Electric power system air conditioner load splitting method and system based on deep learning and ensemble learning

Abstract

The invention discloses an electric power system air conditioner load splitting method and system based on deep learning and integrated learning. The method comprises the following steps that historical total load data and meteorological temperature data are obtained and subjected to cleaning and alignment processing; identifying an air conditioner operation interval through a temperature threshold value, and determining non-air-conditioning season data for establishing a reference load model; a long short-term memory network (LSTM) algorithm is adopted, historical loads, date types and temperatures are used as features, and a dynamic reference load prediction model is obtained through training; comparing the total load of the target day with a reference load to obtain a temperature-sensitive load; and inputting the temperature sensitive load into an extreme gradient boost tree (XGBoost) model for nonlinear fitting mapping, and outputting an air conditioner load splitting result. According to the method disclosed by the invention, the problem that the reference load and the air-conditioning load are difficult to accurately separate in the existing method can be solved, and the decomposition precision is enhanced by utilizing workday and holiday influences and meteorological variables.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of power system load analysis, and particularly relates to a power system air conditioner load splitting method and system based on deep learning and ensemble learning. BACKGROUND

[0002] With the development of economic society, the proportion of air conditioning equipment in residential and commercial electricity is increasing, and air conditioning load has become a major component of peak load in winter and summer of the power system. Its volatility and uncertainty bring great challenges to the planning, scheduling, demand side response, etc. of the power grid. Therefore, accurately separating the air conditioning load component from the total load is crucial for understanding load characteristics, implementing precise load control, and improving prediction accuracy. Traditional air conditioning load splitting methods are mostly based on empirical statistics or simple statistical analysis, which is difficult to accurately distinguish between baseline load and air conditioning load, resulting in insufficient precision of the splitting result and limiting the subsequent prediction and optimization application. In addition, after the parameters of the traditional mathematical statistical model are set, it lacks the ability to learn and optimize itself according to new data, and lacks an effective post-evaluation and correction mechanism for the rationality of the splitting result. SUMMARY

[0003] To solve the problem that the air conditioning load is difficult to accurately separate in the prior art, the present application provides a power system air conditioner load splitting method and system based on deep learning and ensemble learning, so as to enhance the decomposition precision.

[0004] To achieve the above purpose, the technical scheme of the present application is as follows:

[0005] A power system air conditioner load splitting method based on deep learning and ensemble learning, comprising the following steps:

[0006] Step one, obtaining historical total load data of a target area and meteorological temperature data , and performing cleaning and time alignment, and dividing the preprocessed data into a training set and a test set;

[0007] Step two, identifying the air conditioning operation interval through the temperature threshold, and selecting the non-air conditioning season load data as the baseline load modeling sample;

[0008] Step three, constructing a baseline load prediction model, training the model using the baseline load data, and saving the trained model;

[0009] Step four, subtracting the baseline load prediction value from the total load of the target day to obtain temperature-sensitive load ;

[0010] Step five, performing temperature threshold identification on the temperature-sensitive load with weather temperature Input extreme gradient boosting tree (XGBoost) model for nonlinear fitting mapping to obtain the final air conditioning load splitting result .

[0011] Further, the data cleaning includes missing value interpolation, outlier removal and duplicate value processing.

[0012] Further, the temperature threshold value identifies the air conditioner running interval method comprising: when the daily maximum temperature is greater than the first threshold value, it is determined as an air conditioner cooling running day; when the daily minimum temperature is lower than the second threshold value, it is determined as an air conditioner heating running day; when the daily temperature does not exceed the first threshold value and is not lower than the second threshold value, it is determined as an air conditioner non-running day.

[0013] Further, the LSTM input layer is used to receive , , ; the hidden layer is one or more layers of LSTM units for extracting time series features; the output layer maps the hidden state to the reference load prediction value .

[0014] Further, the quasi-load prediction model is a long short-term memory network (LSTM), and the input features include historical load , date type and weather temperature , and the output is a dynamic reference load prediction value ; the LSTM deep learning network includes an input layer, an LSTM hidden layer and an output layer

[0015] Further, the input layer is used to receive , , ; the hidden layer is one or more layers of LSTM units for extracting time series features; the output layer maps the hidden state to the reference load prediction value .

[0016] Further, the state update of each unit of the LSTM hidden layer is represented by the following formula:

[0017]

[0018]

[0019]

[0020]

[0021]

[0022]

[0023] wherein, , , are input gate, forget gate and output gate respectively, is the cell state, is the hidden state, is the input vector, is the Sigmoid activation function, is the element-wise multiplication.

[0024] Further, the LSTM output layer maps the hidden state of the last time step of the LSTM to the hourly reference load forecast value using linear transformation:

[0025]

[0026] wherein, is the hidden state vector of the last time step of the LSTM, including , , time series features; is the output layer weight matrix, used to map the hidden state vector to ; is the output layer bias vector, used to adjust the forecast value to fit the actual load data.

[0027] Further, the input of the XGBoost model is the temperature-sensitive load and the meteorological temperature , and the output is the air conditioning load split result ;

[0028] Further, the XGBoost model is an ensemble learning model based on gradient boosting trees, including multiple regression trees, and the prediction of each tree is updated by the residual of the previous tree, and the final air conditioning load split result is obtained by weighted summation, which is represented by the formula:

[0029]

[0030] wherein, is the prediction value of the mth regression tree.

[0031] Further, the leaf node prediction value of each regression tree is updated by minimizing the objective function:

[0032] ,

[0033]

[0034] wherein, is the true load value, is the model predicted value, is the number of regression trees, is the number of leaf nodes of each tree, is the leaf node weight, and is the regularization parameter.

[0035] In a second aspect, the present application provides a power system air conditioner load splitting system based on deep learning and ensemble learning, comprising:

[0036] A data acquisition and preprocessing module is configured to acquire historical total load data and meteorological temperature data of a target area, and perform missing value interpolation, outlier removal and time alignment;

[0037] A baseline load modeling module is configured to identify an air conditioner operating interval based on a temperature threshold, select non-air conditioner season load data as a sample, construct and train an LSTM model, and output a baseline load prediction result;

[0038] A temperature-sensitive load calculation module is configured to subtract the baseline load prediction value from the total load of a target day to obtain a temperature-sensitive load;

[0039] An air conditioner load splitting module is configured to input the temperature-sensitive load and meteorological features into an XGBoost model to obtain an air conditioner load splitting result.

[0040] Compared with the prior art, the present application has the following beneficial effects:

[0041] Firstly, the present application introduces LSTM to dynamically model the baseline load, which can fully utilize historical load and meteorological data, capture long-term dependencies in time series, achieve hour-level fine-grained prediction, and improve the accuracy of baseline load fitting.

[0042] Secondly, the present application uses XGBoost to perform nonlinear mapping of temperature-sensitive load and meteorological features, which can effectively separate air conditioner load and other temperature-sensitive electricity, and improve the robustness and generalization ability of air conditioner load splitting.

[0043] Thirdly, the method of the present application can not only be used for daily operation load splitting, but also can provide high-precision air conditioner load data support for load prediction, demand response and energy efficiency analysis of the power system, and has engineering application value. BRIEF DESCRIPTION OF DRAWINGS

[0044] The accompanying drawings, which form a part of this specification, are included to provide a further understanding of the application and are incorporated in and constitute a part of this specification. The embodiments of the application, and their

[0045] Figure 1 Flow chart of the power system air conditioning load splitting method based on deep learning and ensemble learning of the application;

[0046] Figure 2 Trend chart of the air conditioning load splitting result of the embodiment of the application in Cangyou line in 2023-2024;

[0047] Figure 3 Comparison chart of total load, benchmark load prediction value and air conditioning load splitting value in Cangyou line in 2023-2024 in the embodiment of the application. DETAILED DESCRIPTION

[0048] The application will be described in detail below with reference to the accompanying drawings and embodiments. It should be noted that the embodiments in the application and the features in the embodiments can be combined with each other without conflict.

[0049] The following detailed description is exemplary and is intended to provide further detailed description of the application. Unless otherwise specified, all technical terms used in the application have the same meaning as understood by those skilled in the art. The terms used in the application are only for the purpose of describing the specific embodiments and are not intended to limit the exemplary embodiments according to the application.

[0050] Embodiment one, with reference to Figure 1 In one embodiment of the application, a power system air conditioning load splitting method based on deep learning and ensemble learning is provided, and the specific steps are as follows:

[0051] Step S1: Obtain historical total load data and meteorological temperature data of the target area, and perform data cleaning and time alignment. Divide the preprocessed data into training set and test set;

[0052] Collect historical total load data of Cangyou substation in Lianyungang City from January 1, 2023 to December 31, 2024 and historical temperature data of Chaoyang substation weather station , data interval 1 hour sampling. Perform missing value interpolation, outlier removal and repeated value processing on the obtained data to ensure data integrity and accuracy.

[0053] Step S2: Identify the air conditioning operation interval by temperature threshold, and select non-air conditioning season load data as benchmark load modeling sample;

[0054] According to the meteorological statistical law, set the first threshold = 26°C, second threshold = 10°C; when the daily maximum temperature is greater than , it is determined as an air conditioning cooling operation day; when the daily minimum temperature is lower than , it is determined as an air conditioning heating operation day; otherwise, it is determined as an air conditioning non-operation day; the load data of the non-operation day is used as the baseline load modeling sample.

[0055] Step S3: constructing a baseline load prediction model, training the model using the baseline load data, and obtaining the baseline load prediction value through the trained model;

[0056] LSTM is used as the baseline load prediction model. The input features are historical load sequence , date type feature , and meteorological temperature sequence ; the network structure is input layer-LSTM hidden layer (2 layers, 64 units per layer)-output layer. The output is the baseline load prediction value .

[0057] The state update of each unit of the LSTM hidden layer is represented by the following formula:

[0058]

[0059]

[0060]

[0061]

[0062]

[0063]

[0064] wherein, , , are the input gate, the forget gate, and the output gate, respectively, is the unit state, is the hidden state, is the input vector, is the Sigmoid activation function, is the element-wise multiplication.

[0065] The output layer uses linear mapping:

[0066]

[0067] wherein, is the hidden state vector of the LSTM at the last time, which contains , , time series features; is an output layer weight matrix for mapping the hidden state vector to ; is an output layer bias vector for adjusting the predicted value to fit the actual load data.

[0068] After training, save the model.

[0069] Step S5: subtract the total load of the target day from the baseline load prediction value to obtain the temperature-sensitive load:

[0070]

[0071] Step S42: subtract the temperature-sensitive load from the meteorological temperature Input the XGBoost model for nonlinear fitting to obtain the final air conditioning load splitting result:

[0072]

[0073] wherein, is the prediction value of the mth regression tree.

[0074] The leaf node prediction value of each regression tree is updated by minimizing the objective function:

[0075]

[0076]

[0077] wherein, is the true load value, is the model prediction value, is the number of regression trees, is the number of leaf nodes of each tree, is the leaf node weight, and are regularization parameters.

[0078] Through the above steps, the air conditioning load splitting is completed, and the air conditioning load is obtained.

[0079] The final air conditioning load splitting result is referred to Figure 2 , Figure 3 . Figure 2 is the air conditioning load splitting result of Cangwu line 2023-2024 in this embodiment. Figure 3The total daily load of the Cangyou line in 2023-2024, the baseline load prediction value, and the air conditioner load splitting value are compared in the embodiment.

[0080] In the above embodiment, the application further provides an air conditioner load splitting system for a power system based on deep learning and ensemble learning, which can realize the air conditioner load splitting method based on deep learning and ensemble learning.

[0081] The data acquisition and preprocessing module is used to acquire historical total load data and meteorological temperature data of a target area, and to perform missing value interpolation, outlier removal, and time alignment.

[0082] The baseline load modeling module is used to identify the air conditioner operation interval based on the temperature threshold, select non-air conditioner season load data as samples, construct and train an LSTM model, and output the baseline load prediction result.

[0083] The temperature-sensitive load calculation module is used to subtract the baseline load prediction value from the total load of the target day to obtain the temperature-sensitive load.

[0084] The air conditioner load splitting module is used to input the temperature-sensitive load and meteorological features into an XGBoost model to obtain the air conditioner load splitting result.

[0085] The method disclosed in the application can solve the problem of accurate separation of baseline load and air conditioner load in the prior art, and can enhance the decomposition accuracy by using workdays, holiday effects, and meteorological variables.

[0086] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application and not to limit them, although the application has been described in detail with reference to the above embodiments, those skilled in the art should understand that the specific embodiments of the application can be modified or replaced, without departing from the spirit and scope of the application, any modification or equivalent replacement, which should be covered within the protection scope of the claims of the application.

Claims

1. A power system air conditioning load splitting method based on deep learning and ensemble learning, characterized in that, The method comprises the steps of: Step 1: obtaining historical total load data and meteorological temperature data of a target area, and performing data cleaning and time alignment, and dividing the preprocessed data into a training set and a test set; Step 2: identifying an air conditioner running interval through a temperature threshold, and selecting non-air conditioner season load data as a baseline load modeling sample; Step 3: constructing a baseline load prediction model, training the model using the baseline load data, and obtaining a baseline load prediction value through the trained model; Step 4: subtracting the baseline load prediction value from the total load of a target day to obtain a temperature-sensitive load; Step 5: inputting the temperature-sensitive load and meteorological temperature into an extreme gradient boosting tree model for nonlinear fitting mapping to obtain a final air conditioner load splitting result.

2. The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 1, wherein, The data cleaning includes missing value interpolation, outlier removal and repeated value processing. 3.The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 1, wherein, The air conditioner running interval is identified through a temperature threshold, including: when the daily maximum temperature is greater than a first threshold, it is determined as an air conditioner cooling running day; when the daily minimum temperature is lower than a second threshold, it is determined as an air conditioner heating running day; when the daily temperature does not exceed the first threshold and is not lower than the second threshold, it is determined as an air conditioner non-running day. 4.The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 1, wherein, The baseline load prediction model adopts an LSTM deep learning network, including an input layer, an LSTM hidden layer and an output layer.

5. The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 4, wherein, The input layer is used to receive historical load , date type , and weather temperature ; the hidden layer is one or more LSTM units for extracting time series features; and the output layer maps the hidden state to the benchmark load prediction value .

6. The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 5, wherein, The output layer adopts linear transformation to map the hidden state of the last time of LSTM to the hourly reference load prediction value To: ； wherein, is the hidden state vector of the last time step of the LSTM, containing , , the time series feature of the previous time step; is the output layer weight matrix, used to map the hidden state vector to ; is the output layer bias vector, used to adjust the predicted value to fit the actual load data.

7. The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 1, wherein, The input of the extreme gradient boosting tree model is temperature-sensitive load With meteorological temperature The output is air conditioning load split result .

8. The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 7, wherein, The extreme gradient boosting tree model includes multiple regression trees, the prediction of each tree is updated by the residual error of the previous tree, and the final air conditioner load splitting result is obtained by weighted summation, represented as: ； where, is the prediction value of the mth regression tree.

9. The deep learning and ensemble learning based power system air conditioning load disaggregation method of claim 8, wherein, predicted value of the mth regression tree update by minimizing an objective function ； ； where, is the true load value, is the model prediction, is the number of regression trees, is the number of leaf nodes per tree, is the leaf node weight, and is the regularization parameter.

10. A deep learning and ensemble learning based power system air conditioning load disaggregation system implementing the method of any one of claims 1-9, characterized in that, The method comprises the steps of: A data acquisition and preprocessing module acquires historical total load data of a target area and meteorological temperature data and performs data cleaning and time alignment, and divides the preprocessed data into a training set and a test set A baseline load modeling module is configured to identify an air conditioner running interval based on a temperature threshold, select non-air conditioner season load data as a sample, construct and train a baseline load prediction model, and output a baseline load prediction value; A temperature-sensitive load calculation module is configured to subtract the baseline load prediction value from the total load of a target day to obtain a temperature-sensitive load; An air conditioner load splitting module is configured to input the temperature-sensitive load and meteorological temperature into an extreme gradient boosting tree model for nonlinear fitting mapping to obtain a final air conditioner load splitting result.

Images