Information processing apparatus, information processing method, and computer program

Abstract

To provide an information processing apparatus, an information processing method, and a computer program for estimating the amount of water flowing into a facility.SOLUTION: An information processing apparatus according to the present embodiment comprises a processing unit that, on the basis of actual result data of the amount of water flowing into a facility and prediction data of meteorological quantity in a past period, generates a first model including an objective variable related to the amount of water flowing into the facility at a target time and a plurality of first explanatory variables related to prediction values of the meteorological quantities in a plurality of first periods earlier than the target time, and on the basis of the first model and the prediction data of the meteorological quantity in a future period later than the past period, estimates the amount of water flowing into the facility at the target time.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present embodiment relates to an information processing apparatus, an information processing method, and a computer program.

Background Art

[0002] Power generation businesses are obliged to generate the same amount of electricity as the planned value at the same time, such as accurately generating the amount of electricity sold. In renewable energy power sources such as solar power generation, wind power generation, and hydroelectric power generation, under the FIT (Feed-in Tariff) system, the obligation of generating the same amount as the planned value at the same time was waived by the grid operator's full purchase. After the end of FIT or in the non-FIT situation, power generation businesses must submit their own power generation and sales plans and achieve the same amount as the planned value at the same time. Therefore, power generation businesses need to accurately predict the power generation amount at the planned time in order to fulfill the obligation of generating the same amount as the planned value at the same time.

[0003] Among renewable energy power sources, a hydroelectric power plant has a mechanism that converts the potential energy of water into the power generation amount. In a hydroelectric power plant, the power generation amount is determined by the inflow rate to the turbine that rotates the hydroelectric generator. At this time, the inflow rate is affected by weather such as precipitation and snowmelt, and the power generation amount changes accordingly. Since the power generation amount depends on future weather, it is considered difficult to accurately predict the power generation amount of hydroelectric power generation.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] The present embodiment provides an information processing apparatus, an information processing method, and a computer program for estimating the inflow rate of water into a facility.

Means for Solving the Problems

[0006] The information processing apparatus according to the present embodiment generates a first model including an objective variable related to the water inflow amount to the facility at the target time, and a plurality of first explanatory variables related to the predicted values of the weather amount in a plurality of first periods before the target time, based on the actual data of the water inflow amount to the facility and the predicted data of the weather amount in the past period, and estimates the water inflow amount to the facility at the target time based on the first model and the predicted data of the weather amount in the future period after the past period, and includes a processing unit.

Brief Description of the Drawings

[0007]

Figure 1

Figure 2

Figure 3A

Figure 3B

Figure 3C

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Embodiments for Carrying Out the Invention

[0008] Hereinafter, this embodiment will be described with reference to the drawings. FIG. 1 is an overall configuration diagram of a hydroelectric power prediction device as an information processing device according to this embodiment.

[0009] The hydroelectric power prediction device in FIG. 1 includes a learning processing unit 100 and a prediction processing unit 200. The learning processing unit 100 generates an inflow model (first model) for estimating the inflow and a power generation conversion model (second model) for estimating the power generation from the inflow by machine learning. The prediction processing unit 200 predicts the inflow at the target date and time (target time) using the inflow model, and estimates the power generation at the target time using the predicted inflow by the power generation conversion model. The target time may be in seconds, minutes, hours, or days. Predicting means estimating the value of a future time based on the current time. Hereinafter, the case of predicting the inflow and the power generation is shown, but this embodiment can also estimate the value of a past time. In this case, the prediction of the inflow and the prediction of the power generation in the following description can also be read as the estimation of the inflow and the estimation of the power generation. Hereinafter, the learning processing unit 100 and the prediction processing unit 200 will be described in detail.

[0010] The learning processing unit 100 includes a past power generation amount achievement data storage unit 110, a past inflow amount achievement data storage unit 120, a past upstream discharge amount data storage unit 130, a past weather prediction data storage unit 140, an inflow amount model learning unit 150, and a power generation amount conversion model learning unit 160. The inflow amount model learning unit 150 and the power generation amount conversion model learning unit 160 correspond to the processing units that perform the processing related to the present embodiment.

[0011] The past power generation amount achievement data storage unit 110 stores past power generation amount achievement data, which is the past power generation amount achievement data of a hydroelectric power plant (power generation facility). In the past power generation amount achievement data, for one or more hydroelectric power plants, the past power generation amount is stored in time units such as days, hours, or minutes. The power generation amount is measured by a measuring instrument of the hydroelectric power plant.

[0012] The past inflow amount achievement data storage unit 120 stores past inflow amount achievement data, which is the past inflow amount achievement data of a hydroelectric power plant. In the past inflow amount achievement data, for one or more hydroelectric power plants, the past inflow amount is stored in time units such as days, hours, or minutes. The inflow amount is measured by a measuring instrument of the hydroelectric power plant.

[0013] The past upstream discharge amount data storage unit 130 stores past upstream discharge amount data, which is the past actual data of the amount of water discharged (upstream discharge amount) from a water storage device such as a dam on the upstream side of a hydroelectric power plant. In the past upstream discharge amount data, for one or more hydroelectric power plants, the past upstream discharge amount is stored in time units such as days, hours, or minutes. When there are multiple dams or the like upstream, the upstream discharge amount for each dam or the like is stored. The upstream discharge amount is measured by a measuring instrument of the hydroelectric power plant.

[0014] Note that a configuration in which the past upstream discharge amount data is not used by the inflow amount model learning unit 150 is also possible. In that case, the past upstream discharge amount data storage unit 130 is unnecessary. For example, when there is no dam or the like upstream of the hydroelectric power plant, or when it is impossible to obtain the past upstream discharge amount data, the past upstream discharge amount data may not be used by the inflow amount model learning unit 150.

[0015] The past weather prediction data storage unit 140 stores past weather prediction data, which is at least weather prediction data for a past period in the target area. The target area is an area where the weather is expected to affect the inflow to the hydroelectric power plant, for example, an area including the upstream river connected to the hydroelectric power plant and its vicinity. In the past weather prediction data, at least predicted values of past weather amounts in the target area are stored for each time at predetermined time intervals. The predetermined time may be a time unit such as 1 hour or 3 hours, a day unit such as 1 day or 10 days, or other units such as minute units.

[0016] The weather prediction data is, for example, data obtained by setting representative points (for example, the center of the mesh) for each mesh obtained by dividing the area within Japan into a mesh pattern and predicting the weather amounts at each point. In the past weather prediction data storage unit 140, at least past prediction data of the weather amounts at each point included in the target area is stored. Examples of weather amounts include solar radiation amount, rainfall amount, snowfall amount, snow depth, temperature, humidity, etc. The prediction of weather amounts is performed by a weather prediction server managed by the Japan Meteorological Agency or other companies based on data observed by a large number of weather observation stations arranged in Japan by the Japan Meteorological Agency.

[0017] The inflow model learning unit 150 reads the past inflow actual data, past upstream discharge data, and past weather prediction data from the storage units 120 to 140 respectively, and generates an inflow model by machine learning. Details will be described later.

[0018] The power generation amount conversion model learning unit 160 reads the past power generation amount actual data and past inflow actual data from the storage units 110 and 120, and generates a power generation amount conversion model by machine learning. Details will be described later.

[0019] Next, the prediction processing unit 200 includes a weather prediction data storage unit 210, an upstream flow rate data storage unit 220, an inflow amount actual data storage unit 230, an inflow amount prediction unit 250, and a power generation amount prediction unit 260. The inflow amount prediction unit 250 and the power generation amount prediction unit 260 correspond to the processing units that perform the processing related to the present embodiment. The processing units (inflow amount model learning unit 150, power generation amount conversion model learning unit 160) of the learning processing unit 100 and the processing units (inflow amount prediction unit 250, power generation amount prediction unit 260) of the prediction processing unit 200 may be physically the same circuit (such as a processor) or separate circuits.

[0020] The weather prediction data storage unit 210 acquires at least the real-time weather prediction data in the target area from an external weather prediction server via a communication unit (see the communication device in FIG. 18) and stores it internally. The weather prediction data is weather prediction data of the weather amount in a future period after the past weather prediction data. The data items included in the weather prediction data may be the same as those in the past weather prediction data. The acquisition frequency of the weather prediction data may be the data update frequency in the weather prediction server, once every V (V is an integer of 1 or more) days, once every U (U is an integer of 1 or more) hours, or the like. The weather prediction data includes prediction values (weather prediction values) of the weather amount at predetermined time intervals within a predetermined prediction period (for example, 39 hours, 78 hours, etc.) at the time of weather prediction. The predetermined time may be a time unit such as 1 hour, 3 hours, a day unit such as 1 day, 10 days, or other units such as minutes. Depending on the prediction period and acquisition frequency of the weather prediction data, the prediction time of the weather prediction data acquired this time may overlap with the prediction period of the weather prediction data acquired last time. In that case, the previous data may be updated with the latest weather prediction data.

[0021] The upstream discharge data storage unit 220 acquires, in real time, the actual performance data (upstream discharge data) of the discharge at the upstream of the hydropower plant from the measuring instruments of the hydropower plant or the server that manages the measurement data of the measuring instruments for one or more hydropower plants via the communication unit (refer to the communication device in FIG. 18), and stores the acquired data internally. The upstream discharge data is the latest upstream discharge data acquired after the past upstream discharge data. Note that a configuration in which the upstream discharge data is not used by the inflow prediction unit 250 is also possible, and in that case, the upstream discharge data storage unit 220 is unnecessary. For example, when the past upstream discharge data is not used in creating the inflow model in the inflow model learning unit 150, when there is no dam or the like upstream of the hydropower plant, or when it is impossible to obtain the upstream discharge data, the upstream discharge data may not be used by the inflow prediction unit 250. The data items included in the upstream discharge data may be the same as those in the past upstream discharge data.

[0022] The actual inflow data storage unit 230 acquires, in real time, the actual performance data (actual inflow data) of the inflow at the hydropower plant from the measuring instruments of the hydropower plant or the server that manages the measurement data of the measuring instruments for one or more hydropower plants via the communication unit (refer to the communication device in FIG. 18), and stores the acquired data internally. The data items included in the actual inflow data may be the same as those in the past actual inflow data. The actual inflow data is the latest actual inflow data acquired after the past actual inflow data.

[0023] The inflow prediction unit 250 reads out the weather prediction data, the upstream discharge data, and the actual inflow data from the storage units 210 to 230 respectively, and also acquires the learned inflow model from the inflow model learning unit 150. The inflow prediction unit 250 performs calculations with the data read out from the storage units 210 to 230 as input to the acquired inflow model, and obtains a predicted value of the inflow as the model output.

[0024] The power generation prediction unit 260 acquires the predicted value of the inflow volume from the inflow volume prediction unit 250, and acquires the power generation conversion model from the power generation conversion model learning unit 160. The power generation prediction unit 260 performs calculations using the predicted value of the inflow volume as an input to the power generation conversion model, and acquires the predicted value of the power generation amount that is the model output. The power generation prediction unit 260 outputs the predicted value of the power generation amount to an external server.

[0025] The external server is, for example, a market bidding plan creation server that creates a bidding plan for the power trading market based on the predicted value of the power generation amount of the renewable energy of the power generation operator (for example, the predicted value of the power generation amount of each of hydropower, solar power generation, and wind power generation). In this embodiment, since the power generation amount of hydropower, which has been conventionally considered difficult to predict, can be predicted with high accuracy, the possibility of fulfilling the obligation of the planned value at the same time and in the same amount can be increased. The output destination server is not limited to the market bidding plan creation server, and may be, for example, a server that analyzes the trend of the transition of the predicted value of the power generation amount for each hydropower plant. When it can be determined using this server that the predicted value of the power generation amount of a certain hydropower plant is sufficiently low compared to the power generation capacity, it can be determined that there is room for an increase in the inflow volume, and measures such as increasing the upstream discharge flow of this hydropower plant can be taken.

[0026] FIG. 2 is a block diagram of the inflow volume model learning unit 150. The inflow volume model learning unit 150 includes a weather quantity extraction unit 1001, a time delay pattern creation processing unit 1002, a machine learning unit 1003, an inflow volume model update unit 1004, and an inflow volume model storage unit 1005.

[0027] The meteorological quantity extraction unit 1001 extracts data of predetermined meteorological quantities that affect the inflow from past meteorological prediction data for each point included in the target area. The predetermined meteorological quantities are, for example, precipitation and snowfall. For example, in addition to rainfall flowing directly into rivers, it also flows out into rivers through the ground surface or underground. When rainfall flows directly into a river, the rainfall is quickly reflected in the inflow, but when rainfall flows into a river through the ground surface or underground, it takes time to be reflected in the inflow. Also, since there is a time lag from when snow falls on the ground surface until it melts and flows out into a river through the ground surface or underground, it also takes time for the snowfall to be reflected in the inflow, and this time is assumed to be longer than in the case of rainfall. Precipitation and snowfall are examples of meteorological quantities, and there are also air temperature, solar radiation, snow depth, humidity, etc. One or any combination of meteorological quantities such as precipitation, snowfall, air temperature, solar radiation, snow depth, humidity, etc. can be used. For example, the higher the air temperature and the greater the solar radiation, the faster the snow melts and flows into the river, so it is quickly reflected in the inflow.

[0028] The time lag pattern creation processing unit 1002 sets the time after a predetermined time from the time of prediction (referred to as the prediction start time A) as the target time (referred to as the prediction target time B). A plurality of periods are set before the prediction target time B, and the values of meteorological quantities in each period are calculated or acquired. The plurality of periods each have a different length, and a frame indicating the time length (size) of each period is called an average window. Here, the plurality of periods are periods going back from the prediction target time, and the average of the meteorological quantities included in each period (average window) is calculated. The plurality of periods may be included between the prediction start time A and the prediction target time B, or one or more of the plurality of periods may include times before the prediction start time A (may straddle the prediction start time A). The average is an example, and other statistical values such as the median may be calculated.

[0029] When it is desired to perform predictions for a plurality of prediction target times at the same prediction start time A, the prediction target time B can be gradually shifted backward in time by gradually increasing the above-mentioned predetermined time by one unit time, and the same processing can be performed. In this case, the average window of each size can be shifted backward in time by one unit time step by step, and the average of the meteorological quantities included in the average window of each size can be calculated.

[0030] Further, the time lag pattern creation processing unit 1002 sets a plurality of periods (average windows) for the upstream flow rate as well as for the meteorological amount based on the past upstream flow rate data, and calculates the average of the upstream flow rate for each period.

[0031] The average window for each size corresponds to a time lag pattern in which a predetermined meteorological amount (rainfall, snowfall, etc.) affects the inflow amount.

[0032] FIG. 3A shows an example in which a plurality of periods (average windows) P1 to PN that gradually increase in length are set so as to go back from the prediction target time B a predetermined time after the prediction start time A, and the average is calculated for each of the average windows P1 to PN. The plurality of average windows are of different sizes, and in this example, the longest period PN coincides with the predetermined time. The size of each average window may be, for example, in units of 3 hours such as 3 hours, 6 hours, 9 hours, or any other arbitrary time unit, or in units of 1 month such as 1 month, 2 months, 3 months, or any arbitrary month unit, or a time determined in other units. In the example of the figure, a period is set within the range from the prediction start time A to the prediction target time B, but some periods may include a period before the prediction start time A or may straddle the prediction start time A.

[0033] FIG. 3B shows an example in which some periods include a period in the past from the prediction start time A. The periods P1 to PN are the same as those in FIG. 3A, but a period PN_1 is added, and the period PN_1 includes a period in the past from the prediction start time A.

[0034] In the examples of FIGS. 3A and 3B, the plurality of periods (average windows) are set so that some of them overlap with each other, but the present invention is not limited to this example.

[0035] FIG. 3C shows an example in which a plurality of periods P1' to PN' are discretely set so as not to overlap. As another example, only some of the periods may overlap with each other, and the remaining periods may not overlap with any other period.

[0036] Once a plurality of average values corresponding to a plurality of average windows are obtained, the prediction start time A is changed (for example, advanced by one unit of time. The prediction target time B is also shifted by the same time unit), and the process of calculating the average with the same plurality of average windows is performed. By repeating this, the number of data sets required for learning can be obtained. One data set includes a plurality of average values corresponding to a plurality of average windows. Note that the prediction start time A and the prediction target time B during learning are past times included in the respective past data of the storage units 120 to 140.

[0037] Figure 4(A) shows an example of the averaging process. In the example of Figure 4(A), when the prediction target time B is 10:00 on a certain past day (referred to as the D1st day), the prediction start time A is set to 11:00 on the D2nd day, which is 24 hours before that time. A plurality of periods (average windows) of different sizes are set within these 24 hours, and the average is calculated for each. Thereby, one data set is obtained. Figure 4(B) shows an example of performing the same calculation by advancing the prediction start time A and the prediction target time B in Figure 4(A) by one unit of time each (while maintaining the time interval between the prediction start time A and the prediction target time B) in order to increase the data sets for learning.

[0038] Based on the plurality of data sets created by the time delay pattern creation processing unit 1002 and the past inflow volume actual data, the machine learning unit 1003 generates an inflow volume model by machine learning. A regression model is generated with the average value of the meteorological quantity for each meteorological quantity and each average window size as an explanatory variable, and further with the average value of the inflow volume for each average window size of the inflow volume actual data (described later) as an explanatory variable, and the inflow volume into the facility as the target variable. In this embodiment, the facility is shown as a hydroelectric power facility (turbine), but it is not limited to this. The type of regression model is not limited to a specific one such as linear regression, multiple regression, neural network, or logistic regression.

[0039] Here, an example of simultaneously performing explanatory variable selection and model generation using Lasso regression (Least absolute shrinkage and selection operator) will be described. The Lasso regression equation (model type) is shown in the following equation (1). When past upstream flow rate data is not used, the term related to the upstream flow rate may be removed from equation (1).

[0040]

Number

[0041] RF i , Snow j , Release h , Inflow k Each corresponds to an explanatory variable and there are as many as the number of average windows in one dataset respectively. The average window and its number may be different or the same for each explanatory variable.

[0042] When predicting the inflow at multiple prediction target times (for example, 10 o'clock, 11 o'clock, 12 o'clock tomorrow, etc.), a model may be generated for each prediction target time.

[0043] Here, an example of calculating the average of the inflow volume actual results is shown. FIG. 5 is a diagram for explaining an example of calculating the average of the inflow volume actual results. The average of the inflow volume actual results is calculated, for example, for a period (for example, 24 hours, 48 hours, etc.) from the prediction start time A back to the time C, using periods (average windows) Q1 to QN of a plurality of sizes. The calculation method of the average may be the same as in the case of the above meteorological quantity. Note that the method of setting the average window can have various variations as in the case of the meteorological quantity.

[0044] FIG. 6 shows an example of the average processing of the inflow volume actual results. In this example, the 11th hour of the D2nd day is set as the prediction start time A, and the previous 24 hours are set as the average period.

[0045] In this example, the machine learning unit 1003 creates a data set of the average values of the inflow volume actual results, but a configuration in which the data set of the average values of the inflow volume actual results is also calculated by the time lag pattern creation processing unit 1002 is also possible.

[0046] The machine learning unit 1003 optimizes the model of formula (1) by L1 regularization. The optimization is performed by minimizing or quasi-minimizing an evaluation function including a loss function and an L1 regularization term. As a result, the values of each w and β are obtained. At this time, some w become zero (0), which means that the explanatory variable corresponding to the w was not selected (the term including the w was deleted). That is, by not including explanatory variables that are not important for prediction in the model, explanatory variables of the time lag (average window) suitable for prediction remain. The explanatory variable with a large w value has a large influence on the inflow volume at the prediction target time. In the case of the example of the average window in FIG. 3A, the fact that w of the explanatory variable with a small window size is large means that the meteorological quantity of the explanatory variable affects the inflow volume in a short time, and when w of the explanatory variable with a large window size is large, it means that the meteorological quantity of the explanatory variable affects the inflow volume over time. Note that, as the loss function, for example, the mean squared error function between the value (predicted value) of formula (1) and the actual value can be used. The L1 regularization term includes, for example, the sum of the absolute values of each w.

[0047] The machine learning unit 1003 provides the inflow prediction model, which is a model generated by machine learning, to the inflow model update unit 1004. Specifically, the inflow prediction model corresponds to the formula (1) after obtaining w and β, with the terms containing the explanatory variables where w is 0 removed.

[0048] The inflow model update unit 1004 stores the inflow prediction model provided by the machine learning unit 1003 in the inflow model storage unit 1005 and outputs the inflow model to the prediction processing unit 200. However, the inflow model learning may be repeated (for example, periodically once a day) in response to the update of the storage units 120 to 140. When the inflow prediction model is generated for the second time or later, the inflow model update unit 1004 updates the inflow model in the inflow model storage unit 1005 with the inflow prediction model provided by the machine learning unit 1003 and outputs the updated inflow model to the prediction processing unit 200. The output inflow prediction model is received by the prediction processing unit 200 and used in the inflow prediction unit 250 in the prediction processing unit 200.

[0049] When at least any one of the storage units 120 to 140 is updated, the learning processing unit 100 may perform the learning of the inflow model. For example, once a day, the storage units 120 to 140 are updated, and each time they are updated, the learning (update) of the inflow model may be performed.

[0050] FIG. 7 is a block diagram of the power generation conversion model learning unit 160. The power generation conversion model learning unit 160 includes a water intake upper limit parameter optimization unit 2001, a water intake upper limit processing unit 2002, a conversion efficiency parameter optimization unit 2003, a power generation conversion model update unit 2004, and a power generation conversion model storage unit 2005.

[0051] The water intake upper limit parameter optimization unit 2001 optimizes or determines the water intake upper limit parameter from the past power generation performance data and the past inflow volume performance data. In a hydropower plant (power generation facility), among the inflowing water, the water exceeding the upper limit volume is discharged into the river or discarded, etc., and the water below the upper limit volume is used for power generation. The value of this upper limit volume is obtained as the water intake upper limit parameter. For example, by analyzing the relationship between the past inflow volume performance data and the past power generation performance data, the value of the inflow volume at which the power generation volume does not increase even when the inflow volume increases is determined as the water intake upper limit parameter. The upper limit volume is basically a fixed value determined by the configuration of the power generation facility, but there may also be a power generation facility whose upper limit volume can be adjusted. The water intake upper limit parameter is provided to the water intake upper limit processing unit 2002 and the power generation volume conversion model update unit 2004.

[0052] The water intake upper limit processing unit 2002 reads out the value at each time from the past inflow volume performance data, determines whether this value exceeds the water intake upper limit parameter, and if it exceeds, provides the value of the water intake upper limit parameter as the inflow volume to the conversion efficiency parameter optimization unit 2003. If it does not exceed the water intake upper limit parameter, the read value is provided as the inflow volume to the conversion efficiency parameter optimization unit 2003.

[0053] Based on the pair of the value of the inflow volume acquired from the water intake upper limit processing unit 2002 and the power generation performance value at the time corresponding to the inflow volume in the past power generation performance data, the conversion efficiency parameter optimization unit 2003 optimizes the power generation conversion efficiency parameter of the power generation volume conversion model.

[0054] The formula of the power generation volume conversion model is shown in the following formula (2).

[0055]

Equation

[0056] The effective head h is the remainder obtained by subtracting the lost head from the total head, and this acts on the water turbine to play a role in power generation. The value of h is given in advance. The power generation conversion efficiency usually falls within the range of about 60 to 85%.

[0057] The value of the above inflow amount corresponds to Q, and the power generation actual result value corresponds to P. Using a plurality of pairs of the above inflow amount value and the power generation actual result value (data for a plurality of time periods), for example, the formula (2) is optimized by the least squares method or the like. Thereby, the power generation conversion efficiency parameter η is determined.

[0058] The power generation amount conversion model update unit 2004 acquires a power generation amount conversion model (optimized power generation amount conversion model) including the power generation conversion efficiency parameter optimized by the conversion efficiency parameter optimization unit 2003, and also acquires the water intake upper limit parameter from the water intake upper limit parameter optimization unit 2001. The power generation amount conversion model update unit 2004 stores these models and parameters in the power generation amount conversion model storage unit 2005 and provides them to the prediction processing unit 200. Note that instead of the power generation amount conversion model, the optimized power generation conversion efficiency parameter may be stored in the power generation amount conversion model storage unit 2005 and provided to the prediction processing unit 200. In this case, it is assumed that the type data of the power generation amount conversion model is held in the prediction processing unit 200.

[0059] The optimization of the water intake upper limit parameter and the power generation conversion efficiency parameter (optimization of the power generation amount conversion model) may be repeated (for example, periodically once a day) in response to the update of the storage units 110 and 120. When the optimization of the water intake upper limit parameter and the power generation conversion efficiency parameter is the second time or later, the power generation amount conversion model update unit 2004 updates the models and parameters in the power generation amount conversion model storage unit 2005 with these models and parameters, and outputs the updated models and the updated parameters to the prediction processing unit 200. The output models and parameters are received by the prediction processing unit 200 and used in the power generation amount prediction unit 260 in the prediction processing unit 200.

[0060] The inflow prediction unit 250 in the prediction processing unit 200 acquires the inflow model generated by the inflow model learning unit 150 of the learning processing unit 100. The inflow prediction unit 250 reads out weather prediction data, upstream discharge data, and inflow actual data from the storage units 210 to 230. The inflow prediction unit 250 calculates the value of the explanatory variable (the average value based on the average window corresponding to the explanatory variable) included in the inflow model for the prediction target time (the prediction target time used in the prediction process is distinguished from the prediction target time used in the learning process and is denoted as the prediction target time T). Then, by giving the calculated value of the explanatory variable as the input to the inflow model, the predicted value of the inflow (inflow predicted value) at the prediction target time T is calculated. For example, if the prediction target time T is 10 o'clock, the predicted value of the inflow at 10 o'clock is calculated. Note that the prediction target time T is a future time after the current time (processing execution time), and is a time after a certain period from the current time. The said certain period is the same length as the period from the prediction start time A to the prediction target time B used in the learning of the inflow model by the learning processing unit 100.

[0061] The inflow prediction unit 250 sends the calculated inflow predicted value to the power generation prediction unit 260.

[0062] The power generation prediction unit 260 receives the inflow predicted value from the inflow prediction unit 250, and also acquires the water intake upper limit parameter and the power generation conversion model from the power generation conversion model learning unit 160 of the learning processing unit 100. The power generation prediction unit 260 predicts the power generation amount obtained by the water of the amount indicated by the inflow predicted value flowing into the hydroelectric power facility based on the water intake upper limit parameter and the power generation conversion model. Since the time required for power generation in hydroelectric power generation is short, the obtained power generation amount can be regarded as the power generation amount at the prediction target time T. The details of the power generation prediction unit 260 will be described with reference to FIG. 8.

[0063] FIG. 8 is a block diagram of the power generation prediction unit 260. The power generation prediction unit 260 includes a water intake upper limit processing unit 3001 and a power generation conversion unit 3002.

[0064] The water intake upper limit processing unit 3001 acquires the water intake upper limit parameter from the power generation amount conversion model learning unit 160, and determines whether the predicted inflow amount from the inflow amount prediction unit 250 exceeds the value of the water intake upper limit parameter. If the predicted inflow amount exceeds the value of the water intake upper limit parameter, the predicted inflow amount is corrected to the value of the water intake upper limit parameter, and the corrected predicted inflow amount is sent to the power generation amount conversion unit 3002. If the predicted inflow amount does not exceed the value of the water intake upper limit parameter, the predicted inflow amount is directly sent to the power generation amount conversion unit 3002. The process of suppressing the predicted inflow amount sent by the water intake upper limit processing unit 3001 to the power generation amount conversion unit 3002 below the water intake upper limit parameter in this way is called the water intake upper limit processing.

[0065] The power generation amount conversion unit 3002 calculates a predicted power generation amount by using the predicted inflow amount acquired from the water intake upper limit processing unit 3001 as an input to the power generation amount conversion model from the power generation amount conversion model learning unit 160. The power generation amount conversion model learning unit 160 outputs the calculated predicted power generation amount, that is, the predicted power generation amount corresponding to the prediction target time T.

[0066] Hereinafter, with reference to FIGS. 9 to 12, the overall operation of the hydroelectric power prediction device in FIG. 1 will be described.

[0067] FIG. 9 is a diagram for explaining the overall operation flow of the hydroelectric power prediction device in FIG. 1. FIG. 9(A) is an operation flowchart of the learning processing unit 100, and FIG. 9(B) is an operation flowchart of the prediction processing unit 200.

[0068] As shown in FIG. 9(A), the learning processing unit 100 acquires past weather prediction data and stores it in the storage unit 140 (S101), and acquires past inflow volume actual data, past upstream discharge data, and past power generation actual data as past actual data and stores them in the storage units 110 to 130 (S102). Then, based on the data stored in the storage units 110 to 140, the learning processing unit 100 performs an inflow volume model learning process (S103) for generating an inflow volume model by machine learning with an arbitrary trigger, determines a water intake upper limit parameter, and performs a power generation conversion model learning process including generating a power generation conversion model (optimizing a power generation conversion efficiency parameter) (S104). The arbitrary trigger includes an instruction from the operator of this device, the arrival of a pre-specified time, the update of any of the storage units 110 to 130, etc.

[0069] As shown in FIG. 9(B), the prediction processing unit 200 acquires weather prediction data from a weather prediction server and stores it in the storage unit 210 (S201), and also acquires upstream discharge data and inflow volume actual data and stores them in the storage units 220 and 230 (S202). When the prediction processing unit 200 detects that the prediction start time (processing execution time) has arrived based on a clock (not shown) provided in the hydroelectric power prediction device or an external clock, it predicts the inflow volume at the prediction target time T based on the data in the storage units 210 to 230 and the inflow volume model generated in the inflow volume model learning process (S203). Subsequently, the prediction processing unit 200 predicts the power generation amount corresponding to the prediction target time T based on the predicted inflow volume, the water intake upper limit parameter acquired in the above learning, and the power generation conversion model, and outputs information indicating the value of the predicted power generation amount (S204).

[0070] FIG. 10 is a flowchart showing the details of the process of step S103 in FIG. 9(A) (inflow volume model learning process).

[0071] In step S111, the weather quantity extraction unit 1001 extracts a weather quantity (for example, rainfall amount, snowfall amount, etc.) determined in advance as a weather quantity that affects the inflow volume among a plurality of weather quantities included in the past weather prediction data.

[0072] In step S112, the time delay pattern creation processing unit 1002 sets a prediction start time A and a prediction target time B after a predetermined time from the prediction start time A, and sets a plurality of periods (average windows of a plurality of sizes) based on the prediction target time B and the prediction start time A. For each period, for each of the extracted meteorological quantities, a plurality of average values are calculated using average windows of a plurality of sizes. As an example, if the meteorological quantities are two, namely rainfall and snowfall, and there are 24 average windows respectively, 24 average values for rainfall and 24 average values for rainfall are obtained. Similarly, for the past upstream flow rate data, a plurality of periods (average windows of a plurality of sizes) are set, and a plurality of average values are calculated. As an example, when the number of average windows is 12, 12 average values are calculated. Thus, one data set is obtained. By repeating the above-described process with the prediction start time A and the prediction target time B each shifted by the same time, the number of data sets required for learning is obtained.

[0073] In step S113, the machine learning unit 1003 generates an inflow rate prediction model by machine learning based on the plurality of data sets acquired in step S112 and the past inflow amount actual data. At this time, as explanatory variables, for each of the above-extracted meteorological quantity and upstream flow rate, a plurality of explanatory variables each representing an average corresponding to a plurality of average windows and a plurality of explanatory variables each representing an average of the inflow amount actual values corresponding to the plurality of average windows (see FIGS. 5 and 6) are used. By performing Lasso regression using the above-described formula (1), it is possible to simultaneously perform the selection of explanatory variables and model generation.

[0074] In step S114, the inflow rate model update unit 1004 stores the inflow rate prediction model generated in step S113 in the inflow rate model storage unit 1005 and outputs it to the prediction processing unit 200. Alternatively, the previously stored model in the inflow rate model storage unit 1005 is updated by the currently generated inflow rate prediction model, and the updated inflow rate prediction model is output to the prediction processing unit 200.

[0075] FIG. 11 is a flowchart showing details of the process of step S104 in FIG. 9(A) (power generation amount conversion model learning process).

[0076] In step S121, the water intake upper limit parameter optimization unit 2001 optimizes or determines the water intake upper limit parameter from the past power generation amount actual data and the past inflow amount actual data.

[0077] In step S122, the water intake upper limit processing unit 2002 reads out the value at each time from the past inflow amount actual data, and restricts this value to be equal to or less than the value of the water intake upper limit parameter. That is, when this value exceeds the water intake upper limit parameter, the value of the water intake upper limit parameter is provided to the conversion efficiency parameter optimization unit 2003 as the inflow amount. When it does not exceed the water intake upper limit parameter, the read value is directly provided to the conversion efficiency parameter optimization unit 2003 as the inflow amount.

[0078] In step S123, based on the pair of the value of the inflow amount acquired from the water intake upper limit processing unit 2002 and the power generation actual value at the time corresponding to the inflow amount in the past power generation amount actual data, the conversion efficiency parameter optimization unit 2003 optimizes the power generation conversion efficiency parameter of the power generation amount conversion model (see Equation (2)). The power generation amount conversion model including the optimized power generation conversion efficiency parameter corresponds to the optimized power generation amount conversion model.

[0079] In step S124, the power generation amount conversion model update unit 2004 acquires the optimized power generation amount conversion model, and also acquires the water intake upper limit parameter from the water intake upper limit parameter optimization unit 2001. The power generation amount conversion model update unit 2004 stores these model and parameter in the power generation amount conversion model storage unit 2005, and provides them to the prediction processing unit 200. When the process of this flowchart is the second time or later, the power generation amount conversion model update unit 2004 updates the previously generated model and parameter in the power generation amount conversion model storage unit 2005 with these model and parameter, and outputs the updated model and the updated parameter to the prediction processing unit 200.

[0080] FIG. 12 is a flowchart showing details of the process (power generation amount prediction process) in step S204 of FIG. 9(B).

[0081] In step S211, the water intake upper limit processing unit 3001 in the power generation amount prediction unit 260 acquires the predicted inflow amount value at the prediction target time T from the inflow amount prediction unit 250, and acquires the water intake upper limit parameter from the power generation amount conversion model learning unit 160.

[0082] In step S212, the water intake upper limit processing unit 3001 performs water intake upper limit processing on the acquired predicted inflow amount value based on the water intake upper limit parameter. That is, when the predicted inflow amount value from the inflow amount prediction unit 250 exceeds the value of the water intake upper limit parameter, the predicted inflow amount value is corrected to the value of the water intake upper limit parameter, and the corrected predicted inflow amount value is sent to the power generation amount conversion unit 3002. When the predicted inflow amount value does not exceed the value of the water intake upper limit parameter, the predicted inflow amount value is sent to the power generation amount conversion unit 3002 as it is.

[0083] In step S213, the power generation amount conversion unit 3002 acquires the power generation amount conversion model from the power generation amount conversion model learning unit 160, and calculates a power generation amount prediction value using the predicted inflow amount value input from the water intake upper limit processing unit 3001 as an input to the power generation amount conversion model.

[0084] In step S214, the power generation amount conversion model learning unit 160 outputs the calculated power generation amount prediction value to an external device.

[0085] This information processing apparatus may visualize the coefficients (weights) of the respective explanatory variables included in the generated inflow amount model and the identification information of the explanatory variables, and present them to the user via a display device. Thereby, the user can easily grasp which explanatory variable affects the inflow amount with what time lag.

[0086] According to this embodiment, by learning the time lag in the reflection of meteorological quantities (such as precipitation and snowfall) on the inflow for each hydroelectric power generation facility or each river, highly accurate power generation prediction can be performed. That is, for meteorological quantities that affect the inflow, a plurality of explanatory variables for the same meteorological quantity are created using a plurality of patterns representing time lags. Machine learning is used to select the explanatory variables (set weights) to obtain an inflow prediction model for power generation prediction. By calculating with the meteorological prediction values for the selected explanatory variables (explanatory variables having the selected time lags) input to this model, highly accurate power generation prediction is realized while reflecting the time lag of the meteorological quantities.

[0087] (Second Embodiment: Another Example of Time Lag Pattern) In the above-described first embodiment, the average of the meteorological prediction values by the average window was used as the time lag pattern of the meteorological quantities related to the influence on the inflow, but another example of the time lag pattern is shown. Hereinafter, the same explanations as those already described will be omitted as appropriate. The block diagram of the information processing apparatus (hydroelectric power generation prediction apparatus) of the second embodiment is the same as that of the first embodiment.

[0088] The time lag pattern creation processing unit 1002 acquires the values at each time (sample time) per unit time (for example, 1 hour) in the period from the prediction start time A to the prediction target time B (referred to as the shift period) for the meteorological quantities for each location extracted by the meteorological quantity extraction unit 1001. Also, the time lag pattern creation processing unit 1002 acquires the values at each time (sample time) per unit time in the shift period for the upstream discharge flow rate as well based on the past upstream discharge flow rate data.

[0089] FIG. 13 shows an example of each time t per unit time in a certain period (shift period) from the prediction start time A to the prediction target time B. Here, x times are set. The number of times may be different or the same for each meteorological quantity and upstream discharge flow rate.

[0090] These values acquired for each time for each of the meteorological quantity and the upstream discharge flow rate are taken as one data set.

[0091] The method for determining the shift period is not limited to the above method, and other methods are also possible, such as setting a fixed period retroactively from a predetermined time or an arbitrary time before the prediction target time B, or setting the period from a time in the past before the prediction start time A to the prediction target time B.

[0092] The prediction start time A and the prediction target time B are each changed by the same amount of time (for example, advanced by 1 unit of time), and the same processing is performed. By repeating this, the required number of data sets for learning can be obtained. Note that the prediction start time A and the prediction target time B during learning are past times included in the past data of the storage units 120 to 140.

[0093] Based on the plurality of data sets created by the time lag pattern creation processing unit 1002 and the past inflow volume actual data, the machine learning unit 1003 generates an inflow volume model by machine learning. Using the values of the meteorological quantity and the upstream discharge at each time in the data set as explanatory variables, and further using the values of the inflow volume actual at each time (described later) as explanatory variables, a regression model is generated to regress the inflow volume prediction value from these explanatory variables. The type of the regression model is not limited to a specific one such as linear regression, multiple regression, neural network, or logistic regression.

[0094] Here, an example of simultaneously performing the selection of explanatory variables and model generation using Lasso regression will be described. The Lasso regression formula (model type) is shown in the following formula (3). When past upstream discharge data is not used, the term related to the upstream discharge may be removed from formula (3).

[0095]

Equation

[0096] RF l , Snow m , Release n , Inflow o correspond to the explanatory variables respectively and each exists for the number of times (sample times) within the shift period. The length of the shift period may be different for each explanatory variable or the same.

[0097] Here, for the value of each time of the actual inflow, for example, taking a certain period (such as 24 hours, etc.) retroactively from the prediction start time A as the shift period, the values of each time within the shift period can be obtained. The method for determining the shift period is not limited to this method, and other methods such as taking a certain period retroactively from the time before a predetermined time of the prediction start time may also be used.

[0098] Through optimization, some w become zero, and the explanatory variables with zero w are excluded. The explanatory variables with large w values mean that they have a large impact on the inflow at the prediction target time. That is, it means that the impact on the inflow appears significantly with a delay of the time difference between the time of that explanatory variable and the prediction target time.

[0099] The processing of the inflow prediction unit 250 is the same except that the definition of the inflow model is different from that of the foregoing embodiments. That is, the inflow prediction unit 250 acquires the inflow model generated by the inflow model learning unit 150 of the learning processing unit 100, and reads out weather prediction data, upstream discharge data, and inflow actual data from the storage units 210 to 230. Then, for the prediction target time (denoted as T), the values of each explanatory variable included in the inflow model (the values of the explanatory variables for each time selected for the weather amount, upstream discharge amount, and inflow amount in the above learning) are acquired. Then, by giving the acquired values of the explanatory variables as the input of the inflow model, the predicted value of the inflow amount at the prediction target time T is calculated. The inflow prediction unit 250 sends the calculated predicted value of the inflow amount to the power generation amount prediction unit 260.

[0100] (Third Embodiment) In the above-described first and second embodiments, the weather amount used as the explanatory variable was the predicted value (past weather prediction data). However, depending on the type of weather variable, it is also possible to use the observed value. For example, when using the snow depth as the weather amount, it takes a longer time until the snow is reflected in the inflow amount to the power generation facility than when rainfall or the like is reflected. Therefore, when the period from the prediction start time to the prediction target time is shorter than the time required until the above snow is reflected, the influence of the melting of the snow on the inflow amount can be appropriately modeled using the observed value. Hereinafter, the third embodiment will be described. The same description as that of the foregoing first and second embodiments will be omitted.

[0101] FIG. 14 is a block diagram of a hydroelectric power prediction device as an information processing device according to the third embodiment. A past weather observation data storage unit 170 for storing past weather observation data is added to the learning processing unit 100 in FIG. 1. Further, a weather observation data storage unit 270 for acquiring and storing real-time weather observation data from a weather observation server is added to the prediction processing unit 200.

[0102] FIG. 15 is a block diagram of the inflow model learning unit 150 according to the third embodiment. The same reference numerals are given to the elements having the same names as those in FIG. 2, and the description will be omitted as appropriate.

[0103] The meteorological quantity extraction unit 1001 extracts meteorological quantities (such as snow depth) to be used in prediction from the past meteorological observation data in the past meteorological observation data storage unit 170.

[0104] The time lag pattern creation processing unit 1002 performs the same processing as in the first embodiment or the second embodiment (averaging by a plurality of average windows or obtaining values at each time within the shift period) on the extracted meteorological quantity (such as snow depth), and includes the result value of the processing in the acquired data set.

[0105] The machine learning unit 1003 generates an inflow model that additionally includes explanatory variables related to the extracted meteorological quantity (such as snow depth) by machine learning.

[0106] The inflow prediction unit 250 of the prediction processing unit 200 calculates the values of the explanatory variables related to the extracted meteorological quantity (such as snow depth) included in the inflow model using the meteorological observation data in the meteorological observation data storage unit 270.

[0107] As described above, according to the present embodiment, the time lag of the reflection of the meteorological quantity on the inflow can be learned using the meteorological observation data. In the case of a meteorological quantity with a large time lag of reflection, highly accurate power generation prediction can be performed.

[0108] (Fourth Embodiment) FIG. 16 is a block diagram of a hydroelectric power generation prediction device as an information processing device according to the fourth embodiment. A power generation amount actual data storage unit 240 is added to the prediction processing unit 200. Elements with the same names as those in FIG. 1 are denoted by the same reference numerals, and the same description is omitted as appropriate.

[0109] The power generation amount actual data storage unit 240 acquires, in real time, actual data (power generation amount actual data) of the power generation amount in the hydroelectric power plant from a measuring instrument of the hydroelectric power plant or a server that manages the measurement data of the measuring instrument via a communication unit (see the communication device in FIG. 18), and stores the acquired data internally. The data items included in the power generation amount actual data may be the same as the past power generation amount actual data.

[0110] After predicting the power generation amount in the same manner as in the first to third embodiments, the power generation amount prediction unit 260 corrects the predicted value as appropriate based on the actual power generation amount data. For example, an error is calculated by subtracting the predicted value of the previous day from the actual value of the previous day for a certain period in the past, the average of this error is calculated, and the predicted power generation amount is corrected by only this average. If the average is positive, the average is added to the predicted power generation amount, and if the average is negative, the average is subtracted from the predicted power generation amount. Thereby, it can be expected to reduce the error of the predicted amount of power generated. The correction method is not limited to this method, and other methods may be used.

[0111] (Fifth Embodiment) The fifth embodiment relates to an aggregation system 400 using the hydroelectric power prediction device according to any one of the first to fourth embodiments.

[0112] FIG. 17 is a schematic block diagram of the aggregation system 400 according to the fifth embodiment.

[0113] The power generation planning unit 410 includes a hydroelectric power prediction device 1, a PV (Photovoltaic) power generation prediction device 2, and a wind power generation prediction device 3. The hydroelectric power prediction device 1 is a hydroelectric power prediction device according to any one of the first to fourth embodiments, and predicts the hydroelectric power generation amount at the prediction target time. The PV power generation prediction device 2 predicts the PV power generation amount at the prediction target time, and the wind power generation prediction device 3 predicts the wind power generation amount at the prediction target time. The PV power generation prediction device 2 and the wind power generation prediction device 3 may perform prediction by any method based on weather prediction data or the like. The power generation planning unit 410 obtains the total power generation amount prediction value at the prediction target time by summing the predicted power generation amounts of the respective prediction devices 1 to 3.

[0114] The market bidding plan creation unit 420 determines the bidding amount for one or more power markets 500 based on the total power generation amount prediction value. Examples of the power market 500 include a one-day market (spot market) where power is traded for the prediction target time on the day before the prediction target time, and a same-day market (time-ahead market) where trading is performed until a certain time before the prediction target time on the day of the prediction target time. There are also other wholesale power trading markets. In addition, there are a capacity market or a supply-demand adjustment market.

[0115] The power market cooperation unit 430 transmits bidding data instructing the bid for the determined bid volume to the power market 500 and acquires the transaction result data from the power market 500.

[0116] The planned value submission unit 460 transmits data regarding the transaction result (assuming the sale of electricity here) and the power generation plan (total power generation forecast value, or power generation volume forecast values by hydropower, PV, and wind power) at the forecast target date and time to the wide-area power operation promotion organization (wide-area organization) 600. The wide-area organization 600 monitors the power supply and demand of each electric utility company that is a member and instructs other members to supply power to members with a deteriorating power supply and demand situation. When a member acquires or generates a transaction result and a power generation plan, it is necessary to submit these transaction results and power generation plans to the wide-area organization 600.

[0117] The adjustable power source operation planning unit 440 creates an operation plan for the adjustable power source 700 such as a storage battery and thermal power generation based on the power generation plan (total power generation forecast value, or power generation volume forecast values by hydropower, PV, and wind power), the transaction result, and the power supply plan for the relative transaction with other predetermined operators. For example, it creates an operation plan to output the insufficient power or charge the surplus power. Note that a part of the above-mentioned total power generation forecast value may be allocated to the power supply for the relative transaction.

[0118] The control unit 450 generates and transmits a control instruction value for avoiding the imbalance between at least one of the operation plan and the power generation plan and the actual power generation to at least one of the adjustable power source 700 and the PV's PCS (power conditioner) 800, thereby remotely controlling the adjustable power source 700 and the PV's PCS 800 in real time. As an example of the control of the PCS 800, control may be performed to suppress the PCS output. As an example of controlling the adjustable power source 700 in real time, control may be performed to output power from the adjustable power source 700 not included in the operation plan. By such an operation of the control unit 450, the simultaneous operation of the planned values of the renewable energy sources is made possible.

[0119] (Hardware Configuration) FIG. 18 shows the hardware configuration of the information processing apparatus according to the above-described embodiment. The information processing apparatus is configured by a computer apparatus 900. The computer apparatus 900 includes a CPU 901, an input interface 902, a display device 903, a communication device 904, a main memory device 905, and an external memory device 906, which are mutually connected by a bus 907.

[0120] The CPU (Central Processing Unit) 901 executes an information processing program, which is a computer program, on the main memory device 905. The information processing program is a program that realizes each of the above-described functional configurations of the information processing apparatus. The information processing program may be realized not by a single program but by a combination of a plurality of programs and scripts. By the CPU 901 executing the information processing program, each functional configuration is realized.

[0121] The input interface 902 is a circuit for inputting operation signals from input devices such as a keyboard, a mouse, and a touch panel into the information processing apparatus. The input interface 902 corresponds to the input unit of the information processing apparatus according to the above-described embodiment.

[0122] The display device 903 displays data output from the information processing apparatus. The display device 903 is, for example, an LCD (Liquid Crystal Display), an organic electroluminescence display, a CRT (Cathode Ray Tube), or a PDP (Plasma Display Panel), but is not limited thereto. The data output from the computer apparatus 900 can be displayed on this display device 903. The display device 903 corresponds to the output unit of the information processing apparatus according to the above-described embodiment.

[0123] The communication device 904 is a circuit for the information processing apparatus to communicate with an external device wirelessly or by wire. Data can be input from the external device via the communication device 904. The data input from the external device can be stored in the main memory device 905 or the external memory device 906. The communication device 904 corresponds to the communication unit of the information processing apparatus according to the above-described embodiment.

[0124] The main memory device 905 stores an information processing program, data necessary for the execution of the information processing program, and data generated by the execution of the information processing program. The information processing program is expanded and executed on the main memory device 905. The main memory device 905 is, for example, RAM, DRAM, SRAM, but is not limited thereto. Each storage unit or database of the information processing device according to the above-described embodiment may be constructed on the main memory device 905.

[0125] The external storage device 906 stores an information processing program, data necessary for the execution of the information processing program, and data generated by the execution of the information processing program. These information processing programs and data are read into the main memory device 905 when the information processing program is executed. The external storage device 906 is, for example, a hard disk, an optical disk, a flash memory, and a magnetic tape, but is not limited thereto. Each storage unit or database of the information processing device according to the above-described embodiment may be constructed on the external storage device 906.

[0126] Note that the information processing program may be pre-installed in the computer device 900, or may be stored in a storage medium such as a CD-ROM. Also, the information processing program may be uploaded on the Internet.

[0127] Also, the information processing device may be configured by a single computer device 900, or may be configured as a system including a plurality of computer devices 900 connected to each other.

[0128] Note that the present invention is not limited to the above-described embodiments as they are, and at the implementation stage, the components can be modified and embodied without departing from the gist thereof. Also, various inventions can be formed by appropriately combining the plurality of components disclosed in the above-described embodiments. Also, for example, a configuration in which some components are deleted from all the components shown in each embodiment is also conceivable. Further, the components described in different embodiments may be appropriately combined.

[0129] This embodiment can also be configured as follows. [Item 1] Based on the actual data of the water inflow rate into the facility and the predicted data of the meteorological quantities in the past period, a first model including an objective variable regarding the water inflow rate into the facility at the target time and a plurality of first explanatory variables regarding the predicted values of the meteorological quantities in a plurality of first periods before the target time is generated, and based on the first model and the predicted data of the meteorological quantities in the future period after the past period, an information processing apparatus including a processing unit that estimates the water inflow rate into the facility at the target time. [Item 2] The plurality of first periods are periods of a plurality of different lengths going back from the target time, and the plurality of first explanatory variables are the averages of the predicted values of the meteorological quantities in the respective plurality of first periods. The information processing apparatus according to Item 1. [Item 3] The plurality of first periods are a plurality of different times before the target time, and the plurality of first explanatory variables are the predicted values of the meteorological quantities at the plurality of different times. The information processing apparatus according to Item 1 or 2. [Item 4] The first model includes a plurality of first weights that are coefficients of the plurality of first explanatory variables, and the processing unit calculates the plurality of first weights by LASSO regression and deletes the terms for which the first weights are zero. The information processing apparatus according to any one of Items 1 to 3. [Item 5] The processing unit generates a second model that estimates the power generation amount in the facility from the water inflow rate into the facility based on the actual data of the power generation amount in the facility and the actual data of the water inflow rate into the facility, and the processing unit estimates the power generation amount in the facility based on the estimated inflow rate and the second model. The information processing apparatus according to any one of Items 1 to 4. [Item 6] The processing unit determines a water intake upper limit parameter regarding the upper limit value of the amount of water that can be processed in the facility based on the actual data of the power generation amount and the actual data of the inflow amount of water into the facility in the facility, When the estimated inflow amount is greater than the water intake upper limit parameter, the processing unit estimates the power generation amount in the facility using the value of the water intake upper limit parameter as the inflow amount. The information processing apparatus according to item 5. [Item 7] The processing unit further generates the first model further including a plurality of second explanatory variables regarding the observed values of the weather amount in a plurality of second periods based on the actual data of the observed values of the weather amount, The plurality of second periods are before the processing execution time for performing the process of estimating the inflow amount, The processing unit acquires the observed data of the weather amount observed after the actual data, and estimates the inflow amount at the target time based on the acquired observed data of the weather amount. The information processing apparatus according to item 5 or 6. [Item 8] The plurality of second periods are periods of a plurality of different lengths going back from the processing execution time, The plurality of second explanatory variables are the averages of the observed values of the weather amount in each of the plurality of second periods. The information processing apparatus according to item 7. [Item 9] The plurality of second periods are a plurality of different times before the processing execution time, The plurality of second explanatory variables are the observed values of the weather amount at the plurality of different times. The information processing apparatus according to item 7. [Item 10] The first model includes a plurality of first weights and a plurality of second weights that are coefficients of the plurality of first explanatory variables and the plurality of second explanatory variables, The processing unit calculates the plurality of first weights and the plurality of second weights by LASSO regression, and deletes the terms where the first weight is zero and the terms where the second weight is zero. The information processing apparatus according to any one of items 7 to 9. [Item 11] Based on the actual discharge data upstream of the facility, the processing unit generates the first model further including a plurality of third explanatory variables related to the discharge amount in a plurality of third periods before the target time. The processing unit acquires the actual discharge data of the latest upstream after the actual data, and estimates the inflow amount at the target time based on the acquired actual discharge data of the latest upstream. The information processing apparatus according to any one of items 5 to 10. [Item 12] The plurality of third periods are periods of a plurality of different lengths going back from the target time. The plurality of third explanatory variables are the averages of the actual values of the discharge amount in each of the plurality of third periods. The information processing apparatus according to item 11. [Item 13] The plurality of third periods are a plurality of different times going back from the target time. The plurality of third explanatory variables are the actual values of the discharge amount at the plurality of different times. The information processing apparatus according to item 11 or 12. [Item 14] The first model includes a plurality of first weights and a plurality of third weights which are the coefficients of the plurality of first explanatory variables and the plurality of third explanatory variables. The processing unit calculates the plurality of first weights and the plurality of third weights by LASSO regression, and deletes the terms with the first weight being zero and the terms with the third weight being zero. The information processing apparatus according to any one of items 11 to 13. [Item 15] Based on the actual inflow data, the processing unit generates the first model further including a plurality of fourth explanatory variables related to the inflow amount in a plurality of fourth periods. The plurality of fourth periods are before the processing execution time for performing the process of estimating the inflow amount. The processing unit acquires the performance data of the latest inflow volume after the performance data, and estimates the inflow volume at the target time based on the acquired performance data of the latest inflow volume. The information processing apparatus according to any one of items 5 to 14. [Item 16] The plurality of fourth periods are periods of a plurality of different lengths retrogressing from the processing execution time. The plurality of fourth explanatory variables are the averages of the performance values of the inflow volume in each of the plurality of fourth periods. The information processing apparatus according to item 15. [Item 17] The plurality of fourth periods are a plurality of different times retrogressing from the processing execution time. The plurality of fourth explanatory variables are the performance values of the inflow volume at the plurality of different times. The information processing apparatus according to item 15 or 16. [Item 18] The first model includes a plurality of first weights and a plurality of fourth weights that are coefficients of the plurality of first explanatory variables and the plurality of fourth explanatory variables. The processing unit calculates the plurality of first weights and the plurality of fourth weights by LASSO regression, and deletes the terms where the first weight is zero and the terms where the fourth weight is zero. The information processing apparatus according to any one of items 15 to 17. [Item 19] The facility is a hydroelectric power facility. The information processing apparatus according to any one of claims 1 to 18. [Item 20] Based on the performance data of the water inflow volume into the facility and the predicted data of the meteorological quantity in the past period, a first model including an objective variable related to the water inflow volume into the facility at the target time and a plurality of first explanatory variables related to the predicted values of the meteorological quantity in a plurality of first periods before the target time is generated, and based on the first model and the predicted data of the meteorological quantity in the future period after the past period, the water inflow volume into the facility at the target time is estimated. Information processing method. [Item 21] Generating a first model including an objective variable related to the water inflow rate to the facility at the target time and a plurality of first explanatory variables related to the predicted values of the weather quantity in a plurality of first periods before the target time, based on the actual data of the water inflow rate to the facility and the predicted data of the weather quantity in the past period; Estimating the water inflow rate to the facility at the target time based on the first model and the predicted data of the weather quantity in the future period after the past period; A computer program for causing a computer to execute the above.

Explanation of Signs

[0130] 1 Hydropower prediction device 2 PV power generation prediction device 3 Wind power generation prediction device 100 Learning processing unit 110 Past power generation amount actual data storage unit 110 Storage unit 120 Past inflow rate actual data storage unit 120 Storage unit 130 Past upstream discharge rate data storage unit 130 Storage unit 140 Past weather prediction data storage unit 140 Storage unit 150 Inflow rate model learning unit 160 Power generation amount conversion model learning unit 170 Past weather observation data storage unit 200 Prediction processing unit 210 Storage unit 210 Weather prediction data storage unit 220 Storage unit 220 Upstream discharge rate data storage unit 230 Storage unit 230 Inflow rate actual data storage unit 240 Power generation amount actual data storage unit 250 Inflow rate prediction unit 260 Power generation amount prediction unit 270 Weather observation data storage unit 400 Aggregation system 410 Power Generation Planning Department 420 Market Bidding Plan Making Department 430 Power Market Linkage Department 440 Adjustable Power Operation Planning Department 450 Control Department 460 Planned Value Submission Department 500 Power Market 600 Power Wide-Area Operation Promotion Institution (Wide-Area Institution) 600 Wide-Area Institution 700 Adjustable Power 900 Computer Device 902 Input Interface 903 Display Device 904 Communication Device 905 Main Memory Device 906 External Memory Device 907 Bus 1001 Meteorological Quantity Extraction Department 1002 Pattern Creation Processing Department 1003 Machine Learning Department 1004 Inflow Model Update Department 1005 Inflow Model Memory Department 2001 Water Intake Limit Parameter Optimization Department 2002 Water Intake Limit Processing Department 2003 Conversion Efficiency Parameter Optimization Department 2004 Power Generation Conversion Model Update Department 2005 Power Generation Conversion Model Memory Department 3001 Water Intake Limit Processing Department 3002 Power Generation Conversion Department

Claims

1. A processing unit that generates a first model including an objective variable related to the water inflow rate into the facility at a target time and a plurality of first explanatory variables related to predicted values of the weather quantity in a plurality of first periods before the target time, based on the actual data of the water inflow rate into the facility and the predicted data of the weather quantity in the past period, and estimates the water inflow rate into the facility at the target time based on the first model and the predicted data of the weather quantity in a future period after the past period. An information processing apparatus comprising the same.

2. The plurality of first periods are periods of different lengths traced back from the target time, and the plurality of first explanatory variables are the averages of the predicted values of the weather quantity in each of the plurality of first periods. The information processing apparatus according to Claim 1.

3. The plurality of first periods are a plurality of different times before the target time, and the plurality of first explanatory variables are the predicted values of the weather quantity at the plurality of different times. The information processing apparatus according to Claim 1.

4. The first model includes a plurality of first weights that are coefficients of the plurality of first explanatory variables, and the processing unit calculates the plurality of first weights by LASSO regression and deletes the terms where the first weights are zero. The information processing apparatus according to Claim 1.

5. The processing unit generates a second model for estimating the power generation amount in the facility from the water inflow rate into the facility based on the actual data of the power generation amount in the facility and the actual data of the water inflow rate into the facility, and the processing unit estimates the power generation amount in the facility based on the estimated inflow rate and the second model. The information processing apparatus according to Claim 1.

6. The processing unit determines a water intake upper limit parameter regarding the upper limit value of the amount of water that can be processed in the facility based on the actual data of the power generation amount in the facility and the actual data of the inflow amount of water into the facility. When the estimated inflow amount is greater than the water intake upper limit parameter, the processing unit uses the value of the water intake upper limit parameter as the inflow amount to estimate the power generation amount in the facility. The information processing apparatus according to claim 5.

7. Based on the actual data of the observed values of the weather quantity, the processing unit further generates the first model further including a plurality of second explanatory variables regarding the observed values of the weather quantity in a plurality of second periods. The plurality of second periods are before the processing execution time when the process of estimating the inflow amount is performed. The processing unit acquires the observed data of the weather quantity observed after the actual data, and estimates the inflow amount at the target time based on the acquired observed data of the weather quantity. The information processing apparatus according to claim 5.

8. The plurality of second periods are periods of a plurality of different lengths retrogressing from the processing execution time. The plurality of second explanatory variables are the averages of the observed values of the weather quantity in each of the plurality of second periods. The information processing apparatus according to claim 7.

9. The plurality of second periods are a plurality of different times before the processing execution time. The plurality of second explanatory variables are the observed values of the weather quantity at the plurality of different times. The information processing apparatus according to claim 7.

10. The first model includes a plurality of first weights and a plurality of second weights which are the coefficients of the plurality of first explanatory variables and the plurality of second explanatory variables. The processing unit calculates the plurality of first weights and the plurality of second weights by LASSO regression, and deletes the terms with the first weight being zero and the terms with the second weight being zero. The information processing apparatus according to claim 7.

11. The processing unit generates the first model further including a plurality of third explanatory variables related to the discharge amount in a plurality of third periods before the target time based on the performance data of the discharge amount upstream of the facility, The processing unit acquires the performance data of the latest upstream discharge amount after the performance data, and estimates the inflow amount at the target time based on the acquired performance data of the latest upstream discharge amount. The information processing apparatus according to claim 5.

12. The plurality of third periods are periods of a plurality of different lengths counted back from the target time, The plurality of third explanatory variables are the averages of the performance values of the discharge amount in each of the plurality of third periods. The information processing apparatus according to claim 11.

13. The plurality of third periods are a plurality of different times counted back from the target time, The plurality of third explanatory variables are the performance values of the discharge amount at the plurality of different times. The information processing apparatus according to claim 11.

14. The first model includes a plurality of first weights and a plurality of third weights that are the coefficients of the plurality of first explanatory variables and the plurality of third explanatory variables, The processing unit calculates the plurality of first weights and the plurality of third weights by LASSO regression, and deletes the terms with the first weight being zero and the terms with the third weight being zero. The information processing apparatus according to claim 11.

15. The processing unit generates the first model further including a plurality of fourth explanatory variables related to the inflow amount in a plurality of fourth periods based on the performance data of the inflow amount, The plurality of fourth periods are before the processing execution time for performing the process of estimating the inflow amount. The processing unit acquires the performance data of the latest inflow volume after the performance data, and estimates the inflow volume at the target time based on the acquired performance data of the latest inflow volume. The information processing apparatus according to claim 5.

16. The plurality of fourth periods are periods of a plurality of different lengths going back from the processing execution time. The plurality of fourth explanatory variables are the averages of the performance values of the inflow volume in each of the plurality of fourth periods. The information processing apparatus according to claim 15.

17. The plurality of fourth periods are a plurality of different times going back from the processing execution time. The plurality of fourth explanatory variables are the performance values of the inflow volume at the plurality of different times. The information processing apparatus according to claim 15.

18. The first model includes a plurality of first weights and a plurality of fourth weights that are the coefficients of the plurality of first explanatory variables and the plurality of fourth explanatory variables. The processing unit calculates the plurality of first weights and the plurality of fourth weights by LASSO regression, and deletes the terms where the first weight is zero and the terms where the fourth weight is zero. The information processing apparatus according to claim 15.

19. The facility is a hydroelectric power facility. The information processing apparatus according to any one of claims 1 to 18.

20. Based on the performance data of the water inflow volume into the facility and the predicted data of the meteorological quantity in the past period, a first model including an objective variable related to the water inflow volume into the facility at the target time and a plurality of first explanatory variables related to the predicted values of the meteorological quantity in a plurality of first periods before the target time is generated, and based on the first model and the predicted data of the meteorological quantity in the future period after the past period, the water inflow volume into the facility at the target time is estimated. Information processing method.

21. A step of generating a first model including an objective variable related to the water inflow rate into the facility at a target time based on the actual data of the water inflow rate into the facility and the predicted data of the meteorological amount in a past period, and a plurality of first explanatory variables related to the predicted values of the meteorological amount in a plurality of first periods before the target time; A step of estimating the water inflow rate into the facility at the target time based on the first model and the predicted data of the meteorological amount in a future period after the past period; A computer program for causing a computer to execute the above.

Images