JP2019140831A - Power demand prediction system, power demand prediction method, and program - Google Patents

Abstract

To provide a power demand prediction system, method, and program, for accurately performing prediction of a power demand of the whole area including power partial supply, using a predetermined model expression.SOLUTION: A power demand prediction system of the present invention includes: an analysis unit 11c that calculates, from past power data, total amount actual power including load following actual power, extracts a day that is earlier than a prediction day by the predetermined number of days or more and at which the maximum power has been consumed, calculates reference power on the basis of the maximum power, subtracts the reference power from the total amount actual power, and takes a result of the subtraction as a reference value; and a prediction unit 11d that substitutes a time zone code for a model expression constituted by a function of the time zone code and a differential value between the reference value and the load following actual power, calculates a differential value calculated in a regressive manner, and takes a value obtained by adding the differential value to the reference value as load following demand prediction power.SELECTED DRAWING: Figure 2

Description

  The present invention relates to a power demand prediction system, a power demand prediction method, and a program for performing power demand prediction related to partial power supply.

  In the electric power market today, in addition to conventional general electric utilities, retail electric utilities who supply electric power as new businesses have entered. And since a retail electric power company borrows the power transmission and distribution network which the existing power transmission and distribution business owns, various duties are imposed.

  For example, payment of consignment fees, which are usage fees for the power transmission and distribution network, securing supply capacity for stable supply of the entire area, and achieving the same amount of planned values.

  In addition, retail electric utilities are required to achieve the same amount of planned value and actual demand submitted to the Organization for Cross-regional Coordination of Transmission Operators (OCCTO). .

  Here, excess or deficiency of power supply and demand when unachieved is referred to as imbalance, “supply amount> demand result” is referred to as surplus imbalance, and “supply amount <demand result” is referred to as insufficient imbalance.

  As shown in FIG. 11, with regard to imbalance, surplus is purchased with a power transmission and distribution company, shortage is replenished, and finally settled. However, if the imbalance deviates greatly, there is a possibility of receiving guidance such as recommendations. Therefore, in order to achieve the same amount simultaneously, it is essential to improve the accuracy of power demand prediction.

  On the other hand, power supply to consumers from two power companies is called partial power supply. There are mainly three types of supply methods, a cross-cut partial supply, a vertical cut partial supply, and a notification-type partial supply. At present, the cross-cut partial supply is the mainstream.

  As shown in FIG. 12, the cross-section partial supply is a supply form in which one of the two power companies supplies a certain amount of base supply, and the other performs load following supply exceeding the base supply power. That means.

  In such power demand forecasting and power supply by partial power supply, as the actual power supply procedure, it is necessary to make a total power demand prediction for each transmission and distribution pipe and submit a plan that determines the power supply amount to the OCCTO There is. In other words, it is necessary to estimate and estimate the total predicted value of contract customers in the entire area regardless of the power supply form.

  Here, as a technique related to partial supply, for example, in Patent Document 1, a plurality of partial supply contracts are performed by a target power setting unit for a time period when there are a plurality of partial supply contract powers A and B in the same time period. The target power W2 is set based on the sum (A + B) of the powers A and B, the target current power W3 is calculated with the target power W2 as a final value, and is compared with the current power W1 by the power comparison unit. In this case, there is disclosed a partial supply compatible demand control device that issues a warning.

JP 2017-63584 A

  The total amount of power used has been predicted by existing knowledge, although there are various characteristics. In other words, in the prior art, for example, as shown in FIGS. 13A and 13B, prediction of power consumption is made for each consumer, cut out based on base contract power, and individual load following supply power is predicted. Things have been done.

  However, errors that seem to be small in individual projects are likely to increase when the number of consumers grows. In addition, the load increases in both hardware and software, making it difficult to operate as a prediction system.

  Therefore, when trying to make a power demand forecast for the entire area including partial power supply after reducing the system load, the load following supply power of all customers in the area is not the case. It is necessary to predict the total value, and for that, it is essential to construct some model formula and develop the system. However, in the prior art including the above-described Patent Document 1, power demand prediction based on the construction of such a model formula has not been performed.

  In addition, in the time period when the usage amount increases, the demand for the entire retail is more likely to fluctuate more greatly in the partial power supply than in the full power supply. For this reason, the current situation is that the demand forecast is unclear, especially in peak hours, and there is a mismatch between the power generation side and the demand side. Therefore, today, it is desired to clarify the power demand prediction related to the unclear power partial supply occurring in such a real environment.

  The present invention has been made in view of such problems, and the object of the present invention is to construct a model formula for power demand prediction related to partial power supply, and perform power demand prediction using the model formula. It is to perform with high accuracy and speed.

  In order to solve the above-mentioned problem, the power demand prediction system according to the first aspect of the present invention analyzes at least past power data and accompanying data by a statistical method and constructs a model formula used for power demand prediction. An analysis unit that performs prediction of the demand predicted power, and a storage unit that stores the model formula, and the analysis unit includes the total amount actual power including the load following actual power from the past power data. And calculating the reference power based on the maximum power, subtracting the reference power from the total actual power, and calculating the result. As a reference value, the prediction unit calculates a differential value calculated recursively by substituting the time zone code into the model formula that is a function of a time zone code and a difference value between the reference value and the tracked actual power. And the standard A value obtained by adding the difference value to said load follow demand predicted power.

  In the power demand prediction method according to the second aspect of the present invention, the analysis unit calculates the total amount actual power including the load following actual power from the past power data, and the maximum power before the predetermined number of days on the prediction date is output. The reference power is calculated based on the maximum power, the reference power is subtracted from the total actual power, the result is used as a reference value, and the prediction unit uses the time zone code, the reference value, and the load. Substituting a time zone code into the model formula consisting of a function of the difference value with the tracked actual power, calculating a recursively calculated difference value, and adding the difference value to the reference value as the load following demand Estimated power.

  According to a third aspect of the present invention, there is provided a program that analyzes at least past power data and accompanying data by a statistical method and constructs a model formula used for power demand prediction, and the demand A prediction unit that performs prediction power prediction and a storage unit that stores the model formula function, and the analysis unit calculates a total amount actual power including a load following actual power from the past power data, and performs prediction Extract the day when the maximum power is more than a predetermined number of days before, calculate the reference power based on the maximum power, subtract the reference power from the total amount actual power, the result as a reference value, The prediction unit assigns the time zone code to the model formula that is a function of the time zone code and the difference value between the reference value and the load tracking actual power, calculates a recursively calculated difference value, and calculates the reference value. To the difference The value obtained by adding to said load follow demand predicted power.

  ADVANTAGE OF THE INVENTION According to this invention, about the power demand prediction which concerns on electric power partial supply, the power demand prediction system which can construct | assemble a model formula and can perform a power demand prediction using the said model formula accurately and rapidly, power demand prediction Methods and programs can be provided.

1 is a configuration diagram of a power demand prediction system according to an embodiment of the present invention. It is a block diagram of the server apparatus of the system. It is a block diagram of the information terminal of the system. It is a flowchart which shows the procedure of the model formula calculation by the system. It is a flowchart which shows the procedure of the electric power demand prediction by the system. It is a figure explaining each area | region which calculates a model formula. It is a figure which shows the relationship between the value which reflected the actual value of electric power, the reference | standard prediction value, and each element. It is a list which shows the influence factor in connection with electric power demand prediction. It is a figure which shows the relationship between the performance value of electric power, and a predicted value. It is a figure which shows the learning cycle of a model type | formula. It is a general electricity supply-demand graph. It is a figure explaining general load following supply electric power. It is a figure which shows the relationship between general base contract electric power and electric power used.

  Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

The power demand prediction system according to the present embodiment has at least the following characteristics.
-A system with a light load on both hardware and software-A system with simple prediction procedures-A system that can be operated regardless of the proportion of partial power supply

  FIG. 1 shows and describes the configuration of a power demand prediction system according to an embodiment of the present invention.

  As shown in the figure, the power demand prediction system includes a server device 1 of a power demand prediction service provider, a client information terminal 2 and a server device 3 of a power company via a network 4 such as the Internet. Are connected and configured to communicate freely.

  In such a configuration, the server device 3 of the electric power company provides data such as contract power for each group, past power consumption, and past load following power supply. The server device 1 acquires various data from the server device 3 of the electric power company, analyzes the data, and constructs a model formula as a basis for calculating demand forecast power, particularly load following demand forecast power.

  When receiving a request for power demand prediction from the information terminal 2 of the client, the server apparatus 1 calculates demand predicted power, in particular, load following demand predicted power using the constructed model formula, and the calculation result Is provided to the information terminal 2.

  In the information terminal 2, it is possible to check the calculation result of the demand forecast power, particularly the load follow-up demand forecast power by the browser function. After that, the server device 2 calculates the degree of influence of each element (influencing factor) such as temperature, day of the week, daylight hours, and precipitation on the calculated demand forecast power, and automatically learns it by machine learning and builds it first Update the model formula as needed.

  In FIG. 2, the structure of the server apparatus in the electric power demand prediction system which concerns on one Embodiment of this invention is shown and demonstrated.

  As shown in the figure, the server device 1 includes a control unit 11 such as a CPU (Central Processing Unit) that performs overall control. The control unit 11 is electrically connected to a communication unit 12 and a storage unit 13. It is connected to the.

  In such a configuration, the communication unit 12 is a communication interface (I / F) for communicating with the client information terminal 2 or the like via the network 4 such as the Internet. The storage unit 13 includes a memory such as a random access memory (RAM) and a read only memory (ROM), a hard disk drive (HDD), and the like, and stores a program 16 executed by the control unit 11. Provide area.

  The storage unit 13 includes a power DB 14 and a model formula DB 15. The power DB 14 stores past power data and the like. Each power data is associated with each element (influencing factor) such as temperature, day of the week, sunshine duration, and precipitation.

  The model formula DB 15 stores model formulas calculated by calculation. The constructed model formula is updated as needed by machine learning, but the model formula DB 15 stores all the past model formulas to the latest model formulas.

  The control unit 11 functions as the main control unit 11a, the input control unit 11b, the analysis unit 11c, the prediction unit 11d, the learning unit 11e, and the display data generation unit 11f by executing the program 16 in the storage unit 13.

  The main control unit 11a performs grouping for each characteristic on the power data. In this example, grouping is performed using a unique industry classification.

  The input control unit 11b receives input of various data used for constructing the model formula. Here, power big data such as contract power corresponding to the group, past used power, and past load following supply power, etc. from the server device 3 of the power company via the communication unit 12 or from the storage unit 13, and It will accept inputs such as weather data.

  The analysis unit 11c analyzes various data acquired through the input control unit 11b from a statistical viewpoint, and constructs a model formula used for power demand prediction. Although the details will be described later, the model formula is a function of the time zone code x and the difference value y between the reference value and the load tracking actual power.

  The prediction unit 11d inputs the total power consumption prediction into the latest model formula among the model formulas stored in the model formula DB 14 of the storage unit 13, and predicts demand prediction power, particularly load following demand prediction power. Become.

  The learning unit 11e updates the model formula constructed by the analysis unit 11c by machine learning. More specifically, the learning unit 11e captures various factors (for example, influencing factors such as temperature, humidity, solar radiation amount, day of the week, holidays, and disasters) from the storage unit 13 as learning factors, and influences them. In order to reflect the influence level of the factor in the already constructed model formula, machine learning is performed by a method such as deep learning, the model formula already constructed is updated, and the model formula DB 15 of the storage unit 13 is updated. The model formula is stored.

  The display data generation unit 11f generates screen data related to demand predicted power calculated by calculation, in particular, load following demand predicted power, and is a client information terminal via the communication unit 12 in the HTML (Hyper Text Markup Language) format or the like. 2 to control transmission.

  In FIG. 3, the structure of the information terminal in the electric power demand prediction system which concerns on one Embodiment of this invention is shown and demonstrated. As shown in the figure, the information terminal 2 includes a control unit 21 such as a CPU that performs overall control. The control unit 21 includes a communication unit 22, an operation unit 23, a display unit 24, and a storage unit 25. Connected with.

  In such a configuration, the communication unit 22 is a communication I / F for communicating with the server device 1 and the like via the network 4 such as the Internet. The operation unit 23 is an operation device such as a mouse or a keyboard. The display unit 24 is a display device such as a liquid crystal display. The operation unit 23 and the display unit 24 may be integrally configured as a touch panel. The storage unit 25 includes a ROM, a RAM, an HDD, and the like, stores a program 26 executed by the control unit 21, and provides a work area when the program 26 is executed.

  And the control part 21 functions as the main control part 21a, the browser part 21b, and the display control part 21c by executing the program 26 of the memory | storage part 25. FIG. The main control unit 21 a plays a general role such as communication with the server device 1. The browser unit 21b enables browsing based on HTML data related to demand predicted power sent from the server device 1, particularly load following demand predicted power. The display control unit 21 c controls display on the display unit 24.

  Hereinafter, the procedure for constructing the basic model formula will be described with reference to the flowchart of FIG.

  When the process is started, the main control unit 11a classifies data (for example, manufacturing industry, health facility, hospital, convenience store, etc.) for each area and each industry (S1).

  Subsequently, the analysis unit 11c extracts the actual demand power (hereinafter referred to as the total actual power) based on the past actual data (S2). Here, the load following performance power is also calculated simultaneously by the analysis unit 11c.

  And the analysis part 11c extracts the day when the maximum electric power before the predetermined number of days (3 days in this example) is given, and calculates the reference electric power B ′ based on the maximum electric power A [kW] ( S3). The reason why the predetermined number of days is set to 3 days in this example is that the forecast is preferably created no later than 2 days before the target date, and in this case, the data necessary for the forecast is only available for 3 days or more.

More specifically, the analysis unit 11c converts the past maximum power A into data A ′ having characteristics of the target date.
A ′ = A × total amount contracted power ratio × monthly load factor ratio For example, even if the contract status and the monthly load factor differ between the acquisition date of the past data and the target date, according to the above calculation, those characteristics are The maximum power A can be reflected. And the analysis part 11c calculates the reference electric power B 'which cuts off the past maximum electric power A by following Formula.
Reference power B ′ = A ′ × (target daily load additional contract power / target daily total contract power)

  Subsequently, the analysis unit 11c cuts the reference power B 'from the total actual power and sets the result as the reference value (S4).

More specifically, the analysis unit 11c calculates the load following contract power ratio r by the following equation in order to reflect the difference between the load following contract state on the day when the maximum power is in the past and the target day.
Load following contract power ratio r = (Target daily load following contract power / Maximum power daily load following contract power)
Then, the reference value is calculated by the following formula.
Reference value = (total amount actual power−reference power B ′) × load following contract power ratio r
Therefore, according to the above calculation, the calculated reference value sufficiently reflects the difference in the load following contract situation between the date when the maximum power is in the past and the target date.

  Next, the analysis unit 11c calculates a difference value (= y) between the reference value and the track-following actual power as shown in FIG. 6 (S5).

  Then, the analysis unit 11c is based on the average value of the difference values y and the time zone code x (x = 1 to 48) every 48 frames (in this example, 24 hours a day is classified into 48 frames in units of 30 minutes). A basic model formula is calculated using a statistical analysis method (S6).

  Here, the basic model equation is basically calculated in an appropriate block unit such as a year unit or a season unit. The basic model formula is calculated for each area and each industry. As a statistical analysis method, a method of obtaining an approximate expression from data distribution is adopted, but the method is not limited to this. In the present embodiment, the basic model formula is defined by a function of the time zone code x and the difference value y, but is not limited to this.

  In this way, as shown in FIG. 6, the entire region of the characteristic is divided into regions (regions a, b, c...) In a time zone where the significance is high, and the basic model formula is determined for each region. The level of significance may be determined mainly using statistical evaluation.

  Next, with reference to the flowchart of FIG.

  To outline this calculation process, the difference value is obtained from the basic model formula obtained earlier based on the time zone code, the load following demand forecast power is calculated, and the temperature, day of the week, sunshine hours, The degree of influence of each element, such as precipitation, is determined and automatically learned by machine learning, and the previously obtained model formula is updated as needed. Details will be described below.

  When entering this process, first, the analysis unit 11c calculates demand prediction power (total amount prediction power) for the entire amount from the past data for each area and each industry (S11). Here, the load following performance power is also calculated simultaneously by the analysis unit 11c.

  Subsequently, the analysis unit 11c calculates a reference value for the prediction date. That is, the day when the maximum power is output more than a predetermined number of days (three days in this example) on the predicted date is extracted, and the reference power B 'is calculated based on the maximum power A [kW] (S12).

Specifically, the analysis unit 11c converts the past maximum power A into data A ′ having characteristics of the target day.
A ′ = A × total amount contracted power ratio × monthly load factor ratio For example, even if the contract status and the monthly load factor differ between the acquisition date of the past data and the target date, according to the above calculation, those characteristics are The maximum power A can be reflected. And the analysis part 11c calculates the reference electric power B 'which cuts off the past maximum electric power A by following Formula.
Reference power B ′ = A ′ × (target daily load additional contract power / target daily total contract power)

  Then, the analysis unit 11c cuts the reference power B 'from the total actual power and sets the result as the reference value (S13).

In detail, the analysis unit 11c calculates the load additional power ratio r by the following equation in order to reflect the difference in the load additional contract state on the day when the past maximum power is output and the target day.
Load additional contract power ratio r = (Target daily load additional power / Maximum daily daily additional power)
And the analysis part 11c calculates a reference value by following Formula.
Reference value = (total amount actual power−reference power B ′) × load additional contract power ratio r

  Next, the predicting unit 11d substitutes the time zone code x in the basic model formula, and calculates a differential value y 'calculated recursively (S14). Then, the prediction unit 11d sets the reference value + difference value y ′ as the load following demand prediction power (S15).

  Then, the learning unit 11e calculates the degree of influence of each element such as temperature, day of the week, daylight hours, precipitation amount, and the like, and automatically learns it by machine learning to update the model formula. (S16).

  Next, update processing by model-type machine learning will be described in detail.

  First, the learning unit 11e determines the importance of each element for each area and each industry. That is, based on the load follow-up demand forecast power (reference forecast value) described above, a forecast value to which an element is applied is calculated every month and compared with the actual value. Here, the “element” means each element such as temperature, day of the week, sunshine duration, and precipitation.

  And the learning part 11e averages the difference which is a comparison result of each 48 frames | frames corresponding to the time slot | zone code x, and makes the factor into which the difference of a track record value and a predicted value fell below a predetermined ratio as an influence factor. .

For example, in the example of FIG.
Average difference between actual value and standard predicted value: 1.375
Average difference between actual value and predicted value reflecting element A: 0.25
Difference average between the actual value and the predicted value reflecting the element B: 5.20833
It becomes. For example, when the predetermined ratio is the difference average between the actual value and the reference predicted value, the element A that is less than that is set as an influence factor, and the element B that is higher than that is excluded from consideration in calculating the predicted value. About the classification of whether or not it is an influence factor, for example, the analysis state of each element is summarized in a list as shown in FIG.

  Subsequently, as illustrated in FIG. 9, the learning unit 11 e performs power demand prediction using, as a predicted value, an average of predicted values to which elements previously determined as influencing factors are applied among predicted values to which elements are applied. . And the prediction value derived | led-out from an electric power demand prediction is accumulate | stored as information, The improvement of prediction accuracy is aimed at by calculating the newest influence according to element from this accumulated information.

  In this way, the learning unit 11e automatically repeats the above steps after the predicted date, that is, the result of the target date, to recalculate the influencing factors, and automatically learns by machine learning based on the calculation results. To update the model formula.

  The recalculation of the predicted value used in the automatic update of the above model formula is realized by repeating the process as shown in FIG. In other words, the learning unit 11e can always update the model formula based on the latest predicted value through a series of “accumulation”, “learning”, and “re-prediction”.

  Specifically, the learning unit 11e performs daily calculations using a calculation formula derived from the elemental influence calculation, accumulates the results as information, and further calculates the latest elemental influences from the accumulated information. Thus, the prediction accuracy is improved, and the process of recalculating the prediction value again based on the learning result is repeated.

Therefore, according to the power demand prediction system according to the present embodiment, the following effects are produced.
・ Further spread of partial supply can be achieved by improving accuracy. This will reduce the barriers to entry of new power, and realize management normalization by improving imbalance and balance.
・ Improve the balance between demand and power generation. That is, it is possible to clarify the power demand that has become unclear due to large fluctuation due to partial power supply, and to improve the balance between the mismatched power generation and demand.

  Although one embodiment of the present invention has been described above, the present invention is not limited to this, and it is needless to say that various improvements and changes can be made without departing from the spirit of the present invention.

  For example, in the learning cycle of model formula machine learning, feedback of demand forecast power actually calculated using model formula, especially load forecast demand forecast power, is optimized and used to update model formula You may do it.

  DESCRIPTION OF SYMBOLS 1 ... Server apparatus, 2 ... Information terminal, 3 ... Server apparatus, 11 ... Control part, 11a ... Main control part, 11b ... Input control part, 11c ... Analysis part, 11d ... Prediction part, 11e ... Learning part, 11f ... Display Data generation unit, 12 ... communication unit, 13 ... storage unit, 14 ... power DB, 15 ... model formula DB, 21 ... control unit, 21a ... main control unit, 21b ... browser unit, 21c ... display control unit, 22 ... communication Part, 23 ... operation part, 24 ... display part, 25 ... storage part.

To solve the above problems, the power demand prediction system in accordance with the first aspect of the present invention, a solution analyzing unit you analyze historical power data even without low, a prediction unit that performs prediction of the demand predicted power A storage unit that stores at least power data , wherein the analysis unit identifies all amount actual power including the load following actual power from the past power data, and outputs the maximum power more than a predetermined number of days before the target date. And converting the maximum power into data A ′ having the characteristics of the target date according to the following equation:
A '= A x Contract power ratio x Monthly load ratio
A reference power B ′ for cutting off the maximum power is calculated by the following equation:
Reference power B ′ = A ′ × (Target daily load following contract power / Target daily total contract power)
In order to reflect the difference between the load following contract state on the day when the maximum power is output and the target day, the load following contract power ratio r is calculated by the following equation:
r = (target daily load following contract power / maximum power daily load following contract power)
The reference value is calculated by the following formula,
Reference value = (total amount actual power−reference power) × r
The prediction unit specifies a time zone code, the difference value by substituting the time zone code a function of the difference value, corresponding to the time period code and said reference value and the load follow performance power, the reference a value obtained by adding the difference value to a value with the load follow forecast power.

In the power demand prediction method according to the second aspect of the present invention, the analysis unit specifies the total amount actual power including the load following actual power from the past power data in the storage unit, and the maximum power before a predetermined number of days before the target date. Is extracted, and the maximum power is converted into data A ′ having the characteristics of the target date according to the following equation:
A '= A x Contract power ratio x Monthly load ratio
A reference power B ′ for cutting off the maximum power is calculated by the following equation:
Reference power B ′ = A ′ × (Target daily load following contract power / Target daily total contract power)
In order to reflect the difference between the load following contract state on the day when the maximum power is output and the target day, the load following contract power ratio r is calculated by the following equation:
r = (target daily load following contract power / maximum power daily load following contract power)
The reference value is calculated by the following formula,
Reference value = (total amount actual power−reference power) × r
Prediction unit, identifies the time zone code, the difference value by substituting the time zone code a function of the difference value, corresponding to the time period code and said reference value and the load follow performance power, the reference value a value obtained by adding the difference value to load follow forecast power.

Program according to the third aspect of the present invention, the computer stores a solution analyzing unit you analyze historical power data even without low, a prediction unit that performs prediction of the demand predicted power, at least power data The storage unit functions as a storage unit, and the analysis unit identifies all amount actual power including the load following actual power from the past power data, and extracts the day when the maximum power is more than a predetermined number of days before the target day. And converting the maximum power into data A ′ having the characteristics of the target date according to the following equation:
A '= A x Contract power ratio x Monthly load ratio
A reference power B ′ for cutting off the maximum power is calculated by the following equation:
Reference power B ′ = A ′ × (Target daily load following contract power / Target daily total contract power)
In order to reflect the difference between the load following contract state on the day when the maximum power is output and the target day, the load following contract power ratio r is calculated by the following equation:
r = (target daily load following contract power / maximum power daily load following contract power)
The reference value is calculated by the following formula,
Reference value = (total amount actual power−reference power) × r
The prediction unit specifies a time zone code, the difference value by substituting the time zone code a function of the difference value, corresponding to the time period code and said reference value and the load follow performance power, the reference a value obtained by adding the difference value to a value with the load follow forecast power.

Claims (12)

Analyzing at least past power data and accompanying data by statistical methods, and constructing a model formula used for power demand prediction,
A forecasting unit for forecasting the demand forecast power;
A storage unit for storing the model formula,
The analysis unit calculates the total amount actual power including the load following actual power from the past power data, extracts the day when the maximum power is more than a predetermined number of days before the forecast date, and based on the maximum power Calculate the power, subtract the reference power from the total actual power, the result as the reference value,
The prediction unit assigns a time zone code to the model formula composed of a function of a time zone code and a difference value between a reference value and a load tracking actual power, calculates a recursively calculated difference value, A power demand prediction system in which a value obtained by adding the difference value to a value is used as the load following demand prediction power. In the calculation of the reference value, the analysis unit calculates the load additional power ratio r by the following equation in order to reflect the difference in the load additional contract state on the day when the past maximum power is output and the target day,
r = (target daily load following contract power / maximum power daily load following contract power)
The reference value is calculated by the following formula
Reference value = (total amount actual power−reference power) × r
The power demand prediction system according to claim 1. The analysis unit
Extract the total amount of actual power including the load following actual power based on the past actual data, extract the day when the maximum power is more than the predetermined number of days before the forecast date, calculate the reference power based on the maximum power, Subtract the reference power from the total actual power, use the result as a reference value, calculate the difference value between the reference value and the load following actual power, and calculate the statistical value based on the average value and the time zone code of the difference value for each predetermined area The power demand prediction system according to claim 1, wherein the model formula is calculated using a statistical analysis method. A learning unit for updating the model formula by machine learning;
The learning unit calculates a predicted value to which the element is applied every month based on the predicted power following the cargo following demand, compares it with the actual value, and calculates a difference that is a comparison result of each area corresponding to the time zone code. Average the difference between the actual value and the predicted value as an influencing factor, and out of the predicted values to which the element is applied, the average of the predicted values to which the element that has been determined as the influencing factor is applied. The cumulative value is accumulated in the storage unit as a predicted value, the elemental influence degree is calculated based on the predicted value of the storage part, automatic learning is performed by machine learning based on the elemental influence degree, and the model formula is updated. 3. The power demand prediction system according to 3. The analysis unit calculates the total amount actual power including the load following actual power from the past power data, extracts the day when the maximum power is more than a predetermined number of days before the forecast date, and calculates the reference power based on the maximum power Calculate, subtract the reference power from the total amount actual power, the result as a reference value,
The prediction unit assigns the time zone code to the model formula that is a function of the time zone code and the difference value between the reference value and the tracked actual power, calculates the recursively calculated difference value, and the reference value A power demand prediction method in which a value obtained by adding the difference value to the load tracking demand prediction power is used. In the calculation of the reference value, the analysis unit calculates the load additional power ratio r by the following formula in order to reflect the difference in the load additional contract state on the day when the maximum power is in the past and the target day,
r = (target daily load following contract power / maximum power daily load following contract power)
The reference value is calculated by the following formula
Reference value = (total amount actual power−reference power) × r
The power demand prediction method according to claim 5. The analysis unit is
Extract the total amount of actual power including the load following actual power based on the past actual data, extract the day when the maximum power is more than the predetermined number of days before the forecast date, calculate the reference power based on the maximum power, Subtract the reference power from the total actual power, use the result as a reference value, calculate the difference value between the reference value and the load following actual power, and calculate the statistical value based on the average value and the time zone code of the difference value for each predetermined area The power demand prediction method according to claim 5 or 6, wherein the model formula is calculated using a statistical analysis method. The learning unit calculates the predicted value to which the element is applied every month based on the predicted power following the cargo follow-up, compares it with the actual value, and averages the difference that is the comparison result of each area corresponding to the time zone code Then, an element whose difference between the actual value and the predicted value falls below a predetermined ratio is used as an influencing factor, and among the predicted values to which the element is applied, an average of the predicted values to which the element that has been used as the influencing factor is predicted 8. The value is accumulated in the storage unit as a value, an elemental influence is calculated based on a predicted value of the storage, automatic learning is performed by machine learning based on the elemental influence, and the model formula is updated. The power demand forecast method described in 1. Computer
Analyzing at least past power data and accompanying data by statistical methods, and constructing a model formula used for power demand prediction,
A forecasting unit for forecasting the demand forecast power;
Function as a storage unit for storing the model formula;
The analysis unit calculates the total amount actual power including the load following actual power from the past power data, extracts the day when the maximum power is more than a predetermined number of days before the forecast date, and based on the maximum power Calculate the power, subtract the reference power from the total actual power, the result as the reference value,
The prediction unit assigns a time zone code to the model formula composed of a function of a time zone code and a difference value between a reference value and a load tracking actual power, calculates a recursively calculated difference value, A program in which a value obtained by adding the difference value to the value is used as the load following demand forecast power. In the calculation of the reference value, the analysis unit calculates the load additional power ratio r by the following equation in order to reflect the difference in the load additional contract state on the day when the past maximum power is output and the target day,
r = (target daily load following contract power / maximum power daily load following contract power)
The reference value is calculated by the following formula
Reference value = (total amount actual power−reference power) × r
The program according to claim 9. The analysis unit
Extract the total amount of actual power including the load following actual power based on the past actual data, extract the day when the maximum power is more than the predetermined number of days before the forecast date, calculate the reference power based on the maximum power, Subtract the reference power from the total actual power, use the result as a reference value, calculate the difference value between the reference value and the load following actual power, and calculate the statistical value based on the average value and the time zone code of the difference value for each predetermined area The program according to claim 9 or 10, wherein the model formula is calculated using a statistical analysis method. Computer
Function as a learning unit that updates the model formula by machine learning;
The learning unit calculates a predicted value to which the element is applied every month based on the predicted power following the cargo following demand, compares it with the actual value, and calculates a difference that is a comparison result of each area corresponding to the time zone code. Average the difference between the actual value and the predicted value as an influencing factor, and out of the predicted values to which the element is applied, the average of the predicted values to which the element that has been determined as the influencing factor is applied. The cumulative value is accumulated in the storage unit as a predicted value, the elemental influence degree is calculated based on the predicted value of the storage part, automatic learning is performed by machine learning based on the elemental influence degree, and the model formula is updated. 11. The program according to 11.