Intelligent heating energy load prediction method

Abstract

This invention relates to the field of load forecasting technology, specifically a smart heating energy load forecasting method, comprising: statistically analyzing heating load data corresponding to each heating area; quantifying the regional load demand of each heating area based on the changing trend of heating load, and configuring a load demand sequence; fitting the load demand sequence with a probability distribution to obtain the probability distribution characteristics of each time period; forming a regional feature library for each heating area based on the probability distribution characteristics of each time period; using the regional feature library as the basis for load fluctuation analysis, extracting the spatial clustering degree of heating areas, and determining the load forecast value of each heating area under load scheduling based on the heating load of adjacent heating areas under spatial clustering; and performing cumulative calculation on each heating area based on the load forecast values ​​of each time period and combined with the indication information under load scheduling to obtain the output load forecast value. This improves the efficiency and accuracy of heating load forecasting.

Description

Technical Field

[0001] This invention relates to the field of load forecasting technology, specifically a smart heating energy load forecasting method. Background Technology

[0002] Changes in heating load are usually closely related to multiple factors. Typically, the patterns of change in heating data are discovered by analyzing multi-dimensional information on regional heating. However, when considering the different heating load demands of different regions, multi-dimensional analysis of historical data alone can easily overlook the differences between regions, making it difficult to apply the discovered patterns to different regions. Ultimately, this affects the accuracy of heating load forecasting.

[0003] For example, Chinese Patent Publication No. CN118569457A discloses a heating load prediction method and system. The method includes: S100: determining the boundary of the heating area, dividing the heating area into several blocks based on the location of the heating station, recording the historical heating and historical temperature of the blocks, establishing a correspondence, collecting the electricity consumption growth and heating load growth of the blocks, and constructing a reference table; S200: creating a prediction tree, and dividing the prediction tree into a left subtree and a right subtree.

[0004] For example, Chinese Patent Publication No. CN120317455A discloses a method and system for predicting heat load in district heating, belonging to the field of data processing technology. The method includes: data acquisition, data preprocessing, district heating data optimization, construction of a heat load prediction model, and intelligent prediction. This scheme combines random forest split gain and mutual information to obtain the global saliency of features and select initial centroids. It fuses Euclidean distance and mutual information, calculates weighted distances and performs clustering, calculates association purity based on mutual information and information entropy, and filters and deletes redundant features according to redundancy conditions. It generates a bimodal basis adjacency matrix through a gating mechanism, generates a hidden feature matrix and a dynamic propagation adjacency matrix based on diffusion graph convolution, fuses the two adjacency matrices to generate a gated adaptive adjacency matrix, extracts spatial features based on diffusion graph convolution, and extracts temporal features based on spatiotemporal embedding and multi-head attention mechanisms.

[0005] Existing technologies predict heating load using random forests or through load multi-attention and gating mechanisms. However, these technologies emphasize local optimization at the predictive model level, which can easily overlook the inconsistency in actual heating demand across different scenarios. This leads to inaccurate quantification of regional differences and deviations between heating supply and demand when the heating load is allocated to different areas, resulting in reduced accuracy in load prediction. Summary of the Invention

[0006] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is: a smart heating energy load prediction method, including: statistically analyzing the heating load data corresponding to each heating area based on the distribution location of the heating area.

[0007] Based on the changing trend of heating load, the regional load demand of each heating area is quantified, and a load demand sequence is configured.

[0008] Based on the values ​​of the load demand sequence in each time period, the probability distribution of the heating load data in each time period is fitted to obtain the probability distribution characteristics of each time period; based on the probability distribution characteristics of each time period, regional difference analysis is performed on each heating area to form a regional feature library for each heating area.

[0009] Based on the regional feature database as the basis for load fluctuation analysis, the spatial clustering degree of heating areas is extracted, and the load forecast value of each heating area under load scheduling is determined according to the heating load of adjacent heating areas under spatial clustering.

[0010] Based on the load forecast values ​​for each time period, and combined with the indication information under load dispatch, the load forecast values ​​for each heating area are accumulated and calculated to obtain the output load forecast values.

[0011] The beneficial effects of this invention are as follows: First, this invention divides the heating network's radiation range into multiple independent minimum heating zones; selects the moving average of the heating load at any given time as the sampling basis to extract the load change trend within each time period; based on the extracted change trend, calculates the first-order and second-order difference values ​​respectively, using the inflection point of the second-order difference as the analysis anchor point to extract multiple feature points reflecting load changes; and configures a load demand sequence based on the time period length and heating load value corresponding to each feature point. This clarifies the source of heating load processing and the basis for trend analysis, laying a data foundation for subsequent regional differentiation analysis.

[0012] II. This invention synchronizes all heating areas under the same scenario, defining a unified analytical data dimension based on the data dimension corresponding to the load demand sequence. For any pair of heating areas and each data dimension, relative entropy is calculated, and the data after relative entropy symmetry processing is used as the final output relative entropy value. A three-dimensional difference matrix is ​​constructed using the range of relative entropy values, and the clustering results based on the three-dimensional difference matrix are used as the output probability distribution features. Based on the heating areas corresponding to the probability distribution features, difference feature labels are established for each heating area, and the data is integrated using these difference feature labels to obtain the final regional feature library. This clarifies the theme of regional difference analysis and provides a foundation for data backtracking using the difference feature labels of each area, enabling predictive heating loads to be adapted to areas with different differences.

[0013] Third, this invention divides heating regions into multiple groups based on scenarios in a regional feature library. Spatial adjacency analysis is performed on the spatial location of each group to determine the degree of spatial clustering. For adjacent heating regions within the same group, Pearson correlation coefficient is used for analysis. When both exhibit common changes, the current group's heating region is used as the basis for prediction to determine the output load forecast. By grouping heating regions according to scenarios in the regional feature library, consistent heating characteristics within the same group are ensured. Furthermore, by focusing on common changes, regional linkages between adjacent heating regions are ensured, thereby improving the accuracy of handling regional differences. Attached Figure Description

[0014] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0015] Figure 1 This is a flowchart illustrating a smart heating energy load prediction method.

[0016] Figure 2 This is a flowchart illustrating step S2 of a smart heating energy load prediction method.

[0017] Figure 3 This is a flowchart illustrating step S3 of a smart heating energy load prediction method.

[0018] Figure 4 This is a flowchart illustrating step S4 of a smart heating energy load prediction method. Detailed Implementation

[0019] The embodiments of the present invention are described in detail below. The embodiments described below are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention. Where specific techniques or conditions are not specified in the embodiments, they shall be performed in accordance with the techniques or conditions described in the literature in the art or in accordance with the product manual.

[0020] See Figure 1 A smart heating energy load prediction method includes: S1, based on the distribution location of heating areas, statistically analyzing the heating load data corresponding to each heating area.

[0021] S2 quantifies the regional load demand of each heating area based on the changing trend of heating load and configures the load demand sequence.

[0022] S3. Based on the values ​​of the load demand sequence in each time period, the probability distribution of the heating load data in each time period is fitted to obtain the probability distribution characteristics of each time period. Based on the probability distribution characteristics of each time period, regional difference analysis is performed on each heating area to form a regional feature library for each heating area.

[0023] S4, based on the regional feature library as the basis for load fluctuation analysis, extracts the spatial clustering degree of heating areas, and determines the load forecast value of each heating area under load scheduling based on the heating load of adjacent heating areas under spatial clustering.

[0024] S5, based on the load forecast values ​​for each time period and combined with the indication information under load dispatch, performs cumulative calculations on each heating area to obtain the output load forecast value.

[0025] Specifically, in the current implementation scenario, load dispatch is used to regulate and control the heating network. Its instruction information includes the heating area covered by the instruction, the branch lines of the network, and typical scenarios such as steady-state operation / low temperature cold wave / sudden temperature rise / holiday / emergency fault, in order to determine the main processing scenarios under the heating load forecast.

[0026] The implementation method for statistically analyzing the heating load data corresponding to each heating area in step S1 includes: based on the radiation range of the heating network, dividing it into multiple independent minimum heating areas, and establishing a mapping relationship between measuring points and heating areas based on the geographical boundary of each heating area.

[0027] Based on the heating load collected at each measuring point, the changing trend of the heating load over any time period is used as the output heating load data.

[0028] For example, based on the topology of the heating network and the radiation range of its branches, the geographical boundaries of the heating network are examined, and it is divided into multiple independent minimum heating zones to ensure that the heating conditions and load changes in each heating zone are highly consistent. Furthermore, multiple measuring points are set up in each heating zone, and the heating load of each heating zone is recorded synchronously based on the temperature, flow rate, and heat measured at these points.

[0029] In one embodiment of the present invention, for a heating area, each heating area is equivalent to an independent heating unit, and the variation of its heating load is constrained by the environment and demand of the corresponding area. After introducing the regional load demand of each area, the trend characteristics of each heating area can be obtained by using the mapping relationship between the regional load demand and the area, thereby realizing the mapping processing of the regional environment.

[0030] like Figure 2 As shown, the implementation method of configuring the load demand sequence in step S2 includes: S21, using the moving average value of the heating load at any time as the sampling basis, extracting the changing trend of the heating load in each time period.

[0031] S22, based on the extracted trend of change, calculate the first-order difference value and the second-order difference value respectively. The first-order difference quantifies the load change amplitude at adjacent time points, indicating whether the heating load is on an upward or downward trend. The second-order difference describes the acceleration of the heating load change, indicating the specific form of the heating load accelerating upward, decelerating upward, or accelerating downward. Using the inflection point of the second-order difference as the feature point for analysis, extract multiple feature points reflecting the change of heating load.

[0032] S23 Quantify the regional load demand of each heating area based on the time period length and heating load value corresponding to each feature point, and configure the corresponding heating load data into a load demand sequence starting from the time period corresponding to the load demand.

[0033] Among them, regional load demand represents the increase or decrease in the demand for heating load under a specific environmental scenario. For example, the load in a certain area surges during a cold wave. Here, the regional load demand in the current scenario is quantified by highlighting the feature points that highlight the trend change. This can quantify the demand for heating in different heating areas under a specific scenario.

[0034] Specifically, when quantifying the regional load demand of each heating area, the implementation method includes: taking the time point corresponding to each characteristic point as the center, extending the time forward and backward, and taking the extended time period as the time period corresponding to the current characteristic point. The extended time period can be marked as 12 hours or the time interval between any two characteristic points can be evenly divided to configure a corresponding time series for each characteristic point, so as to explain the heating load variation pattern over a long period of time.

[0035] Examine the feature description and scene type corresponding to each feature point, and map the time period in which the feature point is located to the scene type. The feature description represents the inflection point that the feature point belongs to, such as an acceleration inflection point or a deceleration inflection point. The scene type is used to explain the environment corresponding to the current heating area. For example, the scene type can be labeled as the starting point of a low temperature cold wave or the point where the load growth rate slows down. The corresponding type is then mapped to the current time period to determine the scene theme for subsequent predictive analysis.

[0036] The average load, peak load, and load fluctuation rate within the current time period are calculated as the basic metrics of load demand. The basic metrics are then numerically corrected according to the environmental factors corresponding to each heating area, and the corrected data is used as the output load demand sequence.

[0037] Specifically, the output load demand sequence includes not only real-time heating load data for the corresponding time period, but also basic metrics corrected for environmental factors based on real-time load data. This quantifies the average, peak, and fluctuation rates of heating load in the corresponding scenario, emphasizing the demand characteristics of different scenarios and enriching the information dimensions during data statistics.

[0038] Furthermore, when making numerical corrections to the basic metrics, environmental factors are used as independent variables, and average load, peak load, and load fluctuation rate are used as dependent variables to construct a multiple linear regression model.

[0039] Average load, peak load, and load fluctuation rate are all calculated using historical data from the same scenario. By fitting the historical data, regression coefficients for the current heating area are obtained, resulting in a fitted multiple linear model. Then, the values ​​of current environmental factors and standard environmental factors are substituted into the fitted model, and the ratio of the two is used as the correction coefficient. The values ​​of standard environmental factors are defined based on the average values ​​of environmental factors in the same scenario, and environmental factors such as outdoor temperature, wind speed, and solar radiation are selected as quantified values. Finally, the product of the correction coefficient and the current baseline measurement is considered the corrected data, representing the heating load values ​​under a specific scenario.

[0040] In one embodiment of the present invention, in order to highlight the value of each time period, the load demand sequence is converted into a probability value according to the probability of its occurrence, and the relative entropy is used to process any two heating areas to quantify the difference in probability distribution between the two areas. At the same time, according to the statistical form of relative entropy, the degree of difference of each heating area is combined into an output area feature library.

[0041] like Figure 3 As shown, when performing probability distribution fitting on the heating load data for each time period in step S3, the implementation method includes: S31, synchronizing all heating areas to the same scenario, and determining the data dimension for the current analysis based on the data dimension corresponding to the load demand sequence. This data dimension includes the average load, peak load, and load volatility rate corrected using basic metrics in step S2. These data are considered as the data dimensions here. The scenario revealed by each data dimension will differ due to its value characteristics. For example, the average value is suitable for analyzing stable scenarios and is biased towards a probabilistic description of the entire scenario; the peak load represents the maximum heating load under that scenario, emphasizing the extreme value of scenario demand; and the load volatility rate emphasizes the fluctuation at each time point, highlighting the changes in each time segment within the overall time period. Therefore, under different data dimensions, the differences between heating areas will present diverse content.

[0042] S32 calculates the relative entropy for any pair of heating zones and each data dimension, and uses the data after relative entropy symmetry processing as the output relative entropy value; the relative entropy ranges from 0 to positive infinity, where a value of 0 represents no difference, and the larger the value, the greater the difference between heating zones.

[0043] Specifically, the range of relative entropy determines the differences between heating areas, and the differences in the three dimensions are quantified according to the range of relative entropy values.

[0044] If we consider the data dimensions used in the current scenario, each data dimension will be selected based on the load requirements of the specific scenario, namely the extreme scenario, stable scenario, and fluctuating scenario corresponding to the load requirements, rather than using three dimensions to process the same scenario at the same time.

[0045] For example, when calculating the relative entropy of average load, taking any two heating areas A and B as examples, the average value and standard deviation of their corresponding areas are used as inputs, and the solution is obtained using a normal distribution method.

[0046] Since there is no upper limit to the value of relative entropy, it is impossible to uniformly quantify and compare the degree of difference between different regions. Therefore, JS divergence (also known as symmetric KL divergence) is used to quantify the current value of relative entropy, so as to give the boundary of relative entropy division, and thus make it easier to clarify the load difference between heating areas.

[0047] The calculation formula can be expressed as follows: ;in, The JS divergence represents the values ​​of relative entropy symmetry processing in the corresponding scenario, indicating the values ​​of heating areas A and B. This is represented as the average distribution of heating areas A and B. In other words, when the average value and standard deviation of the two are used as inputs, the average of their inputs is regarded as the input data for this area M. and Let A and M represent the relative entropy of heating areas A and M, respectively, and let B and M represent the relative entropy of heating areas B and M.

[0048] Furthermore, the relative entropy of heating zones A and M can be expressed as follows.

[0049] ;in, and This represents the average value and standard deviation of heating area A; and The mean and standard deviation of region M are represented by the following. Similarly, the solution for the heating region B is obtained, thus yielding the data after relative entropy symmetry processing.

[0050] This completes the calculation of the relative entropy of the average value relative to the dimension. Furthermore, when the JS divergence is base 2, its value range boundary can be synchronized to the value range of 0 to 1, which can measure the significance of differences under each dimension, thereby driving regional difference matching in heating scenarios.

[0051] Therefore, when synchronizing to the data dimension corresponding to the peak load, the essence is to measure the information loss in the dimension where the peak load is located. Its relative entropy can be calculated using the generalized Pareto distribution. The approximate value of the relative entropy is calculated based on the probability density ratio of sample points in two heating areas. The relative entropy is then processed according to the above JS divergence calculation process to obtain the data after symmetry processing.

[0052] Furthermore, the generalized Pareto distribution is expressed as: ;in, It represents the probability density of the heating area, which represents the probability density of heating area A or B in the corresponding scenario, and the logarithm of the ratio of its probability densities represents the relative entropy in the current scenario. This represents the input parameter, specifically the heating load value corresponding to the peak load in the current scenario; This represents the location parameter, which explains the range of values ​​for its heating load. It can be used as a threshold defined in the current scenario, and can be set to 0 in the extreme scenario corresponding to the peak load to characterize the probability density distribution under relatively extreme conditions. This represents the scale parameter, which is a real number greater than 0. This represents the shape parameter; its scale parameter and shape parameter can be configured here based on the current scene, using Monte Carlo sampling, with the average value calculated from a large sample as the parameter.

[0053] Furthermore, since the relative entropy represented by the generalized Pareto distribution represents the expected value under multiple sample values, the average value of the probability density ratio of heating area A and heating area B under multiple samples will be introduced to obtain the output relative entropy.

[0054] As for the scenario corresponding to load volatility, since load volatility represents the relative entropy under continuous probability, it can be quantified by adopting the form of gamma distribution to complete its symmetry processing.

[0055] The probability density of the gamma distribution can be expressed as follows: ;in, This represents the probability density of the gamma distribution; , and These represent the input parameter, shape parameter, and scale parameter, respectively. The relative definitions of the input parameter are the same as those in the generalized Pareto distribution. The shape parameter indicates the sharpness of the load to explain non-negative, right-skewed load data. The scale parameter is proportional to the load mean to explain the relative level of the heating load. This represents the gamma function. The shape and scale parameters are selected using maximum likelihood estimation to determine the relative differences between heating areas under continuous load fluctuations. Then, the same probability density ratio calculation method as described above is used to calculate the relative entropy of the corresponding area.

[0056] S33. Utilizing the range of relative entropy values, a three-dimensional difference matrix is ​​constructed with all heating regions as rows and columns. Each element records the significance of differences between each pair of heating regions in the three dimensions. The clustering results based on the three-dimensional difference matrix are used as the output probability distribution features. The significance of differences in the three dimensions is determined by the value of relative entropy. The average relative entropy of the current heating region pair in historical data is selected as the basis for judging the significance of differences. Data exceeding this value are considered significant differences, while those not exceeding it are considered insignificant differences, thus completing the labeling and construction of the three-dimensional difference matrix.

[0057] Secondly, when clustering grouping results based on the three-dimensional difference matrix, the implementation method includes: based on the set three-dimensional difference matrix, the relative entropy value of each pair of heating areas is used as the clustering index, and clustering is performed according to its dimension to form multiple groups of clustered data. This ensures that the heating areas in each group are not different in all dimensions, and the heating areas between different groups are different in at least one dimension. The grouped data is regarded as the output clustering grouping results.

[0058] Specifically, hierarchical clustering or K-means clustering algorithms are used, treating the values ​​of each pair of heating areas in three dimensions as an index value, and then clustering based on this data. Based on the clusters formed, it is ensured that there are no significant differences between the multiple pairs of heating areas within each cluster; that is, the relative entropy values ​​are required to be similar. For heating area pairs between different clusters, there needs to be a descriptive difference to classify different combinations of similar and different regions. Then, the differences under average, peak, and continuous fluctuation scenarios are quantified, and finally, the regional feature library is formed for subsequent output.

[0059] S34. Based on the probability distribution characteristics corresponding to the heating areas, differential feature labels are established for each heating area. These labels are then used to integrate the data, resulting in an output regional feature library. The differential feature labels explain the differences between different groups under average, peak, and continuous fluctuation scenarios. For example, Group 1 has high average demand, large peak demand, and low fluctuation stability; Group 2 has medium average demand, small peak demand, and high fluctuation stability. This characterizes scenarios where there is no difference within a group but at least one dimension of difference exists between different groups. Additionally, based on the relative entropy value under clustering, additional annotations are used to indicate whether there are significant differences relative to historical data, thus completing the feature statistics of the heating areas.

[0060] Furthermore, when forming the regional feature database for each heating area, the implementation method also includes: for the acquired probability distribution features, the probability distribution features are divided into data in multiple time periods through progressive time windows; the progressive time windows are divided into the basic window for short-term load forecasting, the scenario window for medium-term load forecasting, and the periodic window for long-term load forecasting, and the probability distribution features are synchronized to multiple time periods with the time length of a single day, a week, and a month, respectively.

[0061] For each time period, the probability distribution characteristics are bound to the location of the heating area to determine the effective time length of each output probability distribution characteristic.

[0062] Based on the effective time length, the probability distribution features of the corresponding time period are combined into the output regional feature library.

[0063] In contrast, the aforementioned effective time length represents the maximum time range within which heating load differences can be obtained for a certain period of time. Furthermore, in the context of heating load configuration, the effective time length is based on the actual heating time period and combined with the average heating time length in historical data for relative scenarios. When the actual heating time falls within the confidence interval of the historical heating time, it is considered as the effective time length and sequentially included in the regional feature library. The confidence interval can be represented by the average time length ± 1.96 times the standard deviation.

[0064] In one embodiment of the present invention, such as Figure 4 As shown, the implementation method of determining the load forecast value of each heating area under load scheduling in step S4 includes: S41, based on the heating area corresponding to the regional feature library, the heating area is divided into multiple groups according to the scenario of the regional feature library, and spatial adjacency analysis is performed on the spatial location of each group of heating areas to determine the spatial clustering degree of the heating area.

[0065] Spatial adjacency represents two heating areas that are geographically adjacent or share the same pipeline branch. If we consider the scenario of heating areas in the regional feature library, that is, areas with different average loads, peak loads and continuous fluctuations, we take each area as a target of the current analysis and check whether there is spatial clustering between the corresponding heating areas.

[0066] Specifically, weights are set for spatially adjacent heating areas. If they are spatially adjacent, the weight is set to 1; otherwise, it is set to 0. After traversing all heating areas step by step, the Moran index is calculated using spatial autocorrelation analysis with the heating load value at the corresponding time and the set weights. The Moran index is then used as the output spatial clustering degree.

[0067] S42, for adjacent heating areas within the same group, analyze the adjacent heating areas using the Pearson correlation coefficient. When both have common changes, use the heating area of ​​the current group as the basis for prediction to determine the output load forecast value.

[0068] The Pearson correlation coefficient is calculated using the heating load values ​​of two heating areas. Only when the Pearson correlation coefficient is greater than 0.7 is it considered that there is a scenario where adjacent heating areas change synchronously when the load changes, thus providing the basis for data prediction.

[0069] Specifically, for regions clustered by average load, peak load, and continuous fluctuation, the load is grouped according to its region, and the heating load values ​​are simultaneously checked for synchronous changes. Only when synchronous changes occur are the current scenario data used as the basis for prediction to improve the accuracy of load forecasting. Secondly, using the data from the prediction basis as input, the load value at the corresponding time is fitted through linear regression to obtain the output load forecast value.

[0070] It should be noted that when performing the prediction process in step S4, the load prediction value for a certain area at a certain moment is biased towards output. When performing the subsequent process in step S5, the load prediction values ​​will be superimposed according to the range that needs to be predicted here, so as to obtain the final output load prediction value.

[0071] Furthermore, after obtaining the spatial aggregation degree, the method for determining the spatial aggregation degree of the heating area in step S41 also includes: aligning the spatial aggregation degree with the heating area based on the range of values ​​for the spatial aggregation degree, and determining the spatial radius corresponding to each group of heating areas.

[0072] By using the Pearson correlation coefficient between spatial radius and spatial clustering, the priority of each heating area is constructed, and the output order of load forecast values ​​is adjusted based on the priority of the heating area.

[0073] In the above processing, the calculated spatial clustering degree is mapped to its related spatial radius. After introducing the calculation method of Pearson correlation coefficient, it can be seen that the similarity of heating area load change is linearly correlated with its corresponding spatial size. Then, the clustering situation under spatial scheduling is determined, and the data is output in descending order of Pearson correlation coefficient to prioritize the processing of areas with strong spatial distribution and clustering correlation, thereby improving the efficiency of heating load prediction.

[0074] In step S5, when performing the cumulative calculation for each heating area, the implementation method includes: based on the load dispatching instruction information, calculating the normalized load forecast value according to the load forecast value for each time period, and adjusting the output order of each time period according to the proportion of the sum of the load forecast values ​​for each time period. Specifically, the normalization calculation essentially distributes all load values ​​to each time period with relative weights, calculates the predicted load value for each time period, and outputs the load forecast values ​​for each time period in descending order of their sum, thereby adjusting the relative order of the load forecast output.

[0075] Secondly, by accumulating the load forecast values, the overall load demand of the region at a specific time is obtained, and the load forecast for the corresponding time period is completed according to the overall load demand.

[0076] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention, which are still covered within the protection scope of the present invention.

Claims

1. A smart heating energy load prediction method, characterized in that, include: Based on the distribution of heating areas, statistical data on the heating load corresponding to each heating area is collected. Based on the changing trend of heating load, quantify the regional load demand of each heating area and configure the load demand sequence; Based on the values ​​of the load demand sequence in each time period, the probability distribution of the heating load data in each time period is fitted to obtain the probability distribution characteristics of each time period. Based on the probability distribution characteristics of each time period, regional differences analysis is conducted for each heating area to form a regional feature database for each heating area. Based on the regional feature database as the basis for load fluctuation analysis, the spatial clustering degree of heating areas is extracted, and the load forecast value of each heating area under load scheduling is determined according to the heating load of adjacent heating areas under spatial clustering. Based on the load forecast values ​​for each time period, and combined with the indication information under load dispatch, the load forecast values ​​for each heating area are accumulated and calculated to obtain the output load forecast values.

2. The intelligent heating energy load prediction method according to claim 1, characterized in that, The methods for collecting heating load data for each heating area include: Based on the radiation range of the heating network, it is divided into multiple independent minimum heating zones. The geographical boundary of each heating zone is used as the measurement basis to establish the mapping relationship between the measuring points and the heating zones. Based on the heating load collected at each measuring point, the changing trend of the heating load over any time period is used as the output heating load data.

3. The intelligent heating energy load prediction method according to claim 1, characterized in that, The implementation methods for configuring load demand sequences include: Using the moving average of the heating load at any given time as the sampling basis, the changing trend of the heating load in each time period is extracted; Based on the extracted trend of change, the first-order difference value and the second-order difference value are calculated respectively. The inflection point of the second-order difference is used as the feature point for analysis, and multiple feature points reflecting the change of heating load are extracted. The regional load demand of each heating area is quantified by the time period length and heating load value corresponding to each feature point. The corresponding heating load data is configured into a load demand sequence, with the time period corresponding to the load demand as the starting point.

4. The intelligent heating energy load prediction method according to claim 3, characterized in that, When quantifying the regional load demand of each heating area, the methods include: Taking the time point corresponding to each feature point as the center, the time is extended forward and backward, and the extended time period is regarded as the time period corresponding to the current feature point. Check the feature description and scene type corresponding to each feature point, and map the time period in which the feature point is located according to the scene type; The average load, peak load, and load fluctuation rate within the current time period are calculated as the basic metrics of load demand. The basic metrics are then numerically corrected according to the environmental factors corresponding to each heating area, and the corrected data is used as the output load demand sequence.

5. The intelligent heating energy load prediction method according to claim 1, characterized in that, When fitting the probability distribution of heating load data for each time period, the methods include: Synchronize all heating areas to the same scenario, and determine the data dimension for the current analysis based on the data dimension corresponding to the load demand sequence; For any pair of heating zones and each data dimension, calculate the relative entropy separately, and use the data after relative entropy symmetry processing as the output relative entropy value; Using the range of relative entropy values, a three-dimensional difference matrix is ​​constructed with all heating areas as rows and columns. The clustering results based on the three-dimensional difference matrix are used as the output probability distribution features. Based on the heating areas corresponding to the probability distribution characteristics, differential feature labels are established for each heating area, and the data are integrated using the differential feature labels to obtain the output regional feature library.

6. The intelligent heating energy load prediction method according to claim 5, characterized in that, When clustering grouping results based on a three-dimensional difference matrix, the implementation methods include: Based on the set three-dimensional difference matrix, the relative entropy value of each pair of heating areas is used as the clustering index. Clustering is performed according to the dimension in which they are located to form multiple groups of clustered data. This ensures that the heating areas within each group are identical in all dimensions, while the heating areas between different groups are identical in at least one dimension. The grouped data is then regarded as the output clustering grouping result.

7. The intelligent heating energy load prediction method according to claim 1, characterized in that, When forming the regional feature database for each heating area, the implementation methods also include: Based on the obtained probability distribution features, the probability distribution features are divided into data within multiple time periods through progressive time windows; For each time period, the probability distribution characteristics are bound to the location of the heating area to determine the effective time length of each output probability distribution characteristic; Based on the effective time length, the probability distribution features of the corresponding time period are combined into the output regional feature library.

8. The intelligent heating energy load prediction method according to claim 1, characterized in that, The methods for determining the load forecast values ​​for each heating area under load dispatch include: Based on the heating areas corresponding to the regional feature database, the heating areas are divided into multiple groups according to the scenarios in the regional feature database. Spatial adjacency analysis is performed on the spatial location of each group of heating areas to determine the degree of spatial clustering of the heating areas. For adjacent heating areas within the same group, the Pearson correlation coefficient is used to analyze the adjacent heating areas. When both have common changes, the heating area of ​​the current group is used as the basis for prediction to determine the output load forecast value.

9. The intelligent heating energy load prediction method according to claim 1, characterized in that, Other methods for determining the spatial concentration of heating areas include: Based on the range of spatial clustering degree values, the spatial clustering degree is spatially aligned with the heating area to determine the spatial radius corresponding to each group of heating areas; By using the Pearson correlation coefficient between spatial radius and spatial clustering, the priority of each heating area is constructed, and the output order of load forecast values ​​is adjusted based on the priority of the heating area.

10. The intelligent heating energy load prediction method according to claim 1, characterized in that, When performing cumulative calculations for each heating area, the implementation methods include: Based on the load forecast values ​​for each time period, normalized load forecast values ​​are calculated for the load forecast values ​​for each time period. The output order of each time period is then adjusted according to the proportion of the sum of the load forecast values ​​for each time period.

Images