Multi-element load analysis and prediction method for integrated energy system
By adopting the RCNN-SVR model in an integrated energy system, taking into account the coupling characteristics between multiple loads, the problem of low accuracy of traditional load prediction methods is solved, and a higher precision load prediction is achieved.
Patent Information
- Application Number
- CN202411565973.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-05
- Publication Date
- 2025-02-25
AI Technical Summary
The traditional comprehensive energy system load prediction method fails to effectively consider the coupling characteristics between multiple loads, resulting in low prediction accuracy.
The multivariate load prediction method based on RCNN-SVR is adopted, and through steps such as data preprocessing, influencing factor correlation analysis, characteristic factor extraction and load prediction, the coupling characteristics between different energy types are considered to improve the prediction accuracy.
By analyzing the coupling characteristics between multiple loads, the accuracy and reliability of load prediction are improved, and more accurate load prediction is achieved.
Smart Images
Figure 00000000_0000_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to a multi-element load analysis and prediction method for an integrated energy system, and in particular to a multi-element load analysis and prediction method for an integrated energy system taking into account the interaction and complementary coupling of multi-energy supply and demand. Background Art
[0002] As an important part of energy transformation, the Integrated Energy System (IES) integrates multiple energy networks and multi-energy coupling devices such as electricity, natural gas, and thermal energy through smart grids as the core hub. This system uses the coupling and complementarity between different energy forms to improve energy utilization efficiency and achieve optimal resource allocation.
[0003] In an integrated energy system, accurate load forecasting is crucial to the stable and flexible operation of the system. It can provide accurate forecasting data and decision support for the power grid. With the development of technology, the interactivity between multiple loads is increasing. However, traditional load forecasting methods often ignore the coupling and complementarity between energy sources, resulting in low accuracy of multiple load forecasting in integrated energy systems.
[0004] Therefore, it is an urgent problem to study and develop new methods that can accurately predict the multi-element load of integrated energy systems. These methods need to be able to consider the coupling characteristics between different energy types and compensate for the load prediction errors accordingly to improve the accuracy and reliability of the prediction. Summary of the invention
[0005] The purpose of the present invention is to overcome the defects of the prior art and provide a multi-load analysis and prediction method for an integrated energy system taking into account the interaction and complementary coupling of multi-energy supply and demand. The method is a multi-load prediction method based on RCNN-SVR, which solves the problem that the load prediction method of the traditional integrated energy system does not consider the coupling characteristics of the multi-loads and has poor accuracy, and can improve the accuracy of load prediction; the coupling characteristics between the multi-loads in the integrated energy system are analyzed, so that each prediction task of the multi-loads obtains redundant implicit information to improve the prediction accuracy.
[0006] The technical solution to achieve the above purpose is: a multi-element load analysis and prediction method for an integrated energy system, comprising the following steps:
[0007] S1, data preprocessing: preprocess the original load data to obtain data that can better reflect the load change law;
[0008] S2, determine the correlation of influencing factors: In the correlation analysis process, the influencing factor data is composed of hourly data of electric load values, cooling / heating load values, hot water load values, temperature, wind speed, humidity, and solar radiation at different times. The Pearson correlation coefficient between each influencing factor and a single type of load is calculated, and the formula is used for t-distribution verification. The influencing factors with a correlation coefficient greater than 0.5 are selected as input;
[0009] S3, feature factor extraction: first normalize the existing load data, use the RCNN model, set relevant parameters, and extract more representative influencing factors. The input of the RCNN model is the input factor and other influencing factor values determined in the previous step, and the output is used as the input of the SVR model;
[0010] S4, load forecasting: Set the relevant parameters in the SVR model. The input in the forecasting model is the output value of the influencing factors of the RCNN model. The output is the predicted load value, and the load forecasting process ends.
[0011] In the above-mentioned method for multivariate load analysis and prediction of an integrated energy system, in step S1, the data preprocessing method includes performing missing data filling, outlier replacement and data smoothing processing to fill missing data and eliminate noise data.
[0012] The above-mentioned multivariate load analysis and prediction method for a comprehensive energy system, wherein missing data filling: the modified Akima cubic Hermite interpolation method is selected to construct a missing data filling model, and the derivation of the interpolation polynomial is as follows:
[0013] When it is known that the mutually distinct nodes x k and x k+1 and the corresponding function value g i and the first-order derivative value g i ′(i=k,k+1), there exists a unique Hermite interpolation polynomial H3(x) at node x k and x k+1 The following conditions are satisfied: H3(x) = g(x), H3′(x) = g′(x);
[0014] According to formula (1), the basis function method can be used to solve H3(x):
[0015]
[0016] The modified Akima cubic Hermite interpolation method produces less fluctuation than the ordinary cubic Hermite interpolation method, is more efficient in real-time applications, and the obtained fill values will not overshoot.
[0017] The above-mentioned multivariate load analysis and forecasting method for an integrated energy system, wherein outlier data replacement: the quartile method is used to detect outliers in historical load data;
[0018] The core of the quartile method is the concept of quartiles. Quartiles are a type of quantile in statistics. All values are arranged from small to large and divided into four equal parts in sequence. The values at the three dividing points are collectively called quartiles, which are the first quartile, the second quartile, and the third quartile from small to large.
[0019] Outliers in the historical load data are defined as elements that are above the first quartile or below the third quartile by more than a specified interquartile range threshold, which is 1.5.
[0020] In the above-mentioned method for multivariate load analysis and prediction of an integrated energy system, after outliers in historical load data are identified, it is also necessary to perform a replacement operation on the outliers, and the outlier replacement operation is still performed using a data filling model established by modified Akima cubic Hermite interpolation.
[0021] The above-mentioned method for multivariate load analysis and prediction of an integrated energy system, wherein data smoothing: a moving average filter is used to smooth noisy historical load data.
[0022] The above-mentioned method for multivariate load analysis and prediction of an integrated energy system, in which the principle of the moving average filter is based on statistical laws, the working process of the filter is as follows: the continuous sampling data to be filtered is regarded as a queue of fixed length. When a new measurement process is completed, the first data in the above queue is removed, and the remaining data are moved forward in sequence, and the new sampling data is inserted as the last data in the new queue; then the specified operation is performed on the obtained new queue, and the result is used as the result of this measurement. The whole process is repeated until the filtering is completed.
[0023] In the above-mentioned multivariate load analysis and prediction method for a comprehensive energy system, in step S2, the method for determining the correlation of the influencing factors is specifically: using the Pearson correlation coefficient method to calculate the correlation between multiple variables in the electricity, cooling and heating loads. The size of the Pearson correlation coefficient can better reflect the strength of the linear correlation between two variables;
[0024] For variable X = [x1, x2, ···, x n ] Τ and Y=[y1,y2,···,y n ] Τ , and the calculation formula of its Pearson correlation coefficient is:
[0025]
[0026] In formula (2): and is the average value of n sample data; the range of correlation coefficient r is (-1,1), that is, |r|≤1; the value of correlation coefficient r is defined as follows: the closer the correlation coefficient |r| is to 1, the higher the correlation between x and y is. If r=1, it means that there is a positive linear complete correlation between x and y; the closer the correlation coefficient |r| is to 0, the lower the correlation between x and y is; if r=0, it means that there is no linear correlation between x and y; the present invention uses a universal coefficient measurement method: when |r|≥0.8, the two variables are highly correlated; when 0.5≤|r|<0.8, the two variables are moderately correlated; when 0.3≤|r|<0.5, the two variables are slightly correlated; when |r|<0.3, it means that the linear correlation between the two variables is weak, which can be regarded as nonlinear correlation.
[0027] In the above-mentioned method for multivariate load analysis and prediction of an integrated energy system, in step S3, the method for extracting characteristic factors is specifically as follows: feature extraction is completed using the RCNN model; the feature extraction process is performed by two convolutional layers and two maximum pooling layers, an excitation layer and a normalization layer; the prediction process is completed by a fully connected layer and a regression layer; the input size (I) of the input layer is assumed to be I×1 influencing factors, the filter size (F) of convolutional layers 1 and 2 is 1×1, the number of filters (N) is 25, and the padding size (P) is 0; the stride size (S) of maximum pooling layers 1 and 2 is 2×2; therefore, after the maximum pooling layer, the feature dimension is divided by 2; and the ReLU unit of the excitation layer reduces the number of traversals to achieve a smaller training error than the traditional tanh unit; the normalization layer increases the generalization of the model and reduces the error rate; in addition, the excitation layer and the normalization layer do not change the size of the feature dimension.
[0028] In the above-mentioned multivariate load analysis and prediction method for an integrated energy system, in step S4, the load prediction method is specifically as follows: the processed load data obtained in step S1 and the influencing factors obtained in step S3 are input into the SVR prediction model to finally output the load prediction result.
[0029] The comprehensive energy system multi-element load analysis and prediction method of the present invention has the following beneficial effects:
[0030] (1) The present invention pre-processes the original data before load forecasting, including missing data filling, outlier replacement, and data smoothing processing, which has the functions of filling missing data and eliminating noise data, thereby improving the accuracy of load forecasting;
[0031] (2) The present invention analyzes the coupling characteristics between multiple loads in the integrated energy system, including hourly data of electric load values, cold / hot load values, hot water load values, temperature, wind speed, humidity, and solar radiation at different times, so as to realize the sharing of coupling information between multiple loads in the energy system, so that each prediction task of the multiple loads can obtain redundant implicit information to improve the prediction accuracy;
[0032] (3) The present invention adopts the RCNN-SVR model to carry out load forecasting, which can effectively improve the load forecasting accuracy. BRIEF DESCRIPTION OF THE DRAWINGS
[0033] Figure 1 It is the flow chart of load forecasting;
[0034] Figure 2 This is a comparison chart of summer power load forecast results;
[0035] Figure 3 This is a comparison chart of summer cooling load prediction results;
[0036] Figure 4 This is a comparison chart of summer hot water load prediction results. DETAILED DESCRIPTION
[0037] In order to enable those skilled in the art to better understand the technical solution of the present invention, the specific implementation methods thereof are described in detail below in conjunction with the accompanying drawings:
[0038] See also Figure 1 According to an embodiment of the present invention, a method for analyzing and predicting multi-loads of an integrated energy system taking into account the interaction and complementary coupling of multi-energy supply and demand comprises the following steps:
[0039] S1, data preprocessing: in the actual data used, missing data filling, outlier replacement, and data smoothing are performed on the historical load data to obtain data that can better reflect the load change law;
[0040] S2, determine the correlation of influencing factors: In the correlation analysis process, the influencing factor data is composed of hourly data of electric load values, cooling / heating load values, hot water load values, temperature, wind speed, humidity, and solar radiation at different times. The Pearson correlation coefficient between each influencing factor and a single type of load is calculated, and the formula is used for t-distribution verification. The influencing factors with a correlation coefficient greater than 0.5 are selected as input;
[0041] S3, feature factor extraction: first normalize the existing data, use the RCNN model, set relevant parameters, and extract more representative influencing factors. The input of the RCNN model is the input factor and other influencing factor values determined in the previous step, and the output is used as the input of the SVR model;
[0042] S4, load forecasting: Set the relevant parameters in the SVR model. The input in the forecasting model is the output value (influencing factor) of the RCNN model. The output is the predicted load value, and the load forecasting process ends.
[0043] In step S1, the data preprocessing method is as follows:
[0044] Step 1.1, missing data filling. The present invention selects the modified Akima cubic Hermite interpolation method to construct a missing data filling model. The derivation of the interpolation polynomial is as follows:
[0045] When it is known that the mutually distinct nodes x k and x k+1 and the corresponding function value g i and the first-order derivative value g i ′(i=k,k+1), there exists a unique Hermite interpolation polynomial H3(x) at node x k and x k+1 It satisfies that: H3(x)=g(x), H3′(x)=g′(x).
[0046] According to the above relationship, H3(x) can be solved by using the basis function method.
[0047]
[0048] The modified Akima cubic Hermite interpolation method produces less fluctuation than the ordinary cubic Hermite interpolation method, is more efficient in real-time applications, and the obtained fill values will not overshoot.
[0049] Step 1.2: Outlier data replacement: The present invention uses the quartile method to detect singular points in multivariate load data to improve the prediction accuracy of multivariate loads.
[0050] The core of the quartile method is the concept of quartiles. The concept of quartiles is introduced as follows: Quartiles are a type of quantile in statistics, that is, all values are arranged from small to large and divided into four equal parts in order. The values at the three dividing points are collectively called quartiles, from small to large, respectively, the first quartile, the second quartile, and the third quartile.
[0051] The quartiles are applied in such a way that outliers are defined as elements that are above the first quartile or below the third quartile by more than the specified interquartile range threshold, which is 1.5 by default.
[0052] After identifying the outliers in the historical load data, it is necessary to replace the outliers. The outlier replacement operation is still performed using the data filling model established by the modified Akima cubic Hermite interpolation.
[0053] Step 1.3, data smoothing. In the initial data, the noise information contained in the historical load data will have a great impact on the accuracy of multi-source load forecasting. The present invention uses a moving average filter to smooth the noisy historical load data to improve the training and prediction accuracy of the prediction model.
[0054] The principle of the moving average filter is based on statistical laws. The working process of the filter is as follows: the continuous sampled data to be filtered is regarded as a queue of fixed length. When a new measurement process is completed, the first data in the above queue is removed, and the remaining data are moved forward in turn, and the new sampled data is inserted as the last data in the new queue; then the specified operation is performed on the obtained new queue, and its result is used as the result of this measurement. The whole process is repeated until the filtering is completed.
[0055] In step 2, the method for determining the correlation of influencing factors is as follows: The present invention uses the Pearson correlation coefficient method to calculate the correlation of multiple variables in the electricity, cooling and heating loads. The size of the Pearson correlation coefficient can better reflect the strength of the linear correlation between two variables. For variable X = [x1, x2, ···, x n ] Τ and Y=[y1,y2,···,y n ] Τ , and the calculation formula of its Pearson correlation coefficient is:
[0056]
[0057] In formula (2): and is the average value of n sample data. The range of the correlation coefficient r is (-1,1), that is, |r|≤1. The value of the correlation coefficient r is defined as follows: the closer the correlation coefficient |r| is to 1, the higher the correlation between x and y. If r=1, it means that there is a positive linear complete correlation between x and y; the closer the correlation coefficient |r| is to 0, the lower the correlation between x and y; if r=0, it means that there is no linear correlation between x and y. The present invention uses a universal coefficient measurement method: when |r|≥0.8, the two variables are highly correlated; when 0.5≤|r|<0.8, the two variables are moderately correlated; when 0.3≤|r|<0.5, the two variables are poorly correlated; when |r|<0.3, it means that the linear correlation between the two variables is weak and can be regarded as a nonlinear correlation.
[0058] In step S3, the feature factor extraction method is: feature extraction is completed using the RCNN model. The feature extraction process is performed by two convolutional layers and two maximum pooling layers, an excitation layer (ReLU) and a normalization layer; the prediction process is completed by a fully connected layer and a regression layer. Assume that the input size (I) of the input layer is I×1 influencing factors, the filter size (F) of convolutional layers 1 and 2 is 1×1, the number of filters (N) is 25, and the padding size (P) is 0; the stride size (S) of maximum pooling layers 1 and 2 is 2×2. Therefore, after the maximum pooling layer, the feature dimension is divided by 2. The ReLU unit of the excitation layer reduces the number of traversals to achieve a smaller training error than the traditional tanh unit. The normalization layer increases the generalization of the model and reduces the error rate. In addition, the excitation layer and the normalization layer do not change the size of the feature dimension.
[0059] In step S4, the method of load forecasting is: input the processed load data obtained in step 1 and the influencing factors obtained in step 3 into the SVR forecasting model to finally output the load forecasting result.
[0060] In this embodiment, the method is implemented based on the MATLAB platform. The example selects the hourly load data of a certain energy center in 2023 as the original sample, and the sample contains three different types of load data, namely, electric load, cooling / heating load, and hot water load.
[0061] The load data and influencing factors from July 1 to July 31, 2023 are used as the original training sample data for summer forecasting, and the corresponding information of typical working days and typical rest days in summer are selected as test sample data for forecasting analysis. In addition to load data, relying on sufficient collection devices of an energy center, an energy center also provides other original data, including relevant data of influencing factors such as temperature, humidity, wind speed, radiation, etc. corresponding to each moment of the system with hourly steps and relevant date type information.
[0062] First, data preprocessing is performed on the collected total historical load data, and then correlation analysis is performed. According to the obtained Pearson coefficient value, after a t-distribution test, the influencing factors with a correlation coefficient greater than 0.3 are selected and normalized as the input training model.
[0063] Taking a typical summer month as an example, the calculated Pearson correlation coefficient values are shown in Tables 1, 2, and 3.
[0064] Table 1, Pearson correlation coefficient between electrical load and influencing factors:
[0065] Type of influencing factors Absolute value of coefficient Electric load in the first hour 0.9511 Cold / heat load in the first hour 0.6415 Hot water load in the first hour 0.2485 Electric load in the first 2 hours 0.8351 Cold / heat load in the first 2 hours 0.6063 Electric load in the first 3 hours 0.7102 temperature 0.4734 radiation 0.1001 Wind speed 0.2818 time 0.1212 Week Type 0.1561 Holidays 0.1694
[0066] Table 2, Pearson correlation coefficient between cooling / heating load and influencing factors:
[0067] Type of influencing factors Absolute value of coefficient Electric load in the first hour 0.6570 Cold / heat load in the first hour 0.9613 Hot water load in the first hour 0.1943 Electric load in the first 2 hours 0.6242 Cold / heat load in the first 2 hours 0.8720 Cold / heat load in the first 3 hours 0.7765 temperature 0.5900 radiation 0.2165 Wind speed 0.1995 time 0.2961 Week Type 0.1164 Holidays 0.1187
[0068] Table 3, Pearson correlation coefficient between hot water load and influencing factors:
[0069] Type of influencing factors Absolute value of coefficient Electric load in the first hour 0.2455 Cold / heat load in the first hour 0.2104 Hot water load in the first hour 0.9730 Electric load in the first 2 hours 0.2304 Cold / heat load in the first 2 hours 0.1990 Hot water load in the first 2 hours 0.9026 Hot water load in the first 3 hours 0.8102 temperature 0.1388 radiation 0.1232 Wind speed 0.1918 time 0.1594 Week Type 0.0859 Holidays 0.0824
[0070] After completing the model training, the actual load data is selected as the test set for load forecasting. Taking a typical summer day as an example, after inputting the various influencing factors with hourly steps into the model, the multivariate load forecasting results are obtained as follows: Figure 2-Figure 4 shown.
[0071] from Figure 2 , 3 As can be seen from Figures 4 and 5, the change trend of the summer prediction result curve obtained by the method proposed in the present invention is close to the actual value curve and has a higher accuracy.
[0072] The accuracy of load forecasting results is measured by mean absolute percentage error, mean accuracy and weighted mean accuracy, and the results are shown in Table 4.
[0073] Table 4, model evaluation indicators:
[0074]
[0075] It can be seen from Table 4 that the prediction results considering the correlation of multiple loads have better prediction accuracy. When the three loads of electricity load, cooling / heating load and hot water load are considered in meteorological factors and social factors and applied to the prediction at the same time, better prediction results can be obtained in this model.
[0076] The feasibility and effectiveness of the model method proposed in the present invention are verified by comparison. By referring to the support vector machine load forecasting model of the integrated energy system considering the influence of natural gas load and meteorological factors, and comparing the average forecast results of the same category of loads, the feasibility of the model of the present invention can be proved, as shown in Table 5.
[0077] Table 5, comparison of model results:
[0078] Model Model of the present invention Comparison Models Mean absolute percentage error 3.57％ 7.56％
[0079] The method of the present invention is compared with the indicators of the electricity, cooling and heating loads when they are predicted separately as follows to prove the effectiveness of the method of the present invention, as shown in Table 6.
[0080] Table 6, comparison of average relative errors:
[0081] Load Type Method of the present invention Electricity, cooling and heating separate forecast Typical solar load in summer 3.6971％ 18.3546％ Typical daily cooling / heating load in summer 3.3424％ 14.6739％ Typical daily hot water load in summer 1.4833％ 19.1367％
[0082] It can be seen from Table 6 that, compared with the method of predicting the electric, cooling and heating loads separately, the prediction method of the present invention that considers the load correlation and influencing factors can reduce the mean absolute error of the electric, cooling and heating loads and improve the model accuracy.
[0083] In summary, the multi-load analysis and prediction method for an integrated energy system taking into account the interaction and complementary coupling of multi-energy supply and demand of the present invention is a multi-load prediction method based on RCNN-SVR, which solves the problem that the load prediction method of the traditional integrated energy system does not consider the coupling characteristics of the multi-loads and has poor accuracy. By fully exploring the correlation between the multi-loads, the accuracy of load prediction can be improved; the coupling characteristics between the multi-loads in the integrated energy system are analyzed, so that each prediction task of the multi-loads obtains redundant implicit information to improve the prediction accuracy.
[0084] Those skilled in the art should recognize that the above embodiments are only used to illustrate the present invention, and are not intended to limit the present invention. As long as they are within the spirit of the present invention, any changes or modifications to the above embodiments will fall within the scope of the claims of the present invention.
Claims
1. A multi-element load analysis and prediction method for an integrated energy system, characterized in that: The following steps are involved: S1, data preprocessing: preprocess the original load data to obtain data that can better reflect the load change law; S2, determine the correlation of influencing factors: In the correlation analysis process, the influencing factor data is composed of hourly data of electric load values, cooling / heating load values, hot water load values, temperature, wind speed, humidity, and solar radiation at different times. The Pearson correlation coefficient between each influencing factor and a single type of load is calculated, and the formula is used for t-distribution verification. The influencing factors with a correlation coefficient greater than 0.5 are selected as input; S3, feature factor extraction: first normalize the existing load data, use the RCNN model, set relevant parameters, and extract more representative influencing factors. The input of the RCNN model is the input factor and other influencing factor values determined in the previous step, and the output is used as the input of the SVR model; S4, load forecasting: Set the relevant parameters in the SVR model. The input in the forecasting model is the output value of the influencing factors of the RCNN model. The output is the predicted load value, and the load forecasting process ends.
2. A multi-element load analysis and prediction method for an integrated energy system according to claim 1, characterized in that: In step S1, the data preprocessing method includes performing missing data filling, outlier replacement and data smoothing processing to fill missing data and eliminate noise data.
3. A method for multi-element load analysis and prediction of an integrated energy system according to claim 2, characterized in that: Missing data filling: The modified Akima cubic Hermite interpolation method is selected to construct the missing data filling model, and the interpolation polynomial is derived as follows: When it is known that the mutually distinct nodes x k and x k+1 and the corresponding function value g i and the first-order derivative value g i ′(i=k,k+1), there exists a unique Hermite interpolation polynomial H3(x) at node x k and x k+1 The following conditions are satisfied: H3(x) = g(x), H3′(x) = g′(x); According to formula (1), the basis function method can be used to solve H3(x): The modified Akima cubic Hermite interpolation method produces less fluctuation than the ordinary cubic Hermite interpolation method, is more efficient in real-time applications, and the obtained fill values will not overshoot.
4. A method for multi-element load analysis and prediction of an integrated energy system according to claim 2, characterized in that: Outlier data replacement: Use quartile method to detect outliers in historical load data; The core of the quartile method is the concept of quartiles. Quartiles are a type of quantile in statistics. All values are arranged from small to large and divided into four equal parts in sequence. The values at the three dividing points are collectively called quartiles, which are the first quartile, the second quartile, and the third quartile from small to large. Outliers in the historical load data are defined as elements that are above the first quartile or below the third quartile by more than a specified interquartile range threshold, which is 1.
5.
5. A method for multi-element load analysis and prediction of an integrated energy system according to claim 4, characterized in that: After identifying the outliers in the historical load data, it is necessary to replace the outliers. The outlier replacement operation is still performed using the data filling model established by the modified Akima cubic Hermite interpolation.
6. A method for multi-element load analysis and prediction of an integrated energy system according to claim 2, characterized in that: Data smoothing: Use moving average filter to smooth noisy historical load data.
7. A method for multi-element load analysis and prediction of an integrated energy system according to claim 6, characterized in that: The principle of the moving average filter is based on statistical laws. The working process of the filter is as follows: the continuous sampled data to be filtered is regarded as a queue of fixed length. When a new measurement process is completed, the first data in the above queue is removed, and the remaining data are moved forward in turn, and the new sampled data is inserted as the last data in the new queue; then the specified operation is performed on the obtained new queue, and its result is used as the result of this measurement. The whole process is repeated until the filtering is completed.
8. The method for multi-element load analysis and prediction of an integrated energy system according to claim 1, characterized in that: In step S2, the method for determining the correlation of the influencing factors is specifically: using the Pearson correlation coefficient method to calculate the correlation of multiple variables in the electricity, cooling and heating loads. The size of the Pearson correlation coefficient can better reflect the strength of the linear correlation between two variables; For variable X = [x1, x2, ···, x n ] Τ and Y=[y1,y2,···,y n ] Τ , and the calculation formula of its Pearson correlation coefficient is: In formula (2): and is the average value of n sample data; the range of correlation coefficient r is (-1,1), that is, |r|≤1; the value of correlation coefficient r is defined as follows: the closer the correlation coefficient |r| is to 1, the higher the correlation between x and y is. If r=1, it means that there is a positive linear complete correlation between x and y; the closer the correlation coefficient |r| is to 0, the lower the correlation between x and y is; if r=0, it means that there is no linear correlation between x and y; the present invention uses a universal coefficient measurement method: when |r|≥0.8, the two variables are highly correlated; when 0.5≤|r|<0.8, the two variables are moderately correlated; when 0.3≤|r|<0.5, the two variables are slightly correlated; when |r|<0.3, it means that the linear correlation between the two variables is weak, which can be regarded as nonlinear correlation.
9. The method for multi-element load analysis and prediction of an integrated energy system according to claim 1, characterized in that: In step S3, the method of feature factor extraction is specifically as follows: feature extraction is completed using the RCNN model; the feature extraction process is performed by two convolutional layers and two maximum pooling layers, an excitation layer and a normalization layer; the prediction process is completed by a fully connected layer and a regression layer; the input layer input size (I) is assumed to be I×1 influencing factors, the filter size (F) of convolutional layers 1 and 2 is 1×1, the number of filters (N) is 25, and the padding size (P) is 0; the stride size (S) of maximum pooling layers 1 and 2 is 2×2; therefore, after the maximum pooling layer, the feature dimension is divided by 2; and the ReLU unit of the excitation layer reduces the number of traversals to achieve a smaller training error than the traditional tanh unit; the normalization layer increases the generalization of the model and reduces the error rate; in addition, the excitation layer and the normalization layer do not change the size of the feature dimension.
10. The method for multi-element load analysis and prediction of an integrated energy system according to claim 1, characterized in that: In step S4, the load forecasting method is specifically as follows: the processed load data obtained in step S1 and the influencing factors obtained in step S3 are input into the SVR forecasting model to finally output the load forecasting result.