QR-TCN-LSTM-based multi-element load prediction method and system for integrated energy system

Abstract

The invention discloses a QR-TCN-LSTM-based multi-element load prediction method and system for an integrated energy system, and relates to the technical field of load prediction. The method comprises the following steps: acquiring various load data with obvious time correlation, including power load data, thermal load data and cold energy load data; calculating a correlation coefficient between each type of load data and the environmental influence factor, and determining a key environmental influence factor according to the correlation coefficient; constructing a data set according to the load data and the key environment influence factors, and dividing the data set into a training set and a test set; a QR-TCN-LSTM model is constructed, and the QR-TCN-LSTM model is trained by using the training set, so that a trained QR-TCN-LSTM model can be obtained; and performing load probability prediction by using the trained QR-TCN-LSTM model to obtain probability prediction results of each load data in different confidence intervals. According to the method and the system, the prediction accuracy and the generalization capability of the multi-element load are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of load forecasting, and more specifically to a multi-element load forecasting method and system for an integrated energy system based on QR-TCN-LSTM. Background Art

[0002] Integrated Energy System (IES) can integrate electricity, heat, cooling and renewable energy to achieve efficient resource utilization and complementarity. Multi-element load forecasting, as a key technology for IES planning and operation, is of great significance to improving the reliability and economy of energy systems. However, due to the complex coupling relationship between various energy forms and the diversity and uncertainty of user needs, accurate load forecasting faces severe challenges. Therefore, studying advanced multi-element load forecasting methods is extremely critical to optimizing IES scheduling and energy storage management.

[0003] Traditional load forecasting methods mainly include statistical methods, machine learning methods and deep learning methods. Statistical methods such as regression analysis and time series analysis perform well when dealing with linear relationships, but are unable to cope with complex nonlinear relationships. Although machine learning methods such as support vector machines and random forests have improved forecasting accuracy, they are still insufficient when dealing with multi-energy coupling and uncertainty. Deep learning methods such as LSTM and CNN have made significant progress in load forecasting, but they are mainly focused on point forecasting, fail to effectively quantify forecast uncertainty, and are difficult to meet IES's requirements for load probability distribution.

[0004] In addition, existing methods often ignore the mutual influence between different energy loads, resulting in inaccurate prediction results.

[0005] Therefore, how to develop a multivariate load probability forecasting method and system that can effectively capture the multi-energy coupling relationship and quantify prediction uncertainty has become a hot topic and difficulty in current research. Summary of the invention

[0006] In view of this, the present invention provides a multi-load forecasting method and system for an integrated energy system based on QR-TCN-LSTM, which is used to forecast multi-loads.

[0007] In order to achieve the above object, the present invention adopts the following technical solution:

[0008] On the one hand, the present invention discloses a multivariate load forecasting method for an integrated energy system based on QR-TCN-LSTM, comprising the following steps:

[0009] Obtain a variety of load data with obvious time correlation, including power load data, thermal load data, and cooling load data;

[0010] Calculate the correlation coefficient between each load data and the environmental influencing factor respectively, and determine the key environmental influencing factor according to the correlation coefficient;

[0011] Constructing a data set according to the load data and key environmental influencing factors, and dividing the data set into a training set and a test set;

[0012] Construct a QR-TCN-LSTM model and use the training set to train the QR-TCN-LSTM model to obtain a trained QR-TCN-LSTM model;

[0013] The trained QR-TCN-LSTM model is used to perform load probability forecasting and obtain the probability prediction results of each load data under different confidence intervals.

[0014] Furthermore, in the step of calculating the correlation coefficient between each load data and the environmental influencing factor, it specifically includes calculating the Pearson correlation coefficient between each load data and the environmental influencing factor as the final correlation coefficient.

[0015] Further, in the step of constructing a QR-TCN-LSTM model, the QR-TCN-LSTM model includes a TCN network layer, an LSTM network layer and a quantile regression prediction layer;

[0016] The TCN layer is used to receive the load data and key environmental influencing factors in the training set, and output a feature sequence containing local features and time dependencies;

[0017] The LSTM network layer is used to process the feature sequence output by the TCN layer to obtain the prediction result of the LSTM network layer;

[0018] The quantile regression prediction layer receives the prediction results of the LSTM network layer, and obtains the final probability prediction results of each load data under different confidence intervals by minimizing the Pinball loss function.

[0019] Furthermore, the Pinball loss function specifically includes the following loss function:

[0020]

[0021] Among them, L q,t Represents the loss value of the target quantile q at time t, represents the qth quantile of the prediction result of the LSTM network layer at time t, y t represents the true value at time t.

[0022] Furthermore, the step of training the QR-TCN-LSTM model using the training set also includes: optimizing the parameters of the QR-TCN-LSTM model using a gradient descent method.

[0023] The present invention also discloses a multi-element load forecasting system for an integrated energy system based on QR-TCN-LSTM, comprising:

[0024] Data collection module, used to collect various load data with obvious time correlation, including power load data, thermal load data and cooling load data;

[0025] A correlation coefficient calculation module, used to calculate the correlation coefficient between each load data and the environmental influencing factor, and determine the key environmental influencing factor according to the correlation coefficient;

[0026] A data processing module, used for constructing a data set according to the load data and key environmental influencing factors, and dividing the data set into a training set and a test set;

[0027] The model training module is used to build a QR-TCN-LSTM model and train the QR-TCN-LSTM model using the training set to obtain a trained QR-TCN-LSTM model;

[0028] The load forecasting module is used to perform load probability forecasting using the trained QR-TCN-LSTM model to obtain the probability forecast results of each load data under different confidence intervals.

[0029] Preferably, in the above correlation coefficient calculation module, the Pearson correlation coefficient between each load data and the environmental influencing factor is calculated as the final correlation coefficient.

[0030] Preferably, in the above-mentioned model training module, the constructed QR-TCN-LSTM model includes a TCN network layer, an LSTM network layer and a quantile regression prediction layer;

[0031] The TCN layer is used to receive the load data and key environmental influencing factors in the training set, and output a feature sequence containing local features and time dependencies;

[0032] The LSTM network layer is used to process the feature sequence output by the TCN layer to obtain the prediction result of the LSTM network layer;

[0033] The quantile regression prediction layer receives the prediction results of the LSTM network layer, and obtains the final probability prediction results of each load data under different confidence intervals by minimizing the Pinball loss function.

[0034] Preferably, the Pinball loss function minimized by the quantile regression prediction layer specifically includes the following loss function:

[0035]

[0036] Among them, L q,t Represents the loss value of the target quantile q at time t, represents the qth quantile of the prediction result of the LSTM network layer at time t, y t represents the true value at time t.

[0037] Preferably, the model training module also includes: optimizing the parameters of the QR-TCN-LSTM model using a gradient descent method.

[0038] It can be seen from the above technical solutions that, compared with the prior art, the present invention discloses a multi-element load forecasting method and system for an integrated energy system based on QR-TCN-LSTM, which has the following beneficial effects:

[0039] The present invention conducts an in-depth analysis of the characteristics of multivariate loads, identifies key features and captures the interdependence between different energy loads. Then, the time dependency in the load data is captured through the trained QR-TCN-LSTM, and probability predictions at different quantile levels are generated through quantile regression, which significantly improves the prediction accuracy and generalization ability of multivariate loads. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0041] Figure 1 A schematic diagram of the overall process of the multi-element load forecasting method for an integrated energy system based on QR-TCN-LSTM provided by the present invention.

[0042] Figure 2 The overall framework flow chart of the QR-TCN-LSTM model provided by the present invention.

[0043] Figure 3 Schematic diagram of the dilated convolution structure of the TCN layer in the QR-TCN-LSTM model provided by the present invention.

[0044] Figure 4 Schematic diagram of the residual structure in the QR-TCN-LSTM model provided by the present invention.

[0045] Figure 5 A schematic diagram of the structure of each LSTM unit in the QR-TCN-LSTM model provided by the present invention.

[0046] Figure 6 A schematic diagram of the calculation results of the Wilson correlation coefficients of the electricity, cooling, heating loads and environmental influencing factors provided in an embodiment of the present invention.

[0047] Figure 7 A schematic diagram of the power load data prediction results obtained by using the method of the present invention provided in an embodiment of the present invention.

[0048] Figure 8 A schematic diagram of the cooling energy load data prediction results obtained by using the method of the present invention provided in an embodiment of the present invention.

[0049] Fig. 9 A schematic diagram of thermal load data prediction results obtained by using the method of the present invention is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] The embodiment of the present invention first discloses a multi-element load forecasting method for an integrated energy system based on QR-TCN-LSTM. Figure 1 As shown, the following steps are included:

[0052] Obtain a variety of load data with obvious time correlation, including power load data, thermal load data, and cooling load data;

[0053] Calculate the correlation coefficient between each load data and environmental influencing factors respectively, and determine the key environmental influencing factors based on the correlation coefficient;

[0054] Constructing a data set according to the load data and key environmental influencing factors, and dividing the data set into a training set and a test set;

[0055] Construct a QR-TCN-LSTM model and use the training set to train the QR-TCN-LSTM model to obtain a trained QR-TCN-LSTM model;

[0056] The trained QR-TCN-LSTM model is used to perform load probability forecasting and obtain the probability prediction results of each load data under different confidence intervals.

[0057] The steps of the present invention are described in further detail below.

[0058] In the step of obtaining multiple load data with obvious time correlation, the multivariate load data used in the step of the present invention comes from the integrated energy system. By performing autocorrelation analysis on the data in the integrated energy system, the power load data, thermal load data and cooling load data with obvious time correlation with the current time are obtained as the load data in the present invention.

[0059] Specifically, the autocorrelation coefficient formula used in the present invention is as follows:

[0060]

[0061] In the formula, τ represents the time delay, Cov(X t ,X t-τ ) is the time series data X representing the load t and X t-τ The covariance between Var(X t ) and Var(X t-τ ) represent the time series data X t and X t-τ The autocorrelation coefficient ranges from -1 to 1, 0 means no correlation, positive value means positive correlation, and negative value means negative correlation. Finally, the power load data, thermal load data, and cooling load data with autocorrelation coefficients within a certain threshold range are selected as load data.

[0062] After obtaining the above load data, the correlation coefficient between each load data and the environmental influencing factors is calculated respectively, and the key environmental influencing factors are determined according to the correlation coefficient. The embodiment of the present invention adopts Pearson correlation analysis to calculate the correlation coefficient between environmental influencing factors. The calculation formula of Pearson correlation coefficient is as follows:

[0063]

[0064] Among them, x i represents the i-th sample in a certain load data, y i represents the i-th sample in a certain type of environmental data, and are the means of load data x and environmental data y, respectively. The Pearson correlation coefficient is calculated by formula (2) to identify environmental factors with significant linear relationship with load change. Specifically, in the embodiment of the present invention, environmental factors with average absolute values ​​of Pearson correlations with each type of load data greater than 0.3 are selected as key environmental factors according to the results of Pearson correlation analysis, including dew point, temperature, precipitation, air pressure and relative humidity.

[0065] A data set is constructed based on the above load data and key environmental influencing factors, and the data set is divided into a training set and a test set.

[0066] The present invention realizes load prediction by constructing a QR-TCN-LSTM model, such as Figure 2 As shown in the figure, the QR-TCN-LSTM model includes a TCN network layer, an LSTM network layer, and a quantile regression prediction layer.

[0067] First, the TCN network layer receives the load data and key environmental factors in the training set as the input layer of the QR-TCN-LSTM model, and converts the load data and key environmental factors into one-dimensional time series data. The TCN network layer captures the local dependencies in the time series by expanding the causal convolution. For the one-dimensional time series input X = (x0, x1, ..., x t ) and convolution kernel f, the extended convolution operation F(·) is defined at sequence element t as follows:

[0068]

[0069] Where d is the expansion rate, which determines the spacing of the elements in the convolution kernel. By introducing the expansion rate d convolution kernel, the receptive field can be expanded without increasing the computational cost. k is the convolution kernel size. The dilated convolution structure is as follows Figure 3 shown.

[0070] As the number of TCN network layers increases, the problems of gradient vanishing and gradient exploding become more prominent. To solve these problems, TCN introduces residual connections, which allow the input signal to be directly passed to the output, allowing the gradient to propagate more efficiently, thereby alleviating the gradient vanishing problem in deep networks. Figure 4As shown in the figure, the residual structure is to connect the dilated causal convolution layer Dilated Causal Conv, the weight normalization layer WeightNorm, and the ReLU activation function Dropout layer through the residual. Each residual block usually consists of two branches, one branch is responsible for performing the conversion operation, which includes the dilated causal convolution layer, the weight normalization layer, the ReLU activation function, and the Dropout layer. The right branch is a residual connection composed of 1*1 convolution, and the two branches form a residual block in parallel. In order to ensure that the network can effectively train the deep architecture, the TCN layer uses a residual connection. Assume that the input of the hth residual block is X (h-1) , then output X (h) for:

[0071] X (h) =σ(X (h-1) +f(X (h-1) )) (4)

[0072] Among them, σ(·) is the activation function.

[0073] Next, the LSTM network layer of the QR-TCN-LSTM model processes the output of the TCN layer to capture long-term dependencies. The key to the LSTM network layer is its unique unit structure. In the embodiment of the present invention, the LSTM network layer is composed of multiple LSTM unit layers stacked together, and the output of the previous LSTM unit is used as the input of the next LSTM unit. This structure can learn more complex features and long-term dependencies in sequence data. Each LSTM unit contains three main gates: input gate, forget gate and output gate. The basic unit of the LSTM unit network is as follows: Figure 5 These gates control the flow of information in the network through different mechanisms to capture long-term dependencies. The first gate in the LSTM unit is the forget gate f t , which determines the memory cell state C from the previous moment t-1 How much information is retained in the forget gate. The state formula of the forget gate at time t is:

[0074] f t =σ(W f ·[h t-1 ,X t ]+b f ) (5)

[0075] Where σ(·) represents the sigmoid activation function; X t is the input at the current moment, mainly including load data and key environmental influencing factors; h t-1 and b f Represent the output vector at time t-1 and the forget gate bias at time t respectively; W f is the weight matrix of the forget gate.

[0076] Next is the input gate i t , which controls how much of the current information is needed as input to generate the current state C t The calculation formula of the input gate is:

[0077] i t =σ(W i ·[h t-1 ,X t ]+b i ) (6)

[0078] At the same time, LSTM generates a new candidate memory cell state The calculation formula is:

[0079]

[0080] Among them, W i and W c are the weight matrices of the input gate and candidate memory cell states, b i and b c is the bias vector and tanh(·) is the activation function.

[0081] Memory cell state C t The update is done by the memory cell state C at the previous moment t-1 It is determined together with the input information at the current moment, and the formula is:

[0082]

[0083] The final stage of the LSTM cell is to calculate how much information can be output. Another control gate acts as the output gate o t :

[0084] o t =σ(W o ·[h t-1 ,X t ]+b o ) (9)

[0085] Hidden state h t Determined by the current output gate and memory cell state, the final hidden layer output of LSTM is h t Defined as:

[0086] h t =o t *tanh(C t ) (10)

[0087] In order to improve the computational efficiency and prediction accuracy of the model, the present invention uses the gradient descent method to optimize the model parameters during the model training process.

[0088] In the present invention, the TCN layer and the LSTM network layer can be used as a TCN-LSTM model. The quantile regression prediction module is connected after the output layer of the TCN-LSTM model. The core of this module is to transform the hidden state vector h t Mapped to the predicted values ​​at different quantile levels. For each quantile τ, define the trainable parameter weight vector W τ and the bias term b τ , the predicted value corresponding to the quantile τ is calculated through linear transformation. The specific expression is as follows:

[0089] Q τ (Y|X)=W τ ·h t +b τ

[0090] Among them, Q(Y|X) is the predicted value corresponding to the quantile τ; h t It is the hidden state output of the TCN-LSTM model. Setting different quantile levels such as τ = 0.05, 0.1, 0.25, 0.5, 0.75, 0.9, 0.95 enables the model to output multivariate load forecast values ​​Q at different probability levels. τ1 (Y|X),Q τ2 (Y|X),...,Q τk (Y|X). The parameters of the quantile regression layer are trained by minimizing the Pinball loss function to measure the deviation between the model prediction value and the true value at different quantile levels to ensure accurate prediction of each probability interval. The loss function is:

[0091]

[0092] Where q represents the target quantile, represents the qth quantile estimated at time t, L q,t It represents the pinball loss at the qth quantile at time t. The smaller the loss value, the better the probability prediction effect.

[0093] Then the quantile regression prediction layer is deeply integrated with the TCN-LSTM model. For any prediction model, it is necessary to train it according to the appropriate loss function and optimization strategy. In the training stage, the parameters of the prediction model are obtained by minimizing the loss function:

[0094]

[0095] In the formula, f q (X t ,W q ) is the predicted value, y t is the true value.

[0096] By dynamically adjusting the learning rate during the training process, the model can be prevented from falling into the local optimal solution and the convergence speed can be accelerated. Using the cross-validation method, the training data set is divided into multiple subsets, and one subset is used as the validation set in turn, and the rest are used as the training set. The model is trained and evaluated for multiple rounds. According to the evaluation indicators such as the root mean square error (RMSE) and average coverage error (ACE) on the validation set, the hyperparameters such as the number of neural network layers and the number of nodes are adjusted to obtain the optimal model performance. Finally, the preprocessed real-time multivariate load data is input into the trained QR-TCN-LSTM model, and the multivariate load forecast values ​​at different quantile levels are output by the quantile regression layer. These forecast results presented in the form of probability distribution can help energy system decision makers formulate scientific and reasonable energy production, distribution and scheduling plans. For example, energy suppliers can reasonably arrange power generation plans based on the power load forecast values ​​at different probability levels to avoid energy shortages or surpluses caused by load fluctuations.

[0097] In order to implement the above method, the present invention also discloses a multi-load forecasting system for an integrated energy system based on QR-TCN-LSTM, comprising:

[0098] Data collection module, used to collect various load data with obvious time correlation, including power load data, thermal load data and cooling load data;

[0099] A correlation coefficient calculation module, used to calculate the correlation coefficient between each load data and the environmental influencing factor, and determine the key environmental influencing factor according to the correlation coefficient;

[0100] A data processing module, used for constructing a data set according to the load data and key environmental influencing factors, and dividing the data set into a training set and a test set;

[0101] The model training module is used to build a QR-TCN-LSTM model and train the QR-TCN-LSTM model using the training set to obtain a trained QR-TCN-LSTM model;

[0102] The load forecasting module is used to perform load probability forecasting using the trained QR-TCN-LSTM model to obtain the probability forecast results of each load data under different confidence intervals.

[0103] The specific implementation process of each module can be found in the method description, which will not be repeated here.

[0104] The effectiveness of the present invention is demonstrated below by specific embodiments:

[0105] The load data of the embodiment of the present invention is sourced from the historical data of electricity, cooling and heating loads in a certain region, and the weather data is sourced from the weather data of the location of the region on the website of the Climate Data Center, including temperature, atmospheric pressure, wind speed and wind direction. The sampling time of all data is from 0:00 on May 2, 2018 to 23:00 on December 31, 2019, and the sampling interval of all data is 1 hour. The present invention uses the above data to jointly predict the electricity, cooling and heating loads of the region in the next week. The specific results and analysis are as follows:

[0106] 1. Load characteristics analysis

[0107] The present invention calculates the correlation coefficients between the collected cold, hot and electric loads and between the environmental factors such as dew point, temperature, relative humidity, wind speed, precipitation and air pressure and the loads. By calculating the Pearson correlation coefficient between each environmental influencing factor and the load data, the Figure 6 shown.

[0108] Depend on Figure 6 It can be seen that there is a strong correlation between environmental factors and different load data, which also shows that the coupling relationship between multiple loads should be fully considered when performing load forecasting in IES, and a multiple load joint forecasting model should be adopted. In order to effectively train the model, this embodiment selects environmental factors that have a significant impact on the forecasting performance. When the absolute value of the correlation coefficient is greater than 0.3, it is considered that there is a strong correlation between the environmental influencing factor and the load data, that is, dew point, temperature, precipitation, air pressure and relative humidity are selected as key environmental factors, and the three load data of electricity, heat and cooling energy and key environmental factors are used as input features of the QR-TCN-LSTM model.

[0109] 2. Contrast method settings

[0110] In order to verify the effectiveness and reliability of the model proposed in this paper, the experimental results of the proposed model were compared with those of five other probability prediction models, including four typical time series prediction models: LSTM, recurrent neural network (RNN), GRU and temporal convolutional network (TCN); and the CNN-LSTM model, which can effectively extract the coupling relationship and time series characteristics between multivariate time series.

[0111] Hyperparameter setting is crucial for the model to achieve good prediction performance. Hyperparameters are determined by empirical method and grid search method. First, the initial values ​​of the hyperparameters of the prediction model are determined based on actual operation experience and the hyperparameters provided by relevant literature. Then, the grid search method is used to find the optimal hyperparameters within the search range. Considering the limitations of machine memory and training speed, the batch size is set to 32 in the experiment, and the model optimizer uses the Adam method with a learning rate of 0.0001. To ensure the credibility of the results, each model is repeated 5 times, and the average value of the evaluation index is taken as the final result.

[0112] In order to evaluate the performance of the constructed QR-TCN-LSTM multivariate load probability forecasting model, the present invention selects Average Coverage Error (ACE), Average Width (AW) and Root Mean Squared Error (RMSE) as evaluation indicators. These indicators can comprehensively reflect the prediction performance of the model in different aspects, among which ACE and AW are mainly used to measure the accuracy and width of the probability distribution, while RMSE is used to evaluate the overall accuracy and deviation of the predicted value. The specific evaluation index expression is as follows:

[0113]

[0114] in, refers to the actual value y i Is it within the prediction interval? Indicator function inside. and represent the upper and lower limits of the prediction interval respectively. is the predicted value, y i is the actual value. q is the confidence interval and N is the sample size.

[0115] 3. Experimental results and analysis

[0116] In order to verify the effectiveness and reliability of the model proposed in the present invention, a comparative experiment was conducted with five other probability prediction models, including four typical time series prediction models: LSTM, recurrent neural network (RNN), GRU and temporal convolutional network (TCN); and a CNN-LSTM model, which can effectively extract the coupling relationship and time series characteristics between multivariate time series.

[0117] To ensure the credibility of the results, the present invention conducts 5 repeated tests on each model and takes the average value of the evaluation index as the final result. The tests are conducted at 50%, 70% and 90% confidence intervals, and the evaluation results are shown in Table 1.

[0118] Table 1 Comparison of evaluation results of different models

[0119]

[0120] As can be seen from Table 1, under the 50% confidence interval, the ACE values ​​of the proposed method on electric load, cooling load and heating load are 0.0119, 0.0952 and 0.0416 respectively, which are significantly lower than other methods, indicating that its prediction interval can cover the true value more accurately, indicating that the proposed method has extremely high coverage accuracy under a lower confidence interval. In addition, the proposed method also performs well on AW. The AW values ​​of electric load, cooling load and heating load are 1020.7, 1003.28 and 226.14 respectively, which are the smallest, showing that its prediction interval is the narrowest and the highest accuracy. Finally, the proposed method also performs well on RMSE. The RMSE values ​​of electric load, cooling load and heating load are 582.75, 551.61 and 186.34 respectively, which are the smallest, indicating that its prediction error is the smallest and the performance is the best.

[0121] Under the 70% confidence interval, the advantages of the method of the present invention are still significant. Its ACE values ​​for electricity, cooling and heating loads are 0.1464, 0.1761 and 0.1781 respectively, continuing to maintain its leading position. At the same time, the AW values ​​of electricity, cooling and heating loads are 1113.14, 1823.38 and 335.54 respectively, indicating that its prediction interval is narrow and the accuracy is high. In terms of RMSE, the RMSE values ​​of the method of the present invention for electricity, cooling and heating loads are 592.34, 530.75 and 155.88 respectively, all of which are the smallest, showing its excellent prediction accuracy.

[0122] Under the 90% confidence interval, the ACE values ​​of the method of the present invention are 0.0761, 0.0821 and 0.0821, respectively, which are significantly better than other methods. The performance of the method of the present invention on AW is still excellent. The AW values ​​of the electric, cold and hot loads are 4428.32, 3553.44 and 449.65, respectively, which are the smallest, indicating that its prediction interval is the narrowest and the accuracy is the highest. In terms of RMSE, the RMSE values ​​of the method of the present invention on electric load, cold load and hot load are 553.75, 511.62 and 131.84, respectively, with high prediction performance.

[0123] Figure 7 The figure shows the prediction results of power load under different confidence intervals of the QR-TCN-LSTM model. Figure 7 As shown in the figure, as the confidence interval increases, the forecast uncertainty also increases. Although there are significant peak fluctuations in the electric load in some periods, the deviations are mostly within the 90% confidence interval, showing the robustness and reliability of the model. The upper limit margin is large and the lower limit margin is small, reflecting the conservatism of the model in dealing with the peak of the electric load. Figure 8 The figure shows the prediction results of cooling energy load under different confidence intervals by the QR-TCN-LSTM model. Figure 8As shown in the figure, the cooling load changes in a clear pattern, and the model captures the actual load changes well within the 50% confidence interval. Although the predicted values ​​deviate from the actual values ​​at some drastic peak-to-valley transitions, most of the cooling loads are still within the 90% confidence interval, showing the model's prediction reliability. The margins of the upper and lower limits of the confidence interval reflect the model's adaptability in dealing with cooling load fluctuations. Fig. 9 The prediction results of thermal load under different confidence intervals of the QR-TCN-LSTM model are shown as follows: Fig. 9 The heat load change trend shown is moderate, but there are also fluctuations. As the confidence interval increases, the prediction interval becomes wider and covers more actual values, reflecting the uncertainty of the prediction results.

[0124] In summary, through the analysis of the test results under different confidence intervals, it can be seen that the method proposed in the present invention shows the best prediction performance in the multivariate load probability forecasting of the integrated energy system. Whether in terms of average coverage error (ACE), average width (AW) or root mean square error (RMSE), the method of the present invention is superior to the other five comparison methods, further verifying the effectiveness and superiority of the proposed method. This result fully demonstrates the potential and practical value of the method of the present invention in multivariate load probability forecasting.

[0125] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0126] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A multi-element load forecasting method for integrated energy system based on QR-TCN-LSTM, characterized in that: The following steps are involved: Obtain a variety of load data with obvious time correlation, including power load data, thermal load data, and cooling load data; Calculate the correlation coefficient between each load data and the environmental influencing factor respectively, and determine the key environmental influencing factor according to the correlation coefficient; Constructing a data set according to the load data and key environmental influencing factors, and dividing the data set into a training set and a test set; Construct a QR-TCN-LSTM model and use the training set to train the QR-TCN-LSTM model to obtain a trained QR-TCN-LSTM model; The trained QR-TCN-LSTM model is used to perform load probability forecasting and obtain the probability prediction results of each load data under different confidence intervals.

2. According to claim 1, a multi-element load forecasting method for an integrated energy system based on QR-TCN-LSTM is characterized in that: The step of calculating the correlation coefficient between each load data and the environmental influencing factor specifically includes calculating the Pearson correlation coefficient between each load data and the environmental influencing factor as the final correlation coefficient.

3. According to the QR-TCN-LSTM-based multi-element load forecasting method for an integrated energy system according to claim 1, it is characterized in that: In the step of constructing a QR-TCN-LSTM model, the QR-TCN-LSTM model includes a TCN network layer, an LSTM network layer and a quantile regression prediction layer; The TCN layer is used to receive the load data and key environmental influencing factors in the training set, and output a feature sequence containing local features and time dependencies; The LSTM network layer is used to process the feature sequence output by the TCN layer to obtain the prediction result of the LSTM network layer; The quantile regression prediction layer receives the prediction results of the LSTM network layer, and obtains the final probability prediction results of each load data under different confidence intervals by minimizing the Pinball loss function.

4. According to claim 3, a multi-element load forecasting method for an integrated energy system based on QR-TCN-LSTM is characterized in that: The Pinball loss function specifically includes the following loss functions: in, L q,t express t Moment target quantile q The loss value, express t The prediction result of the LSTM network layer at the moment q Quantile, yt express t The true value of the moment.

5. According to claim 1, a multi-element load forecasting method for an integrated energy system based on QR-TCN-LSTM is characterized in that: The step of training the QR-TCN-LSTM model using the training set also includes: optimizing the parameters of the QR-TCN-LSTM model using a gradient descent method.

6. A multi-element load forecasting system for integrated energy system based on QR-TCN-LSTM, characterized in that: include: Data collection module, used to collect various load data with obvious time correlation, including power load data, thermal load data and cooling load data; A correlation coefficient calculation module, used to calculate the correlation coefficient between each load data and the environmental influencing factor, and determine the key environmental influencing factor according to the correlation coefficient; A data processing module, used for constructing a data set according to the load data and key environmental influencing factors, and dividing the data set into a training set and a test set; The model training module is used to build a QR-TCN-LSTM model and train the QR-TCN-LSTM model using the training set to obtain a trained QR-TCN-LSTM model; The load forecasting module is used to perform load probability forecasting using the trained QR-TCN-LSTM model to obtain the probability forecast results of each load data under different confidence intervals.

7. The multi-element load forecasting system for an integrated energy system based on QR-TCN-LSTM according to claim 6 is characterized in that: In the correlation coefficient calculation module, the Pearson correlation coefficient between each load data and the environmental influencing factor is calculated as the final correlation coefficient.

8. The multi-element load forecasting system for an integrated energy system based on QR-TCN-LSTM according to claim 6 is characterized in that: In the model training module, the constructed QR-TCN-LSTM model includes a TCN network layer, an LSTM network layer and a quantile regression prediction layer; The TCN layer is used to receive the load data and key environmental influencing factors in the training set, and output a feature sequence containing local features and time dependencies; The LSTM network layer is used to process the feature sequence output by the TCN layer to obtain the prediction result of the LSTM network layer; The quantile regression prediction layer receives the prediction results of the LSTM network layer, and obtains the final probability prediction results of each load data under different confidence intervals by minimizing the Pinball loss function.

9. The QR-TCN-LSTM-based integrated energy system multi-load forecasting system according to claim 8, characterized in that: The Pinball loss function minimized by the quantile regression prediction layer specifically includes the following loss functions: in, L q,t express t Moment target quantile q The loss value, express t The prediction result of the LSTM network layer at the moment q Quantile, yt express t The true value of the moment.

10. The QR-TCN-LSTM-based integrated energy system multi-load forecasting system according to claim 6, characterized in that: The model training module also includes: optimizing the parameters of the QR-TCN-LSTM model using the gradient descent method.

Images