A multi-model fusion short-term load prediction method and device

Abstract

The application provides a multi-model fusion short-term load prediction method and device, and relates to the technical field of short-term load prediction. The method comprises the following steps: acquiring load data of a target power system at a current time, and acquiring date attributes, meteorological attributes and seasonal attributes at the current time and a target time respectively; inputting the load data at the current time, the date attributes, the meteorological attributes and the seasonal attributes at the current time, and the date attributes, the meteorological attributes and the seasonal attributes at the target time into a short-term load prediction model, and outputting predicted load data of the target power system at the target time, wherein the short-term load prediction model is constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network. The application can improve the accuracy of prediction.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of short-term load forecasting, in particular to a short-term load forecasting method and device based on multi-model fusion. BACKGROUND

[0002] With the development of economy and technology and the improvement of people's living standards, electric energy has become an essential secondary energy in people's production and life, bringing endless convenience to people's production and life. Therefore, ensuring the stable and reliable supply of electric energy has become one of the most important tasks of the power system.

[0003] Short-term load forecasting is one of the important tasks of the power system. Accurate and reliable short-term load forecasting results can effectively help the power system to make more scientific operation plans and scheduling plans, thereby effectively assisting the safe and stable operation of the power system. Therefore, for the power system, accurate and reliable short-term load forecasting is of great significance.

[0004] For short-term load forecasting in the power system, the traditional method generally uses a single algorithm based on statistics, machine learning, artificial intelligence, etc. However, due to the high complexity and uncertainty of load data, the traditional prediction method often fails to accurately capture its change rule. Moreover, a single neural network model often fails to comprehensively capture the complex characteristics of load data, resulting in limited prediction accuracy. SUMMARY

[0005] The present application provides a short-term load forecasting method and device based on multi-model fusion to solve the problem of inaccurate short-term load forecasting of the power system by a single algorithm in the prior art.

[0006] In a first aspect, the present application provides a short-term load forecasting method based on multi-model fusion, comprising:

[0007] Obtaining load data of a target power system at a current time, and obtaining date attributes, weather attributes and seasonal attributes at the current time and a target time respectively;

[0008] Inputting the load data at the current time, the date attributes, the weather attributes and the seasonal attributes at the current time, and the date attributes, the weather attributes and the seasonal attributes at the target time into a short-term load forecasting model, and outputting predicted load data of the target power system at the target time, wherein the short-term load forecasting model is constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network.

[0009] In a second aspect, the present application provides a short-term load forecasting device based on multi-model fusion, comprising:

[0010] The data acquisition module is configured to acquire load data of a target power system at a current time, and acquire date attributes, meteorological attributes and seasonal attributes at the current time and a target time respectively.

[0011] The load prediction module is configured to input the load data at the current time, the date attributes, the meteorological attributes and the seasonal attributes at the current time, and the date attributes, the meteorological attributes and the seasonal attributes at the target time into a short-term load prediction model, and output predicted load data of the target power system at the target time, wherein the short-term load prediction model is constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network.

[0012] The application provides a multi-model fusion short-term load prediction method and device. The load data of a target power system at a current time is acquired, and the date attributes, the meteorological attributes and the seasonal attributes at the current time and a target time are acquired respectively. The load data at the current time, the date attributes, the meteorological attributes and the seasonal attributes at the current time, and the date attributes, the meteorological attributes and the seasonal attributes at the target time are input into a short-term load prediction model, and predicted load data of the target power system at the target time is output. The short-term load prediction model is constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network. The application fuses the BP neural network, the LSTM neural network and the Logistic regression analysis network. The method can fully utilize the advantages of each model and capture complex features in the load data. The multi-model fusion strategy enables the prediction model to more comprehensively understand the load data, thereby improving the prediction accuracy. Since multiple models are combined, the application exhibits stronger robustness in dealing with load prediction requirements of different power systems, different time periods and different weather conditions, can better adapt to the volatility and uncertainty of load data, and improves the reliability of prediction. BRIEF DESCRIPTION OF DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the application, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only some embodiments of the application, and other drawings can be obtained by those skilled in the art without creative labor.

[0014] Figure 1 is the implementation flowchart of the multi-model fusion short-term load prediction method provided by the embodiments of the application;

[0015] Figure 2 is a structural schematic diagram of the short-term load prediction model provided by the embodiments of the application;

[0016] Figure 3is a structural schematic diagram of a multi-model fusion short-term load prediction device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0017] In the following description, for the purpose of explanation and not limitation, specific details are set forth, such as particular system configurations, techniques, etc., in order to provide a thorough understanding of the embodiments of the present application. However, it will be apparent to those skilled in the art that the present application can be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present application with unnecessary detail.

[0018] In order to make the purpose, technical scheme and advantages of the present application clearer, specific embodiments will be described below with reference to the accompanying drawings.

[0019] Traditional short-term load prediction is usually predicted by using a single neural network model, but a single neural network model often has difficulty in comprehensively capturing the complex characteristics of load data, resulting in limited load prediction accuracy. The present application proposes a multi-model fusion short-term load prediction method, which aims to fuse multiple neural network models by using an innovative model fusion strategy to fuse the above multiple different types of neural network models, dynamically adjust the weight parameters of each model, and continuously update the model parameters to obtain the prediction result for power system operation management. The present application not only provides accurate prediction results for power grid dispatching, energy management, and power market transactions, but also helps optimize resource allocation, improve economic efficiency, and ensure the safe and stable operation of the power grid.

[0020] Figure 1 The implementation flowchart of the multi-model fusion short-term load prediction method provided by the embodiments of the present application is described in detail as follows:

[0021] In step 101, the load data of the target power system at the current time is obtained, and the date attribute, weather attribute, and season attribute at the current time and the target time are obtained respectively.

[0022] Among them, the factors affecting the load data generally include date attribute, weather attribute, and season attribute.

[0023] The date attribute includes weekdays, weekends, and holidays. Under different date attributes, the load data generated in the power system is also different. For example, a residential community, when it is a weekday, the residents all go out to work, and the corresponding load data is relatively small; when it is a weekend, they all stay at home, and the load data generated at this time is correspondingly increased, and different situations may occur during holidays.

[0024] The meteorological attributes include temperature, humidity, wind speed, wind direction, and atmospheric pressure. The load data in the power system under different temperatures, different humidities, different wind speeds, different wind directions, and different atmospheric pressures are also different.

[0025] The seasonal attributes include spring, summer, autumn, and winter. Due to different average temperatures in different seasons and different natural phenomena, such as summer peak electricity consumption and winter heating demand, the load data of the same power system will also be different.

[0026] In the embodiments of the present application, when performing short-term load forecasting for a certain power system, the load data of the power system at the current time, the date attribute, the meteorological attribute, and the seasonal attribute, and the date attribute, the meteorological attribute, and the seasonal attribute of the required forecasting date are needed.

[0027] The embodiments of the present application can more comprehensively reflect the load change of the power system by considering the date attribute, the meteorological attribute, and the seasonal attribute of the current time and the target time. Different dates, meteorological conditions, and seasons will have a significant impact on the load. Including these attributes in the prediction model can significantly improve the adaptability of the model to different environments and conditions.

[0028] In step 102, the load data at the current time, the date attribute, the meteorological attribute, and the seasonal attribute, and the date attribute, the meteorological attribute, and the seasonal attribute of the target time are input into the short-term load forecasting model, and the predicted load data of the target power system at the target time is output. The short-term load forecasting model is constructed based on BP neural network, LSTM neural network, and Logistic regression analysis network.

[0029] Among them, the BP neural network performs well in handling nonlinear problems, the LSTM neural network is good at capturing long-term dependencies in time series data, and the Logistic regression analysis network helps to understand the logical relationship between variables and load.

[0030] In the embodiments of the present application, the load data at the current time, the date attribute, the meteorological attribute, and the seasonal attribute obtained in step 101, and the date attribute, the meteorological attribute, and the seasonal attribute of the target time are input into the short-term load forecasting model which has been trained, and the predicted load data at the target time is output.

[0031] Among them, the short-term load forecasting model which has been trained is constructed based on BP neural network, LSTM neural network, and Logistic regression analysis network.

[0032] The embodiments of the present application combine the advantages of BP neural network, LSTM neural network and Logistic regression analysis network to form a powerful prediction model. This multi-model fusion method can consider multiple factors to improve the accuracy of load prediction.

[0033] In addition, although multi-model fusion increases the complexity of the model, through reasonable model design and optimization, an efficient calculation process can be realized. In addition, accurate load prediction helps the power system to make scheduling and planning in advance, avoids power shortage or surplus caused by load fluctuation, and improves the operation efficiency of the whole system.

[0034] In a possible implementation, the short-term load prediction model can include a first weight model and a second weight model constructed based on a BP neural network, a first calculation model and a second calculation model constructed based on an LSTM neural network, and a regression analysis model constructed based on a Logistic regression analysis network.

[0035] The output end of the first weight model is connected to the first input end of the first calculation model, and the output end of the second weight model is connected to the first input end of the second calculation model.

[0036] The output end of the first calculation model is connected to the first input end of the regression analysis model, and the output end of the second calculation model is connected to the second input end of the regression analysis model.

[0037] Optionally, the short-term load prediction model in the embodiments of the present application is composed of the first weight model, the second weight model, the first calculation model, the second calculation model and the regression analysis model. The first weight model and the second weight model are both constructed based on a BP neural network, the first calculation model and the second calculation model are both constructed based on an LSTM neural network, and the regression analysis model is constructed based on a Logistic regression analysis network.

[0038] Referring to Figure 2 The output end of the first weight model 21 is connected to the first input end of the first calculation model 23, and the input end of the first weight model 21 inputs the date attribute of the current time and the target time.

[0039] The output end of the second weight model 22 is connected to the first input end of the second calculation model 24, and the input end of the second weight model 22 inputs the meteorological attribute and the seasonal attribute of the current time and the target time.

[0040] The output end of the first calculation model 23 is connected to the first input end of the regression analysis model 25, and the output end of the second calculation model 24 is connected to the second input end of the regression analysis model 25.

[0041] When the short-term load forecasting model in the embodiment of the present application is used to perform load forecasting, the date attribute of the current time and the target time is input into the first weight model, the meteorological attribute and the seasonal attribute of the current time and the target time are input into the second weight model, and the load data of the current time is input into the second input end of the first calculation model and the second input end of the second calculation model respectively, and the predicted load data of the target time is output from the output end of the regression analysis model.

[0042] The first weight model and the second weight model based on the BP neural network are constructed in the embodiment of the present application, so that the contribution of different input features to the prediction result can be flexibly adjusted, and the importance of the features can be accurately measured; the first calculation model and the second calculation model constructed by the LSTM neural network can fully utilize the characteristics of the time series data, and capture the long-term dependence relationship and the short-term fluctuation mode of the load data; the regression analysis model constructed by the Logistic regression analysis network can further analyze the nonlinear relationship between each input variable and the load, and improve the accuracy and robustness of the prediction. In addition, the introduction of the Logistic regression analysis model provides more interpretability for the prediction result, and through the regression analysis, the contribution degree of each input variable to the load prediction result can be understood, thereby helping the user to better understand the prediction logic of the model.

[0043] Meanwhile, the embodiment of the present application can more comprehensively capture the complex information in the load data through the complementation and cooperation between different models, thereby improving the accuracy and stability of the prediction; the combination of the weight model and the calculation model enables the model to dynamically adjust the weight distribution in the prediction process according to the importance of different input features, and further improves the prediction effect.

[0044] In a possible implementation manner, the training process of the short-term load forecasting model can be as follows:

[0045] The historical load data of each historical current time of the target power system and the historical date attribute, the historical meteorological attribute and the historical seasonal attribute corresponding to each historical load data are obtained, and the historical load data, the historical date attribute, the historical meteorological attribute and the historical seasonal attribute of the corresponding historical target time are obtained;

[0046] The first weight model and the second weight model are constructed by using the BP neural network, the first calculation model and the second calculation model are constructed by using the LSTM neural network, and the regression analysis model is constructed by using the Logistic regression analysis network;

[0047] The historical date attribute of each historical current time and the corresponding historical target time is taken as an input of the first weight model, the historical meteorological attribute and the historical seasonal attribute of each historical current time and the corresponding historical target time are taken as an input of the second weight model, and the historical load data of each historical current time is taken as an input of the first calculation model and the second calculation model respectively, and the historical load data of the corresponding historical target time is taken as an output of the regression analysis model, so as to train the short-term load prediction model.

[0048] Optionally, the training process of the short-term load prediction model in the embodiment of the application is as follows:

[0049] The historical load data, the historical date attribute, the historical meteorological attribute and the historical seasonal attribute of the target power system at each historical current time and the corresponding historical target time are acquired. Then, the first weight model and the second weight model are constructed by using the BP neural network, the first calculation model and the second calculation model are constructed by using the LSTM neural network, and the regression analysis model is constructed by using the Logistic regression analysis network. Then, the historical date attribute of each historical current time and the corresponding historical target time is taken as an input of the first weight model, the historical meteorological attribute and the historical seasonal attribute of each historical current time and the corresponding historical target time are taken as an input of the second weight model, and the historical load data of each historical current time is taken as an input of the first calculation model and the second calculation model respectively, and the historical load data of the corresponding historical target time is taken as an output of the regression analysis model, so as to train the short-term load prediction model.

[0050] The embodiment of the application trains different models (such as the first weight model, the second weight model, the first calculation model, the second calculation model and the regression analysis model), so that the trained short-term load prediction model can more flexibly adapt to different power load change scenarios; and the design of the sub-modules enables the short-term load prediction model to more quickly adjust the prediction strategy when facing different seasons, weather or special events, thereby improving the robustness and adaptability of the prediction. At the same time, by using efficient algorithms such as neural networks and regression analysis, the training process of the model can be more quickly performed, thereby shortening the preparation time of the prediction.

[0051] In a possible implementation, after the historical load data of each historical current time of the target power system and the historical date attribute, the historical meteorological attribute and the historical seasonal attribute corresponding to each historical load data are acquired, and the historical load data, the historical date attribute, the historical meteorological attribute and the historical seasonal attribute of the corresponding historical target time are acquired, the method can further include:

[0052] The missing part in the collected historical load data is filled by using the mean filling method;

[0053] The maximum-minimum value normalization method is used to standardize the historical date attribute, the historical meteorological attribute and the historical season attribute of each historical current time and the corresponding historical target time.

[0054] Optionally, the historical load data acquired before training of the short-term load prediction model may have missing, error and other situations at different times. The mean filling method is used to fill the missing part of the collected historical load data. The mean value of the missing data is calculated, and the average value is used to fill the missing data. That is, the data of the missing part is calculated by a first formula, and the first formula is:

[0055]

[0056] wherein, is the data of the filled missing part, is the number of data of the missing part, is whether the historical load data needs to be filled, is the numerical value of the historical power load data.

[0057] In addition, due to the difference in the numerical value units of the acquired historical date attribute, historical meteorological attribute and historical season attribute, in order to ensure the accuracy of the output predicted load data, the maximum-minimum value normalization method is used to standardize the historical date attribute, the historical meteorological attribute and the historical season attribute of each historical current time and the corresponding historical target time. That is, the data value of the historical date attribute, the historical meteorological attribute and the historical season attribute of each historical current time and the corresponding historical target time after standardization is calculated by a second formula, and the second formula is:

[0058]

[0059] wherein, is the data value of the historical date attribute, the historical meteorological attribute and the historical season attribute of each historical current time and the corresponding historical target time after standardization, is the historical date attribute, the historical meteorological attribute and the historical season attribute of each historical current time and the corresponding historical target time, is the minimum value of the data in the given data value standard range, is the maximum value of the data in the given data value standard range.

[0060] The data processed by the method can reduce input noise of the model, improve training effect and prediction accuracy of the model, and has important significance for constructing a power system load prediction model based on historical data, optimizing a scheduling model, and the like. In addition, the mean filling method and the maximum-minimum value normalization method are relatively simple and easy to implement, which reduces the technical threshold of data processing, so that more people can participate in the process of data preprocessing and model construction. The missing part in the collected historical load data is filled by using the mean filling method, which can effectively solve the data missing problem, ensure the integrity and continuity of the data, maintain the overall trend and distribution characteristics of the data, and avoid analysis deviation caused by data missing. The historical date attribute, the historical weather attribute, and the historical season attribute are standardized by using the maximum-minimum value normalization method, which can eliminate the dimensional difference between different attributes, so that these data can be compared and analyzed on the same scale, which is helpful to improve the accuracy and efficiency of subsequent data analysis.

[0061] In a possible implementation, after training the short-term load prediction model, the method can further include:

[0062] The particle swarm algorithm is used to optimize the parameters of the trained short-term load prediction model.

[0063] Optionally, the particle swarm algorithm is used to optimize the parameters of the trained short-term load prediction model, and the optimization is specifically as follows:

[0064] Step 1.1, all parameters in the short-term load prediction model are taken as a particle swarm, and the particle swarm is initialized, wherein the parameters include parameters in a first weight model, a second weight model, a first calculation model, a second calculation model, and a regression analysis model in the short-term load prediction model.

[0065] Step 1.2, the fitness of each particle is calculated.

[0066] Step 1.3, according to the fitness of each particle, the speed and position of the corresponding particle are updated.

[0067] Step 1.4, it is judged whether the iteration number corresponding to the current time reaches a preset maximum iteration number, if not, the speed and position of each particle after being updated are updated to the speed and position of each particle after being initialized, and step 1.2 is performed; if yes, the particle swarm corresponding to the speed and position of each particle after being updated is output.

[0068] Based on the above optimization process, the optimal parameters in the short-term load prediction model provided in the embodiment are obtained, and then the optimal short-term load prediction model is obtained.

[0069] The global search capability of the particle swarm algorithm is used in the embodiments of the present application to help find the parameter combination that makes the short-term load prediction model have the highest prediction accuracy, thereby improving the prediction accuracy. In addition, the particle swarm algorithm is used to optimize the parameters, which can make the short-term load prediction model better adapt to different load change modes and data characteristics, enhance the generalization capability of the model, and mean that the model can still maintain high prediction performance when facing new and unseen data, i.e., the short-term load prediction model after parameter optimization can exhibit stronger robustness and stability when facing fluctuations and abnormalities in load data. This means that the model can more reliably provide accurate prediction results in actual application.

[0070] In a possible implementation, inputting the load data of the current time, the date attribute, the weather attribute and the season attribute of the current time, and the date attribute, the weather attribute and the season attribute of the target time into the short-term load prediction model, and outputting the predicted load data of the target power system at the target time can include:

[0071] Inputting the date attribute of the current time and the date attribute of the target time into the first weight model, and outputting the first weight;

[0072] Inputting the weather attribute and the season attribute of the current time and the weather attribute and the season attribute of the target time into the second weight model, and outputting the second weight;

[0073] Inputting the first weight and the load data of the current time into the first calculation model, and outputting the first data;

[0074] Inputting the second weight and the load data of the current time into the second calculation model, and outputting the second data;

[0075] Inputting the first data and the second data into the regression analysis model, and outputting the predicted load data of the target power system at the target time.

[0076] Optionally, referring to Figure 2 Inputting the date attribute of the current time and the target time into the first weight model 21 in the short-term load prediction model, and outputting the first weight a; inputting the weather attribute and the season attribute of the current time and the target time into the second weight model 22 in the short-term load prediction model, and outputting the second weight b; inputting the first weight a and the load data of the current time into the first calculation model 23, and outputting the first data X1; inputting the second weight b and the load data of the current time into the second calculation model 24, and outputting the second data X2. Finally, inputting the first data X1 and the second data X2 into the regression analysis model 25, and outputting the predicted load data Y of the target power system at the target time, i.e., Y=m*X1+n*X2+C. Wherein, m and n are different coefficients, and C is a constant.

[0077] The embodiment of the application decomposes the complex load prediction problem into multiple sub-problems (such as date attribute weight calculation, meteorological / season attribute weight calculation, etc.), and processes them through special calculation models, simplifying the overall prediction process. The method allows parallel processing of different sub-problems, which can shorten the time required for prediction and improve calculation efficiency.

[0078] The application provides a multi-model fusion short-term load prediction method, which comprises the following steps: acquiring load data of a target power system at a current time, and acquiring date attributes, meteorological attributes and seasonal attributes at the current time and a target time respectively; inputting the load data at the current time, the date attributes, the meteorological attributes and the seasonal attributes at the current time, and the date attributes, the meteorological attributes and the seasonal attributes at the target time into a short-term load prediction model to output predicted load data of the target power system at the target time, wherein the short-term load prediction model is constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network. The application fuses the BP neural network, the LSTM neural network and the Logistic regression analysis network, which can fully utilize the advantages of each model, capture complex features in the load data, and improve the prediction accuracy. Since multiple models are combined, the application has stronger robustness in dealing with load prediction requirements of different power systems, different time periods and different weather conditions, can better adapt to the volatility and uncertainty of load data, and improve the reliability of prediction.

[0079] It should be understood that the size of the serial number of each step in the above embodiment does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiment of the application.

[0080] The following is the device embodiment of the application. For details not described in detail, please refer to the corresponding method embodiments described above.

[0081] Figure 3 The structure schematic diagram of the multi-model fusion short-term load prediction device provided by the embodiment of the application is shown. For the convenience of description, only the part related to the embodiment of the application is shown, and the details are as follows:

[0082] As shown in Figure 3 The multi-model fusion short-term load prediction device 3 comprises:

[0083] The data acquisition module 31 is configured to acquire load data of a target power system at a current time, and acquire date attributes, meteorological attributes and seasonal attributes at the current time and a target time respectively.

[0084] The load prediction module 32 is configured to input the load data at the current time, the date attribute, the weather attribute and the season attribute at the current time and the target time, and the date attribute, the weather attribute and the season attribute at the target time into a short-term load prediction model, and output the predicted load data of the target power system at the target time, wherein the short-term load prediction model is constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network.

[0085] The application provides a multi-model fusion short-term load prediction device. The load data of a target power system at a current time is obtained, and the date attribute, the weather attribute and the season attribute at the current time and the target time are obtained respectively. The load data at the current time, the date attribute, the weather attribute and the season attribute at the current time, and the date attribute, the weather attribute and the season attribute at the target time are input into a short-term load prediction model, and the predicted load data of the target power system at the target time is output. The short-term load prediction model is constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network. The application fuses the BP neural network, the LSTM neural network and the Logistic regression analysis network. The method can make full use of the advantages of each model and capture the complex characteristics in the load data. The multi-model fusion strategy makes the prediction model more comprehensively understand the load data, thereby improving the prediction accuracy. Since multiple models are combined, the application has stronger robustness in dealing with the load prediction demand of different power systems, different time periods and different weather conditions, can better adapt to the volatility and uncertainty of the load data, and improve the reliability of the prediction.

[0086] In a possible implementation, the short-term load prediction model comprises a first weight model and a second weight model constructed based on a BP neural network, a first calculation model and a second calculation model constructed based on an LSTM neural network, and a regression analysis model constructed based on a Logistic regression analysis network.

[0087] The output end of the first weight model is connected with the first input end of the first calculation model, and the output end of the second weight model is connected with the first input end of the second calculation model.

[0088] The output end of the first calculation model is connected with the first input end of the regression analysis model, and the output end of the second calculation model is connected with the second input end of the regression analysis model.

[0089] In a possible implementation, the load prediction module can be configured to:

[0090] The date attribute at the current time and the date attribute at the target time are input into the first weight model, and the first weight is output.

[0091] inputting the meteorological attribute and the seasonal attribute of the current time and the meteorological attribute and the seasonal attribute of the target time into a second weight model, and outputting a second weight;

[0092] inputting the first weight and the load data of the current time into a first calculation model, and outputting first data;

[0093] inputting the second weight and the load data of the current time into a second calculation model, and outputting second data;

[0094] inputting the first data and the second data into a regression analysis model, and outputting the predicted load data of the target power system at the target time.

[0095] In a possible implementation, the training process of the short-term load prediction model is as follows:

[0096] obtaining historical load data of each historical current time of the target power system, and historical date attributes, historical meteorological attributes and historical seasonal attributes corresponding to each historical load data, and obtaining historical load data, historical date attributes, historical meteorological attributes and historical seasonal attributes of a corresponding historical target time;

[0097] constructing the first weight model and the second weight model by using a BP neural network, constructing the first calculation model and the second calculation model by using an LSTM neural network, and constructing the regression analysis model by using a Logistic regression analysis network;

[0098] taking the historical date attributes of each historical current time and the corresponding historical target time as the input of the first weight model, taking the historical meteorological attributes and the historical seasonal attributes of each historical current time and the corresponding historical target time as the input of the second weight model, and taking the historical load data of each historical current time as the input of the first calculation model and the second calculation model respectively, and taking the historical load data of the corresponding historical target time as the output of the regression analysis model, and training the short-term load prediction model.

[0099] In a possible implementation, the device can further include a preprocessing module, which can be configured to:

[0100] filling the missing part in the collected historical load data by using a mean filling method;

[0101] standardizing the historical date attributes, the historical meteorological attributes and the historical seasonal attributes of each historical current time and the corresponding historical target time by using a maximum-minimum value normalization method.

[0102] In a possible implementation, the preprocessing module can be specifically configured to:

[0103] calculating the data of the missing part by using a first formula, the first formula being:

[0104]

[0105] wherein, is the data of the filled missing part, is the number of data of the missing part, is whether the historical load data needs to be filled, is the numerical value of the historical power load data.

[0106] In a possible implementation, the preprocessing module can be specifically configured to:

[0107] calculate the data value of the historical date attribute, the historical weather attribute and the historical season attribute of each historical current time and the corresponding historical target time after standardization processing by a second formula, the second formula being:

[0108]

[0109] wherein, is the data value of the historical date attribute, the historical weather attribute and the historical season attribute of each historical current time and the corresponding historical target time after standardization processing, is the historical date attribute, the historical weather attribute and the historical season attribute of each historical current time and the corresponding historical target time, is the minimum value of the data in the standard range of the given data value, is the maximum value of the data in the standard range of the given data value.

[0110] In a possible implementation, the apparatus can further include a parameter optimization module, which can be configured to:

[0111] perform parameter optimization on the trained short-term load prediction model by using a particle swarm algorithm.

[0112] In the above embodiments, the description of each embodiment has its own focus, and the parts not described or recorded in a certain embodiment can be referred to the relevant description of other embodiments.

[0113] Those skilled in the art can realize that the templates, units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0114] The modules / units, if realized in the form of software function units and sold or used as independent products, can be stored in a computer readable storage medium. Based on such understanding, all or part of the processes in the above-mentioned embodiment methods can also be completed by a computer program instructing related hardware, and the computer program can be stored in a computer readable storage medium. When the computer program is executed by a processor, the steps of each of the above-mentioned short-term load prediction methods based on multi-model fusion can be implemented. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms, etc. The computer readable medium can include any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory, random access memory, electrical carrier signal, telecommunication signal and software distribution medium, etc.

[0115] The above-mentioned embodiments are only used to illustrate the technical solutions of the present application, rather than limit them. Although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features. These modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application, and should be included in the protection scope of the present application.

Claims

1. A multi-model fusion short-term load forecasting method, characterized in that, The utility model relates to a kind of short-term load forecasting method and system of target power system, including: Obtain the load data of target power system of current time, and respectively obtain the date attribute, meteorological attribute and season attribute of current time and target time, the date attribute includes weekday, holiday and holiday, the meteorological attribute includes temperature, humidity, wind speed, wind direction and atmospheric pressure, the season attribute includes spring, summer, autumn and winter; Current time load data, date attribute, meteorological attribute and season attribute and target time date attribute, meteorological attribute and season attribute are input into short-term load forecasting model, and the predicted load data of the target power system at target time is output, the short-term load forecasting model is based on BP neural network, LSTM neural network and Logistic regression analysis network construction obtains; Wherein, the short-term load forecasting model includes the first weight model and the second weight model based on BP neural network construction, the first calculation model and the second calculation model based on LSTM neural network construction and the regression analysis model based on Logistic regression analysis network construction; The output end of the first weight model is connected with the first input end of the first calculation model, and the output end of the second weight model is connected with the first input end of the second calculation model; The output end of the first calculation model is connected with the first input end of the regression analysis model, and the output end of the second calculation model is connected with the second input end of the regression analysis model; Wherein, the current time load data, date attribute, meteorological attribute and season attribute and target time date attribute, meteorological attribute and season attribute are input into short-term load forecasting model, and the predicted load data of the target power system at target time is output, including: Current time date attribute and target time date attribute are input into the first weight model, and first weight is output; Current time meteorological attribute and season attribute and target time meteorological attribute and season attribute are input into the second weight model, and second weight is output; The first weight and current time load data are input into the first calculation model, and first data is output; The second weight and current time load data are input into the second calculation model, and second data is output; The first data and the second data are input into the regression analysis model, and the predicted load data of the target power system at target time is output. 2.The multi-model fused short-term load forecasting method of claim 1, wherein, The training process of the short-term load forecasting model is: Obtain the historical load data of each historical current time of target power system and the historical date attribute, historical meteorological attribute and historical season attribute corresponding to each historical load data, and obtain the historical load data, historical date attribute, historical meteorological attribute and historical season attribute of corresponding historical target time; First weight model and second weight model are constructed using BP neural network, first calculation model and second calculation model are constructed using LSTM neural network, and regression analysis model is constructed using Logistic regression analysis network; The historical date attribute of each historical current time and the corresponding historical target time is taken as an input of the first weight model, the historical meteorological attribute and the historical seasonal attribute of each historical current time and the corresponding historical target time are taken as an input of the second weight model, and the historical load data of each historical current time is taken as an input of the first calculation model and the second calculation model respectively, and the historical load data of the corresponding historical target time is taken as an output of the regression analysis model, so as to train the short-term load prediction model. 3.The multi-model fused short-term load forecasting method of claim 2, wherein, After the historical load data of each historical current time of the target power system and the historical date attribute, the historical meteorological attribute and the historical seasonal attribute corresponding to each historical load data are acquired, and the historical load data, the historical date attribute, the historical meteorological attribute and the historical seasonal attribute of the corresponding historical target time are acquired, the method further comprises: The missing part in the collected historical load data is filled by using a mean filling method; The historical date attribute, the historical meteorological attribute and the historical seasonal attribute of each historical current time and the corresponding historical target time are standardized by using a maximum-minimum value normalization method.

4. The multi-model fused short-term load forecasting method according to claim 3, characterized in that, The missing part in the collected historical load data is filled by using a mean filling method, which comprises: The data of the missing part is calculated by a first formula, and the first formula is: wherein, is the data of the missing part filled in, is the number of data of the missing part, is whether the historical load data needs to be filled in, is the numerical value of the historical power load data.

5. The multi-model fused short-term load forecasting method according to claim 3, characterized in that, The historical date attribute, the historical meteorological attribute and the historical seasonal attribute of each historical current time and the corresponding historical target time are standardized by using a maximum-minimum value normalization method, which comprises: The data value of the historical date attribute, the historical meteorological attribute and the historical seasonal attribute of each historical current time and the corresponding historical target time after the standardization is calculated by a second formula, and the second formula is: wherein, a data value of a historical date attribute, a historical weather attribute and a historical season attribute for each historical current time and a corresponding historical target time after standardization processing, a data value of a historical date attribute, a historical weather attribute and a historical season attribute for each historical current time and a corresponding historical target time, a data minimum value within a standard range of a given data value, a data maximum value within a standard range of a given data value. 6.The multi-model fusion short-term load forecasting method according to claim 2, characterized in that, After the short-term load prediction model is trained, the method further comprises: The trained short-term load prediction model is optimized by using a particle swarm algorithm.

7. A multi-model fusion short-term load forecasting device, characterized by, It comprises: A data acquisition module is configured to acquire load data of a target power system at a current time, and acquire date attributes, meteorological attributes and seasonal attributes at the current time and a target time respectively, the date attributes comprising weekdays, holidays and holidays, the meteorological attributes comprising temperature, humidity, wind speed, wind direction and atmospheric pressure, and the seasonal attributes comprising spring, summer, autumn and winter; A load prediction module is configured to input the load data, the date attributes, the meteorological attributes and the seasonal attributes at the current time and the date attributes, the meteorological attributes and the seasonal attributes at the target time into a short-term load prediction model, and output predicted load data of the target power system at the target time, the short-term load prediction model being constructed based on a BP neural network, an LSTM neural network and a Logistic regression analysis network; The short-term load prediction model comprises a first weight model and a second weight model constructed based on the BP neural network, a first calculation model and a second calculation model constructed based on the LSTM neural network, and a regression analysis model constructed based on the Logistic regression analysis network. An output end of the first weight model is connected with a first input end of the first calculation model, and an output end of the second weight model is connected with a first input end of the second calculation model; An output end of the first calculation model is connected with a first input end of the regression analysis model, and an output end of the second calculation model is connected with a second input end of the regression analysis model; The load prediction module is configured to: input a date attribute of a current time and a date attribute of a target time into the first weight model to output a first weight; input meteorological attributes and seasonal attributes of the current time and meteorological attributes and seasonal attributes of the target time into the second weight model to output a second weight; input the first weight and load data of the current time into the first calculation model to output first data; input the second weight and load data of the current time into the second calculation model to output second data; input the first data and the second data into the regression analysis model to output predicted load data of the target power system at the target time.

Images