Heating load prediction method based on similar daily clustering and convolution gating circulation

Abstract

The invention discloses a heating load prediction method based on similar daily clustering and convolution gating circulation, and relates to the technical field of heating load prediction of a heat exchange station of a centralized heat system, and the method comprises the steps: obtaining side heating load historical data and meteorological data, carrying out the preprocessing, carrying out the slicing of the data according to the number of days, and dividing the data into a training set and a verification set; according to the heating load change condition, rough classification of weather types is carried out on the training set, then fine classification is carried out on the weather types, and features of different weather types are obtained; in combination with grey correlation degree analysis and a Pearson product-moment correlation coefficient, obtaining a value with the maximum correlation degree between the target day and each weather type, and selecting a corresponding data set as input; training is conducted through the combined neural network model, and a heat supply load prediction result is obtained. Therefore, by adopting the heating load prediction method based on similar daily clustering and convolution gating circulation, the prediction precision of the heat load of the heat exchange station of the central heating system can be improved, and the method has higher flexibility and universality.

Description

Technical Field

[0001] The invention relates to the technical field of heating load prediction of a heat exchange station in a centralized heating system, and in particular to a heating load prediction method based on similar day clustering and convolution gated loop. Background Art

[0002] The central heating system is an important part of the livelihood project in northern China. However, the pipe network of large central heating systems usually stretches for several kilometers, and it takes several hours or even longer for the heat to reach the heat users from the heat source, which is manifested as a prominent temperature transmission delay characteristic. This characteristic causes serious mismatch between source and load supply and demand and energy waste, reducing the economic and environmental benefits of the central heating system. The accurate prediction of the heat load of the heat exchange station node can provide a reference for the operation and control personnel of the central heating system, and then adjust and optimize the next operation strategy of the central heating system, so as to achieve the purpose of improving the source-load matching degree of the central heating system and improving the economic and environmental benefits of the central heating system.

[0003] Traditional research on heat load prediction of heat exchanger stations in centralized heat exchange systems mostly adopts methods based on statistical theory. This method has strong interpretability and simple calculation, but low prediction accuracy and poor robustness. With the rapid development and widespread application of machine learning technology, it has become possible to accurately predict the heat load of each node in the heat exchanger station using big data. For example, Xin Tan proposed a room heat load prediction calculation method based on the combination of analytic hierarchy process (AHP) and back propagation (BP) neural network, and took 8 relevant environmental factors as the input of the model to ensure the accuracy and stability of the prediction results. Aliyu Ibrahim discussed the role of CNN in residential building load modeling. By considering the shape characteristics of the building to make multivariate predictions on one-dimensional data, not only rapid modeling was achieved, but also the prediction accuracy was improved. Ling Z used convolutional neural networks and random forest algorithms based on deep learning technology to establish a heat exchanger station prediction model, and analyzed the data of a heat exchanger station in Handan City during the heating season throughout the year, which improved the accuracy of the heat load prediction of the heat exchanger station. Hossein MF proposed and constructed a building heat load prediction model, using real-time updated data to ensure the validity of information, which shows that artificial neural networks can achieve accurate predictions and have practical application value.

[0004] In order to further improve the prediction accuracy, some studies are devoted to exploring methods for heat load prediction using multiple neural networks. For example, CNN and LSTM are combined to form a hybrid neural network, and the hybrid neural network is further enhanced by combining wavelet decomposition for short-term heat load prediction. Compared with a single neural network model, the hybrid neural network model has higher prediction accuracy, thus better meeting the requirements of the heating system. On the other hand, there are also studies that have improved the shortcomings of neural networks. For example, a combined neural network based on an optimization method improves the accuracy of the load energy consumption prediction calculation results; Junni S et al. introduced adversarial neural networks in the load prediction of integrated energy systems, and used LSTM as a generator and discriminator, which effectively improved the load prediction accuracy; Yaohui H et al. used Ac-GRN (Active Graph Recurrent Network) to predict regional heat loads, effectively enhancing the interpretability of the prediction results.

[0005] Although the neural network method based on deep learning technology has shown great potential in the field of heat load forecasting, it has high requirements on the quality of the data set and cannot capture long-sequence features. In addition, the heat load of the heat exchange station node is affected by many factors such as time, environment and human factors. There is a mismatch between source and load supply and demand and energy waste in the centralized heating system. Therefore, it is necessary to provide a method based on environmental parameters and similar day clustering analysis to further improve the model prediction efficiency and accuracy, and provide a reference for the operation of the centralized heating system. Summary of the invention

[0006] The purpose of the present invention is to provide a heating load forecasting method based on similar day clustering and convolution gated loop, which can improve the accuracy of heat load forecasting of heat exchange stations in centralized heating systems and has stronger flexibility and universality.

[0007] To achieve the above object, the present invention provides a heating load forecasting method based on similar day clustering and convolution gated loop, comprising the following steps:

[0008] S1. Obtain the historical data of heating load and meteorological data, perform data preprocessing, normalize the heating load and various meteorological parameters, slice them by day, and divide them into training set and validation set;

[0009] S2. According to the change of heating load, the training set is first roughly classified into weather types, and then the weather types are finely classified to obtain the characteristics of different weather types;

[0010] S3. According to the change of heating load under each weather type, the correlation between the target day and each weather type is calculated by combining grey correlation analysis and Pearson product-moment correlation coefficient, and the weather type data set with high correlation is selected as the input data set;

[0011] S4. Based on the selected input data set, the combined neural network model is trained and predicted to obtain the prediction results of the target day heating load change.

[0012] Preferably, the DBSCAN clustering algorithm is used for coarse classification, and the K-Means algorithm is used for fine classification.

[0013] Preferably, data preprocessing includes marking and filling missing values ​​in the data using linear interpolation; detecting outliers using the interquartile range of the box plot, setting them as null values, and then filling them by linear interpolation; and smoothing the data using an SG filter.

[0014] Preferably, the weather types are classified according to the change of heating load, as follows:

[0015] H=f(x1,x2,···,x F );

[0016] Where H is the heating load, f represents the functional relationship between each environmental parameter and the heating load, x1,x2,···,x F For various environmental parameters.

[0017] Preferably, calculating the correlation between the target day and each weather type includes:

[0018] First, the Pearson product-moment correlation coefficient between each environmental parameter and heating load under different weather types is calculated, and then the correlation between the target day and each weather type is obtained through grey correlation analysis, as follows:

[0019]

[0020] In the formula, ζ i (k) represents the grey correlation coefficient of the kth weather type of the ith meteorological factor, ρ represents the resolution coefficient, G min represents the global minimum value of the difference between the target day and the historical day, G max Indicates the global maximum value of the difference between the target day and the historical day, x i (k) represents the normalized value of the kth weather type of the i-th meteorological factor, and x0(k) represents the normalized value of the corresponding meteorological factor on the target day;

[0021] Then, the calculated Pearson product-moment correlation coefficient is used to perform weighted average on the grey correlation coefficients of various meteorological factors under different weather types, thereby obtaining the correlation coefficients between the target day and each weather type.

[0022] Preferably, the combined neural network model includes a convolutional gated recurrent unit and a gated recurrent unit, and introduces an attention mechanism.

[0023] Preferably, the convolutional gated recurrent unit performs feature extraction by combining a convolutional neural network and a gated recurrent unit as follows:

[0024]

[0025] Where σ represents the sigmoid activation function, W z , W r , W represents the weight matrix of update gate, reset gate, candidate hidden layer state, z t 、r t , represents the output of the update gate, reset gate, and candidate hidden layer state, h t 、h t-1 They represent the output values ​​of the state unit at time t and t-1 respectively, and × represents the convolution operation.

[0026] Therefore, the present invention adopts the above-mentioned heating load prediction method based on similar day clustering and convolution gated cycle, which has the following technical effects:

[0027] (1) Cluster analysis is performed on the training set to obtain the characteristics of different weather types. Similar day analysis is then used to obtain a data set with a high correlation with the target day, providing a high-precision data set for subsequent model training, preventing the model from overfitting and enhancing robustness.

[0028] (2) The use of combined neural networks can improve the ability of neural network feature extraction. At the same time, the introduction of the attention mechanism improves the learning efficiency of the model and effectively improves the accuracy of heating load prediction of the heat exchanger station in the centralized heating system.

[0029] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flow chart of a heating load forecasting method based on similar day clustering and convolution gated loop;

[0031] Figure 2 is a Pearson product-moment correlation coefficient between each environmental parameter and the heating load under different weather types in an embodiment of a heating load prediction method based on similar day clustering and convolution gated loop;

[0032] Figure 3 The invention is a method for predicting heating load based on similar day clustering and convolution gated loop, and includes a grey correlation coefficient between each environmental parameter on a target day and different weather types;

[0033] Figure 4is a correlation coefficient between a target day and each weather type in an embodiment of a heating load prediction method based on similar day clustering and convolution gated loop;

[0034] Figure 5 The present invention is a schematic diagram of a combined neural network structure in an embodiment of a heating load prediction method based on similar day clustering and convolution gated loop. DETAILED DESCRIPTION

[0035] The present invention can be explained in more detail by the following examples. The purpose of disclosing the present invention is to protect all changes and improvements within the scope of the present invention. The present invention is not limited to the following examples.

[0036] like Figure 1 As shown, the present invention provides a heating load prediction method based on similar day clustering and convolution gated loop, comprising the following steps:

[0037] Embodiment 1

[0038] S1. In this embodiment, the historical data of the primary side heating load and the meteorological data of a heat exchange station from November 2022 to March 2023 are obtained, and the time step is 10 minutes.

[0039] Since there are missing values ​​or outliers in the acquired raw data, it is necessary to preprocess it first. In this embodiment, for the missing values ​​in the data, linear interpolation is used to mark and fill the missing values; for outliers, the box plot method is used to identify, that is, the interquartile range of the box plot is used to detect outliers. After finding the outliers, these outliers are first set to null values, and then linearly filled.

[0040] After completing the processing of abnormal values ​​and missing values, this embodiment also uses an SG filter (SavitzkyGolay filter) to smooth the data.

[0041] At the same time, due to the differences in dimensions of the heating load and various influencing parameters, it is necessary to normalize the heating load and various meteorological parameters, slice them by days, and take the first 90% of the dates as the training set and the last 10% of the dates as the validation set.

[0042] S2. The heating load usually has a nonlinear relationship with weather conditions such as ambient temperature and humidity, such as H = f(x1, x2, ···, x F ), where H is the heating load, f represents the functional relationship between each environmental parameter and the heating load, x1,x2,···,x F Therefore, the weather type can be classified according to the change of heating load.

[0043] In this embodiment, the DBSCAN clustering algorithm is first used to roughly classify the weather types of the training set according to the changes in the heating load, and then the K-Means algorithm is used for fine classification to obtain the changes in the heating load under 10 weather types.

[0044] S3. To further illustrate the difference in heating load changes under different weather types, in this embodiment, the Pearson product-moment correlation coefficient between meteorological factors and heating load is calculated, and it is found that there are different Pearson product-moment correlation coefficients between meteorological parameters and heating load in different weather types, such as Figure 2 shown.

[0045] Take a certain day in the validation set as the target day, and use grey correlation analysis to obtain the correlation between the target day and each weather type. The calculation formula is as follows:

[0046]

[0047] In the formula, ζ i (k) represents the grey correlation coefficient of the kth weather type of the ith meteorological factor, ρ represents the resolution coefficient, G min represents the global minimum value of the difference between the target day and the historical day, G max Indicates the global maximum value of the difference between the target day and the historical day, x i (k) represents the normalized value of the kth weather type of the ith meteorological factor, and x0(k) represents the normalized value of the corresponding meteorological factor on the target day.

[0048] Then, the calculated Pearson product-moment correlation coefficient is used to analyze the grey correlation coefficients of various meteorological factors under different weather types ( Figure 3 ) to obtain the correlation coefficient between the target day and each weather type, and select the weather type data set with high correlation as the input data set, such as Figure 4 shown.

[0049] S4. Based on the selected input data set, the combined neural network model is trained and predicted to obtain the prediction results of the target day heating load change.

[0050] like Figure 5 As shown in Figure 1, the combined neural network model includes a convolutional gated recurrent unit and a gated recurrent unit, and introduces an attention mechanism. The convolutional gated recurrent unit extracts features by combining a convolutional neural network and a gated recurrent unit, and the expression is as follows:

[0051]

[0052] Where σ represents the sigmoid activation function, W z , Wr , W represents the weight matrix of update gate, reset gate, candidate hidden layer state, z t 、r t , represents the output of the update gate, reset gate, and candidate hidden layer state, h t 、h t-1 They represent the output values ​​of the state unit at time t and t-1 respectively, and × represents the convolution operation.

[0053] In order to verify the effectiveness of the similar day analysis method, in this embodiment, the data sets of various weather types are used as input for training and prediction, and the accuracy is evaluated by mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), and mean absolute percentage error (MAPE), as shown in Table 1.

[0054] Table 1 Training accuracy of different weather type datasets

[0055] Weather Type MSE(kW) RMSE(kW) MAE(kW) MAPE(%) Type 1 117.61 10.85 8.45 1.14 Type 2 238.20 15.43 12.22 1.65 Type 3 175.76 13.26 11.09 1.48 Type 4 124.49 11.16 8.28 1.11 Type 5 160.74 12.68 10.26 1.36 Type 6 220.99 14.87 12.72 1.72 Type 7 227.34 15.08 13.56 1.82 Type 8 360.10 18.98 16.32 2.21 Type 9 212.16 14.57 12.03 1.61 Type 10 160.75 12.68 10.57 1.41

[0056] As can be seen from Table 1, the combined neural network model performs best in all evaluation indicators when the data set corresponding to type 4 is used as input for training. Figure 3 The results are consistent, and the correlation coefficient between the target day and type 4 is the largest. This means that the method of this embodiment can effectively select the input training set and improve the accuracy and stability of the prediction.

[0057] On the other hand, this embodiment further verifies the accuracy and reliability of the method of this embodiment by using the entire data set as input for training (predicting heating load 1), using the existing similar day method for training (predicting heating load 2), and using the method of this embodiment for training (predicting heating load 3), and analyzing the prediction results.

[0058] The prediction results obtained by training with different methods are shown in Table 2. It can be seen that compared with the results of the two existing training methods, the MSE of the method used in this embodiment for predicting heating load 1 and predicting heating load 2 is reduced by 339.15kW (27.61%) and 140.16kW (48.00%), the RMSE is reduced by 3.36kW (14.91%) and 7.41kW (27.88%), the MAE is reduced by 2.58kW (15.82%) and 5.47kW (28.49%), the MAPE is reduced by 0.34% (15.25%) and 0.75% (28.41%), and the R 2 The increases were 0.12 (17.14%) and 0.28 (40.00%) respectively.

[0059] Table 2. The accuracy of training results using different methods

[0060] Evaluation indicators Forecast heating load1 Forecast heating load 2 Forecast heating load3 MSE(kW) 706.63 507.64 367.48 RMSE(kW) 26.58 22.53 19.17 MAE(kW) 19.20 16.31 13.73 MAPE(%) 2.64 2.23 1.89 <![CDATA[R 2 ]]> 0.42 0.58 0.70

[0061] Embodiment 2

[0062] In order to verify the performance of the training data set selected by the method of this embodiment on different models, the existing CNN, LSTM, GRU and CMT models are used for comparative analysis. The prediction results of different models are shown in Table 3.

[0063] As can be seen from Table 3, the improved neural networks such as ConvGRU-GRU, CMT, PVT, and CPVT all show relatively high fitting accuracy (R 2 ), and with the continuous optimization of the neural network, the prediction accuracy also increases. This shows that the method of this embodiment uses the data set corresponding to the weather type with a high correlation with the target day as the input of the neural network, which has higher flexibility and universality, effectively improves the accuracy of the heat load prediction of the heat exchange station in the centralized heating system, and saves the calculation cost at the same time, and improves the practicability in the project. Therefore, the neural network model can be optimized or replaced according to actual needs.

[0064] Table 3. Accuracy of training results of different models

[0065]

[0066] Therefore, the present invention adopts the above-mentioned heating load prediction method based on similar day clustering and convolution gated cycle, which effectively improves the prediction accuracy of the heating load of the heat exchange station of the centralized heating system.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A heating load forecasting method based on similar day clustering and convolution gated loop, characterized in that: The following steps are involved: S1. Obtain the historical data of heating load and meteorological data, perform data preprocessing, normalize the heating load and various meteorological parameters, slice them by day, and divide them into training set and validation set; S2. According to the change of heating load, the training set is first roughly classified into weather types, and then the weather types are finely classified to obtain the characteristics of different weather types; S3. According to the change of heating load under each weather type, the correlation between the target day and each weather type is calculated by combining grey correlation analysis and Pearson product-moment correlation coefficient, and the weather type data set with high correlation is selected as the input data set; S4. Based on the selected input data set, the combined neural network model is trained and predicted to obtain the prediction results of the target day heating load change.

2. A heating load forecasting method based on similar day clustering and convolution gated loop according to claim 1, characterized in that: Data preprocessing includes marking and filling missing values ​​in the data using linear interpolation; using the interquartile range of the box plot to detect outliers, setting them as null values, and then filling them through linear interpolation; and using the SG filter to smooth the data.

3. A heating load forecasting method based on similar day clustering and convolution gated loop according to claim 1, characterized in that: The DBSCAN clustering algorithm is used for coarse classification, and the K-Means algorithm is used for fine classification.

4. A heating load forecasting method based on similar day clustering and convolution gated loop according to claim 1, characterized in that: Weather types are classified according to the changes in heating load as follows: H=f(x1,x2,···,x F ); Where H is the heating load, f represents the functional relationship between each environmental parameter and the heating load, x1,x2,···,x F For each environmental parameter.

5. The method for heating load forecasting based on similar day clustering and convolution gated loop according to claim 1 is characterized in that: Calculate the correlation between the target day and each weather type, including: First, the Pearson product-moment correlation coefficient between each environmental parameter and the heating load under different weather types is calculated, and then the correlation between the target day and each weather type is obtained through grey correlation analysis, as follows: In the formula, ζ i (k) represents the grey correlation coefficient of the kth weather type of the ith meteorological factor, ρ represents the resolution coefficient, G min represents the global minimum value of the difference between the target day and the historical day, G max Indicates the global maximum value of the difference between the target day and the historical day, x i (k) represents the normalized value of the kth weather type of the i-th meteorological factor, and x0(k) represents the normalized value of the corresponding meteorological factor on the target day; Then, the calculated Pearson product-moment correlation coefficient is used to perform weighted average on the grey correlation coefficients of various meteorological factors under different weather types, thereby obtaining the correlation coefficients between the target day and each weather type.

6. A heating load forecasting method based on similar day clustering and convolution gated loop according to claim 1, characterized in that: The combined neural network model includes convolutional gated recurrent units and gated recurrent units, and introduces an attention mechanism.

7. A heating load forecasting method based on similar day clustering and convolution gated loop according to claim 1, characterized in that: The convolutional gated recurrent unit performs feature extraction by combining a convolutional neural network and a gated recurrent unit as follows: Where σ represents the sigmoid activation function, W z , W r , W represents the weight matrix of update gate, reset gate, candidate hidden layer state, z t 、r t , represents the output of the update gate, reset gate, and candidate hidden layer state, h t 、h t-1 They represent the output values ​​of the state unit at time t and t-1 respectively, and × represents the convolution operation.

Images