Power operation system and method

Abstract

To improve accuracy of a prediction scenario, and therefore, improve a processing result in a system which performs processing based on the prediction scenario.SOLUTION: A plurality of error factors are defined in prediction data of a prediction target. An electric power operation system calculates error distribution of the error factors, about each of the plurality of defined error factors, and calculates one or more pieces of error data from the error distribution. About each error factor, the error data are time series data of errors determined from the error distribution of the error factors. The electric power operation system generates a prediction scenario which reflects predicted error data according to an error dataset in prediction data, about each of one or more error datasets. About each error dataset, the error dataset includes one error data about each of the plurality of error factors. The predicted error data includes time series of errors of a predicted value of the prediction target. The prediction scenario includes the time series of values which reflect the errors in the predicted value.SELECTED DRAWING: Figure 3

Description

[Technical Field]

[0001] The present invention generally relates to an electric power operation system and method, and is suitable for application to, for example, an operation system that creates an operation plan based on forecast data, in cases where forecast errors occur in the forecast data due to various factors. [Background technology]

[0002] In the energy business sector, such as the electricity and gas businesses, the communications business sector, and the transportation business sector, such as taxis and delivery services, demand forecasts are made to operate facilities and allocate resources in accordance with consumer demand.

[0003] For example, in the field of electric power generation, there is a physical constraint that the amount of electricity generated must always match the amount of electricity demand, so it is necessary to have a sufficient number of generators on standby in advance.

[0004] Power demand forecasts are used to create operation plans that keep sufficient generators on standby in advance. Because forecast data (for example, time-series data of predicted power demand values) usually contains prediction errors, methods are used to create operation plans that take prediction errors into account in advance.

[0005] Patent Document 1 discloses a method for estimating a variance-covariance matrix from time series data of past forecast errors and generating simulated values ​​of the error time series as a multidimensional normal distribution with the same dimensions as the number of time points within the forecast period. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Japanese Patent Application Laid-Open No. 2010-20442 Summary of the Invention [Problem to be solved by the invention]

[0007] A supply and demand management equipment operation system is known, which is a system for controlling one or more power facilities including at least one of a generator, a power storage facility, and a switchgear. The supply and demand management equipment operation system generates an operation plan (control plan) for the power facilities based on a forecast scenario, and controls the one or more power facilities based on the generated operation plan. The forecast scenario is data including a time series of values ​​(e.g., values ​​representing power demand) for a forecast target (e.g., power demand). The forecast scenario is data that can be generated by reflecting forecast error data (time series data of predicted errors) in forecast data.

[0008] In order to improve the results of processing by the supply and demand management facility operation system, it is necessary to improve the accuracy of the forecast scenario. This problem may also apply to other types of systems that perform processing based on forecast scenarios instead of or in addition to the supply and demand management facility operation system. [Means for solving the problem]

[0009] A plurality of error factors are determined for the forecast data of the forecast target. The power operation system calculates the error distribution (e.g., the variance or standard deviation of the error) of each of the determined plurality of error factors and calculates one or more error data from the error distribution. For each error factor, the error data is data including an error determined from the error distribution of the error factor for each of one or more time periods. The power operation system generates a forecast scenario for each of one or more error datasets, which is data that reflects the forecast error data according to the error dataset in the forecast data. For each of the one or more error datasets, the error dataset includes one error data for each of the plurality of error factors, and the forecast error data includes an error of the forecast value of the forecast target for each of one or more time periods, and the forecast scenario includes a value in which the error is reflected in the forecast value for each of the one or more time periods. [Effects of the Invention]

[0010] According to the present invention, the accuracy of a prediction scenario can be improved, thereby improving the results of processing in a system that performs processing based on a prediction scenario. [Brief explanation of the drawings]

[0011] [Figure 1] 1 is a diagram illustrating a device configuration of a power operation system according to a first embodiment. [Figure 2] 1 is a diagram illustrating an internal configuration of a device in a power operation system according to a first embodiment. [Figure 3] FIG. 2 is a diagram showing a data flow of the power operation system according to the first embodiment. [Figure 4] FIG. 2 is a diagram showing a processing flow of the power operation system according to the first embodiment. [Figure 5] FIG. 4 is a diagram showing a data flow of a model-dependent error calculation unit according to the first embodiment. [Figure 6A] FIG. 4 is a diagram illustrating an outline of processing by a model-dependent error calculation unit according to the first embodiment. [Figure 6B] FIG. 4 is a diagram illustrating an outline of processing by a model-dependent error calculation unit according to the first embodiment. [Figure 7] FIG. 10 is a diagram showing a data flow of a data-precision-dependent error calculation unit according to the first embodiment. [Figure 8A] FIG. 10 is a diagram illustrating an outline of processing by a data-precision-dependent error calculation unit according to the first embodiment. [Figure 8B] FIG. 10 is a diagram illustrating an outline of processing by a data-precision-dependent error calculation unit according to the first embodiment. [Figure 9] FIG. 10 is a diagram showing a data flow of a data granularity dependent error calculation unit according to the first embodiment. [Figure 10A] FIG. 10 is a diagram illustrating an outline of processing by a data granularity dependent error calculation unit according to the first embodiment. [Figure 10B] FIG. 10 is a diagram illustrating an outline of processing by a data granularity dependent error calculation unit according to the first embodiment. [Figure 11] FIG. 2 is a diagram showing a data flow of a predictive scenario generation unit according to the first embodiment. [Figure 12]FIG. 4 is a diagram showing a processing flow of a correlation estimation unit according to the first embodiment. [Figure 13] FIG. 4 is a diagram showing a processing flow of an error initial value generation unit according to the first embodiment. [Figure 14] FIG. 10 is a diagram showing a processing flow of an error series generation unit according to the first embodiment. [Figure 15] FIG. 4 is a diagram showing a processing flow of a scenario synthesis unit according to the first embodiment. [Figure 16] FIG. 10 is a diagram illustrating an outline of the effects of the first embodiment. [Figure 17] FIG. 11 is a diagram showing a data flow of a predictive scenario generation unit according to the second embodiment. [Figure 18] A diagram showing the data flow of the classification scenario selection unit related to the second embodiment. [Figure 19] FIG. 11 is a diagram showing a data flow of a predictive scenario generation unit according to the third embodiment. [Figure 20] FIG. 11 is a diagram showing a data flow of a comparative scenario selection unit according to the third embodiment. [Figure 21] FIG. 13 is a diagram showing a data flow of a predictive scenario generation unit according to the fourth embodiment. [Figure 22] A diagram showing the data flow of the scenario expansion unit according to the fourth embodiment. [Figure 23] FIG. 13 is a diagram showing a data flow of a power operation system according to a fifth embodiment. [Figure 24] FIG. 13 is a diagram showing a data flow of a power operation system according to a seventh embodiment. [Figure 25] FIG. 19 is a diagram showing a data flow of a power operation system according to an eighth embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0012] In the following description, an "interface apparatus" may refer to one or more interface devices, which may be at least one of the following: One or more I / O (Input / Output) interface devices. The I / O (Input / Output) interface devices are interface devices for at least one of the I / O devices and a remote display computer. The I / O interface device for the display computer may be a communications interface device. The at least one I / O device may be a user interface device, for example, either an input device such as a keyboard and a pointing device, or an output device such as a display device. One or more communication interface devices. The one or more communication interface devices may be one or more homogeneous communication interface devices (e.g., one or more NICs (Network Interface Cards)) or two or more heterogeneous communication interface devices (e.g., an NIC and an HBA (Host Bus Adapter)).

[0013] In the following description, "memory" refers to one or more memory devices, typically a primary storage device. At least one memory device in the memory may be a volatile memory device or a non-volatile memory device.

[0014] In the following description, a "persistent storage device" refers to one or more persistent storage devices. A persistent storage device is typically a non-volatile storage device (e.g., an auxiliary storage device), and specifically, for example, an HDD (Hard Disk Drive) or an SSD (Solid State Drive).

[0015] In the following description, the term "storage device" may refer to at least one memory, including memory and persistent storage device.

[0016] In the following description, a "processor" refers to one or more processor devices. The at least one processor device is typically a microprocessor device such as a CPU (Central Processing Unit), but may also be other types of processor devices such as a GPU (Graphics Processing Unit). The at least one processor device may be a single-core or multi-core. The at least one processor device may also be a processor core. The at least one processor device may also be a processor device in a broader sense, such as a hardware circuit (e.g., an FPGA (Field-Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit)) that performs part or all of the processing.

[0017] Furthermore, in the following description, functions are sometimes described using the expression "yyy unit." However, the functions may be realized by one or more computer programs executed by a processor, by one or more hardware circuits (e.g., FPGAs or ASICs), or by a combination thereof. When a function is realized by a program executed by a processor, the specified processing is performed using a storage device and / or an interface device, etc., as appropriate, and therefore the function may be at least a part of the processor. Processing described using a function as the subject may also be processing performed by a processor or a device having the processor. A program may be installed from a program source. The program source may be, for example, a computer from which the program is distributed or a computer-readable recording medium (e.g., a non-transitory recording medium). The description of each function is merely an example; multiple functions may be combined into one function, or one function may be divided into multiple functions.

[0018] Hereinafter, several embodiments of the present invention will be described in detail with reference to the drawings. (1) First embodiment (1-1) Configuration of the Power Operation System According to the Present Embodiment

[0019] FIG. 1 shows the overall device configuration of the power operation system according to this embodiment.

[0020] For example, when applied to the electric power business, the data processing system 1 analyzes past observed values ​​of electric power demand, predicts the amount of electric power demand and trading price for a given target period, and estimates one or more possible prediction errors, and calculates one or more forecast scenarios that combine the predicted values ​​and prediction errors. Based on the forecast scenarios, it becomes possible to manage electric power supply and demand, such as formulating and executing generator operation plans and formulating and executing power procurement trading plans from other electric power companies.

[0021] The data processing system 1 is composed of a user 2, a power operation system 3, an observation data provider 4, an observation data storage device 5, an external data provider 6, an external data storage device 7, a supply and demand management facility 8, a supply and demand management facility operation system 9, and a communication path 10. The communication path 10 may be a communication network such as a LAN (Local Area Network) or a WAN (Wide Area Network), and is a communication path that connects the various devices and terminals that make up the data processing system 1 so that they can communicate with each other. The supply and demand management facility operation system 9 uses one or more forecast scenarios calculated by the power operation system 3 to create and execute plans for the operation, control, market transactions, etc. of facilities such as generators.

[0022] As a specific example, the supply and demand management equipment 8 includes control devices such as generators, storage equipment, and switches, and the supply and demand management equipment operation system 9 is an operation system that performs, for example, market trading management, generator control, storage equipment control, and switch control.

[0023] The observation data storage device 5 is an example of a data source and stores a group of observation data. The group of observation data includes a plurality of pieces of observation data. The observation data is typically periodic data. The observation data may be time-series data of past observation values ​​that serve as input values ​​(input data) for generating a forecast scenario.

[0024] The observation data may be, for example, data including at least one of the following: The observation data may be data for each measuring device (for example, data for each smart meter), or may be data representing the total of multiple measuring devices (for example, observation data representing the average value of the observation data of all smart meters belonging to a specific area). Power consumption data. Energy production data such as solar and wind power. - Data on trading prices of energy traded on wholesale exchanges. -Communication volume data measured at communication base stations, etc. - Historical data on the location of any moving vehicle.

[0025] The observation data storage device 5 searches for and / or transmits observation data in response to a data acquisition request from another device.

[0026] The external data storage device 7 is an example of a data source and stores a group of external data. The group of external data includes a plurality of external data. The external data may be past external data (e.g., data including values ​​observed or forecasted in the past) or future external data (e.g., data including forecasted values ​​in the future) that serve as input values ​​(input data) for generating a forecast scenario.

[0027] The external data may be factor data, which is data including observed or forecasted values ​​for factors (e.g., temperature and humidity) that may affect the value (observed value or predicted value) of the target to be predicted, for each of one or more times. The external data may be, for example, data including at least one of the following: Any of the following data may be time-series data. Any of the following data may be referred to as factor data. Weather data such as temperature, humidity, solar radiation (amount of solar radiation), wind speed, and air pressure. Calendar data such as date, day of the week, and flag values ​​that indicate the type of day you specify. -Data showing whether or not sudden events such as typhoons or other events have occurred. - Data on industrial trends (for example, the number of energy consumers, attributes indicating the type of energy consumer (factory, office, household, etc.), the industry of energy consumers, and production figures and sales amounts by industry and company). Data showing location information and regional topographical or climatic characteristics. -Data such as the number of communication devices connected to a communication base station. -Previously published weather forecast data. - Past observation data itself.

[0028] In response to a data acquisition request from another device, the external data storage device 7 searches for and / or transmits external data.

[0029] The power operation system 3 generates a prediction scenario using the observation data and external data acquired from the observation data storage device 5 and the external data storage device 7. (1-2) Device internal configuration

[0030] FIG. 2 shows the device configuration of the power operation system 3, the observation data storage device 5, and the external data storage device 7 included in the data processing system 1.

[0031] The power operation system 3 is composed of an input device 32, an output device 33, a communication device 34, a storage device 35, and a CPU (Central Processing Unit) 31 connected to them. The power operation system 3 is an information processing system such as a personal computer, a server computer, or a handheld computer. The power operation system 3 may be such a physical computer system (one or more physical computers), or may be a logical computer system based on a physical computer system (for example, a system as a cloud computing service). The input device 32 and the output device 33 may be absent. The communication device 34 is an example of an interface device, and the CPU 31 is an example of a processor.

[0032] The input device 32 is configured, for example, by a keyboard or a mouse, and the output device 33 is configured by a display or a printer. The input device 32 and the output device 33 may be included in a remote client with the power operation system 3 as a server.

[0033] The communication device 34 is configured to include, for example, a NIC (Network Interface Card) for connecting to a wireless LAN or a wired LAN. The storage device 35 is a storage medium such as a RAM (Random Access Memory) or a ROM (Read Only Memory). Output results and intermediate results from each processing unit may be output via the output device 33 as appropriate.

[0034] The storage device 35 stores one or more computer programs for realizing functions such as a data prediction unit 351, a model-dependent error calculation unit 352, a data precision-dependent error calculation unit 353, a data granularity-dependent error calculation unit 354, and a prediction scenario generation unit 355. These functions are realized by the CPU 31 executing these computer programs.

[0035] The storage device 35 also has a storage area 356 for storing a prediction error DB (database) 356A. The prediction error data 356A includes database information and text information used to generate prediction scenarios, and also includes data representing prediction errors separated into components due to various factors, such as model dependence, data accuracy dependence, and data granularity dependence. Of the prediction scenarios generated in the past, the prediction error data 356A includes a set of error components constituting the scenario that takes the closest value to the actually observed value (e.g., a time series of power values) for the prediction target (e.g., power demand).

[0036] The observation data storage device 5 is made up of at least a communication device 51, a storage device 52, and a CPU 53 connected to them. The storage device 52 has a storage area 521 for storing an observation data group 521A. The observation data group 521A includes a plurality of observation data.

[0037] The external data storage device 7 is composed of at least a communication device 71, a storage device 72, and a CPU 73 connected to them. The storage device 72 has a storage area 721 for storing an external data group 721A. The external data group 721A includes a plurality of external data. (1-3) Processing and data flow of power operation system 3

[0038] The data flow and processing flow of power operation system 3 will be described with reference to FIGS.

[0039] First, the data flow of the power operation system 3 will be described with reference to FIG.

[0040] Power operation system 3 acquires observation data group 521A and external data group 721A from observation data storage device 5 and external data storage device 7, respectively.

[0041] The observed data group 521A and the external data group 721A are input to a data prediction unit 351, a model-dependent error calculation unit 352, a data-precision-dependent error calculation unit 353, and a data-granularity-dependent error calculation unit 354, respectively.

[0042] The data prediction unit 351 constructs a prediction model for future values ​​of observed data by any modeling method using mutually linked observed data and external data. The data prediction unit 351 inputs external data linked to the prediction period into the constructed prediction model, and calculates prediction data (time-series data of predicted values ​​(predicted observed values)) for the prediction period.

[0043] The model-dependent error calculation unit 352 constructs multiple prediction models using the input observation data group 521A and external data group 721A, and calculates the variance of the error data of the training error of the prediction model itself by inputting the same external data to each prediction model. The error that depends on the prediction model itself (error caused by the prediction model) can be called the "model-dependent error." The time-series data of the model-dependent error can be called the "model-dependent error data." The variance of the model-dependent error data can be called the "model-dependent error distribution."

[0044] The data precision-dependent error calculation unit 353 constructs a prediction model using the input observation data group 521A and external data group 721A, and calculates the variance of prediction errors caused by errors in the input data itself using the deviation of the model outputs when using forecast values ​​(time series data of forecast values) and observed values ​​(time series data of observed values ​​as the correct answer to the time series data of forecast values) in the external data. Errors that depend on the precision of the input data (errors caused by the precision of the input data) can be called "data precision-dependent errors." Data precision-dependent error time series data can be called "data precision-dependent error data." The variance of data precision-dependent error data can be called "data precision-dependent error distribution."

[0045] The data granularity-dependent error calculation unit 354 constructs a prediction model using the input observation data group 521A and external data group 721A, and calculates overall error data as the difference between the model output (time series data of predicted values) using the forecast values ​​(time series data of forecast values) in the external data and the observed values ​​(time series data of observed values). This "overall error data" is time series data of the overall error. The "overall error" is an error that includes errors due to the various error factors described above (model dependence, data accuracy dependence, and data granularity dependence). The variation in the overall error data can be called the "overall error distribution." After calculating the overall error data, the data granularity-dependent error calculation unit 354 subtracts the error data of the model-dependent error component (time series data of error) and the error data of the data accuracy-dependent error component from the overall error data. The error data of the remaining error component is the error data of the data granularity-dependent error component. The data granularity dependent error calculation unit 354 separates the error data of the data granularity dependent error component based on the strength of the correlation with each data item included in the external data, and calculates the error variation caused by the minute behavior that cannot be reflected in the model output due to the aggregation granularity of each item of the external data.

[0046] The forecast data, model-dependent error data (error data with model dependence as an error component), data accuracy-dependent error data (error data with data accuracy dependence as an error component), and data granularity-dependent error data (error data with data granularity dependence as an error component) are each input to the forecast scenario generation unit 355 along with the forecast error DB 356A. The forecast scenario generation unit 355 generates one or more sets of forecast error data (time-series data of predicted errors) for the forecast period based on the correlation between error factors (e.g., relationships estimated from past data for each factor) and the error variation (error distribution) due to various factors such as model dependence, data accuracy dependence, and data granularity dependence. For each of the one or more sets of generated forecast error data, the forecast scenario generation unit 355 generates a forecast scenario by combining the forecast error data with the forecast data. As a result, one or more forecast scenarios are generated. A "forecast scenario" is time-series data of values ​​in which predicted errors are reflected in predicted observed values.

[0047] At least one of the one or more forecast scenarios generated by the forecast scenario generation unit 355 is input to the supply and demand management equipment operation system 9. The input forecast scenario is used to operate power supply and demand management equipment such as generators, power storage equipment, and switches.

[0048] Next, the processing flow of the power operation system 3 will be described with reference to FIG.

[0049] This process is initiated when an input operation from the user 2 is received via the input device 32 provided in the power operation system 3, or when an execution timing separately set in the storage device 35 is reached, and the power operation system 3 executes the processes from S301 to S306. Note that the actual process is performed by the CPU 31 provided in the power operation system 3 executing a computer program stored in the storage device 35.

[0050] First, the data prediction unit 351 acquires the observed data group 521A and the external data group 721A, and calculates prediction data for a specified period of time to be predicted using an arbitrary prediction model (S301).

[0051] Next, the model-dependent error calculation unit 352 acquires the observation data group 521A and the external data group 721A, and constructs multiple prediction models from multiple data sets obtained by extracting portions of the mutually linked observation data and external data. The model-dependent error calculation unit 352 calculates the variance of the model output values ​​obtained by inputting the same external data for prediction into each prediction model, and calculates the variance of the model-dependent error (S302).

[0052] Next, the data-accuracy-dependent error calculation unit 353 acquires the observation data group 521A and the external data group 721A, and divides the linked observation data and external data into two to create data sets 1 and 2. The data-accuracy-dependent error calculation unit 353 constructs a prediction model using data set 1. The data-accuracy-dependent error calculation unit 353 calculates model outputs using the observation data and forecast data of the external data in data set 2, and calculates the variance in the difference between the outputs as the variance in the data-accuracy-dependent error (S303).

[0053] Next, the data granularity dependent error calculation unit 354 acquires the observation data group 521A and the external data group 721A, and creates data sets 1 and 2 by dividing the linked observation data and external data into two. The data granularity dependent error calculation unit 354 constructs a prediction model using data set 1. The data granularity dependent error calculation unit 354 calculates the variance of the overall error from the difference between the model output using the forecast values ​​of the external data in data set 2 and the corresponding observed values ​​(S304).

[0054] Next, the data granularity dependent error calculation unit 354 calculates the residual by subtracting the model dependent error variance and the data precision dependent error variance from the overall error variance. The data granularity dependent error calculation unit 354 calculates the relative importance between the external data items that make up the residual using the observed values ​​of the external data in dataset 2, and divides the residual by the importance to calculate the data granularity dependent error variance (S305).

[0055] Finally, the forecast scenario generator 355 models the error variance of each individual factor and the correlations between factors based on the error data of various error factors, such as model dependence, data accuracy dependence, and data granularity dependence, and generates one or more sets of forecast error data that conform to the model.For each forecast error data, the forecast scenario generator 355 generates a forecast scenario by adding the forecast error data to the forecast data calculated in S301 (S306).

[0056] This embodiment will be described in more detail below. (1-4) Details of each component (1-4-1) Data prediction unit 351

[0057] The data prediction unit 351 constructs a prediction model using any modeling method that uses linked observation data and external data. The prediction model may be a model such as regression, classification, or clustering, or a combination of these, or a model that uses basic statistics such as the mean, median, or mode of the observation data. The prediction target of the model may have any number of data points, such as a single instantaneous value or each time cross-section during a certain period in units of minutes, hours, days, weeks, or years. In other words, in this embodiment, the number of data points (the number of values ​​in the data) in the "time series data" may be a single instantaneous value at a certain time. Furthermore, the prediction model may be a model that predicts the entire period to be predicted at once, or a model that divides the period to be predicted into one or more points and makes predictions.

[0058] The data prediction unit 351 inputs external data associated with the prediction period into the constructed prediction model and calculates prediction data. The "external data associated with the prediction period" may include forecast data (e.g., forecast data on weather, temperature, humidity, etc.) for the prediction period (e.g., the next day) and observation data for a certain period in the past. The "prediction data" may be predicted observation data (e.g., time-series data of predicted power demand values) for the prediction period (e.g., the next day). (1-4-2) Model-dependent error calculation unit 352

[0059] An embodiment of the model-dependent error calculation unit 352 will be described with reference to FIGS. 5 to 6B.

[0060] FIG. 5 shows the data flow inside the model-dependent error calculation unit 352.

[0061] The model-dependent error calculation unit 352 includes a sampling unit 3521 , a modeling unit 3524 , an output calculation unit 3525 , and an output distribution calculation unit 3526 .

[0062] The sample extraction unit 3521 receives the observation data group 521A and the external data group 721A as input data, and generates multiple data sets by extracting multiple sets of linked observation data and external data as model samples (data as samples for modeling) 3522A. The extraction may be random sampling, or it may be extraction of a portion of data from sets previously classified using flag data indicating the classification of the entity from which the observation data was observed (e.g., attribute information such as a factory, commercial facility, or household), or it may be extraction of a portion of data from sets previously classified using statistics such as the average, maximum, minimum, and variance of data trends. The amount of data included in each data set and the number of data sets to be generated may be determined arbitrarily by the user 2, or may be determined mechanically using an index such as a value that causes the fluctuation in the output value of the model-dependent error calculation unit 352 to fall within a certain threshold. Additionally, sample extraction unit 3521 generates one data set to be input to the model constructed by model sample 3522A as input sample (data as a sample to be input to the model) 3523A. It is preferable to use data similar to the data input to the prediction model by data prediction unit 351 as input sample 3523A, but other data may also be used.

[0063] The modeling unit 3524 constructs a prediction model for the value to be predicted using each data set included in the model sample 3522A generated by the sample extraction unit 3521. The number of models to be constructed is the same as the number of data sets included in the model sample 3522A. The prediction model construction method preferably uses the same modeling method as that used by the data prediction unit 351, but any method may be used, such as a model of regression, classification, clustering, or a model combining these, or a model using basic statistics such as the mean, median, or mode of the observed data. Modeling is performed so that the number of data points in the data output from the model (the number of data points (values) in the output data) matches the number of data points in the prediction data output from the data prediction unit 351.

[0064] The output calculation unit 3525 inputs the input sample 3523A to a plurality of models constructed by the modeling unit 3524, and calculates the output data (time-series data of the output value) of each model.

[0065] The output distribution calculation unit 3526 uses each model output data output by the output calculation unit 3525 to calculate the variance of the model output value for each data point included in the period for which the model is calculated. The variance may be based on a general index that indicates the characteristics of the data distribution, such as standard deviation or variance. The variance of the model outputs (time-series data) of multiple models is output to the data granularity-dependent error calculation unit 354 and the forecast scenario generation unit 355 as a model-dependent error distribution.

[0066] The processing contents of model-dependent error calculation section 352 will be described more specifically with reference to FIGS. 6A and 6B.

[0067] For example, the sample extraction unit 3521 generates a model sample 3522A from data for each day from the beginning of the year to one year (an example of past data) in the data table 352A, and generates data for the beginning of the next year (an example of future data) in the data table 352A as an input sample 3523A. The data table 352A includes, as the data for each day from the beginning of the year to one year, observation data for each day from the beginning of the year to one year, and external data for each day from the beginning of the year to one year. The data table 352A also includes external data for the beginning of the next year as the data for the beginning of the next year.

[0068] In this case, the period of data indicating the model output value is set to a period of 24 hours from midnight. Demand data for each hour (data indicating power demand) corresponds to the observed data, and data indicating temperature, solar radiation, and day type corresponds to the external data. Note that, although the power demand is the target of prediction in this example, the present invention is not limited to power demand as a target of prediction (this also applies to this embodiment and other embodiments).

[0069] First, the sample extraction unit 3521 generates a dataset by extracting a portion of the daily data for the entire year of 2020 through simple random sampling. As an example, we will assume that the extraction process is repeated three times, but in reality, more (or fewer) datasets may be extracted. Also, there may be overlapping data between the datasets. Datasets 1, 2, and 3 generated in this manner are model samples 3522A. In addition, the sample extraction unit 3521 extracts data for January 1, 2021 as input sample 3523A.

[0070] The modeling unit 3524 constructs models 1, 2, and 3 from data sets 1, 2, and 3 included in the model sample 3522A, respectively. In this example, the modeling unit 3524 constructs a model that receives data representing temperature, solar radiation, and day type as input and outputs power demand for each hour of the day.

[0071] Output calculation unit 3525 inputs input sample 3523A to each of models 1, 2, and 3, and calculates the output value from each model.

[0072] The output distribution calculation unit 3526 calculates the variation in model-dependent error from the variation in the outputs of models 1, 2, and 3. Specifically, calculation results of the demand amount for each time indicated by each data point are obtained from models 1, 2, and 3, and the output distribution calculation unit 3526 calculates the variation in model-dependent error by calculating the standard deviation of the output value of each model for each time. In this example, the standard deviation is used as an index of variation. (1-4-3) Data precision dependent error calculation unit 353

[0073] An embodiment of the data precision dependent error calculation unit 353 will be described with reference to FIGS. 7 to 8B.

[0074] FIG. 7 shows the data flow of the data precision dependent error calculation unit 353.

[0075] The data precision dependent error calculation unit 353 includes a sampling unit 3531 , a modeling unit 3535 , an observed value use calculation unit 3536 , a forecast value use calculation unit 3537 , and an output deviation distribution calculation unit 3538 .

[0076] The sample extraction unit 3531 receives the observation data group 521A and the external data group 721A as input values, and divides the sets of the observation data and the external data linked to each other into two data sets. A model sample 3532A is generated from one data set (data set 1), and an input sample is generated from the other data set (data set 2).

[0077] Furthermore, the sample extraction unit 3531 generates, from the input sample, an observation sample 3533A, which is data made up of observed values ​​in the external data, and a forecast sample 3534A, which is data made up of forecast values ​​in the external data (for example, forecast values ​​related only to specific items for evaluating the impact of data accuracy on model output). At this time, since the external data in the dataset must have both forecasted values ​​and actually observed values, all of the external data is data from the past before the calculation processing of the power operation system 3 is executed.

[0078] The modeling unit 3535 receives the model sample 3532A as an input value and constructs a prediction model for the value to be predicted. It is preferable to use the same modeling method as that used by the data prediction unit 351 as the method for constructing the prediction model, but any method may be used, such as a regression, classification, clustering, or other model, a model combining these, or a model using basic statistics such as the mean, median, or mode of the observed data. Modeling is performed so that the number of data points in the data output from the model matches the number of data points in the prediction data output from the data prediction unit 351. Alternatively, instead of the model constructed by the modeling unit 3535, any of the multiple models constructed by the data prediction unit 351 may be adopted (in this case, the modeling unit 3535 and the model sample 3532A may be omitted).

[0079] The observation value use calculation unit 3536 inputs the observation sample 3533A to the model constructed by the modeling unit 3535, thereby calculating the model output value when external data as observed values ​​are input to the model.

[0080] The forecast value use calculation unit 3537 inputs the forecast sample 3534A to the model constructed by the modeling unit 3535, thereby calculating the model output value when external data as a forecasted value is input to the model.

[0081] The output deviation distribution calculation unit 3538 calculates the deviation of the model output values ​​of the forecast value use calculation unit 3537 and the observed value use calculation unit 3536, and calculates the variation of the deviation for each time corresponding to each data point. The index of the variation may be a general index that indicates the characteristics of the data distribution, such as standard deviation or variance.

[0082] The data-accuracy-dependent error calculation unit 353 performs the above series of processes for each item of external data for which data accuracy affects the model output. For example, if the items for which data accuracy affects the model output are temperature and solar radiation, then the output deviation distribution calculation unit 3538 finds the deviations between the model output values ​​of the forecast value usage calculation unit 3537 and the observed value usage calculation unit 3536 for each of the temperature forecast sample 3534A and the solar radiation forecast sample 3534A, and calculates the variation in the deviations for each time corresponding to each data point.

[0083] The output deviation distribution calculation unit 3538 outputs the variation in the model output values ​​to the data granularity dependent error calculation unit 354 and the forecast scenario generation unit 355 as the variation in the data precision dependent error.

[0084] The processing contents of the data precision dependent error calculation unit 353 will be described more specifically with reference to FIGS. 8A and 8B.

[0085] As an example, the sample extraction unit 3521 divides one year's worth of daily data in the data table 353A (past observation data and external data) into two. The sample extraction unit 3521 generates a model sample 3532A from one data set, and generates an observation sample 3533A and a forecast sample 3534A from the other data set. The model sample 3532A does not need to include the forecast values ​​of data set 1.

[0086] In this case, the data period indicated by the model output value is a 24-hour period from midnight. The hourly demand data corresponds to the observed data, while the temperature (forecast), temperature (observed), solar radiation (forecast), solar radiation (observed), and daily data correspond to the external data. In this example, data table 353A is divided equally so that the data dates are consecutive, and model sample 3532A is generated using one half. Observed sample 3533A and forecast sample 3534A are generated from the remaining half. However, the data does not necessarily have to be divided equally and consecutively in time; any division method can be used. For example, by dividing the data into model sample 3532A, observed sample 3533A, and forecast sample 3534A on a daily basis, it is expected that seasonal differences can be negated.

[0087] In this example, to calculate the impact of forecast errors in temperature on the forecast error of the forecast data, the sample extraction unit 3521 uses the observed value of temperature as the observed sample 3533A and the forecast value of temperature as the forecast sample 3534A. When calculating the impact of forecast errors in solar radiation, although not shown, the sample extraction unit 3521 separately prepares a data set in which the forecast value of temperature included in the forecast sample 3534A is replaced with the forecast value of solar radiation (prepares the solar radiation forecast sample 3534A), and repeats the following processing.

[0088] The modeling unit 3535 constructs a prediction model using the model sample 3532A. In this example, the prediction model is a model that receives data representing temperature, solar radiation, and day type as input and outputs data representing power demand for each hour of the day.

[0089] The observation value use calculation unit 3536 inputs the observation sample 3533A into the model constructed by the modeling unit 3535, and calculates the model output when the observation value is used 353B.

[0090] The forecast value use calculation unit 3537 inputs the forecast sample 3534A into the model constructed by the modeling unit 3535, and calculates the model output when the forecast value is used 353C.

[0091] The output deviation distribution calculation unit 3538 calculates the deviation between the model output when using forecast values ​​353C and the model output when using observed values ​​353B, and calculates the variation of the deviation for each time (variation of data precision dependent error). In this example, the standard deviation is used as an index of variation. (1-4-4) Data granularity dependent error calculation unit 354

[0092] An embodiment of the data granularity dependent error calculation unit 354 will be described with reference to FIGS. 9 to 10B.

[0093] FIG. 9 shows the data flow of the data granularity dependent error calculation unit 354.

[0094] The data granularity dependent error calculation unit 354 includes a sample extraction unit 3541 , a modeling unit 3545 , an overall error calculation unit 3546 , a relative importance calculation unit 3547 , and a distribution separation unit 3548 .

[0095] The sample extraction unit 3541 receives the observation data group 521A and the external data group 721A as input values, and divides the sets of the observation data and the external data linked to each other into two data sets. A model sample 3542A is generated from one data set (data set 1), and an input sample is generated from the other data set (data set 2).

[0096] Furthermore, the sample extraction unit 3541 generates, from the input sample, an observation sample 3543A, which is data composed of observed values ​​of the external data, and a forecast sample 3544A, which is composed of forecast values ​​of the external data (for example, forecast values ​​for items whose data accuracy affects the model output). At this time, the forecast sample 3544A also includes the observed values ​​(observation data) of the prediction target in order to calculate the deviation from the model output. Furthermore, since the external data in the dataset must have both forecasted values ​​and actually observed values, all of the external data is data from the past before the calculation processing of the power operation system 3 is executed.

[0097] The modeling unit 3545 receives the model sample 3542A as an input value and constructs a prediction model for the value to be predicted. It is preferable to use the same modeling method as that used by the data prediction unit 351 as the method for constructing the prediction model, but any method may be used, such as models such as regression, classification, or clustering, or a model combining these, or a model using basic statistics such as the mean, median, or mode of the observed data. Modeling is performed so that the number of data points in the data output from the model matches the number of data points in the prediction data output from the data prediction unit 351.

[0098] The overall error calculation unit 3546 inputs the forecast sample 3544A into the model constructed by the modeling unit 3545, and calculates the variance of the overall error including various factors such as model dependence, data accuracy dependence, and data granularity dependence. An interval estimation method such as quantile regression may be used to calculate the variance.

[0099] The relative importance calculation unit 3547 receives as input the observed sample 3543A and a value extracted as the overall component of the data granularity-dependent error by subtracting the model-dependent error variance and the data-precision-dependent error variance from the overall error variance. The relative importance calculation unit 3547 performs modeling using the observed sample 3543A for each time included in the data of the data granularity-dependent error, and calculates the relative importance between each external data item included in the observed sample 3543A. It is preferable to use a method for calculating relative importance while canceling the influence of multicollinearity between data items, such as dominance analysis or RWA (Relative Weight Analysis). However, it is also possible to simply use coefficients obtained by linear regression as the relative importance. Alternatively, it is also possible to use an index of variable importance, such as Gini Importance or Split Importance. The sum of the relative importance of multiple external data items may be used as a single relative importance.

[0100] The distribution separation unit 3548 separates components of the variation in the data granularity-dependent error according to the ratio of relative importance calculated by the relative importance calculation unit 3547. The object to be separated may be the most recent value or a value for a period when the external data conditions are similar from the overall value of the data granularity-dependent error calculated by the relative importance calculation unit 3547. The object to be separated may also be a model output obtained by inputting the same data as input to the model used in the data prediction unit 351 into a model of the overall data granularity-dependent error constructed by the relative importance calculation unit 3547. Alternatively, the distribution separation unit 3548 may calculate variations such as standard deviation and variance from the difference between the model output obtained by inputting the forecast sample 3544A into the model constructed by the modeling unit 3545 and the observation data corresponding to the model output, and use a value obtained by subtracting the respective variations of the model-dependent error and the data accuracy-dependent error from the calculated variations.

[0101] Through the above processing, the data granularity dependent error calculation unit 354 calculates the variance of the data granularity dependent error and outputs this variance to the forecast scenario generation unit 355.

[0102] The processing contents of the data granularity dependent error calculation unit 354 will be described more specifically with reference to FIGS. 10A and 10B.

[0103] As an example, the sample extraction unit 3541 divides one year's worth of daily data in the data table 354A (past observation data and external data) into two. The sample extraction unit 3541 generates a model sample 3542A from one data set, and generates an observation sample 3543A and a forecast sample 3544A from the other data set. At this time, the data period indicated by the model output value is a 24-hour period from midnight. The demand data for each hour corresponds to the observation data, and the temperature (forecast), temperature (observation), solar radiation (forecast), solar radiation (observation), and day-specific data correspond to the external data. In this example, the data table 354A is divided equally so that the data dates are consecutive, and one half is used to generate the model sample 3542A, and the other half is used to generate the observation sample 3543A and the forecast sample 3544A. However, the division does not necessarily have to be continuous and equal in time; any division method can be used. For example, by dividing the data into model sample 3542A, observation sample 3543A, and forecast sample 3544A on a daily basis, it is expected that seasonal differences can be canceled out.

[0104] Of the external data, all data recorded as observed values ​​are assigned to the observed sample 3543A, and all data recorded as forecasted values ​​are assigned to the forecast sample 3544A.

[0105] The modeling unit 3545 constructs a prediction model using the model sample 3542A. In this example, the prediction model is a model that receives data representing temperature, solar radiation, and day type as input and outputs data representing power demand for each hour of the day.

[0106] The overall error calculation unit 3546 obtains the model output and prediction interval by inputting the forecast sample 3544A into the model constructed by the modeling unit 3545. In this example, we consider the case where a 68% prediction interval is calculated in accordance with 1σ. The overall error calculation unit 3546 then obtains the overall error by subtracting the demand data included in the forecast sample 3544A from the model output. The overall error calculation unit 3546 calculates the standard deviation from the value of the overall error for each time period to obtain the variance of the overall error.

[0107] The overall error calculation unit 3546 subtracts the model-dependent error variance and the data-precision-dependent error variance from the prediction interval representing the overall error and the overall error variance to obtain the residual of the prediction interval and the residual of the overall error variance, and inputs these residuals to the relative importance calculation unit 3547.

[0108] The relative importance calculation unit 3547 calculates the relative importance between external data included in the observed sample 3543A when the residuals of the prediction intervals are modeled using the observed sample 3543A. Fig. 10B shows an example in which the residuals of the prediction intervals are modeled using temperature, solar radiation, and day type, and the relative importance is 0.6, 0.1, and 0.3, respectively.

[0109] The distribution separation unit 3548 separates the residual of the overall error variability based on the relative importance between the external data included in the observed sample 3543A.

[0110] In this way, the data granularity dependent error calculation unit 354 obtains the variation of the data granularity dependent error for each external data item and for each time. (1-4-5) Forecast scenario generation unit 355

[0111] An embodiment of the forecast scenario generator 355 will be described with reference to FIGS.

[0112] FIG. 11 shows the data flow of the forecast scenario generator 355.

[0113] The forecast scenario generator 355 includes a correlation estimator 3551 , an error initial value generator 3552 , an error series generator 3553 , and a scenario synthesizer 3554 .

[0114] The correlation estimation unit 3551 receives as input the variance of model-dependent errors, the variance of data precision-dependent errors, the variance of data granularity-dependent errors, and the prediction error DB 356A (data including past errors of model-dependent, data precision-dependent, and data granularity-dependent factors). The correlation estimation unit 3551 calculates the magnitude of error variance for each factor from the error variance of each of the model-dependent, data precision-dependent, and data granularity-dependent factors. In addition, the correlation estimation unit 3551 calculates the interrelationship between factors and the interrelationship between factors and time using past prediction error data for each factor in the prediction error DB 356A. Variance-covariance matrices may be used to represent the magnitude of variance of each factor, the interrelationship between factors, and the interrelationship between factors and time. When calculating the interrelationship between times, the correlation estimation unit 3551 may calculate the interrelationship using error values ​​from multiple past times. If the prediction error DB 356A does not include past prediction error data for each factor, or if the amount of data does not exceed a certain threshold, only the magnitude of the variation of each factor may be calculated.

[0115] The error initial value generator 3552 generates an initial value for the prediction error for each factor based on the magnitude of variance of each factor and the interrelationships between the factors. If the prediction error DB 356A does not contain past prediction error data for each factor, or if the amount of data does not exceed a certain threshold, an initial value for the prediction error may be generated independently for each factor based only on the magnitude of variance of each factor. For each factor, the initial value may be generated as a random number following a multivariate normal distribution using a variance-covariance matrix that represents the magnitude of variance of the factor and the interrelationships between the factors, or may be generated using a general probability distribution including an asymmetric distribution such as the beta distribution.

[0116] The error series generator 3553 sequentially generates prediction error values ​​using the magnitude of variance for each factor, the correlation values ​​between factors and between times, and the error values ​​generated for the previous time point. If the prediction error DB 356A does not contain past prediction error data for each factor, or if the amount of data does not exceed a certain threshold, initial prediction error values ​​may be generated independently for each factor and time point based only on the magnitude of variance for each factor. The error series generator 3553 converts the variance-covariance matrix representing the magnitude of variance for each factor and the correlation between factors and between times calculated by the correlation estimation unit 3551 into a variance-covariance matrix representing the magnitude of variance for each factor and the correlation between factors for the target time point for which prediction errors are to be generated, using the error values ​​for each factor at the previous time point. The error series generator 3553 generates random numbers following a multivariate normal distribution using the converted variance-covariance matrix as prediction error values. Not limited to normal distributions, general probability distributions, including asymmetric distributions such as the beta distribution, may be used to generate prediction error values.

[0117] The scenario synthesis unit 3554 receives as input the error series for each factor (error data for each factor) generated by the error series generation unit 3553 and the prediction data, and generates a prediction scenario by adding up the values ​​for each time.

[0118] The forecast scenario generation unit 355 may generate multiple forecast scenarios by repeating the above series of processes or by repeating the processes for each processing block. Through the above processes, the forecast scenario generation unit 355 generates one or multiple forecast scenarios and outputs the forecast scenarios to the supply and demand management equipment operation system 9. Of the generated forecast scenarios, the forecast scenario generation unit 355 adds the error values ​​constituting the scenario that was closest to the actually observed value of the forecast target to the forecast error DB 356A as forecast error data for each factor.

[0119] The processing flow of the correlation estimation unit 3551 will be described with reference to FIG.

[0120] First, the correlation estimation unit 3551 determines whether the amount of prediction error data including past error values ​​for each factor in the prediction error DB 356A exceeds a certain threshold (S3551A). The threshold may be a value arbitrarily set by the user 2.

[0121] If the amount of data exceeds a certain threshold (S3551A: Yes), the correlation estimation unit 3551 generates a variance-covariance matrix for each time point, with the prediction error for each factor as a random variable (S3551B). Specifically, for example, the correlation estimation unit 3551 uses error data for each factor that is model-dependent, data accuracy-dependent, and data granularity-dependent, which are included in the prediction error DB 356A, to generate a variance-covariance matrix whose elements are the average for each error factor at the time that corresponds to the start of the prediction period, and the variance of each factor and the covariance between factors.

[0122] Next, the correlation estimation unit 3551 generates a variance-covariance matrix for each time period, with the prediction errors for each factor at the current time and the previous time period as random variables (S3551C). Specifically, for example, the correlation estimation unit 3551 uses error data for each factor dependent on the model, data accuracy, and data granularity included in the prediction error DB 356A to calculate the variance of the error value for each factor at the current time period and the variance of the error value for each factor at the time period prior to the current time period. In addition, the correlation estimation unit 3551 calculates the covariance of the error value for each factor at the current time period and the previous time period. The correlation estimation unit 3551 generates a variance-covariance matrix having the above variance and covariance as elements for each time period in the prediction period. Note that values ​​for any number of times may be included in the variance-covariance matrix.

[0123] If the data amount is equal to or less than a certain threshold (S3551A: No), the correlation estimation unit 3551 converts the error variance calculated for each factor for each time into the error variance of the factor (S3551D). Specifically, for example, the correlation estimation unit 3551 converts the value of the model-dependent error variance, the value of the data-precision-dependent error variance, and the value of the data-granularity-dependent error variance into variance values ​​for each time.

[0124] The correlation estimation unit 3551 outputs the variance-covariance matrix or variance values ​​calculated by the above processing to the error initial value generation unit 3552. The processing of the correlation estimation unit 3551 ends.

[0125] The processing flow of the error initial value generating unit 3552 will be described with reference to FIG.

[0126] First, the error initial value generation unit 3552 determines whether the amount of prediction error data, including past error values ​​for each factor, stored in the prediction error DB 356A exceeds a certain threshold (S3552A). The threshold may be a value arbitrarily set by the user 2. The determination result of S3552A may be the same as the determination result of S3551A in FIG. 12.

[0127] If the amount of data exceeds a certain threshold (S3552A: Yes), the error initial value generation unit 3552 converts the variance-covariance matrix, in which the prediction errors for each factor are used as random variables, into a correlation matrix by dividing it by the standard deviation of the errors for each factor (S3552B). Specifically, for example, the error initial value generation unit 3552 extracts the standard deviation of the errors for each factor from the diagonal elements of the variance-covariance matrix calculated in S3551B, and converts the variance-covariance matrix into a correlation matrix by dividing it by the standard deviation.

[0128] The error initial value generation unit 3552 calculates the standard deviation from the values ​​of the error variation due to model-dependent, data accuracy-dependent, and data granularity-dependent factors at the time that corresponds to the start of the prediction period, and by multiplying the standard deviation by the correlation matrix obtained in S3552B, converts the correlation matrix into a variance-covariance matrix that represents the magnitude and interrelationship of each error factor at the time that corresponds to the start of the prediction period (S3552C).

[0129] The error initial value generation unit 3552 generates an initial value of the error for each factor according to the variance-covariance matrix of S3552C (S3552D). Specifically, for example, the error initial value generation unit 3552 generates a random number that follows a multivariate normal distribution for each factor based on the variance-covariance matrix. The random number is the initial value of the error for each factor.

[0130] If the amount of data is equal to or less than a certain threshold (S3552A: No), the error initial value generation unit 3552 generates an initial value of the error for each factor according to the variance of the error calculated for each factor (S3552E). Specifically, for example, the error initial value generation unit 3552 generates random numbers according to the variance of the error for each factor calculated in S3551A of the correlation estimation unit 3551, and sets the random number for each factor as the initial value of the error due to that factor. The mean may be 0, but any value may be used, and the probability distribution when generating the random numbers may be any probability distribution such as a normal distribution or a beta distribution.

[0131] The initial value of the prediction error for each factor generated by the above processing is output by the initial error value generation unit 3552 to the error series generation unit 3553. The processing by the initial error value generation unit 3552 is then completed.

[0132] The processing flow of the error series generation unit 3553 will be described with reference to FIG.

[0133] First, the error series generation unit 3553 determines whether the amount of prediction error data, including past error values ​​for each factor, stored in the prediction error DB 356A exceeds a certain threshold (S3553A). The threshold may be a value arbitrarily set by the user 2. The determination result of S3553A may be the same as the determination result of S3551A in FIG. 12 or S3552A in FIG. 13.

[0134] If the amount of data exceeds a certain threshold (S3553A: Yes), the error series generation unit 3553 repeats the following processes from S3553C to S3553F (generation of correlated error series values) until an error series with the same number of points as the number of data points in the predicted data is generated. For example, if the predicted data is made up of N+1 points of data, the error series generation unit 3553 repeats the processes from S3553C to S3553F N times (the number of times obtained by subtracting from N+1 the value of the initial error value that has already been generated) (S3553B).

[0135] The error series generation unit 3553 extracts the standard deviation of the errors for each factor and time from the diagonal components of the variance-covariance matrix calculated in S3551C of Figure 12 (a variance-covariance matrix in which the errors for each factor between the current time and the previous time are used as random variables), and converts the variance-covariance matrix into a correlation matrix by dividing the variance-covariance matrix by the standard deviation (S3553C).

[0136] The error series generation unit 3553 calculates the standard deviation from the value of the variance for each factor at the target time for generating the prediction error and the value of the variance for each factor at the previous time, and converts each standard deviation into a variance-covariance matrix that represents the magnitude and interrelationship of each error factor at the target time for generating the prediction error and the previous time by multiplying each standard deviation by the correlation matrix of S3553C (S3553D).

[0137] The error series generation unit 3553 uses values ​​generated as prediction errors at a time previous to the target time for prediction error generation to convert the variance-covariance matrix of S3553D into a variance-covariance matrix that represents the magnitude and interrelationship of each factor only at the target time for prediction error generation (S3553E). Specifically, for example, the error series generation unit 3553 expresses the mean and variance-covariance matrix of the former mean and variance-covariance matrix so that the parameters when a variable value corresponding to the error at the previous time in a probability distribution having the error between the target time for prediction error generation and the previous time as random variables match the parameters of a conditional probability density distribution for the target time conditioned on the error at the previous time.

[0138] The error series generation unit 3553 generates random numbers that follow a multivariate normal distribution based on the variance-covariance matrix (conditional variance-covariance matrix) of S3553E, and sets these as prediction errors due to each factor at the target time (S3553F). As a result, errors for each error factor at the target time for error generation are generated.

[0139] If the data amount is equal to or less than a certain threshold (S3553A: No), the error series generation unit 3553 repeats the following process of S3553H (generation of independent error series values) until an error series with the same number of points as the number of data points in the predicted data is generated. For example, if the predicted data is made up of N+1 points of data, the error series generation unit 3553 repeats the process of S3553H N times (the number of times obtained by subtracting from N+1 the value of the initial error value that has already been generated) (S3553G).

[0140] The error series generation unit 3553 generates an error for each factor according to the variance of the error for each factor at the target time for error generation (S3553H). Specifically, for example, the error series generation unit 3553 generates a random number according to the value at the target time for prediction error generation, among the variance of the error for each factor for each time calculated in S3551D of FIG. 12, and sets this as the prediction error for each factor at the target time. The probability distribution used for generating the random number may be any probability distribution, such as a normal distribution or a beta distribution. The correlation between times may be reflected in the random number generation, for example, by using the value of the error at the immediately preceding time as the average of the distribution at the target time for prediction error generation.

[0141] The error series generator 3553 outputs the series values ​​of the prediction errors for each factor generated by the above processing to the scenario combiner 3554. The processing of the error series generator 3553 ends.

[0142] The processing flow of the scenario synthesis unit 3554 will be described with reference to FIG.

[0143] The scenario synthesis unit 3554 repeats the following process of S3554B (synthesis of prediction data and prediction error data) the number of times equal to the number of sets of prediction error series for each factor that have been generated (S3554A). For example, if M sets of error series have been generated, the process of S3554B is repeated M times. One set of error series is made up of prediction error series for each factor (error data for each factor).

[0144] The scenario synthesis unit 3554 adds the set of prediction error series for each factor to the prediction data output from the data prediction unit 351, and generates one prediction scenario (S3554B).

[0145] The scenario synthesis unit 3554 outputs the M forecast scenarios generated by the above processing to the supply and demand management equipment operation system 9. The processing of the scenario synthesis unit 3554 ends. The scenarios output to the supply and demand management equipment operation system 9 may be M+1 scenarios including the forecast data itself.

[0146] With the above processing, the first calculation processing in this embodiment is completed, and the calculation processing in power operation system 3 in this embodiment is also completed. (1-5) Explanation of the Effects of the Present Embodiment

[0147] The effects of power operation system 3 according to this embodiment will be described with reference to FIG.

[0148] FIG. 16 is a conceptual diagram showing the effect of generating a series of prediction errors while taking into consideration the interrelationships between errors for each error factor.

[0149] For example, the overall forecast error is separated into a model-dependent error, an error dependent on the data accuracy of the temperature forecast, an error dependent on the data accuracy of the insolation forecast, an error dependent on the data granularity related to the deviation of the temperature from the point, an error dependent on the data granularity related to the deviation of the insolation from the point, and an error dependent on the data granularity related to the inconsistency between each consumer by day. Note that the thick dashed line 1601 represents the forecast data, and the range 1602 including the thick dashed line 1601 represents the variation in the forecast error.

[0150] In the example shown in FIG. 16, the prediction errors as a whole have similar characteristics of variation, but the characteristics of each factor are different, such as characteristics 1 and 2 of the errors for each factor.

[0151] In the comparative example, the correlation between factors that make up the prediction error is not calculated, so if the prediction errors as a whole have similar characteristics, a prediction error series with a similar tendency is generated.

[0152] According to this embodiment, the generation of a forecast error sequence reflects not only the interrelationships between time points, but also the magnitude of the time-to-time variability of various error factors and the interrelationships between how the behavior of one factor affects the behavior of other factors. Therefore, even if the forecast errors as a whole have similar characteristics, it is possible to generate a more realistic forecast error sequence that reflects the differences in the influence of each factor.

[0153] Therefore, a more realistic forecast scenario can be generated, and the facility operation plan generated by the supply and demand management facility operation system 9 using such a forecast scenario will contribute to more effective maintenance of supply and demand balance.

[0154] The above description of the first embodiment can be summarized as follows: Note that the following summary may include supplementary explanations and explanations of modifications to the description of the first embodiment.

[0155] The power operation system 3 includes a communication device 34 (an example of an interface device), a storage device 35 (an example of a storage device) in which multiple pieces of observation data and multiple pieces of external data are stored, and a CPU (an example of a processor) connected to the communication device 34 and the storage device 35. A supply and demand management equipment operation system 9 is connected to the communication device 34. The supply and demand management equipment operation system 9 is a system (typically a computer system) that controls one or more pieces of power equipment, including at least one of a generator, a power storage device, and a switch, based on an operation plan created based on a forecast scenario. The multiple pieces of observation data stored in the storage device 35 may be all or part of the observation data group 521A, and the multiple pieces of external data stored in the storage device 35 may be all or part of the external data group 721A. Each piece of observation data is data including values ​​observed at one or more times for the forecast target during a period corresponding to the observation data (i.e., time series data of the observation values ​​for the forecast target). For each of the plurality of external data, the external data is data including values ​​at one or more times for factors that may affect the value (observed value or predicted value) of the prediction target during the period corresponding to the external data (i.e., time series data of values ​​for the factors). The plurality of external data includes past external data, which is external data including values ​​at one or more times in the past, and future external data, which is external data including values ​​at one or more times in the future.

[0156] The CPU 31 inputs data including one or more future external data into a prediction model for the prediction target, and calculates prediction data including predicted values ​​for the prediction target at each of one or more future times (i.e., time series data of the predicted values ​​for the prediction target).

[0157] The CPU 31 calculates the error distribution of each of the multiple error factors defined for the prediction data, and calculates one or more error data from the error distribution. Examples of the multiple error factors are the above-mentioned model dependence, data accuracy dependence, and data granularity dependence, and the multiple error factors may include other error factors instead of or in addition to at least one of the model dependence, data accuracy dependence, and data granularity dependence. An "error factor" is a factor that causes an error in the predicted value for the prediction target. For each error factor, the error data (error system) is data containing an error (error time series) determined from the error distribution of the error factor for each of one or more times.

[0158] For each of one or more error datasets, the CPU 31 generates a forecast scenario, which is data in which forecast error data according to the error dataset is reflected in the forecast data. For each of one or more error datasets, the error dataset includes one error data for each of multiple error factors, and the forecast error data includes errors in the forecast value of the forecast target for each of one or more future times (for example, it is composite data of the multiple error data that make up the error dataset), and the forecast scenario includes values ​​in which the errors are reflected in the forecast value for each of the one or more future times. The CPU 31 outputs one or more forecast scenarios to the supply and demand management facility operation system 9.

[0159] In this way, multiple error factors of the forecast data are determined, and the error dataset, which is the basis of the forecast error data reflected in the forecast data to generate the forecast scenario, is composed of error data calculated from the error distribution determined for each of the multiple error factors. This improves the accuracy of the forecast scenario, thereby improving the results of processing in the supply and demand management facility operation system 9 (an example of a system that performs processing based on the forecast scenario).

[0160] The multiple error factors may include an error factor called data accuracy dependence. Data accuracy dependence may refer to dependence on data accuracy, in which values ​​contained in future external data used to calculate forecast data for a forecast target are values ​​for future times, and therefore the future external data itself may contain errors. In cases where the forecast target is electricity demand, future external data, for example, data containing "forecasted" (or "predicted") values ​​of weather, etc., is input to the forecast model. In other words, the data input to the forecast model in inference may itself contain errors. In light of this, the error factor called data accuracy dependence is taken into consideration, and error data that forms the basis of forecast error data is calculated from the error distribution for such error factors. This can be expected to further improve the accuracy of forecast scenarios.

[0161] The CPU 31 may input two or more pieces of past external data in the observation sample 3533A of the input sample (an example of a second dataset) into a prediction model constructed for a prediction target using the model sample 3532A (an example of a first dataset), and calculate a first model output, which is data including values ​​for each of the two or more pieces of past external data in the observation sample 3533A at one or more times for the prediction target. The model sample 3532A may be a dataset composed of two or more pieces of observation data corresponding to a first period and two or more pieces of past external data. The input sample may be a dataset composed of a forecast sample 3534A (an example of two or more pieces of future external data corresponding to a second period different from the first period) and an observation sample 3533A (an example of two or more pieces of past external data corresponding to the second period, which corresponds to the correct solutions for the two or more future external data in the forecast sample 3534A). The CPU 31 may input two or more future external data sets in the forecast sample 3534A into a prediction model constructed for the prediction target using the model sample 3532A, and calculate a second model output for each of the two or more future external data sets, the second model output being data including values ​​for the prediction target at one or more times. The CPU 31 may calculate the difference between the first model output and the second model output for each pair of past external data and future external data sets in the input sample. Based on the variability of multiple differences obtained for multiple pairs of past external data and future external data sets in the input sample, the CPU 31 may determine an error distribution dependent on data accuracy. In this manner, the data accuracy-dependent error distribution is obtained. Note that both the first period and the second period may be continuous periods or a set of discrete times. In an example of the former period, the first period may be August, and the second period may be September. In an example of the latter period, the first period may be some days of the week in August, and the second period may be the remaining days of the week in the same month.

[0162] The multiple error factors may further include an error factor called model dependence and an error factor called data granularity dependence. Model dependence may refer to dependence on model accuracy, in which the accuracy of a predictive model for a prediction target varies depending on the data used for construction, even if the construction method (e.g., the algorithm for constructing the predictive model) is the same. Data granularity dependence may refer to dependence on data granularity, in which the accuracy of the model output of a predictive model varies depending on the aggregation granularity of external data input into the predictive model for a prediction target (e.g., how many prefectures or cities' data are used to make a prediction for a certain area). By determining such error factors, improvement in the accuracy of prediction scenarios is expected.

[0163] The CPU 31 may input two or more pieces of future external data in a forecast sample 3544A of an input sample (an example of a second dataset) into a prediction model constructed for a prediction target using a model sample 3542A (an example of a first dataset), and calculate a model output, which is data including values ​​for the prediction target at one or more times, for each of the two or more pieces of future external data. The model sample 3542A may be a dataset consisting of two or more pieces of observation data corresponding to a first period and two or more pieces of past external data (which may be the same as or different from the model sample 3532A). The forecast sample 3544A of the input sample may be a dataset consisting of two or more pieces of observation data corresponding to a second period different from the first period and two or more pieces of future external data. The CPU 31 may calculate, for each of two or more pieces of observation data in an observation sample 3543A of the input sample, a difference between the observation data and the model output obtained by inputting the future external data corresponding to the observation data, and determine an overall error distribution based on the variance of the difference. The CPU 31 may calculate the data granularity-dependent error distribution by subtracting the error distribution of error factors other than data granularity-dependent (for example, the model-dependent error distribution and the data accuracy-dependent error distribution) from the overall error distribution. Data granularity-dependent elements are likely to be more numerous or complex than other error factors, but because the data granularity-dependent error distribution can be obtained by subtracting the error distribution of other error factors from the overall error distribution in this way (for example, the residual obtained by such subtraction can be determined as the data granularity-dependent error distribution), the data granularity-dependent error distribution can be determined efficiently.

[0164] The CPU 31 may calculate, for each of the two or more factors, a relative importance, which is the relative degree of influence on the occurrence of data granularity-dependent errors, based on the data granularity-dependent error distribution and external data (e.g., past external data) for each of the two or more factors in the input sample (e.g., the observation sample 3543A). The CPU 31 may divide the data granularity-dependent error distribution into two or more error distributions based on the calculated relative importance of each of the previous or subsequent factors. The CPU 31 may calculate one or more error data from the error distribution for each of the two or more divided error distributions for data granularity-dependent errors. This is expected to improve the accuracy of the forecast error data based on the error data for each error distribution, and thus to improve the accuracy of the forecast scenario.

[0165] The CPU 31 may construct a prediction model for a prediction target using two or more different datasets. For each of the two or more different datasets, the dataset may be composed of one or more observation data and, for each of the one or more observation data, past external data corresponding to the same period as the observation data. The CPU 31 may calculate two or more model outputs by inputting the same one or more future external data into two or more prediction models constructed using the two or more different datasets. Each of the two or more model outputs may be data including values ​​for the prediction target at one or more times. The CPU 31 may determine a model-dependent error distribution based on the variability of the difference between the model outputs. In this manner, the model-dependent error distribution is obtained.

[0166] For each error distribution, the CPU 31 may determine the error at each of one or more future times based on the correlation between error factors including the error factors corresponding to the error distribution and the correlation between the error factors and the times. For each error distribution, the error data may include the error determined for each of one or more future times. This is expected to result in highly accurate error data.

[0167] With respect to the above-described embodiment, the following can be stated as an example.

[0168] When generating a simulation value of the prediction error (generating prediction error data), by taking into account the correlation between times, it is possible to generate a simulation value that takes into account the dynamic characteristics of the prediction error. However, the prediction error includes a training error component of the prediction model itself, an error component caused by the tendency of the fine-grained prediction target that cannot be reflected in the model output due to the aggregation granularity of the input data, and a prediction error component caused by the error of the input data itself to the prediction model. In the above-mentioned embodiment, these various error factors are referred to as model dependence, data granularity dependence, and data accuracy dependence, respectively.

[0169] Each error factor may have a different error distribution and interrelationships, and in addition, the error distributions and interrelationships may be non-stationary.

[0170] Therefore, in the above-described embodiment, even when the various error factors that make up the forecast error have a non-stationary error distribution or interrelationships, the dynamic characteristics of the forecast error that take non-stationarity into account are modeled, and realistic simulation values ​​of the forecast error are generated, thereby generating realistic forecast scenarios and improving the effectiveness of maintaining the supply and demand balance through operational plans.

[0171] Specifically, for example, the power operation system 3 includes a data prediction unit 351, a model-dependent error calculation unit 352, a data-precision-dependent error calculation unit 353, a data-granularity-dependent error calculation unit 354, and a prediction scenario generation unit 355. The data prediction unit 351 calculates prediction data for a prediction target period using external data (e.g., future external data) and a prediction model. The model-dependent error calculation unit 352 constructs multiple prediction models using two or more different data sets (one data set includes observed data and past external data), and calculates the variance of the model output obtained by inputting the same one or more future external data sets into each prediction model as an error distribution of the model-dependent error. The data precision-dependent error calculation unit 353 obtains a model sample 3532A (observation data and past external data corresponding to a first period) and an input sample (observation sample 3533A and forecast sample 3534A) by dividing multiple observation data and multiple external data, constructs a prediction model using the model sample 3532A, calculates model output (output of the prediction model constructed using the model sample 3532A) using the observation sample 3533A (past external data corresponding to a second period different from the first period) and the forecast sample 3534A (past and future data corresponding to the second period) as input, and calculates the variance in the difference in the model output as a data precision-dependent error distribution. The data granularity-dependent error calculation unit 354 divides multiple pieces of observation data and multiple pieces of external data to obtain a model sample 3542A (observation data and past external data corresponding to period A) and an input sample (observation sample 3543A and forecast sample 3544A), calculates the overall prediction error variance from the model output (output of a prediction model constructed using the model sample 3542A) using the forecast sample 3544A (observation data and future external data corresponding to period B different from period A) as input, then calculates the residual obtained by subtracting the model-dependent error distribution and the data-accuracy-dependent error distribution from the overall prediction error variance (overall error distribution) as a data granularity-dependent error distribution, determines the relative importance of the external data items (factors) that make up the data granularity-dependent error distribution using the external data of the input sample (e.g., past external data), and divides the data granularity-dependent error distribution by relative importance to calculate the data granularity-dependent error distribution for each external data item (factor).The prediction scenario generation unit 355 models the interrelationships (e.g., correlations) between factors and the interrelationships between factors and between time for each error factor (error distribution) from the error distributions of various error factors, including model-dependent error distribution, data accuracy-dependent error distribution, and data granularity-dependent error distribution, generates one or more error series (error time series data) that follow the model, and generates a prediction scenario by combining the prediction error data (error data based on the error data of each of the multiple error factors) with prediction data.

[0172] The power operation system 3 estimates the correlation between error factors from the error series for each error factor estimated from past forecast error data, and uses the error distribution (error variation) for each error factor during the forecast period to simulate an error series that takes into account the non-stationary error distribution and interrelationships for each error factor. Therefore, realistic simulation values ​​for the error series can be generated, and a realistic forecast scenario can be generated. In this way, it is possible to generate a realistic forecast scenario that reflects the non-stationary error distribution and interrelationships for each error factor, thereby improving the effectiveness of maintaining the supply and demand balance through operational planning.

[0173] The second to ninth embodiments will be described below, focusing mainly on the differences from the first embodiment, and explanations of the commonalities with the first embodiment will be omitted or simplified. (2) Second embodiment (addition of classification-type scenario selection processing in the forecast scenario generation unit)

[0174] In the first embodiment, all forecast scenarios generated by the forecast scenario generation unit 355 are output to the supply and demand management equipment operation system 9, but in this embodiment, a forecast scenario is selected, and the selected forecast scenario is output to the supply and demand management equipment operation system 9. In other words, the number of forecast scenarios output to the supply and demand management equipment operation system 9 can be appropriately reduced.

[0175] This embodiment will be specifically described with reference to FIGS.

[0176] 17, the forecast scenario generation unit 355 has a classification scenario selection unit 3555. The classification scenario selection unit 3555 selects some of the scenarios generated by the scenario synthesis unit 3554, and outputs the selected scenarios to the supply and demand management equipment operation system 9.

[0177] As shown in FIG. 18, the classification-type scenario selection unit 3555 includes a scenario classification unit 35551 and a scenario reduction unit 35552.

[0178] The scenario classification unit 35551 receives as input multiple forecast scenarios generated by the scenario synthesis unit 3554 and groups scenarios with similar features. Any index may be used as the feature, such as the value of one or more specific points on the series, statistics such as the mean, median, maximum, minimum, variance, or standard deviation of the series values, or a value converted into an index representing periodicity using a Fourier transform or wavelet transform. The similarity between the features may be a general distance measure that satisfies the distance axioms, such as Euclidean distance, Mahalanobis distance, Manhattan distance, Chebyshev distance, or Minkowski distance, or a similarity measure such as cosine similarity. The actual classification process into groups may be performed using a hierarchical clustering method such as Ward's method, single-link method, complete-link method, or centroid method, a clustering method as a neighborhood optimization method such as k-means, EM algorithm, or spectral clustering, or a clustering method as a discrimination boundary optimization method such as unsupervised SVM (Support Vector Machine), VQ algorithm, or SOM (Self-Organizing Maps).

[0179] The scenario reduction unit 35552 extracts one or more scenarios from each group of scenarios classified by the scenario classification unit 35551. Scenarios for each group may be extracted by selecting any number of scenarios for each group, or by extracting a representative scenario by calculating the average or median of one or more scenarios belonging to each group.

[0180] The forecast scenarios with the number reduced by the above processing are output to the supply and demand management facility operation system 9.

[0181] In this way, the classification-type scenario selection unit 3555 may classify each forecast scenario into one or more groups according to the similarity of the feature quantities for each forecast scenario, and select some forecast scenarios from each group. The one or more forecast scenarios output may be selected from the one or more forecast scenarios generated. By using the classification-type scenario selection unit 3555, similar forecast scenarios are thinned out, making it possible to reduce processing loads related to equipment operation while maintaining the effectiveness of supply and demand management in the supply and demand management equipment operation system 9. (3) Third embodiment (addition of comparative scenario selection process in the forecast scenario generation unit)

[0182] In this embodiment, as in the second embodiment, a forecast scenario is selected, and the selected forecast scenario is output to the supply and demand management facility operation system 9.

[0183] This embodiment will be specifically described with reference to FIGS.

[0184] 19, the forecast scenario generation unit 355 has a comparative scenario selection unit 3557. The comparative scenario selection unit 3557 outputs a forecast scenario that has been reduced ex post using the generated scenario 3556A and data observed after the start time of the forecast period to the supply and demand management equipment operation system 9. The generated scenario 3556A is data, and may include, for example, for each generated forecast scenario, a breakdown of the forecast scenario (e.g., forecast data, forecast error data, and a breakdown of the forecast error data (error data for each error factor (each error distribution))).

[0185] As shown in FIG. 20, the comparison-type scenario selection unit 3557 includes an output difference calculation unit 35571, a scenario comparison unit 35572, and a scenario reduction unit 35573.

[0186] The output difference calculation unit 35571 receives the prediction model and prediction data of the prediction target from the data prediction unit 351, and acquires the observed values ​​of the external data used as the forecast value from the external data group 721A used to calculate the prediction data. The output difference calculation unit 35571 calculates the model output by inputting the observed values ​​of the external data observed from the start time of the prediction target period to the start time of the processing of the comparative scenario selection unit 3557 into the prediction model of the data prediction unit 351. The output difference calculation unit 35571 obtains the value of the actually occurring data accuracy-dependent error by comparing the model output (data output from the prediction model using observed data for a specified period as input) with the received prediction data (data output from the prediction model using forecast data for the same specified period as input). At this time, the data accuracy-dependent error for each external data item using the forecast value may be calculated using a method similar to that of the data accuracy-dependent error calculation unit 353.

[0187] The scenario comparison unit 35572 compares the actual data-precision-dependent error calculated by the output difference calculation unit 35571 with the series values ​​related to the data-precision-dependent error in the forecast error series for each factor included in the generated scenario 3556A output by the error series generation unit 3553, and calculates the similarity of the feature quantity between the actual data-precision-dependent error and each error series. The feature quantity may be any index, such as the value of one or more specific points on the series, statistical quantities such as the mean, median, maximum, minimum, variance, and standard deviation of the series values, or values ​​converted into an index representing periodicity using Fourier transform or wavelet transform. The similarity of the feature quantities may be a general distance measure that satisfies the distance axioms, such as Euclidean distance, Mahalanobis distance, Manhattan distance, Chebyshev distance, or Minkowski distance, or a similarity such as cosine similarity. Note that when calculating the similarity, it is not necessary to calculate it for all items using forecast values ​​included in the external data; only specific items may be used.

[0188] The scenario reduction unit 35573 extracts, from the generated scenario 3556A, one or more data-accuracy-dependent errors that have been determined to be similar using the calculated similarity and the forecast scenarios associated with each of them. Of the forecast scenarios generated by the scenario synthesis unit 3554, the scenario reduction unit 35573 outputs only the extracted forecast scenario to the supply and demand management equipment operation system 9.

[0189] In this way, the comparative scenario selection unit 3557 may calculate data-accuracy-dependent error data based on the difference between the first forecast data (forecast data from the data prediction unit 351) and the second forecast data obtained by inputting past external data, including values ​​observed from the start of the forecast period until a certain time has elapsed, into a forecast model (a forecast model constructed by the data prediction unit 351), and select a forecast scenario that includes data-accuracy-dependent error data similar to the calculated error data as a component. The one or more forecast scenarios output may be selected from one or more generated forecast scenarios. By using the comparative scenario selection unit 3557, only scenarios with a high probability of realization can be output to the supply and demand management equipment operation system 9 while referring to actually observed data from the generated forecast scenarios. This makes it possible to improve the effectiveness of supply and demand management while reducing processing loads related to equipment operation. (4) Fourth embodiment (addition of scenario expansion processing in the forecast scenario generation unit)

[0190] In this embodiment, the forecast scenario generated by the forecast scenario generation unit 355 is expanded, and the expanded scenario (increased scenario) is output to the supply and demand management facility operation system 9.

[0191] This embodiment will be specifically described with reference to FIGS.

[0192] 21, the forecast scenario generation unit 355 has a scenario expansion unit 3558. The scenario expansion unit 3558 outputs a forecast scenario that has been expanded ex post using the generated scenario 3556A and data observed after the start time of the forecast period to the supply and demand management equipment operation system 9.

[0193] As shown in FIG. 22, the scenario expanding unit 3558 includes an output difference calculating unit 35581, a scenario comparing unit 35582, and a new scenario generating unit 35583.

[0194] The output difference calculation unit 35581 receives the prediction model and prediction data of the prediction target from the data prediction unit 351, and acquires the observed values ​​of the external data used as forecast values ​​in the calculation of the prediction data. The output difference calculation unit 35581 calculates the model output by inputting the observed values ​​of the external data observed from the start time of the prediction target period to the start time of processing by the scenario expansion unit 3558 into the prediction model of the data prediction unit 351. The output difference calculation unit 35581 obtains the value of the actually occurring data accuracy-dependent error by comparing the model output (data output from the prediction model using observed data for a specified period as input) with the received prediction data (data output from the prediction model using forecast data for the same specified period as input). At this time, the data accuracy-dependent error for each external data item using the forecast value may be calculated using a method similar to that of the data accuracy-dependent error calculation unit 353.

[0195] The scenario comparison unit 35582 compares the actual data-precision-dependent error calculated by the output difference calculation unit 35581 with the series values ​​related to the data-precision-dependent error in the forecast error series for each factor included in the generated scenario 3556A that is the output of the error series generation unit 3553, and calculates the similarity of the feature value with the actual data-precision-dependent error for each error series. Note that when calculating the similarity, it is not necessarily necessary to calculate it for all items that use forecast values ​​included in the external data, and it is also possible to use only specific items.

[0196] The new scenario generation unit 35583 generates one or more new forecast error series for each factor using one or more data-precision-dependent errors determined to be similar using the calculated similarity. Specifically, for example, a method similar to the series of processes from the correlation estimation unit 3551 to the error series generation unit 3553 of the forecast scenario generation unit 355 may be used, but when a variance-covariance matrix is ​​used in the error initial value generation unit 3552 and the error series generation unit 3553, the value of the data-precision-dependent error may be fixed by the value of the data included in the generated scenario 3556A, and each matrix may be converted.

[0197] The scenario expanding unit 3558 outputs one or more forecast scenarios newly generated by the above processing to the supply and demand management facility operation system 9.

[0198] In this way, the scenario expansion unit 3558 may calculate data-accuracy-dependent error data based on the difference between the first forecast data (forecast data from the data prediction unit 351) and the second forecast data obtained by inputting past external data, including values ​​observed from the start of the forecast period until a certain time has elapsed, into a forecast model (a forecast model constructed by the data prediction unit 351), select a forecast scenario that includes data-accuracy-dependent error data similar to the calculated data-accuracy-dependent error data as a component, and regenerate a forecast scenario based on the calculated data-accuracy-dependent error data, error data of other error factors (e.g., model-dependent and data granularity-dependent) in the selected forecast scenario, and the first forecast data, thereby increasing the number of forecast scenarios. Using the scenario expansion unit 3558, it is possible to identify a scenario with a high probability of realization while referring to actually observed data, and to operate the supply and demand management equipment operation system 9 by taking into account the value transition patterns of the scenario in more detail, thereby improving the effectiveness of supply and demand management. (5) Fifth embodiment (use of feedback data from supply and demand management equipment operation system)

[0199] In this embodiment, data generated by the supply and demand management facility operation system 9 is fed back to the power operation system 3.

[0200] This embodiment will be specifically described with reference to Fig. 23. As shown in Fig. 23, data representing the processing results in the supply and demand management equipment operation system 9 is fed back to the power operation system 3. The power operation system 3 uses the feedback data.

[0201] As a specific example, the power operation system 3 generates forecast data and forecast scenarios for the power market price, and the supply and demand management equipment operation system 9 plans and manages market transactions. If the market trading plan is changed in accordance with the market price forecast scenario, the market price itself may fluctuate as a result. As shown in Figure 23, by feeding back data representing changes to the trading plan formulated by the supply and demand management equipment operation system 9 to the power operation system 3, the data prediction unit 351 can input at least a portion of the data and newly calculate forecast data for the market price, and the power operation system 3 can generate a new forecast scenario for the market price based on the forecast data.

[0202] As described above, the data prediction unit 351 may acquire data (e.g., data representing the results of processing based on a prediction scenario) output from the supply and demand management equipment operation system 9 (an example of a computer system external to the power operation system 3 that uses a prediction scenario), recalculate the prediction data using the acquired data and a prediction model, and the prediction scenario generation unit 355 may regenerate a prediction scenario based on the prediction data. By repeating the processing of the prediction scenario and feedback of the processing results in the supply and demand management equipment operation system 9, and the regeneration of the prediction scenario in the power operation system 3, it is possible to maintain the operation plan for the supply and demand management equipment as an appropriate plan that follows fluctuations in the value of the prediction target. (6) Sixth embodiment (changing the timing for calculating the variations of various factors)

[0203] In this embodiment, the execution timing of the process for calculating error variations for each factor and the process for generating prediction errors is asynchronous with the execution timing of the data prediction unit 351. Specifically, for example, the execution of each of the series of processes from the correlation estimation unit 3551 to the error series generation unit 3553 of the model-dependent error calculation unit 352, the data-precision-dependent error calculation unit 353, the data granularity-dependent error 354, and the prediction scenario generation unit 355 may be triggered by the passage of a certain period arbitrarily set by the user 2, the execution of another process a certain number of times, or a certain amount of increase in the observation data group 521A or the external data group 721A. When the execution of processes is asynchronous with the data prediction unit 351, the prediction scenario generation unit 355 uses the execution result of the most recent execution.

[0204] By changing the execution timing of each process in the series of processes from the model-dependent error calculation unit 352, the data precision-dependent error calculation unit 353, the data granularity-dependent error 354, and the correlation estimation unit 3551 to the error series generation unit 3553 of the forecast scenario generation unit 355, the processing load can be reduced. (7) Seventh embodiment (use of prediction scenarios to calculate other prediction data)

[0205] In this embodiment, the forecast scenario generated by the power operation system 3 is also used in other types of systems (for example, computer systems that forecast data) that are different from the supply and demand management facility operation system 9.

[0206] This embodiment will be described in detail with reference to Fig. 24. As shown in Fig. 24, a prediction scenario is also output to the other type of system 11. In this embodiment, there may be no output from the power operation system 3 to the supply and demand management equipment operation system 9, or data representing the results in the other type of system 11 may be output to the supply and demand management equipment operation system 9. The output of the other type of system 11 may be a single predicted value, or may be output in the form of one or more prediction scenarios, as with the power operation system 3. Alternatively, values ​​from the other type of system 11 may be further used to calculate other prediction data or scenarios, and this may be output to the supply and demand management equipment operation system 9.

[0207] As a specific example, the power operation system 3 generates forecast data and forecast scenarios for power demand, and the other system 11 generates forecast data or forecast scenarios for the market price of power. Typically, the market price of power is correlated with power demand, so the power demand for a certain future period is often used when predicting future market prices. However, since it is difficult to accurately grasp the value of future power demand in advance, it is common to use forecast values. Therefore, by using the values ​​generated in the form of forecast scenarios to predict the market price, the power operation system 3 can generate forecast data or forecast scenarios for market prices that take into account the forecast error of power demand in advance.

[0208] When operating other types of equipment using other types of forecast data (for example, forecast data on the market price of electricity), it is possible to take into account in advance the forecast errors that may occur in the forecast values ​​of that data, making it possible to formulate operational plans that are robust against forecast errors. (8) Eighth embodiment (warning to user 2 according to the magnitude of error)

[0209] In this embodiment, the output content of output device 33 is changed according to the output result of power operation system 3.

[0210] This embodiment will be described in detail with reference to Fig. 25. As shown in Fig. 25, power operation system 3 has output determination unit 357. Output determination unit 357 outputs data to supply and demand management equipment operation system 9 and output device 33.

[0211] The output determination unit 357 receives as input the forecast scenarios generated by the forecast scenario generation unit 355 and determines the output content of the output device 33. As a specific example, if the values ​​of a certain percentage of the forecast scenarios exceed a certain threshold, the output determination unit 357 outputs a warning to the user 2 to the output device 33. If the supply and demand management equipment operation system 9 also has a device equivalent to the output device 33, a similar warning may be output by that device of the supply and demand management equipment operation system 9. For example, the warning may be output when a certain percentage or more of the forecast scenarios include scenarios in which the peak value of electricity demand exceeds a certain threshold or scenarios in which the market price of electricity rises above a certain price.

[0212] The input to the output determination unit 357 is not limited to the forecast scenario generated by the forecast scenario generation unit 355, but may also include values ​​of error variance for each factor calculated by the model-dependent error calculation unit 352, data precision-dependent error calculation unit 353, and data granularity-dependent error 354, and a warning may be displayed if the error variance exceeds a threshold value.

[0213] In this way, the output determination unit 357 may use the forecast scenario or error data of at least one error factor (e.g., model dependence, data accuracy dependence, or data granularity dependence) to output a signal (e.g., data indicating the relaxation of constraints) that relaxes constraints on the operation plan generated by the supply and demand management equipment operation system 9 based on the forecast scenario. By using the output determination unit 357, it is possible to make a decision to temporarily relax constraints on the equipment operated by the supply and demand management equipment operation system 9 in emergencies such as power shortages, thereby improving the effectiveness of maintaining supply and demand balance in emergencies. Constraints include, for example, constraints on the number of starts and stops of a generator within a certain period of time and constraints on the duration of the stoppage. (9) Ninth embodiment (initial value generation process in error series generation unit)

[0214] In this embodiment, the processing of the error initial value generating unit 3552 is skipped, and the error series generating unit 3553 generates values ​​sequentially from the initial value of the error series.

[0215] In this case, the error value for each factor actually obtained immediately before the forecast period is used as the error value for the time before the forecast error generation, and the mean and variance-covariance matrix are calculated as in S3553E, and the initial value of the error for each factor is generated as in S3553F.

[0216] By generating the initial error values ​​for each factor through the above process, it becomes possible to generate more realistic error series and forecast scenarios that maintain the time correlation with past data.

[0217] Although several embodiments of the present invention have been described above, these are merely illustrative examples of the present invention and are not intended to limit the scope of the present invention to these embodiments. The present invention can be embodied in various other forms. For example, a form in which two or more of the above-described embodiments are used in combination may be adopted. For example, the power operation system 3 may be called a data prediction system because it calculates prediction data. [Explanation of symbols]

[0218] 1...Data processing system, 3...Power operation system

Claims

1. an interface device connected to a supply and demand management equipment operation system that controls one or more electric power facilities including at least one of a generator, a power storage facility, and a switchgear based on an operation plan created based on a forecast scenario; a storage device for storing a plurality of pieces of observation data and a plurality of pieces of external data; a processor connected to the interface device and the storage device; Equipped with For each of the plurality of observation data, the observation data is data including values ​​observed at one or more times for a prediction target during a period corresponding to the observation data, For each of the plurality of external data, the external data is data including values ​​at one or more times of factors that may affect the value of the prediction target during a period corresponding to the external data, the plurality of external data include past external data, which is external data including values ​​at one or more past times, and future external data, which is external data including values ​​at one or more future times, the processor inputs data including one or more future external data into a prediction model for the prediction target, thereby calculating prediction data including a predicted value for the prediction target at each of one or more future times; the processor calculates an error distribution of each of the plurality of error factors determined for the prediction data, and calculates one or more error data from the error distribution; For each error factor, the error data is data including an error determined from the error distribution of the error factor for each of one or more times; the processor generates, for each of one or more error data sets, a prediction scenario that is data in which prediction error data according to the error data set is reflected in the prediction data; For each of the one or more error data sets: the error data set includes one error data for each of the plurality of error factors; the prediction error data includes an error in the predicted value of the prediction target for each of one or more future times, the prediction scenario includes, for each of the one or more future times, a value in which an error is reflected in a predicted value; The processor outputs one or more forecast scenarios to the supply and demand management equipment operation system. Power operation system.

2. the plurality of error factors includes an error factor of data accuracy dependence, The dependence on data accuracy is a dependence on data accuracy in that values ​​included in future external data used to calculate prediction data for the prediction target are values ​​for a future time, and therefore the future external data itself may contain errors. The power operation system according to claim 1 .

3. the processor inputs two or more pieces of past external data from a second dataset into a prediction model constructed for the prediction target using a first dataset, and calculates a first model output, for each of the two or more pieces of past external data, which is data including values ​​for the prediction target at each of one or more times; the first data set is a data set including two or more pieces of observation data corresponding to a first period and two or more pieces of past external data; the second data set is a data set including two or more pieces of future external data corresponding to a second period different from the first period, and two or more pieces of past external data corresponding to the second period and corresponding to the correct answers of the two or more pieces of future external data, the processor inputs the two or more future external data items included in the second dataset into a prediction model constructed for the prediction target using the first dataset, and calculates, for each of the two or more future external data items, a second model output that is data including values ​​for the prediction target at each of one or more times; the processor calculates a difference between a first model output and a second model output for each pair of past external data and future external data in the second data set, and determines the data precision-dependent error distribution based on the variance of the difference; The power operation system according to claim 2 .

4. the plurality of error factors further include a model-dependent error factor and a data granularity-dependent error factor; The model dependence is a dependence on model accuracy in that the accuracy of the model differs when the data used for construction is different even if the construction method of the prediction model for the prediction target is the same, The data granularity dependency is a dependency on data granularity in which the accuracy of the model output of the prediction model varies depending on the aggregation granularity of external data input to the prediction model for the prediction target. The power operation system according to claim 2 .

5. The processor inputs two or more pieces of future external data from the second dataset into a prediction model constructed for the prediction target using the first dataset, and calculates, for each of the two or more pieces of future external data, a model output that is data including values ​​for the prediction target at each of one or more times; the first data set is a data set including two or more pieces of observation data corresponding to a first period and two or more pieces of past external data; the second dataset is a dataset composed of two or more pieces of observation data and two or more pieces of future external data corresponding to a second period different from the first period, the processor calculates, for each of the two or more observation data in the second data set, a difference between the observation data and a model output obtained using future external data corresponding to the observation data as input, and determines an overall error distribution based on the variance of the difference; the processor calculates the data granularity-dependent error distribution by subtracting the error distribution of error factors other than the data granularity-dependent error distribution from the overall error distribution. The power operation system according to claim 4 .

6. the processor calculates, based on the data granularity-dependent error distribution and external data for each of two or more factors in the second data set, a relative importance, which is a relative degree of influence on occurrence of the data granularity-dependent error, for each of the two or more factors; the processor divides the data granularity-dependent error distribution into two or more error distributions based on the calculated relative importance of each of the two or more factors; the processor calculates one or more error data from the error distribution for each of the two or more divided error distributions with respect to the data granularity dependency; The power operation system according to claim 5 .

7. The processor constructs a prediction model for the prediction target using each of two or more different datasets; For each of the two or more different data sets, the data set is composed of one or more pieces of observation data and, for each of the one or more pieces of observation data, past external data corresponding to the same period as the observation data; the processor calculates two or more model outputs by inputting the same one or more future external data into two or more predictive models constructed using the two or more different datasets; Each of the two or more model outputs is data including values ​​of the prediction target at one or more times, the processor determines the model-dependent error distribution based on the variability of model output differences. The power operation system according to claim 4 .

8. The processor determines, for each error distribution, an error at each of the one or more future times based on interrelationships between error factors including the error factors corresponding to the error distribution and interrelationships between the error factors and the times; for each error distribution, for each of the one or more error data, the error data includes an error determined for each of the one or more future times; The power operation system according to claim 1 .

9. the processor classifies each prediction scenario into one or more groups according to the similarity of the feature quantities for each prediction scenario, and selects a portion of the prediction scenarios from each group; the one or more forecast scenarios that are output are selected from the one or more generated forecast scenarios; The power operation system according to claim 1 .

10. The processor calculates data accuracy-dependent error data based on a difference between the first prediction data, which is the prediction data, and second prediction data obtained by inputting past external data including values ​​observed until a certain time has elapsed from the start time of a prediction period into the prediction model, and selects a prediction scenario including data accuracy-dependent error data similar to the calculated error data as a component; the one or more forecast scenarios that are output are selected from the one or more generated forecast scenarios; The power operation system according to claim 2 .

11. the processor calculates data accuracy-dependent error data based on the difference between the first prediction data, which is the prediction data, and second prediction data obtained by inputting past external data including values ​​observed from the start time of the prediction period until a certain time has elapsed from the start time of the prediction period into the prediction model, selects a prediction scenario that includes data accuracy-dependent error data similar to the calculated error data as a component, and increases the number of prediction scenarios by regenerating a prediction scenario based on the calculated data accuracy-dependent error data, error data of other error factors in the selected prediction scenario, and the first prediction data; The power operation system according to claim 2 .

12. the processor acquires data output from the supply and demand management equipment operation system, re-calculates forecast data using the acquired data and the forecast model, and re-generates a forecast scenario based on the forecast data; The power operation system according to claim 1 .

13. The processor outputs a signal to relax constraints on the operation plan using the forecast scenario or error data of at least one error factor. The power operation system according to claim 1 .

14. a computer stores a plurality of observation data and a plurality of external data; For each of the plurality of observation data, the observation data is data including values ​​observed at one or more times for a prediction target during a period corresponding to the observation data, For each of the plurality of external data, the external data is data including values ​​at one or more times of factors that may affect the value of the prediction target during a period corresponding to the external data, the plurality of external data include past external data, which is external data including values ​​at one or more past times, and future external data, which is external data including values ​​at one or more future times, a computer inputs data including one or more future external data into a prediction model for the prediction target, thereby calculating prediction data including a predicted value for the prediction target at each of one or more future times; a computer calculates an error distribution of each of the plurality of error factors of the prediction data, and calculates one or more error data from the error distribution; For each error factor, the error data is data including an error determined from the error distribution of the error factor for each of one or more times; a computer generates, for each of one or more error datasets, a prediction scenario that is data in which prediction error data according to the error dataset is reflected in the prediction data; For each of the one or more error data sets: the error data set includes one error data for each of the plurality of error factors; the prediction error data includes an error in the predicted value of the prediction target for each of one or more future times, the prediction scenario includes, for each of the one or more future times, a value in which an error is reflected in a predicted value; a computer outputs one or more forecast scenarios to a supply and demand management equipment operation system that controls one or more electric power facilities, including at least one of a generator, a power storage facility, and a switchgear, based on an operation plan created based on the forecast scenarios; Electric power operation method.

15. a supply and demand management equipment operation system that controls one or more electric power facilities including at least one of a generator, a power storage facility, and a switchgear; A power operation system that stores multiple observation data and multiple external data. Equipped with For each of the plurality of observation data, the observation data is data including values ​​observed at one or more times for a prediction target during a period corresponding to the observation data, For each of the plurality of external data, the external data is data including values ​​at one or more times of factors that may affect the value of the prediction target during a period corresponding to the external data, the plurality of external data include past external data, which is external data including values ​​at one or more past times, and future external data, which is external data including values ​​at one or more future times, The power operation system calculates prediction data including a predicted value for the prediction target at each of one or more future times by inputting data including one or more future external data into a prediction model for the prediction target; The power operation system calculates an error distribution of each of a plurality of error factors determined in the prediction data, and calculates one or more error data from the error distribution; For each error factor, the error data is data including an error determined from the error distribution of the error factor for each of one or more times; The power operation system generates, for each of one or more error data sets, a forecast scenario that is data in which forecast error data according to the error data set is reflected in the forecast data; For each of the one or more error data sets: the error data set includes one error data for each of the plurality of error factors; the prediction error data includes an error in the predicted value of the prediction target for each of one or more future times, the prediction scenario includes, for each of the one or more future times, a value in which an error is reflected in a predicted value; The power operation system outputs one or more forecast scenarios to the supply and demand management equipment operation system, the supply and demand management equipment operation system generates an operation plan based on at least one of the one or more forecast scenarios, and controls the one or more power equipment based on the operation plan. Data processing system.

Images