor prediction data, which includes the predicted values ​​for each of the one or more factors, and the factor observation data. The prediction system according to claim 1.

7. With respect to at least one of the one or more prediction targets, the processor: Based on the degree of difference calculated for the target of prediction, a sample weight value, which represents the relative importance of the sample data relative to other sample data, is set for each sample data for the target of prediction. Based on the set sample weighting values, identify the corresponding prediction model for the target subject. The prediction system according to claim 1.

8. With respect to at least one of the one or more prediction targets, the processor: Using the corresponding identified prediction model for the target subject, the predicted value for the target subject is calculated. If the degree of difference in the prediction model exceeds a predetermined value, the calculated prediction value is corrected and a new prediction value is calculated. The aforementioned correction process is performed as follows: A process that calculates the difference between the average value of the predicted values ​​over a predetermined observation period and the current observed value, and adds the calculated difference to the calculated predicted value to obtain the new predicted value, or This process involves extracting observational values ​​that show a similar trend to the most recent observational value, ensuring that the time range of the predicted value is included, and then calculating a new predicted value by taking a weighted average of the predicted value and the extracted observational values ​​that show a similar trend. The prediction system according to claim 1.

9. The processor calculates the similarity between prediction targets based on the prediction target observation data, classifies the multiple prediction targets into a predetermined number of groups based on the similarity so that prediction targets with high similarity belong to the same group, and integrates the sample data of all prediction targets belonging to each group. In the aforementioned update data, for each of the one or more prediction targets, the sample data for each observation period is the integrated sample data obtained for the group to which the prediction target belongs. The prediction system according to claim 1.

10. For at least one of the one or more prediction targets, the processor sets a value obtained by multiplying the standard deviation of the distance between sample data in the first sample group by a predetermined multiple as the threshold for the degree of difference. The prediction system according to claim 1.

11. The aforementioned processor, A determination is made as to whether the re-evaluation start condition is met, which is that the number of prediction models that have been determined to need updating exceeds the upper limit of the number of prediction models that can be updated within a predetermined time. If the re-evaluation start condition is met, the data representing the prediction models with a relatively high degree of difference among the prediction models determined to require updating, which are less than or equal to the upper limit of the number of models, will be output as the result of the update necessity determination. The prediction system according to claim 1.

12. The aforementioned one or more factors include at least one of weather and the amount of energy supplied by renewable energy, Each of the one or more forecast targets mentioned above is the power demand of a substation. The prediction system according to claim 1.

13. In a prediction method that uses a prediction model, The computer generates update data from the prediction target observation data and factor observation data. The aforementioned prediction target observation data includes, for each of one or more prediction targets, time-series data of past observation values ​​obtained for said prediction target. The factor observation data includes, for each of the one or more factors that may influence the one or more prediction targets, time-series data of past observations obtained for that factor. The update data includes, for each of the one or more prediction targets, sample data for each observation period. For each prediction target, the sample data for each observation period includes one or more observed values ​​obtained for that prediction target during that observation period, and the observed values ​​obtained for each of the one or more factors during that observation period. The computer calculates the degree of difference between the first sample group and the second sample group for each prediction target. For each prediction target, The first sample group is one or more sample data used to identify the prediction model corresponding to the target of prediction. The second sample group is one or more sample data that have not yet been used to identify the corresponding prediction model for the target of prediction. For each prediction target, the prediction model takes factor prediction values, which are data containing predicted values ​​for each of the one or more factors for a future period, as input, and outputs predicted values ​​of observed values ​​obtained for the prediction target during that future period. The computer performs an update necessity determination for each prediction target, which is a determination of whether the degree of difference is above a threshold, and outputs a first determination result as a result of said update necessity determination. A predictive model in which the above update necessity determination result is true is a predictive model that indicates an update is necessary. A predictive model for which the above update necessity determination result is false is a predictive model that does not require updating. The computer updates the prediction model that it determines needs updating from among the one or more prediction models corresponding to each of the aforementioned prediction targets. Computers As a prediction result for determining the re-determination start condition, the system outputs a prediction result indicating whether the predicted value output from the current prediction model, which is input to factor prediction values ​​that include the predicted values ​​of each of the one or more factors, deviates from a predetermined range of values. A determination is made, based on the prediction results for determination, to determine whether the re-determination start condition is met, which is that there are prediction targets that do not require updating and whose output prediction values ​​do not deviate from the predetermined range. If the re-determination start condition is met, a second determination result is generated as a result of performing the update necessity determination on each current prediction model again using the first determination result and other update determination criteria, and data representing the prediction models, excluding the current prediction models corresponding to prediction targets that do not require updating from the current prediction models that were determined to require updating because the degree of difference is greater than or equal to the threshold, is output as the result of the update necessity determination. The aforementioned other update determination criteria determine that, among the prediction models determined to require updating in the first determination result, prediction models whose predicted values ​​output for a predetermined target date and time do not deviate from the predetermined range of values, when referring to the prediction results for determination, do not require updating, while the remaining prediction models and prediction models determined to require updating in the first determination result require updating. Prediction method.

14. System equipment in the power grid, A prediction system that outputs predicted values ​​obtained for each of the one or more prediction targets, A system control calculation device that generates control data including control instructions based on the predicted values ​​output for each of the one or more prediction targets, and outputs the control data, A system equipment management device that controls the system equipment based on the control data. Equipped with, The aforementioned prediction system, An interface device that accepts input of prediction target observation data and factor observation data, A storage device in which the prediction target observation data and the factor observation data are stored, A processor connected to the interface device and the storage device, which updates one or more prediction models that are determined to require updating, among the one or more prediction models corresponding to one or more prediction targets. Equipped with, For each prediction target, the prediction model takes factor prediction values, which are data containing predicted values ​​for future periods for each of the one or more factors that may influence the one or more prediction targets, as input, and outputs predicted values ​​of observed values ​​obtained for the prediction target during that future period. The aforementioned prediction target observation data includes, for each of the one or more prediction targets, time-series data of past observation values ​​obtained for that prediction target. The factor observation data includes, for each of the one or more factors, time-series data of past observation values ​​obtained for that factor. The processor generates update data from the prediction target observation data and the factor observation data. The update data includes, for each of the one or more prediction targets, sample data for each observation period. For each prediction target, the sample data for each observation period includes one or more observed values ​​obtained for that prediction target during that observation period, and the observed values ​​obtained for each of the one or more factors during that observation period. The processor calculates the degree of difference between the first sample group and the second sample group for each prediction target. For each prediction target, The first sample group is one or more sample data used to identify the prediction model corresponding to the target of prediction. The second sample group is one or more sample data that have not yet been used to identify the corresponding prediction model for the target of prediction. The processor performs an update necessity determination for each prediction target, which is a determination of whether the degree of difference is above a threshold, and outputs a first determination result as a result of the update necessity determination. A predictive model in which the above update necessity determination result is true is a predictive model that indicates an update is necessary. A predictive model for which the above update necessity determination result is false is a predictive model that does not require updating. The aforementioned processor, As a prediction result for determining the re-determination start condition, the system outputs a prediction result indicating whether the predicted value output from the current prediction model, which is input to factor prediction values ​​that include the predicted values ​​of each of the one or more factors, deviates from a predetermined range of values. A determination is made, based on the prediction results for determination, to determine whether the re-determination start condition is met, which is that there are prediction targets that do not require updating and whose output prediction values ​​do not deviate from the predetermined range. If the re-determination start condition is met, a second determination result is generated as a result of performing the update necessity determination on each current prediction model again using the first determination result and other update determination criteria, and data representing the prediction models, excluding the current prediction models corresponding to prediction targets that do not require updating from the current prediction models that were determined to require updating because the degree of difference is greater than or equal to the threshold, is output as the result of the update necessity determination. The aforementioned other update determination criteria determine that, among the prediction models determined to require updating in the first determination result, prediction models whose predicted values ​​output for a predetermined target date and time do not deviate from the predetermined range of values, when referring to the prediction results for determination, do not require updating, while the remaining prediction models and prediction models determined to require updating in the first determination result require updating. System management system.

Images