Park transferable load prediction method and device based on machine learning
Through a machine learning-based method, combining long and short-term memory networks and artificial neural network models, using load influencing factors and adjacent park data, the problem of insufficient power load prediction accuracy is solved, and more accurate transferable load prediction is achieved, supporting grid stability and power generation planning optimization.
Patent Information
- Application Number
- CN202510281532.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-11
- Publication Date
- 2025-07-25
AI Technical Summary
The existing power load prediction model has insufficient prediction accuracy under the influence of complex factors, especially after distributed energy resources and renewable energy are connected to the power grid, it is difficult to achieve accurate prediction.
Using a machine learning-based method, the first neural network model is used to predict the proportion of transferable load, and the second neural network model combines load influencing factors and adjacent campus data, and comprehensive predictions are made through long-term and short-term memory networks and artificial neural network models, combining external factors and historical data.
It improves the prediction accuracy of transferable loads, can better and reasonably schedule distributed resources, ensure stable operation of the power grid and optimize the power generation plan of the power supply company.
Smart Images
Figure CN120377225A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of power systems, and in particular, to a method and device for predicting transferable loads in a park based on machine learning. Background Art
[0002] With the global emphasis on environmental protection and sustainable development, the energy structure is shifting from traditional fossil fuels to clean energy. As the core link in energy conversion and distribution, the stability and efficiency of the power system are crucial. In the case of large-scale access of renewable energy (such as solar energy, wind energy) to the power grid, the volatility of power loads has increased significantly. At the same time, the penetration rate of distributed energy resources (such as household solar panels, small wind turbines) and energy storage devices in the power grid is continuously increasing, and various electrical equipment is emerging continuously. The access of these devices has changed the traditional power flow and load characteristics. Accurately predicting load changes is beneficial to the reasonable scheduling of these distributed resources to ensure the safe and stable operation of the power grid, and is also beneficial to power supply enterprises to reasonably arrange power generation plans and optimize the start-stop sequence and operation time of generating units.
[0003] In terms of traditional statistical prediction models, autoregressive moving average models (ARMA) and their extended autoregressive integrated moving average models (ARIMA) are mostly used. Since they are more suitable for processing stationary load sequences, and actual power loads are often affected by various complex factors, the prediction accuracy is limited. Summary of the Invention
[0004] Embodiments of the present invention provide a method and device for predicting transferable loads in a park based on machine learning to solve the problem of poor accuracy in existing power load prediction.
[0005] In a first aspect, embodiments of the present invention provide a method for predicting transferable loads in a park based on machine learning, including:
[0006] Inputting the load data of each device in the target power park in the current period into a first neural network model, and outputting a predicted value of the transferable load ratio corresponding to the prediction period, where the predicted value of the transferable load ratio is the percentage of the transferable load in the total load;
[0007] Inputting the total load and load influencing factors of the target power park in the current period into a second neural network model, and outputting a predicted value of the total load of the target power park in the prediction period;
[0008] Obtaining a matrix of transferable load percentages of adjacent parks of the target power park in the last N periods; the matrix of transferable load percentages includes the load percentages of each transferable load device in the adjacent parks in the last N periods;
[0009] Based on the predicted value of the proportion of transferable load, the predicted value of the total load in the prediction period of the target power park, and the matrix of the percentages of transferable load in the adjacent parks in the most recent N periods, the transferable load data of the target power park in the prediction period is obtained.
[0010] In one embodiment, before inputting the load data of each device in the target power park in the current period into the first neural network model, the method further includes:
[0011] Obtain the load data of each device in the target power park in the current period;
[0012] Divide the load data of each device in the target power park in the current period by the total load in the current period to obtain the load percentage of each device in the target power park in the current period;
[0013] Select the load percentages of the devices within the preset numerical range as the preprocessed load percentages;
[0014] Correspondingly, inputting the load data of each device in the target power park in the current period into the first neural network model and outputting the predicted value of the proportion of transferable load in the prediction period includes:
[0015] Input the preprocessed load percentages of each device in the current period into the first neural network model and output the predicted value of the proportion of transferable load in the prediction period.
[0016] In one embodiment, the first neural network model is a long short-term memory network model, and the long short-term memory network model includes one layer of unidirectional long short-term memory network and one layer of bidirectional long short-term memory network.
[0017] In one embodiment, the load influencing factors include the behaviors of electricity users and environmental information; inputting the total load and the load influencing factors of the target power park in the current period into the second neural network model and outputting the predicted value of the total load of the target power park in the prediction period includes:
[0018] Adopt the Pearson correlation method to extract the main influencing factors in the load influencing factors of the target power park;
[0019] Use the K-Means algorithm to classify the main influencing factors in the load influencing factors to obtain the dominant load influencing factors and the non-dominant load influencing factors;
[0020] Input the total load of the target power park in the current period and the dominant load influencing factors into the second neural network model and output the predicted value of the total load of the target power park in the prediction period.
[0021] In one embodiment, the second neural network model includes an ANN model.
[0022] In one embodiment, obtaining the transferable load data of the target power park in the prediction period based on the predicted value of the proportion of transferable load, the predicted value of the total load in the prediction period of the target power park, and the matrix of transferable load percentages of the adjacent parks in the most recent N periods includes:
[0023]
[0024] Wherein, represents the matrix of transferable load percentages, represents the percentage of the transferable load of the kth transferable device at the tth moment on the ith day relative to the total load of the kth transferable device at the tth moment on the ith day; the total load of the transferable device is the sum of the transferable load and the non-schedulable load; LV(t) represents the predicted value of the proportion of transferable load corresponding to the tth moment in the prediction period, and TL(t) represents the predicted value of the total load corresponding to the tth moment in the prediction period; h represents the number of moments in a single period; represents the transferable load data of the target power park at the tth moment in the prediction period.
[0025] In one embodiment, after inputting the total load and load influencing factors of the target power park in the current period into the second neural network model and outputting the predicted value of the total load of the target power park in the prediction period, the method further includes:
[0026] Obtaining the true value of the total load of the target power park in the prediction period;
[0027] Based on the formula Determining the prediction evaluation result of the second neural network model, and updating the hyperparameters of the second neural network model based on the prediction evaluation result of the second neural network model;
[0028] Wherein, y i represents the true value of the total load, represents the predicted value of the total load, represents the average value of the predicted values of the total load, and n represents the number of samples.
[0029] In one embodiment, after outputting the predicted value of the proportion of transferable load corresponding to the prediction period, the method further includes:
[0030] Obtaining the true value of the proportion of transferable load corresponding to the prediction period;
[0031] Based on the formula Determine the prediction evaluation result of the first neural network model, and update the hyperparameters of the first neural network model based on the prediction evaluation result of the first neural network model;
[0032] where x i represents the true value of the transferable load ratio, represents the predicted value of the transferable load ratio, represents the average value of the predicted values of the transferable load ratio, and n represents the number of samples.
[0033] In a second aspect, an embodiment of the present invention provides a prediction device for transferable load in a park based on machine learning, including:
[0034] A module for obtaining the predicted value of the transferable load ratio, configured to input the load data of each device in the target power park in the current period into the first neural network model, and output the predicted value of the transferable load ratio corresponding to the prediction period, where the predicted value of the transferable load ratio is the percentage of the transferable load in the total load;
[0035] A module for obtaining the predicted value of the total load, configured to input the total load and load influencing factors of the target power park in the current period into the second neural network model, and output the predicted value of the total load of the target power park in the prediction period;
[0036] A module for obtaining the transferable load percentage matrix, configured to obtain the transferable load percentage matrix of the adjacent parks of the target power park in the most recent N periods; the transferable load percentage matrix includes the load percentages of each transferable load device in the adjacent parks in the most recent N periods;
[0037] A transferable load prediction module, configured to obtain the transferable load data of the target power park in the prediction period based on the predicted value of the transferable load ratio, the predicted value of the total load of the target power park in the prediction period, and the transferable load percentage matrix of the adjacent parks in the most recent N periods.
[0038] This embodiment provides a method and device for predicting transferable load in a park based on machine learning. The method first inputs the load data of each device in the target power park in the current period into the first neural network model, and outputs the predicted value of the transferable load ratio in the prediction period. The predicted value of the transferable load ratio is the percentage of the transferable load in the total load. Then, the total load and load influencing factors of the target power park in the current period are input into the second neural network model, and the predicted value of the total load of the target power park in the prediction period is output. Obtain the transferable load percentage matrix of the adjacent parks of the target power park in the recent N periods. The transferable load percentage matrix includes the load percentages of each transferable load device in the adjacent parks in the recent N periods. Finally, based on the predicted value of the transferable load ratio, the predicted value of the total load of the target power park in the prediction period, and the transferable load percentage matrix of the adjacent parks in the recent N periods, the transferable load data of the target power park in the prediction period is obtained. Among them, the first neural network model uses load time series data to predict the change of transferable load, the second neural network model combines influencing factors and historical load data for comprehensive total load prediction, and at the same time supplements the load information of transferable devices with the load data of adjacent parks. The three work together to improve the prediction accuracy of transferable load. BRIEF DESCRIPTION OF THE DRAWINGS
[0039] Figure 1 is a flowchart of the implementation of the method for predicting transferable load in a park based on machine learning provided by an embodiment of the present invention;
[0040] Figure 2 is a schematic structural diagram of the device for predicting transferable load in a park based on machine learning provided by an embodiment of the present invention;
[0041] Figure 3 is a schematic diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS
[0042] The following will describe the embodiments of the present invention in detail with reference to the drawings.
[0043] See Figure 1 , which shows a flowchart of the implementation of the method for predicting transferable load in a park based on machine learning provided by an embodiment of the present invention, and is described in detail as follows:
[0044] S101: Input the load data of each device in the target power park in the current period into the first neural network model, and output the predicted value of the transferable load ratio in the prediction period. The predicted value of the transferable load ratio is the percentage of the transferable load in the total load.
[0045] In this embodiment, the target power park may be a nano-grid park. A nano-grid is a small, self-sufficient power system that can operate independently of the traditional power grid and is typically used to supply power to a building or a single load. It is a small-scale micro-grid. The shiftable load is the load data of equipment whose electricity demand can be shifted from one time period to another through incentives. The equipment with shiftable load includes dryers, washing machines, dishwashers, etc.
[0046] A single cycle can be one week, one day, or N hours. Preferably, a single cycle is one day.
[0047] First, obtain the time series of load data of the target power park in the current cycle, and preprocess the time series of load data to remove outliers and noise data, so that the prediction result will not be skewed. The prediction of the time series of load data is performed on the premise that the load data is stationary, indicating that the statistical parameters (i.e., the mean and variance) will not change. To test this hypothesis, the data is analyzed; the mean and variance are calculated, and the stationary property is verified.
[0048] Specifically, the time series of load data in the current cycle includes load data at multiple moments, and the time interval between each moment can be 1 minute. To improve the calculation efficiency, after obtaining the time series of load data, the single cycle is divided into multiple time periods, and the load data at each moment within each time period is averaged to obtain the load data corresponding to each time period, resulting in a time series of load data in units of time periods. Among them, a time period can be 1 hour, 2 hours, 3 hours, etc.
[0049] In a possible implementation manner, before S101, the method provided in this embodiment further includes:
[0050] Obtain the load data of each device in the target power park in the current cycle;
[0051] Divide the load data of each device in the target power park in the current cycle by the total load in the current cycle to obtain the load percentage of each device in the target power park in the current cycle;
[0052] Select the load percentages of the devices within the preset numerical range as the preprocessed load percentages;
[0053] Correspondingly, the specific implementation process of S101 includes:
[0054] Input the preprocessed load percentage of each device in the current cycle into the first neural network model, and output the predicted value of the proportion of the shiftable load in the prediction cycle.
[0055] In this embodiment, each load data is divided by the total load of the target power park in the current cycle to obtain a load percentage. Then, the load percentages of the devices within a preset value range are selected as the preprocessed load percentages, and other load percentages are excluded because these loads will cause the basic load to always be in the on state. Specifically, 2.5% - 97.5% is used as the preset value range, and only the load data between 2.5% and 97.5% is considered, as Figure 2 shown, which includes transferable loads and non-schedulable loads. The generated load percentage time series is used as the input of the first neural network module.
[0056] Specifically, after obtaining the load data time series of all devices in the current cycle, the data set is divided in a ratio of 80:20 for the training and testing purposes of the first neural network model.
[0057] In a possible implementation manner, the first neural network model is a long short-term memory network model, and the long short-term memory network model includes a layer of unidirectional long short-term memory network and a layer of bidirectional long short-term memory network.
[0058] In this embodiment, the first neural network model can be a long short-term memory network model SLSTM, which sequentially includes a long short-term memory network (Long Short-Term Memory, LSTM), a first dropout layer, a bidirectional long short-term memory network (Bi-LSTM), a second dropout layer, a first dense layer, and a second dense layer. The Bi-LSTM processes the input load percentage time series forward and backward; compared with the LSTM, it captures the time-dependent relationship more accurately. Therefore, the SLSTM model is a combination of LSTM and Bi-LSTM, which is used to accurately predict the load of the nanogrid that exhibits highly fluctuating properties.
[0059] The SLSTM model predicts the load change of the next day including the transferable load and non-schedulable load of the nanogrid park. A dropout layer with a value of 10% is introduced after the LSTM and bidirectional LSTM layers to reduce overfitting. Both the LSTM and bidirectional LSTM include a forget gate layer, an input gate layer, a memory cell layer, an output gate layer, and a dense layer. The Sigmoid function is used in its forget gate layer, and the hyperbolic tangent function is used as its output gate. The Rectified Linear Unit is used as the activation function in the dense layer. The input data of the SLSTM model is converted into an m*n-dimensional matrix, where m is proportional to the length of the load data time series, and n is the "window size". This window size determines how much past load data should be combined to predict the output. Specifically, the output of the SLSTM model is represented as the predicted value LV(t) of the proportion of the transferable load, where t ranges from 1 to 24, representing the hours of a day.
[0060] In a possible implementation, after S101, the method provided in this embodiment further includes:
[0061] Obtaining the true value of the transferable load ratio corresponding to the prediction period;
[0062] Based on the formula Determining the prediction evaluation result of the first neural network model, and updating the hyperparameters of the first neural network model based on the prediction evaluation result of the first neural network model;
[0063] where x i represents the true value of the transferable load ratio, represents the predicted value of the transferable load ratio, represents the average value of the predicted values of the transferable load ratio, and n represents the number of samples.
[0064] Specifically, when any one of the above R 2 , MAPE, and MSE does not meet the corresponding conditions, it is considered that the prediction evaluation result of the first neural network model is unqualified, and the hyperparameters of the first neural network model need to be updated. When the above R 2 , MAPE, and MSE all meet the corresponding conditions, it is considered that the first neural network model has been successfully trained, and the subsequent evaluation steps may not be continued.
[0065] S102: Based on the load data of each device in the target power park in the current period, determining the total load of the target power park in the current period, and inputting the total load of the target power park in the current period and the load influencing factors into the second neural network model, and outputting the predicted value of the total load of the target power park in the prediction period.
[0066] Since LSTM mainly depends on the historical time series of load data and cannot effectively incorporate sudden load changes that may be caused by external extreme weather conditions or highly fluctuating user behavior, which may significantly change the power load. If such situations are not considered in the load data, it may lead to inaccurate results. Therefore, the second neural network model provided in this embodiment includes the above sudden changes and is integrated with the SLSTM model, thereby embedding external sudden changes and historical loads together.
[0067] Specifically, the influencing factors include environmental information and user electricity consumption behavior, where the environmental information includes meteorological factors such as wind speed, humidity, temperature, and atmospheric pressure. The second neural network model includes an ANN (Artificial Neural Network) model. The ANN model includes an input layer, a first hidden layer, a second hidden layer, a third hidden layer, and an output layer. Both the first hidden layer and the second hidden layer include 30 fully connected neurons, and the third hidden layer includes 20 fully connected neurons. In addition, the sigmoid activation function is used together with the Adam optimizer to update the weights of each layer.
[0068] In a possible implementation, the load influencing factors include electricity user behavior and environmental information; the specific implementation process of S102 includes:
[0069] Adopt the Pearson correlation method to extract the main influencing factors among the load influencing factors of the target power park;
[0070] Use the K-Means algorithm to classify the main influencing factors among the load influencing factors to obtain dominant load influencing factors and non-dominant load influencing factors;
[0071] Input the total load of the target power park in the current period and the dominant load influencing factors into the second neural network model, and output the predicted value of the total load of the target power park in the prediction period.
[0072] Specifically, for the load data, preprocess the load data of each device in the current period in the target power park to eliminate outliers, fill in missing values, and average the load data in terms of time to the load data of each time period. Finally, add the load data of each device in the same time period in the current period to obtain the total load of each time period in the current period of the target power park. At the same time, for the influencing factors, use the Pearson correlation method to extract the main influencing factors from the influencing factors. Use the K-Means algorithm to divide the main influencing factors into dominant and non-dominant, and then retain the dominant main factors. Group and shuffle the processed load data and dominant influencing factors, that is, group the total load and dominant influencing factors in the same time period into one group and input them into the ANN model for predicting the total load of the target power park.
[0073] In a possible implementation, after S102, the implementation method of this embodiment further includes:
[0074] Obtain the true value of the total load of the target power park in the prediction period;
[0075] Based on the formula Determine the prediction evaluation result of the second neural network model, and update the hyperparameters of the second neural network model based on the prediction evaluation result of the second neural network model;
[0076] where y i represents the true value of the total load, represents the predicted value of the total load, represents the average value of the predicted values of the total load, and n represents the number of samples.
[0077] Specifically, when any one of the above R 2 , MAPE, and MSE does not meet the corresponding conditions, it is considered that the prediction evaluation result of the second neural network model is unqualified, and the hyperparameters of the second neural network model need to be updated. When the above R 2 , MAPE, and MSE all meet the corresponding conditions, it is considered that the second neural network model has been successfully trained, and the subsequent evaluation steps may not be continued.
[0078] S103: Obtain the transferable load percentage matrix of the adjacent parks of the target power park in the most recent N cycles; the transferable load percentage matrix includes the load percentages of each transferable load device in the adjacent parks in the most recent N cycles.
[0079] In this embodiment, an SEM module is set up. Inside this SEM module, a matrix is developed using the transferable load data from adjacent power parks, and the adjacent power parks have load curves comparable and similar to those of the target power park. The reason for integrating a separate "transferable device matrix" is to popularize the transferable load prediction method. In this way, even if the transferable load data of a specific nanogrid is not available, users can implement this method by integrating other similar transferable load data from nearby nanogrid parks or online repositories.
[0080] The transferable load percentage matrix of the adjacent power parks includes multiple sub - matrices, and each sub - matrix corresponds to the transferable load percentage matrix of the adjacent park in one cycle. The elements in the transferable load percentage matrix represent the percentage of the average value of the transferable load data of a single device corresponding to each time period to the total transferable load. This operation is repeated for all transferable load devices to generate a stack of matrices corresponding to each device. Since its operation plan will not deviate too much due to response influencing factors, thus making the result unaffected by changes, even if the average value of the transferable load percentage is used, the accuracy of the prediction result will not be endangered.
[0081] S104: Obtain the transferable load data of the target power park during the prediction period based on the predicted value of the proportion of transferable load, the predicted value of the total load of the target power park during the prediction period, and the matrix of the percentages of transferable load of the adjacent parks in the most recent N periods.
[0082] In a possible implementation manner, the specific implementation process of S104 includes:
[0083]
[0084] Among them, represents the matrix of percentages of transferable load, represents the percentage of the transferable load relative to the total load of the kth transferable device at the tth moment on the ith day; the total load of the transferable device is the sum of the transferable load and the non-schedulable load; LV(t) represents the predicted value of the proportion of transferable load corresponding to the tth moment during the prediction period, and TL(t) represents the predicted value of the total load corresponding to the tth moment during the prediction period; h represents the number of moments in a single period; represents the transferable load data of the target power park at the tth moment during the prediction period.
[0085] Specifically, the SLSTM model mainly focuses on the time-series analysis of load data, processes the input load data, first performs data preprocessing, eliminates outliers and extracts the change amount, and then inputs the processed load data percentage into the SLSTM model to output the "predicted value of the proportion of transferable load". This value captures the dynamic characteristics of load changes and provides an adjustment factor for short-term load prediction.
[0086] The ANN model focuses on predicting the total load from a global perspective. The ANN model uses the historical total load of the nano-grid and the exogenous influencing factors related to the load, models the total load through an artificial neural network model, and finally generates the predicted value of the total load, providing a control benchmark for the total amount in the prediction of transferable load.
[0087] The core of the SEM module is to extract the time characteristics and distribution rules of device loads through historical load data. First, preprocess the historical data, convert the transferable load into a transferable load matrix, which represents the load distribution in each time period of N periods. Subsequently, standardize this matrix into a matrix of percentages of transferable load, describing the load proportion of the devices with transferable load at different time periods. The goal of the SEM module is to supplement the fine-grained information of time and device distribution in load prediction through historical feature extraction, thereby improving the prediction accuracy of transferable load.
[0088] It should be understood that the sequence numbers of the steps in the above embodiments do not imply the order of execution. The execution order of each process should be determined according to its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.
[0089] The following is an apparatus embodiment of the present invention. For the details not described in detail herein, reference may be made to the corresponding method embodiments above.
[0090] Figure 2 The structural schematic diagram of the park transferable load prediction device based on machine learning provided by the embodiment of the present invention is shown. For the sake of convenience of description, only the parts related to the embodiment of the present invention are shown and are described in detail as follows:
[0091] As Figure 2 shown, the park transferable load prediction device 100 based on machine learning includes:
[0092] A transferable load ratio prediction value acquisition module 110, configured to input the load data of each device in the target power park in the current period into a first neural network model, and output a predicted transferable load ratio value corresponding to the prediction period, where the predicted transferable load ratio value is the percentage of the transferable load in the total load;
[0093] A total load prediction value acquisition module 120, configured to determine the total load of the target power park in the current period based on the load data of each device in the target power park in the current period, and input the total load and load influencing factors of the target power park in the current period into a second neural network model, and output a predicted total load value of the target power park in the prediction period;
[0094] A transferable load percentage matrix acquisition module 130, configured to acquire a transferable load percentage matrix of adjacent parks of the target power park in the most recent N periods; the transferable load percentage matrix includes the load percentages of each transferable load device in the adjacent parks in the most recent N periods;
[0095] A transferable load prediction module 140, configured to obtain the transferable load data of the target power park in the prediction period based on the predicted transferable load ratio value, the predicted total load value of the target power park in the prediction period, and the transferable load percentage matrix of the adjacent parks in the most recent N periods.
[0096] In a possible implementation manner, the park transferable load prediction device 100 based on machine learning further includes a data preprocessing module, configured to:
[0097] Acquire the load data of each device in the target power park in the current period;
[0098] Divide the load data of each device in the target power park in the current period by the total load in the current period to obtain the load percentage of each device in the target power park in the current period;
[0099] Select the load percentages of devices within the preset numerical range as the preprocessed load percentages;
[0100] Correspondingly, the transferable load ratio prediction value acquisition module 110 is used for:
[0101] Input the preprocessed load percentages of each device in the current period into the first neural network model, and output the predicted transferable load ratio prediction value in the prediction period.
[0102] In a possible implementation manner, the first neural network model is a long short-term memory network model, and the long short-term memory network model includes a one-way long short-term memory network layer and a bidirectional long short-term memory network layer.
[0103] In a possible implementation manner, the load influencing factors include electricity user behavior and environmental information; the total load prediction value acquisition module 120 includes:
[0104] Adopt the Pearson correlation method to extract the main influencing factors among the load influencing factors of the target power park;
[0105] Use the K-Means algorithm to classify the main influencing factors among the load influencing factors to obtain the dominant load influencing factors and non-dominant load influencing factors;
[0106] Input the total load of the target power park in the current period and the dominant load influencing factors into the second neural network model, and output the total load prediction value of the target power park in the prediction period.
[0107] In a possible implementation manner, the second neural network model includes an ANN model.
[0108] In a possible implementation manner, the transferable load prediction module 140 includes:
[0109]
[0110] Wherein, represents the transferable load percentage matrix, represents the percentage of the transferable load relative to the total load of the k-th transferable device at the t-th moment of the i-th day; the total load of the transferable device is the sum of the transferable load and the non-schedulable load; LV(t) represents the predicted value of the proportion of the transferable load corresponding to the t-th moment within the prediction period, and TL(t) represents the predicted value of the total load corresponding to the t-th moment within the prediction period; h represents the number of moments in a single period; represents the transferable load data of the target power park at the t-th moment within the prediction period.
[0111] In a possible implementation, the park transferable load prediction device 100 based on machine learning further includes a second model parameter update module, which is used for:
[0112] Obtain the true value of the total load of the target power park within the prediction period;
[0113] Based on the formula Determine the prediction evaluation result of the second neural network model, and update the hyperparameters of the second neural network model based on the prediction evaluation result of the second neural network model;
[0114] where y i represents the true value of the total load, represents the predicted value of the total load, represents the average value of the predicted values of the total load, and n represents the number of samples.
[0115] In a possible implementation, the park transferable load prediction device 100 based on machine learning further includes a first model parameter update module, which is used for: obtaining the true value of the proportion of the transferable load corresponding to the prediction period;
[0116] Based on the formula Determine the prediction evaluation result of the first neural network model, and update the hyperparameters of the first neural network model based on the prediction evaluation result of the first neural network model;
[0117] where x i represents the true value of the proportion of the transferable load, represents the predicted value of the proportion of the transferable load, represents the average value of the predicted values of the proportion of the transferable load, and n represents the number of samples.
[0118] Figure 3 is a schematic diagram of the electronic device provided by the embodiment of the present invention. As Figure 3As shown, the electronic device 3 of this embodiment includes: a processor 30 and a memory 31. The memory 31 stores a computer program 32. When the processor 30 executes the computer program 32, the steps in each of the above method embodiments are implemented. Alternatively, when the processor 30 executes the computer program 32, the functions of each module / unit in each of the above device embodiments are implemented.
[0119] Exemplarily, the computer program 32 can be divided into one or more modules / units. The one or more modules / units are stored in the memory 31 and executed by the processor 30 to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing specific functions, and these instruction segments are used to describe the execution process of the computer program 32 in the electronic device 3.
[0120] The electronic device 3 may include, but is not limited to, a processor 30 and a memory 31. Those skilled in the art can understand that Figure 3 merely examples of the electronic device 3, which do not constitute a limitation on the electronic device 3. It may include more or fewer components than shown in the figure, or combine certain components, or different components. For example, the electronic device 3 may also include input / output devices, network access devices, buses, etc.
[0121] For the convenience and brevity of description, only the above division of each functional module / unit is used as an example. In actual application, the above functions can be allocated to different functional modules / units according to needs. The above modules / units can be implemented in the form of hardware, or in the form of software, or in the form of a combination of hardware and software.
[0122] In the above embodiments, the descriptions of each embodiment have their own emphases. For the parts not detailed or recorded in a certain embodiment, reference can be made to the relevant descriptions of other embodiments. Without special instructions and logical conflicts, the terms and / or descriptions between different embodiments are consistent and can be referenced to each other. The technical features in different embodiments can be combined to form new embodiments according to their inherent logical relationships.
[0123] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and 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 each embodiment of the present invention, and should all be included in the protection scope of the present invention.
Claims
1. A method for predicting transferable loads in a park based on machine learning, characterized in that, Including: Inputting the load data of each device in the target power park in the current period into the first neural network model to output the predicted value of the transferable load ratio corresponding to the prediction period, where the predicted value of the transferable load ratio is the percentage of the transferable load in the total load; inputting the total load and load influencing factors of the target power park in the current period into the second neural network model to output the predicted value of the total load of the target power park in the prediction period; Obtaining the transferable load percentage matrix of the adjacent parks of the target power park in the most recent N periods; the transferable load percentage matrix includes the load percentages of each transferable load device in the adjacent parks in the most recent N periods; Based on the predicted value of the transferable load ratio, the predicted value of the total load in the prediction period of the target power park, and the transferable load percentage matrix of the adjacent parks in the most recent N periods, obtaining the transferable load data of the target power park in the prediction period.
2. The method for predicting transferable load in a park based on machine learning according to claim 1, wherein Before inputting the load data of each device in the target power park in the current period into the first neural network model, the method further includes: Obtaining the load data of each device in the target power park in the current period; Dividing the load data of each device in the target power park in the current period by the total load in the current period to obtain the load percentage of each device in the target power park in the current period; Selecting the load percentages of the devices within the preset numerical range as the preprocessed load percentages; Correspondingly, the step of inputting the load data of each device in the target power park in the current period into the first neural network model to output the predicted value of the transferable load ratio corresponding to the prediction period includes: Inputting the preprocessed load percentages of each device in the current period into the first neural network model to output the predicted value of the transferable load ratio corresponding to the prediction period.
3. The method for predicting transferable loads in a park based on machine learning according to claim 1, characterized in that, The first neural network model is a long short-term memory network model, and the long short-term memory network model includes one layer of unidirectional long short-term memory network and one layer of bidirectional long short-term memory network.
4. The method for predicting transferable loads in a park based on machine learning according to claim 1, wherein The load influencing factors include the behaviors of electricity users and environmental information; the step of inputting the total load and load influencing factors of the target power park in the current period into the second neural network model to output the predicted value of the total load of the target power park in the prediction period includes: Adopting the Pearson correlation method to extract the main influencing factors in the load influencing factors of the target power park; Using the K-Means algorithm to classify the main influencing factors in the load influencing factors to obtain the dominant load influencing factors and non-dominant load influencing factors; Inputting the total load of the target power park in the current period and the dominant load influencing factors into the second neural network model to output the predicted value of the total load of the target power park in the prediction period.
5. The method for predicting transferable loads in a park based on machine learning according to claim 1, wherein The second neural network model includes an ANN model.
6. The method for predicting transferable load in a park based on machine learning according to claim 1, wherein Based on the predicted value of the transferable load ratio, the predicted value of the total load in the prediction period of the target power park, and the transferable load percentage matrix of the adjacent park in the most recent N periods, obtaining the transferable load data of the target power park in the prediction period, including: Among them, represents the transferable load percentage matrix, represents the percentage of the transferable load of the k-th transferable device at the t-th moment on the i-th day relative to the total load of the k-th transferable device at the t-th moment on the i-th day; the total load of the transferable device is the sum of the transferable load and the non-schedulable load; LV(t) represents the predicted value of the proportion of the transferable load corresponding to the t-th moment in the prediction period, and TL(t) represents the predicted value of the total load corresponding to the t-th moment in the prediction period; h represents the number of moments in a single period; P i s (t) represents the transferable load data of the target power park at the t-th moment in the prediction period.
7. The method for predicting transferable loads in a park based on machine learning according to claim 1, wherein After inputting the total load and load influencing factors of the target power park in the current period into the second neural network model and outputting the predicted value of the total load of the target power park in the prediction period, the method further includes: Obtaining the true value of the total load of the target power park in the prediction period; Based on the formula Determine the prediction evaluation result of the second neural network model, and update the hyperparameters of the second neural network model based on the prediction evaluation result of the second neural network model; Among them, y i represents the true value of the total load, represents the predicted value of the total load, represents the average value of the predicted values of the total load, and n represents the number of samples.
8. The method for predicting transferable load in a park based on machine learning according to claim 1, wherein After outputting the predicted value of the transferable load ratio corresponding to the prediction period, the method further includes: Obtaining the true value of the transferable load ratio corresponding to the prediction period; Based on the formula Determine the prediction evaluation result of the first neural network model, and update the hyperparameters of the first neural network model based on the prediction evaluation result of the first neural network model; Among them, x i represents the true value of the transferable load ratio, represents the predicted value of the transferable load ratio, represents the average value of the predicted values of the transferable load ratio, and n represents the number of samples.
9. A park transferable load prediction device based on machine learning, characterized in that, Including: A transferable load ratio predicted value obtaining module, configured to input the load data of each device in the target power park in the current period into the first neural network model, and output the predicted value of the transferable load ratio corresponding to the prediction period, where the predicted value of the transferable load ratio is the percentage of the transferable load in the total load; A total load predicted value obtaining module, configured to input the total load and load influencing factors of the target power park in the current period into the second neural network model, and output the predicted value of the total load of the target power park in the prediction period; A transferable load percentage matrix obtaining module, configured to obtain the transferable load percentage matrix of the adjacent park of the target power park in the most recent N periods; the transferable load percentage matrix includes the load percentages of each transferable load device in the adjacent park in the most recent N periods; A transferable load prediction module, configured to obtain the transferable load data of the target power park in the prediction period based on the predicted value of the transferable load ratio, the predicted value of the total load in the prediction period of the target power park, and the transferable load percentage matrix of the adjacent park in the most recent N periods.
10. The device for predicting transferable load in a park based on machine learning according to claim 9, characterized in that, The device further includes a data preprocessing module, configured to: Obtain the load data of each device in the target power park in the current period; Dividing the load data of each device in the target power park in the current period by the total load in the current period to obtain the load percentage of each device in the target power park in the current period; Selecting the load percentages of the devices within a preset numerical range as the preprocessed load percentages; Correspondingly, the transferable load ratio predicted value obtaining module is configured to: Input the preprocessed load percentages of each device in the current period into the first neural network model, and output the predicted value of the transferable load ratio corresponding to the prediction period.