JPWO2019038890A1 - Electric power demand prediction apparatus and electric power demand prediction method - Google Patents

Abstract

The power demand peak prediction unit (3) predicts the peak value of the daily power demand based on the linear regression model. The power demand transition prediction unit (6) predicts a relative value with respect to the peak value of the power demand for each time zone based on the nonlinear regression model. The time zone power demand prediction unit (7) predicts the power demand for each time zone based on the peak value of the daily power demand and the relative value to the peak value of the power demand for each time zone.

Description

  The present invention relates to a power demand prediction apparatus and a power demand prediction method for predicting power demand in a factory or the like.

  Conventionally, a method for predicting electric power demand using a machine learning method of a neural network is known. For example, Patent Document 1 predicts the power load of a plant using a neural network that is learned by inputting weather conditions, day of the week, distinction between weekdays and holidays, operation state of a plant, operation pattern of plant components, and the like. How to do is described.

JP 2003-84805 A

  Since the power demand of the factory depends on the production amount of the product, a production plan value is required to predict the power demand of the factory. The production plan value may be significantly changed due to a change in the factory system. In this case, the prediction method described in Patent Literature 1 has a problem in that extrapolation prediction using a neural network cannot be performed accurately on data outside the learning range.

  For example, an RBF (Radial Basis Function) network, which is one of neural networks, is excellent in the approximation capability of a nonlinear function, but a prediction function is not formed for data outside the learning range. For this reason, when data having a value far from the learning data is input to the network, the prediction result becomes “0”.

  Further, in order to obtain an acceptable prediction accuracy using a neural network such as an RBF network, it is necessary to accumulate a certain amount of learning data. For this reason, if the production plan is changed significantly, it is impossible to quickly predict the power demand.

  This invention solves the above-mentioned problem, and even when the production plan of the factory is changed, the power demand can be predicted for each time zone, and the change of the production plan can be followed quickly. An object of the present invention is to obtain a power demand prediction device and a power demand prediction method.

  The power demand prediction apparatus according to the present invention includes a first prediction unit, a second prediction unit, and a third prediction unit. The first prediction unit predicts the peak value of the daily power demand in the factory based on a linear regression model in which the peak value of the daily power demand in the factory changes linearly with respect to the production plan value. The second prediction unit predicts a relative value with respect to the peak value of the power demand for each time zone in the factory based on the nonlinear regression model. The third prediction unit is based on a peak value of the daily power demand predicted by the first prediction unit and a relative value to the peak value of the power demand for each time period predicted by the second prediction unit, Predict power demand by time of day.

  According to this invention, since the relative value with respect to the peak value of the power demand for each time zone is predicted based on a nonlinear regression model such as a neural network, the relative value can be predicted with high accuracy. Since the peak value of the daily power demand is predicted based on the linear regression model, the predicted value of the daily power demand peak value changes linearly with respect to the production plan value. Thereby, even if the production plan value is changed, it is possible to predict without excessively removing the power demand for each time value. In addition, since the peak value of power demand is based on a linear regression model, parameters necessary for prediction can be quickly learned with a small amount of learning data even when the production plan value is significantly changed.

It is a block diagram which shows the structure of the electric power demand prediction apparatus which concerns on Embodiment 1 of this invention. FIG. 2A is a block diagram showing a hardware configuration for realizing the function of the power demand prediction apparatus according to Embodiment 1. FIG. 2B is a block diagram illustrating a hardware configuration that executes software that implements the functions of the power demand prediction apparatus according to Embodiment 1. 3 is a flowchart showing an operation of a power demand peak learning unit in the first embodiment. 3 is a flowchart showing an operation of a power demand transition learning unit in the first embodiment. 3 is a flowchart showing an operation of a power demand peak prediction unit in the first embodiment. 3 is a graph showing an outline of daily peak demand prediction in Embodiment 1. 3 is a flowchart illustrating an operation of a power demand transition prediction unit in the first embodiment. 3 is a flowchart showing an operation of a time period power demand prediction unit in the first embodiment. 4 is a graph showing an outline of power demand prediction for each time zone in the first embodiment.

Hereinafter, in order to describe the present invention in more detail, modes for carrying out the present invention will be described with reference to the accompanying drawings.
Embodiment 1 FIG.
FIG. 1 is a block diagram showing a configuration of a power demand prediction apparatus 1 according to Embodiment 1 of the present invention. The power demand prediction device 1 is a device that predicts power demand in a factory. As shown in FIG. 1, a learning data storage unit 2, a power demand peak learning unit 3, a power demand transition learning unit 4, a power demand peak prediction. Unit 5, power demand transition prediction unit 6, and time zone power demand prediction unit 7.

The learning data storage unit 2 stores learning data used for learning by the power demand peak learning unit 3 and the power demand transition learning unit 4.
The learning data is a combination of data relating to power consumption, weather information, production plan or production result, day of the week, and time in the factory.

  For example, the learning data accumulation unit 2 accumulates learning power consumption data a1 as learning data for power consumption in a factory, accumulates learning weather information a2 as learning data for weather information, and records production results in the factory. Learning production results a3 are accumulated as learning data, learning day information a4 is accumulated as day learning data, and learning time information a5 is accumulated as time learning data.

  Further, the learning data storage unit 2 includes the current day weather information b1 indicating the weather condition of the prediction target day, the current day production plan b2 indicating the production plan value of the prediction target day, the current day information b3 indicating the day of the prediction target day, Current day time information b4 indicating the time on the prediction target day and current day power consumption data b5 indicating the power consumption in the factory on the prediction target day are stored. These data are stored in the learning data storage unit 2 to become the learning data described above.

  The power demand peak learning unit 3 generates the parameter c necessary for predicting the peak value of the daily power demand in the factory based on the learning data a1 to a4 stored in the learning data storage unit 2. 1 learning unit. The parameter c necessary for predicting the peak value of the daily power demand is a parameter of a prediction formula for predicting the peak value of the daily power demand using a linear regression model. For example, the coefficient of the explanatory variable in the prediction formula.

The prediction formula is a formula depending on at least the production plan of the factory, and may be a formula depending on weather information including the maximum daily temperature. Further, considering that the influence on the power demand varies depending on the product, a plurality of explanatory variables related to the production plan may be prepared.
For example, since the use frequency of the air controller in a factory changes according to the maximum temperature, the power demand also changes accordingly. Further, when the production plan value is high, the power demand increases accordingly. The power demand peak learning unit 3 learns the dependency of the weather information and the production plan on the peak value of the daily power demand, and generates the parameter c that reflects the dependency of the weather information and the production plan.

  The power demand transition learning unit 4 is necessary for predicting the relative value with respect to the peak value of the power demand for each time zone in the factory based on the learning data a1, a4, a5 stored in the learning data storage unit 2. It is the 2nd learning part which produces | generates the parameter d. The parameter d necessary for predicting the relative value with respect to the peak value of power demand for each time zone is a parameter for constructing a prediction model which is a nonlinear regression model such as a neural network.

The power demand peak prediction unit 5 predicts the peak value of the daily power demand in the factory based on a linear regression model in which the peak value of the daily power demand in the factory changes linearly with respect to the production plan value. 1 prediction unit.
For example, the power demand peak prediction unit 5 includes the parameter c generated by the power demand peak learning unit 3, the current day weather information b1 indicating the weather condition of the prediction target day, and the same day production plan b2 indicating the production plan value of the prediction target day. Based on the current day information b3 indicating the day of the week to be predicted, the peak value e of the daily power demand is predicted from the prediction formula.

The power demand transition prediction unit 6 is a second prediction unit that predicts a relative value f with respect to a peak value of power demand for each time zone in a factory based on a nonlinear regression model typified by a neural network.
For example, the power demand transition prediction unit 6 is based on the parameter d generated by the power demand transition learning unit 4, the current day information b3 indicating the day of the prediction target day, and the current day time information b4 indicating the time on the prediction target day. Thus, the relative value f with respect to the peak value of the power demand for each time zone is predicted from the prediction model.

  The time zone power demand forecasting unit 7 is relative to the peak value e of the daily power demand predicted by the power demand peak forecasting unit 5 and the peak value of power demand for each time zone forecasted by the power demand transition forecasting unit 6. It is the 3rd prediction part which predicts electric power demand for every time slot in a factory based on value f. The predicted value g of the power demand for each time zone in the factory is output from the time zone power demand prediction unit 7.

Up to now, a power demand prediction comprising a learning data storage unit 2, a power demand peak learning unit 3, a power demand transition learning unit 4, a power demand peak prediction unit 5, a power demand transition prediction unit 6, and a time zone power demand prediction unit 7 Although the apparatus 1 was shown, the electric power demand prediction apparatus 1 which concerns on Embodiment 1 is not limited to this structure.
For example, the learning data storage unit 2 may have a configuration included in an external device provided separately from the power demand prediction device 1.
In this case, the power demand peak learning unit 3 and the power demand transition learning unit 4 perform learning using the learning data input from the external device.

The learning data storage unit 2, the power demand peak learning unit 3, and the power demand transition learning unit 4 may be configured in an external device provided separately from the power demand prediction device 1.
In this case, the power demand peak prediction unit 5 and the power demand transition prediction unit 6 perform prediction using the parameter c input from the external device.
That is, the power demand prediction apparatus 1 according to Embodiment 1 only needs to include the power demand peak prediction unit 5, the power demand transition prediction unit 6, and the time period power demand prediction unit 7, and the learning data storage unit 2, The power demand peak learning unit 3 and the power demand transition learning unit 4 may be omitted.

FIG. 2A is a block diagram illustrating a hardware configuration that realizes the function of the power demand prediction apparatus 1. FIG. 2B is a block diagram illustrating a hardware configuration that executes software that implements the functions of the power demand prediction apparatus 1.
In the power demand prediction device 1, the learning data storage unit 2, the power demand peak learning unit 3, the power demand transition learning unit 4, the power demand peak prediction unit 5, the power demand transition prediction unit 6, and the time zone power demand prediction unit 7. Each function is realized by a processing circuit.
That is, the power demand prediction apparatus 1 includes a processing circuit for executing each process of the flowcharts shown in FIGS.
The processing circuit may be dedicated hardware or a CPU (Central Processing Unit) that executes a program stored in the memory.

When the processing circuit is the dedicated hardware shown in FIG. 2A, the processing circuit 100 includes, for example, a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an ASIC (Application Specific Integrated Circuit), an FPGA (FPGA). Field-Programmable Gate Array) or a combination thereof.
Each function of learning data storage unit 2, power demand peak learning unit 3, power demand transition learning unit 4, power demand peak prediction unit 5, power demand transition prediction unit 6 and time zone power demand prediction unit 7 is processed separately. These functions may be realized by a circuit, or these functions may be realized by a single processing circuit.

  When the processing circuit is the processor 101 shown in FIG. 2B, the learning data storage unit 2, the power demand peak learning unit 3, the power demand transition learning unit 4, the power demand peak prediction unit 5, the power demand transition prediction unit 6, and the time zone Each function of the power demand prediction unit 7 is realized by software, firmware, or a combination of software and firmware. Software or firmware is described as a program and stored in the memory 102.

The processor 101 reads out and executes a program stored in the memory 102 to thereby execute a learning data storage unit 2, a power demand peak learning unit 3, a power demand transition learning unit 4, a power demand peak prediction unit 5, and a power demand transition. The respective functions of the prediction unit 6 and the time zone power demand prediction unit 7 are realized. That is, the power demand prediction apparatus 1 includes a memory 102 for storing a program in which each process of the flowcharts shown in FIGS. 3 to 8 is executed as a result when executed by the processor 101.
These programs include the procedures of the learning data storage unit 2, the power demand peak learning unit 3, the power demand transition learning unit 4, the power demand peak prediction unit 5, the power demand transition prediction unit 6, and the time zone power demand prediction unit 7 or The method is executed by a computer.

  The memory 102 includes, for example, a nonvolatile or volatile semiconductor such as a RAM (Random Access Memory), a ROM (Read Only Memory), a flash memory, an EPROM (Erasable Programmable Read Only Memory), an EEPROM (Electrically-EPROM), or the like. Magnetic disks, flexible disks, optical disks, compact disks, mini disks, DVDs, and the like are applicable.

A part of each function of the learning data storage unit 2, the power demand peak learning unit 3, the power demand transition learning unit 4, the power demand peak prediction unit 5, the power demand transition prediction unit 6 and the time zone power demand prediction unit 7 It may be realized by dedicated hardware and a part may be realized by software or firmware.
For example, the learning data storage unit 2, the power demand peak learning unit 3, and the power demand transition learning unit 4 have their functions realized by the processing circuit 100 as dedicated hardware, and the power demand peak prediction unit 5, power demand The functions of the transition prediction unit 6 and the time zone power demand prediction unit 7 may be realized by the processor 101 reading and executing a program stored in the memory 102.
Thus, the processing circuit can realize each of the above functions by hardware, software, firmware, or a combination thereof.

Next, the operation will be described.
FIG. 3 is a flowchart showing the operation of the power demand peak learning unit 3.
First, the power demand peak learning unit 3 reads, from the learning data storage unit 2, a combination of data regarding power consumption, weather information, production plan or production result, and day of the week in the factory. For example, the power demand peak learning unit 3 reads learning power consumption data a1, learning weather information a2, learning production results a3, and learning day-of-week information a4 from the learning data storage unit 2.

Based on the data a1 to a4 read from the learning data storage unit 2, the power demand peak learning unit 3 sets a parameter c necessary for predicting the peak value of the daily power demand based on the linear regression model. Generate (step ST1).
The power demand peak learning unit 3 outputs the generated parameter c to the power demand peak prediction unit 5 (step ST2).

FIG. 4 is a flowchart showing the operation of the power demand transition learning unit 4.
First, the power demand transition learning unit 4 reads a combination of data regarding power consumption, day of the week, and time in the factory from the learning data storage unit 2. For example, the power demand transition learning unit 4 reads learning power consumption data a1, learning weather information a2, learning day information a4, and learning time information a5 from the learning data storage unit 2.

  Based on the data a2, a4, and a5 read from the learning data storage unit 2, the power demand transition learning unit 4 predicts the relative value f with respect to the peak value of the power demand for each time zone based on the nonlinear regression model. Therefore, a parameter d necessary for this is generated (step ST1a).

  For example, in the factory where the power demand transition learning unit 4 inputs the learning time information a5, the production temporarily stops when the worker changes in shift work and the power consumption in the factory decreases. Learn time dependency of power demand.

  In addition, the power demand transition learning unit 4 inputs the learning day information a4 so that there is no late-night work on the previous day at the beginning of the week, so the power consumption in the early morning is low in the factory, or the late-day work is not on the weekend. Therefore, it learns the day-time dependence of the power demand in the factory, such as low power consumption at midnight in the factory.

  The power demand transition learning unit 4 outputs the generated parameter d to the power demand transition prediction unit 6 (step ST2a).

FIG. 5 is a flowchart showing the operation of the power demand peak prediction unit 5.
The power demand peak predicting unit 5 uses the parameter c necessary for predicting the peak value of the daily power demand, the weather information on the forecast target day, the day of the week, and the production plan, and the peak value e of the daily power demand in the factory. Is predicted (step ST1b).

  The power demand peak prediction unit 5 predicts the peak value e of the daily power demand using a prediction formula based on the linear regression model. The prediction formula is a formula depending on, for example, weather information and a production plan. However, in a factory, even if the production plan value on a Saturday, which is generally a holiday, is “0”, production may actually be performed and power consumption may increase. In this case, power demand is predicted using only weather information for the prediction formula.

Therefore, the power demand peak prediction unit 5 may switch the prediction formula according to the day of the week.
For example, in addition to the prediction formula depending on the weather information and the production plan, the prediction formula depending on the weather information, the production plan and the day of the week is set in the power demand peak prediction unit 5. Thereby, the power demand peak prediction unit 5 can predict the peak value of the power demand that is not in the production plan as described above by predicting the peak value of the power demand using the prediction formula corresponding to the day of the week. it can.

The power demand peak prediction unit 5 outputs the peak value e of the daily power demand predicted as described above to the time period power demand prediction unit 7 (step ST2b).
FIG. 6 is a graph showing an overview of the peak value prediction of the daily power demand in the first embodiment. In FIG. 6, data A is data indicating the actual value of the daily power demand in the factory. The power demand peak prediction unit 5 predicts the peak value e of the power demand that is the maximum value B of the data A.

FIG. 7 is a flowchart showing the operation of the power demand transition prediction unit 6.
The power demand transition prediction unit 6 determines the peak of the power demand for each time zone based on the parameter d required for predicting the relative value with respect to the peak value of the power demand for each time zone, the day of the week and the time to be forecasted in the factory. A relative value f with respect to the value is predicted (step ST1c).

The power demand transition prediction unit 6 predicts a relative value f with respect to the peak value of the power demand for each time zone based on a nonlinear regression model such as a neural network.
For example, the relative value f may be a ratio or a difference for each time period with respect to the peak value of the daily power demand.

  The power demand transition prediction unit 6 outputs the relative value f with respect to the peak value of the power demand for each time zone predicted as described above to the time zone power demand prediction unit 7 (step ST2c).

FIG. 8 is a flowchart showing the operation of the time zone power demand prediction unit 7.
The time period power demand prediction unit 7 includes a peak value e of the daily power demand predicted by the power demand peak prediction unit 5 and a peak value of power demand for each time period predicted by the power demand transition prediction unit 6. Based on the relative value f with respect to the power demand for each time zone in the factory is predicted (step ST1d).

  FIG. 9 is a graph showing an outline of power demand prediction for each time zone in the first embodiment. In FIG. 9, data A1 is data indicating the actual value of the daily power demand in the factory. The time zone power demand prediction unit 7 predicts the predicted value data B1 of the daily power demand.

  For example, when the relative value f with respect to the peak value of the power demand for each time zone is a ratio for each time zone with respect to the peak value of the power demand, the time zone power demand prediction unit 7 peaks the relative value f for each time zone. By multiplying the value, a predicted value g of power demand for each time zone is calculated.

Moreover, when the relative value f with respect to the peak value of the power demand for each time zone is a difference for each time zone with respect to the peak value of the power demand, the time zone power demand prediction unit 7 peaks the relative value f for each time zone. By subtracting from the value, the predicted value g of the power demand for each time zone is calculated.
The time zone power demand prediction unit 7 outputs the predicted power demand g for each time zone to an output device such as a display monitor (step ST2d). Thereby, the manager of power demand can check the predicted value g of power demand for each time zone in the factory.

As described above, the power demand prediction apparatus 1 according to the first embodiment predicts the relative value f with respect to the peak value of the power demand for each time zone using a nonlinear regression model typified by a neural network. Predict with high accuracy. Since the peak value e of the daily power demand is predicted by the linear regression model with respect to the production plan value of the factory, the predicted value of the peak value e of the daily power demand changes linearly with respect to the production plan value. Thereby, even if the production plan value is changed, it is possible to predict without excessively removing the power demand for each time value.
In addition, since the peak value of power demand is based on a linear regression model, parameters necessary for prediction can be quickly learned with a small amount of learning data even when the production plan value is significantly changed.

  Note that the present invention is not limited to the above-described embodiment, and any component of the embodiment can be modified or any component of the embodiment can be omitted within the scope of the present invention.

  The power demand prediction apparatus according to the present invention can predict the power demand for each time zone and can quickly follow the change of the production plan even when the production plan of the factory is changed. Therefore, it can be used for a power demand prediction device in various factory facilities.

  DESCRIPTION OF SYMBOLS 1 Power demand prediction apparatus, 2 Data storage part for learning, 3 Power demand peak learning part, 4 Power demand transition learning part, 5 Power demand peak prediction part, 6 Power demand transition prediction part, 7 Time zone power demand prediction part, 100 Processing circuit, 101 processor, 102 memory.

Claims (4)

A first prediction unit that predicts a peak value of the daily power demand in the factory based on a linear regression model in which the peak value of the daily power demand in the factory changes linearly with respect to the production plan value;
A second prediction unit that predicts a relative value with respect to a peak value of power demand for each time zone in the factory based on a nonlinear regression model;
Based on the peak value of daily power demand predicted by the first prediction unit and the relative value to the peak value of power demand for each time zone predicted by the second prediction unit, the time zone in the factory A power demand prediction apparatus comprising: a third prediction unit that predicts power demand for each. A first learning unit that generates a parameter necessary for predicting a peak value of a daily power demand based on a combination of power consumption, weather information, a production plan or production result, and data relating to a day of the week in the factory;
A second learning unit that generates a parameter necessary for predicting a relative value with respect to a peak value of power demand for each time zone based on a combination of data on power consumption, day of the week, and time in the factory;
The first prediction unit predicts a peak value of a daily power demand based on the parameters generated by the first learning unit, weather information on a prediction target day, day of the week, and production plan,
The second predicting unit predicts a relative value with respect to a peak value of power demand for each time zone based on the parameter generated by the second learning unit, the day of the week and the time of prediction. The power demand prediction apparatus according to claim 1. A learning data storage unit for storing a combination of data relating to power consumption, weather information, production plan or production results, day of the week and time in the factory,
The first learning unit reads the power demand of the day based on a combination of power consumption, weather information, a production plan or production result, and data relating to a day of the week read from the learning data storage unit. Generate the parameters needed to predict peak values,
The second learning unit reads a relative value with respect to a peak value of power demand for each time zone based on a combination of data related to power consumption, day of the week, and time in the factory read from the learning data storage unit. The parameter required for prediction is produced | generated. The electric power demand prediction apparatus of Claim 2 characterized by the above-mentioned. A first prediction unit predicting a peak value of the daily power demand in the factory based on a linear regression model in which the peak value of the daily power demand in the factory changes linearly with respect to the production plan value; When,
A second predicting unit predicting a relative value with respect to a peak value of power demand for each time zone in the factory by a non-linear regression model;
The third prediction unit is based on the peak value of the daily power demand predicted by the first prediction unit and the relative value with respect to the peak value of the power demand for each time zone predicted by the second prediction unit. A method for predicting power demand for each time period.

Images