Demand prediction apparatus, demand prediction method, and program

Abstract

To calculate a correction amount in consideration of a degree of influence of a demand factor, thereby making an accurate demand prediction.SOLUTION: A demand prediction apparatus includes: an extraction unit which extracts, as a similar date, a date similar to a prediction target date from among past dates; an analysis unit which analyzes, based on past actual demands on the past dates and past actual values of factors for the past actual demands, degrees of influence of the factors for the past actual demands; a correction unit which calculates a correction amount from a factor having a higher influence degree, based on index values representing the degrees of influence, to correct the past actual demand on the similar date with the correction amount; and an output unit which outputs the corrected past actual demand, as a predicted demand on the prediction target date.SELECTED DRAWING: Figure 5

Description

[Technical field]

[0001] The present disclosure relates to a demand forecasting device, a demand forecasting method, and a program. [Background technology]

[0002] In a power system, in order to stabilize the frequency and voltage, it is necessary to always match the power demand with the amount of power generation. For this reason, power demand forecasting plays an important role in ensuring a stable supply of power. As a conventional technology related to power demand forecasting, for example, a technology is known that can obtain a power demand forecast value with reduced error by correcting the power demand forecast value on a target forecast date using price elasticity corresponding to factors such as the unit price of electricity and weather conditions on the target forecast date (Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6094369 Summary of the Invention [Problem to be solved by the invention]

[0004] However, in conventional technology, the degree of influence of factors on electricity demand (hereinafter also referred to as demand factors) is not taken into account, so it may not be possible to calculate an appropriate correction amount, and it may not be possible to obtain an accurate electricity demand forecast value for the forecast date.

[0005] The present disclosure has been made in consideration of the above points, and aims to realize accurate demand forecasting by calculating a correction amount that takes into account the degree of influence of demand factors. [Means for solving the problem]

[0006] A demand forecasting device according to one aspect of the present disclosure includes an extraction unit that extracts days similar to the forecast date from among past days as similar days; an analysis unit that analyzes the degree of influence of factors on the demand actual based on the demand actual for the past days and actual values ​​of the factors on the demand actual; a correction unit that calculates a correction amount from the factors with a high degree of influence based on an index value representing the degree of influence and corrects the demand actual for the similar day with the correction amount; and an output unit that outputs the corrected demand actual as a demand forecast for the forecast date. Effect of the Invention

[0007] By calculating the correction amount taking into account the degree of influence of demand factors, it is possible to realize accurate demand forecasting. [Brief description of the drawings]

[0008] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of the demand prediction device according to the present embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of a functional configuration of a demand prediction device according to the present embodiment. [Diagram 3] FIG. 2 is a diagram showing an example of demand record data stored in a demand record DB. [Figure 4] FIG. 2 is a diagram showing an example of weather record data stored in a weather record DB. [Diagram 5] 10 is a flowchart illustrating an example of a demand forecasting process according to the present embodiment. [Figure 6] 10 is a flowchart illustrating an example of a factor analysis process according to the embodiment. [Figure 7] FIG. 13 is a diagram illustrating an example of learning data for a decision tree. [Figure 8] FIG. 13 is a diagram illustrating an example of variable importance of each demand factor. [Figure 9] 13 is a flowchart illustrating an example of a similar day demand correction process according to the embodiment. [Figure 10] FIG. 13 is a diagram illustrating an example of learning data for a prediction model. [Figure 11]FIG. 13 is a diagram showing another example (part 1) of the variable importance of each demand factor. [Figure 12] FIG. 13 is a diagram showing another example (part 2) of the variable importance of each demand factor. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0009] An embodiment of the present invention will be described in detail below with reference to the drawings. In the following, a demand prediction device 10 that can realize accurate power demand prediction by calculating a correction amount that takes into account the degree of influence of demand factors will be described for the case of predicting power demand. However, predicting power demand is just one example, and the demand prediction device 10 according to this embodiment can be similarly applied to, for example, predicting heat demand, predicting gas demand, predicting any other demand, and the like.

[0010] For this reason, the demand forecasting device 10 according to this embodiment forecasts the power demand according to the following (1) to (3).

[0011] (1) Extract similar dates, which are past dates similar to the prediction date.

[0012] (2) Analyze the degree of impact of demand factors on electricity demand.

[0013] (3) A correction amount is calculated from the demand factor that has the greatest influence, and the electricity demand on similar days is corrected using this correction amount.

[0014] This makes it possible to calculate an appropriate correction amount that takes into account the degree of influence of demand factors on the power demand, making it possible to obtain an accurate power demand forecast value for the target forecast date.

[0015] For example, the degree of influence of maximum temperature (an example of a demand factor) on power demand is generally considered to be relatively large in summer and relatively small in winter. Similarly, the degree of influence of minimum temperature (an example of a demand factor) on power demand is generally considered to be relatively large in winter and relatively small in summer. In this way, the degree of influence of demand factors on power demand may vary seasonally. In addition to seasonal variations, the degree of influence of demand factors on power demand may vary due to economic, social, and political factors such as changes in fuel prices. Therefore, according to the demand forecasting device 10 of this embodiment, even if the degree of influence of demand factors on power demand varies due to some cause, it is possible to calculate an appropriate correction amount for power demand on a similar day, and as a result, it is possible to accurately calculate the power demand forecast value for the forecast target day.

[0016] In the following, as an example, it is assumed that meteorological information (e.g., temperature and humidity in each time period) is used as a demand factor, and when predicting the power demand for a target prediction day, weather forecast data including the date (year, month, date, and day of the week) of the target prediction day and a predicted value of the weather information for that target prediction day is provided to the demand prediction device 10. However, the demand factor is not limited to meteorological information, and the demand factor can be any factor that affects the power demand (e.g., fuel price, electricity bill, etc.). Furthermore, in addition to the temperature and humidity in each time period, the meteorological information may include any weather-related information such as wind speed, maximum temperature, minimum temperature, and indices that express the meteorological information based on some standard (discomfort index, laundry index, sensible temperature index, umbrella index, etc.).

[0017] <Hardware Configuration Example of Demand Forecasting Device 10> An example of the hardware configuration of a demand prediction device 10 according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the hardware configuration of the demand prediction device 10 according to this embodiment.

[0018] 1, a demand forecasting device 10 according to this embodiment includes an input device 101, a display device 102, an external I / F 103, a communication I / F 104, a RAM (Random Access Memory) 105, a ROM (Read Only Memory) 106, an auxiliary storage device 107, and a processor 108. Each of these pieces of hardware is connected to each other via a bus 109 so as to be able to communicate with each other.

[0019] The input device 101 is, for example, a keyboard, a mouse, a touch panel, a physical button, etc. The display device 102 is, for example, a display, a display panel, etc. Note that the demand forecasting device 10 does not necessarily have to include at least one of the input device 101 and the display device 102, for example.

[0020] The external I / F 103 is an interface with an external device such as a recording medium 103a. Examples of the recording medium 103a include a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), and a USB (Universal Serial Bus) memory card.

[0021] The communication I / F 104 is an interface for the demand prediction device 10 to communicate with other devices, other equipment, other terminals, etc. The RAM 105 is a volatile semiconductor memory (storage device) that temporarily stores programs and data. The ROM 106 is a non-volatile semiconductor memory (storage device) that can store programs and data even when the power is turned off. The auxiliary storage device 107 is a non-volatile storage device such as a hard disk drive (HDD), a solid state drive (SSD), or a flash memory, in which programs and data are stored. The processor 108 is various types of arithmetic devices such as a central processing unit (CPU) or a graphics processing unit (GPU).

[0022] The demand prediction device 10 according to the present embodiment has the hardware configuration shown in FIG. 1, and thus can realize the demand prediction process described below. However, the hardware configuration shown in FIG. 1 is merely an example, and the demand prediction device 10 according to the present embodiment is not limited thereto. For example, the demand prediction device 10 according to the present embodiment may have a plurality of auxiliary storage devices 107 and a plurality of processors 108, or may have various hardware other than the hardware shown in the figure. In addition, the demand prediction device 10 according to the present embodiment may be realized by one device or by a plurality of devices. When the demand prediction device 10 is realized by a plurality of devices, it may be called a demand prediction system or the like.

[0023] <Example of functional configuration of demand forecasting device 10> An example of the functional configuration of the demand prediction device 10 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the functional configuration of the demand prediction device 10 according to this embodiment.

[0024] As shown in FIG. 2, the demand prediction device 10 according to this embodiment includes an input unit 201, a similar day extraction unit 202, a demand factor analysis unit 203, a demand correction unit 204, and an output unit 205. Each of these units is realized, for example, by a process in which one or more programs installed in the demand prediction device 10 are executed by the processor 108 or the like. The demand prediction device 10 according to this embodiment also includes a demand record DB 206 and a weather record DB 207. Each of these DBs (databases) is realized, for example, by a storage area of ​​the auxiliary storage device 107 or the like. However, at least one of these DBs may be realized by a storage area of ​​a storage device or the like included in a database server communicably connected to the demand prediction device 10.

[0025] When weather forecast data for the prediction target day is given, the input unit 201 inputs the weather forecast data. The weather forecast data for the prediction target day includes the date (year, month, date, and day of the week) of the prediction target day and a predicted value of the weather information for the prediction target day (hereinafter also referred to as a weather predicted value).

[0026] The similar day extraction unit 202 extracts similar days that represent dates similar to the prediction target date from dates included in the demand record data stored in the demand record DB 206. The similar day extraction unit 202 may extract similar days using any method, but may, for example, use weather forecast data and actual weather data stored in the weather record DB to extract dates with identical or similar weather information as similar days.

[0027] Here, actual demand data refers to data that represents actual past electricity demand, for example, data that associates a past date with actual electricity demand values ​​for each time period (for example, hourly time periods) on that date. Also, actual weather data refers to data that represents actual weather information in the past, for example, data that associates a past date with actual weather information values ​​for each time period (for example, hourly time periods) on that date. Specific examples of actual demand data and actual weather data will be described later.

[0028] The demand factor analysis unit 203 analyzes the degree of influence of demand factors (e.g., temperature and humidity in each time period) on the power demand using the demand record data stored in the demand record DB 206 and the weather record data stored in the weather record DB. At this time, the demand factor analysis unit 203 analyzes the degree of influence of the demand factors on the power demand in each time period, for example, by using a decision tree.

[0029] For example, it is known that electricity demand has a nonlinear relationship (quadratic correlation) with temperature. Meanwhile, decision trees are capable of analyzing nonlinear data. Therefore, by using decision trees, it is possible to appropriately analyze nonlinear relationships, including the relationship between electricity demand and temperature.

[0030] The demand correction unit 204 calculates a correction amount for the actual demand value in each time slot on a similar day, using actual demand data stored in a demand record DB 206 and actual weather data stored in a weather record DB 207. At this time, the demand correction unit 204 constructs a prediction model for predicting the power demand for each time slot from the demand factors that have a large influence on the power demand in that time slot (in other words, the demand factors that have a high correlation with the power demand in that time slot), and then calculates the correction amount using this prediction model.

[0031] Furthermore, the demand correction unit 204 corrects the demand result values ​​in each time period of the demand result data including similar days, among the demand result data stored in the demand result DB 206, by each correction amount.

[0032] The output unit 205 outputs each corrected actual demand value (i.e., a value obtained by correcting the actual demand value in each time slot of a similar day by each correction amount) to a predetermined output destination as a demand forecast value in each time slot of the forecast target day. Note that the output destination is not limited to a specific output destination, and can be any output destination. For example, the output destination can be a display device 102 such as a display, or another device, equipment, terminal, etc. connected to the demand forecasting device 10 so as to be able to communicate with the demand forecasting device 10.

[0033] The demand record DB 206 stores actual demand data for a predetermined period in the past. In the following description, it is assumed that actual demand data for a certain month is stored in the demand record DB 206. The actual demand value for electricity is assumed to be the actual demand value for each hourly time slot. However, these are only examples, and the predetermined period is not limited to one month, and can be any period such as three months or six months. In addition, it is also an example that the time span of each time slot is one hour, and it is possible to set the time span of each time slot to any time span such as 30 minutes, two hours, or 24 hours.

[0034] The weather record DB 207 stores the weather record data for the same period as the demand record data stored in the demand record DB 206. Note that the weather information is assumed to be weather information for each hourly time slot, similar to the demand record values.

[0035] <<Example of demand record data stored in the demand record DB 206>> An example of the demand record data stored in the demand record DB 206 will be described with reference to Fig. 3. Fig. 3 is a diagram showing an example of the demand record data stored in the demand record DB 206.

[0036] 3, actual demand data for a predetermined period in the past is stored in the demand result DB 206. In each item of actual demand data, a date is associated with an actual demand value for each time period on that date.

[0037] For example, the actual demand data for the date "Thursday, September 1, 2022" includes the actual demand value "554" in the time period "0:00-1:00" to the actual demand value "716" in the time period "23:00-24:00".

[0038] In this way, each demand result data includes the demand result value for the time slot from (n-1) to n, for n = 1,...,24. Hereinafter, the time slot from (n-1) to n will be referred to as "time slot n," and the demand result value for time slot n will be referred to as "n-hour demand."

[0039] <<Specific example of weather record data stored in the weather record DB 207>> An example of the weather record data stored in the weather record DB 207 will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of the weather record data stored in the weather record DB 207.

[0040] 4, the weather record DB 207 stores weather record data for a predetermined period in the past. In each piece of weather record data, a date is associated with a record value of the weather information for each time period on that date (hereinafter, also referred to as the weather record value).

[0041] For example, actual weather data for the date "Thursday, September 1, 2022" includes actual weather values ​​from the time period "0:00-1:00" (such as actual temperature value "22.4" and actual humidity value "69") to the time period "23:00-24:00" (such as actual temperature value "21.9" and actual humidity value "66").

[0042] Thus, each weather record data contains the record values ​​of the weather information for time period n, for n=1, . . . , 24.

[0043] <Demand forecast processing> Hereinafter, an example of the demand forecasting process according to this embodiment will be described with reference to Fig. 5. Fig. 5 is a flowchart showing an example of the demand forecasting process according to this embodiment. The demand forecasting process shown in Fig. 5 is repeatedly executed every time weather forecast data for a target forecast date is provided to the demand forecasting device 10.

[0044] First, the input unit 201 inputs given weather forecast data (step S101).

[0045] Next, the similar day extraction unit 202 extracts similar days from the dates included in the demand record data stored in the demand record DB 206 (step S102). For example, the similar day extraction unit 202 may identify, from among the weather record data stored in the weather record DB, the weather record data whose weather record value is most similar to the weather forecast value included in the weather forecast data, and then extract the dates included in the identified weather record data as similar days. At this time, any scale may be used as a measure of the similarity between the weather record value and the weather forecast value, but for example, it is possible to use a distance such as Euclidean distance or a similarity of cosine similarity after vector expression of the weather information value.

[0046] The above-mentioned method of extracting similar days is merely an example, and similar day extraction unit 202 may extract similar days using any method. For example, similar day extraction unit 202 may extract similar days using the method described in Reference 1, or other methods. Other methods include, for example, a method of extracting a date of the same month and date as a date included in weather forecast data as a similar day, a method of extracting a date of the same day of the week as a date included in weather forecast data as a similar day, and a method of extracting a date with the same maximum temperature (or maximum humidity) as the maximum temperature (or maximum humidity) included in weather forecast data as a similar day. Also, for example, a user may manually extract similar days.

[0047] Next, the demand factor analysis unit 203 analyzes the degree of influence of the demand factors on the power demand using the demand record data stored in the demand record DB 206 and the weather record data stored in the weather record DB (step S103). Details of the process of this step (demand factor analysis process) will be described later. In the following, the explanation will be continued assuming that the degree of influence of the demand factors in time slot n is analyzed for n=1,...,24.

[0048] Next, the demand correction unit 204 calculates a correction amount for each time period using the demand record data stored in the demand record DB 206 and the weather record data stored in the weather record DB 207, and corrects the demand record value for the time period on similar days by the correction amount (step S104). Details of the process of this step (similar day demand correction process) will be described later. The following description will be continued assuming that the demand at n hours on similar days is corrected for n=1,...,24.

[0049] Then, the output unit 205 outputs each corrected n hour demand as a demand forecast value for the forecast target date to a predetermined output destination (step S105). That is, the output unit 205 outputs the corrected 1 hour demand to the corrected 24 hour demand as a demand forecast value for the forecast target date to a predetermined output destination.

[0050] <Factor analysis processing> An example of the factor analysis process in step S103 above will be described below with reference to Fig. 6. Fig. 6 is a flowchart showing an example of the factor analysis process according to this embodiment. The factor analysis process shown in Fig. 6 is executed for each time period, that is, for each time period n where n = 1,...,24. A case where the factor analysis process is executed for a certain time period n will be described below.

[0051] First, the demand factor analysis unit 203 creates learning data for a decision tree related to time slot n using the demand record data stored in the demand record DB 206 and the weather record data stored in the weather record DB (step S201). That is, the demand factor analysis unit 203 creates learning data for each date, associating the date, the demand at n hour on that date, and each demand factor on that date. This creates learning data for the number of dates included in a predetermined past period. Hereinafter, the set of learning data for these dates will be referred to as the "learning data set for the decision tree related to time slot n."

[0052] Here, the actual weather value for the date is set as the value of the demand factor. Specifically, each actual weather value for time zone 1 to time zone 24 for the date is set as the value of the demand factor. However, this is only an example, and for example, each actual weather value for time zone n for the date may be set as the value of the demand factor, or each actual weather value for time zone n and the time zones before and after the time zone for the date may be set as the value of the demand factor.

[0053] As an example, learning data for a decision tree regarding time period n=1 will be described with reference to Fig. 7. Fig. 7 is a diagram showing an example of learning data for a decision tree.

[0054] As shown in FIG. 7, the learning data of the decision tree for time slot n is associated with a date, the demand at n o'clock on that date, and each demand factor on that date. The example shown in FIG. 7 shows an example in which 30 pieces of learning data were created for the dates "Thursday, September 1, 2022" to "Friday, September 30, 2022". In addition, in the example shown in FIG. 7, 20 actual weather values ​​of "demand factor 1" to "demand factor 20" are set as the demand factor values ​​in each learning data. In the example shown in FIG. 7, demand factor 1 is the temperature in time slot n=1, and demand factor 2 is the humidity in time slot n=1.

[0055] Then, the demand factor analysis unit 203 uses a learning data set of a decision tree for time slot n to learn a decision tree with the demand factors included in each learning data as explanatory variables and the n-hour demand included in the learning data as a response variable, thereby calculating the variable importance of the demand factor (step S202). The variable importance is an index for evaluating how much the demand factor (explanatory variable) contributes to the classification of the demand performance value (response variable). For this reason, in this embodiment, the variable importance is adopted as an index value for measuring the degree of influence of the demand factor on the electricity demand. Note that the variable importance can be calculated, for example, based on the Gini impurity used when classifying the response variable by the decision tree.

[0056] From the above, the variable importance (%) of each demand factor for each time slot n is obtained. As an example, the variable importance of "demand factor 1" to "demand factor 20" in time slot 1 to time slot 24 will be described with reference to Fig. 8. Fig. 8 is a diagram showing an example of the variable importance of each demand factor.

[0057] As shown in Figure 8, the variable importance (%) of each demand factor in each time slot is obtained for each time slot. This makes it possible to obtain the degree of influence of each demand factor on the demand actual value (demand at n hours) for each time slot n as the variable importance.

[0058] <Similar day demand correction processing> An example of the similar day demand correction process in step S104 above will be described below with reference to Fig. 9. Fig. 9 is a flowchart showing an example of the similar day demand correction process according to this embodiment. The similar day demand correction process shown in Fig. 9 is executed for each time slot, that is, for each time slot n where n = 1,...,24. A case where the similar day demand correction process is executed for a certain time slot n will be described below.

[0059] First, the demand correction unit 204 extracts demand factors with high variable importance (in other words, demand factors with high correlation with the demand at time n) using the variable importance in the time slot n (step S301). For example, n Then, the demand correction unit 204 calculates the variable importance as m n For all n = 1, , 24, we extract the demand factors that satisfy m n are the same, that is, m=m n may be also possible.

[0060] However, the above-mentioned method for extracting the demand factors is merely an example, and the demand factors with high variable importance may be extracted by other methods. For example, the demand correction unit 204 extracts M n (However, M n For all n = 1, . . . , 24, M n are the same, that is, M=M n may be also possible.

[0061] Hereinafter, in step S301 above, n Assuming that demand factors are extracted, let us divide these demand factors into x1 (n) ,···,xk n (n) Let us assume that.

[0062] Next, the demand correction unit 204 calculates the demand result data stored in the demand result DB 206, the weather result data stored in the weather result DB 207, and the demand factor x1 (n) ,···,xk n (n)In other words, the demand correction unit 204 creates learning data for a prediction model for the time slot n using the above (step S302). (n) ,···,xk n (n) Data that associates the date and time is created as learning data. As a result, learning data is created for the number of dates included in a specified period in the past. Hereinafter, the set of learning data for these dates will be referred to as the "learning data set of the prediction model for time slot n."

[0063] As an example, learning data for a prediction model for time period n=1 will be described with reference to FIG. 10. FIG. 10 is a diagram showing an example of learning data for a prediction model. In the example shown in FIG. 10, k 1 = 3 and x1 (1) = demand factor 1, x2 (1) = demand factor 2, x3 (1) This shows the case where demand factor is 9.

[0064] As shown in Figure 10, the learning data for the forecasting model for time slot n includes the date, the demand at n hour on that date, and the demand factor x1 (n) ,···,xk n (n) The example shown in FIG. 10 shows an example in which 30 pieces of learning data were created for the dates "Thursday, September 1, 2022" to "Friday, September 30, 2022". In addition, in the example shown in FIG. 10, three actual weather values, "demand factor 1", "demand factor 2", and "demand factor 9", are set as demand factor values ​​in each learning data.

[0065] Next, the demand correction unit 204 uses a learning data set for the prediction model for the time slot n to learn a prediction model for the time slot n, with the demand factors included in each learning data set as explanatory variables and the n-hour demand included in the learning data set as a response variable (step S303). That is, the demand correction unit 204 uses the learning data set for the prediction model for the time slot n to learn y (n) =f (n) (x1(n) ,···,xk n (n) ) where y (n) is the demand at n hours, f (n) represents the prediction model for time slot n.

[0066] In addition, the prediction model f (n) As for the demand factor, x1 (n) ,···,xk n (n) Demand for n hours from y (n) Any model for predicting f can be used. For example, a regression model represented by a regression formula, a machine learning model such as a neural network, etc. can be used as a prediction model f (n) It is possible to use it as:

[0067] For example, the prediction model f (n) When using a regression equation as the prediction model f (n) is expressed as follows:

[0068] f (n) (x1 (n) ,···,xk n (n) )=a1 (n) ×x1 (n) +···+ak n (n) ×xk n (n) +b (n) Here, a1 (n) ,···,ak n (n) is the regression coefficient, b (n) represents the intercept.

[0069] In addition, the prediction model f (n) The learning method may be an appropriate method depending on the type of the prediction model.

[0070] Next, the demand correction unit 204 calculates a correction amount for the demand at n hour on a similar day by using the weather forecast data for the prediction target day and the weather record data for a similar day (step S304). Hereinafter, the correction amount for the demand at n hour on a similar day is referred to as c (n)In addition, the demand factor x1 included in the weather forecast data for the forecast target day (n) ,···,xk n (n) The value of X1 (n) ,···,Xk n (n) , Demand factors included in the weather record data for similar days x1 (n) ,···,xk n (n) The value of X'1 (n) ,···,X'k n (n) At this time, the demand correction unit 204 calculates the correction amount c (n) Calculate.

[0071] c (n) =f (n) (X1 (n) ,···,Xk n (n) )-f (n) (X'1 (n) ,···,X'k n (n) ) In the above, the correction amount c (n) f (n) (X1 (n) ,···,Xk n (n) ) and f (n) (X'1 (n) ,···,X'k n (n) ) is calculated by the difference between the correction amount c and the correction amount c . (n) For example, the function for calculating the correction amount may be g, and c (n) =g(f (n) (X1 (n) ,···,Xk n (n) ),f (n) (X'1 (n) ,···,X'k n (n) )) may be used to calculate the correction amount.

[0072] Then, the demand correction unit 204 corrects the demand at time n on the similar day by a correction amount c (n)(Step S305). Hereinafter, the demand at n hour included in the weather record data of a similar day is expressed as Y' (n) In this case, the demand correction unit 204 calculates the demand Y' at time n on a similar day as follows: (n) Correct the following.

[0073] Y (n) :=Y' (n) +c (n) This will result in the corrected demand for similar days at time n, Y (n) The demand for this n hour Y (n) is the demand at n hour on the forecast target day. Note that in the above, the demand at n hour on a similar day, Y' (n) Correction amount c (n) However, this is only an example, and other methods may be used for correction. For example, let h be the correction function, and Y (n) =h(Y' (n) ,c (n) ) Demand at time n after correction Y (n) may be calculated.

[0074] <Modification> Modifications of the above embodiment will now be described.

[0075] <Variation 1> After calculating the variable importance of each demand factor for each time period n in the factor analysis process shown in Figure 6, statistics (e.g., average value, maximum value, mode, etc.) of the variable importance of the demand factor for each demand factor for time period 1 to time period 24 may be calculated.

[0076] As an example, a case where an average value of the variable importance of each demand factor shown in Fig. 8 is calculated will be described with reference to Fig. 11. Fig. 11 is a diagram showing another example (part 1) of the variable importance of each demand factor.

[0077] As shown in FIG. 11, the average value of the variable importance of each demand factor for time slot 1 to time slot 24 is taken as the variable importance of that demand factor. In the example shown in FIG. 11, the average value of demand factor 1 for time slot 1 to time slot 24, "30", is calculated as the variable importance of demand factor 1 for one day. Similarly, the average value of demand factor 2 for time slot 1 to time slot 24, "20", is calculated as the variable importance of demand factor 2 for one day. Similarly, the average value of demand factor 20 for time slot 1 to time slot 24, "35", is calculated as the variable importance of demand factor 20 for one day.

[0078] In this modification, when demand factors with high variable importance are extracted in step S301 of FIG. 9, the demand factors with high variable importance are extracted using the variable importance common to each time slot n.

[0079] <<Variation 2>> In the above modification 1, the statistics of the variable importance of the demand factors for one day are calculated, but after calculating the variable importance of each demand factor for each time slot n, multiple time slots n may be combined into one time slot, and the statistics of the variable importance of the demand factors for that time slot may be calculated. Hereinafter, a time slot combining multiple time slots n will be represented as "time slot p" (p=1, , P), where P is the total number of time slots p combining multiple time slots n.

[0080] As an example, a case where an average value of the variable importance of each demand factor for the time slots of 1:00 to 6:00, 7:00 to 12:00, 13:00 to 18:00, and 19:00 to 24:00 will be described with reference to FIG. 12. FIG. 12 is a diagram showing another example (part 2) of the variable importance of each demand factor. In the example shown in FIG. 12, P=4, and time slot p=1 is the "time slot of 1:00 to 6:00", time slot p=2 is the "time slot of 7:00 to 12:00", time slot p=3 is the "time slot of 13:00 to 18:00", and time slot p=4 is the "time slot of 19:00 to 24:00".

[0081] As shown in FIG. 12, the average values ​​of the variable importance of each demand factor for time slot 1 to time slot 6 are taken as the variable importance of each demand factor for the time slot from 1:00 to 6:00. Similarly, the average values ​​of the variable importance of each demand factor for time slot 7 to time slot 12 are taken as the variable importance of each demand factor for the time slot from 7:00 to 12:00. Similarly, the average values ​​of the variable importance of each demand factor for time slot 13 to time slot 18 are taken as the variable importance of each demand factor for the time slot from 13:00 to 18:00. Similarly, the average values ​​of the variable importance of each demand factor for time slot 19 to time slot 24 are taken as the variable importance of each demand factor for the time slot from 19:00 to 24:00.

[0082] In the above, the average value of the variable importance of each demand factor for time slot n to time slot n+5 is taken as the variable importance of each demand factor for the time slot from n to n+5, but this is just an example and is not limited to this. For example, a day may be divided into three time slots, "morning", "afternoon", and "night", and the average value of the variable importance of each demand factor for time slot n belonging to "morning" and the average value of the variable importance of each demand factor for time slot n belonging to "afternoon" and the average value of the variable importance of each demand factor for time slot n belonging to "night" may be calculated.

[0083] In this modified example, when extracting demand factors with high variable importance in step S301 of Figure 9, demand factors with high variable importance are extracted using the variable importance of each demand factor for time period p to which time period n belongs.

[0084] <Summary> As described above, the demand forecasting device 10 according to the present embodiment calculates the correction amount for the demand for a day similar to the target forecast date by using only factors that have a high degree of influence on the demand. This makes it possible to calculate an appropriate correction amount that takes into account the degree of influence of factors on the demand, thereby making it possible to obtain a highly accurate demand forecast value for the target forecast date.

[0085] The present invention is not limited to the above-described embodiments specifically disclosed, and various modifications, changes, combinations with known technologies, etc. are possible without departing from the scope of the claims.

[0086] [References] Reference 1: Patent No. 6187003 [Explanation of symbols]

[0087] 10 Demand forecasting device 101 Input Device 102 Display device 103 External I / F 103a Recording media 104 External I / F 105 RAM 106 ROM 107 Auxiliary storage 108 processors 109 Bus 201 Input section 202 Similar date extraction part 203 Demand Factor Analysis Department 204 Demand Correction Department 205 Output section 206 Demand performance DB 207 Weather Record DB

Claims

1. an extraction unit that extracts dates similar to the prediction target date from among past dates as similar dates; an analysis unit that analyzes the degree of influence of the factors on the demand record based on the demand record for the past day and actual values ​​of the factors on the demand record; a correction unit that calculates a correction amount from the factor having a high degree of influence based on the index value representing the degree of influence, and corrects the demand record for the similar day with the correction amount; an output unit that outputs the corrected actual demand result as a demand forecast for the forecast date; A demand forecasting device having the above configuration.

2. The analysis unit includes: The demand forecasting device according to claim 1 , wherein the degree of influence of the factors is analyzed by learning a decision tree using the demand record as a response variable and the factors as explanatory variables.

3. The demand forecasting device according to claim 2 , wherein the index value is a variable importance of the explanatory variable when the decision tree is trained.

4. The correction unit is A prediction model is trained using the factors with high influence as explanatory variables and the demand record as a target variable; The demand forecasting device according to claim 1 , wherein the correction amount is calculated based on a forecast value of the forecast model on the target forecast date and a forecast value of the forecast model on the similar date.

5. The demand forecasting device according to claim 4 , wherein the factors having a high degree of influence are factors with respect to the demand record whose index value is equal to or greater than a predetermined threshold value, or a predetermined number of factors with high index values.

6. The day is made up of multiple time periods, The analysis unit includes: Analyzing the degree of influence of the factor on the demand record for each of the time periods based on the demand record for the time period; The correction unit is 2. The demand forecasting device of claim 1, wherein the correction amount is calculated for each time period from the factor with the highest degree of influence based on the index value for the time period or a statistical amount of the index value for each of the multiple time periods over a specified period, and the actual demand for the similar days is corrected with the correction amount.

7. The demand forecasting device according to claim 6 , wherein the statistics include an average value, a maximum value, and a mode value of the index value in the period.

8. an extraction step of extracting dates similar to the prediction target date from among past dates as similar dates; an analysis step of analyzing the degree of influence of the factors on the demand actual result based on the demand actual result for the past day and actual values ​​of the factors on the demand actual result; a correction step of calculating a correction amount from the factor having a high degree of influence based on the index value representing the degree of influence, and correcting the demand record for the similar day with the correction amount; an output step of outputting the corrected actual demand result as a demand forecast for the forecast target date; A demand forecasting method implemented by a computer.

9. an extraction step of extracting dates similar to the prediction target date from among past dates as similar dates; an analysis step of analyzing the degree of influence of the factors on the demand actual result based on the demand actual result for the past day and actual values ​​of the factors on the demand actual result; a correction step of calculating a correction amount from the factor having a high degree of influence based on the index value representing the degree of influence, and correcting the demand record for the similar day with the correction amount; an output step of outputting the corrected actual demand result as a demand forecast for the forecast target date; A program that causes a computer to execute the following.

Images