A method for evaluating the operation efficiency of heating pipe networks based on data prediction

Abstract

The present invention relates to the technical field of heating pipe network operation efficiency evaluation, and discloses a heating pipe network operation efficiency evaluation method based on data prediction. The method includes: constructing a structural transformation result matrix, establishing a graph modeling structure, extracting disturbance features and constructing a graph propagation modulation mechanism, generating a structural disturbance response representation, identifying the structural behavior type, combining a basic predictor to complete node flow prediction, and comparing and analyzing the prediction results with the actual operation data to determine the degree of fit between the node operation status and the target heating demand, thereby outputting the operation efficiency evaluation results of the heating pipe network. By introducing a disturbance-driven graph propagation mechanism and a multi-mode structure recognition strategy, this method effectively improves the modeling capability and prediction accuracy of nonlinear structural responses in complex heating systems, and provides support for real-time evaluation of system operation status and efficiency management.

Description

Technical Field

[0001] The present invention relates to the field of intelligent evaluation of the operating status of a heating system, and in particular to a method for evaluating the operating efficiency of a heating pipe network based on data prediction. Background Art

[0002] At present, with the continuous expansion of urban centralized heating, the heating pipeline network, as a key energy transmission infrastructure connecting the heat source and the user terminal, plays a core supporting role in ensuring the temperature of people's livelihood, optimizing the energy structure, and improving the resilience of urban operation. The pipeline network operation efficiency is a key indicator reflecting the system's heat energy transmission capacity and energy consumption level. Its evaluation results directly affect the formulation of the heat source side scheduling strategy, user-side service response and the improvement of overall energy utilization efficiency. Traditional heating efficiency evaluation methods mostly rely on manual inspections, historical experience or static model calculations, which are difficult to reflect in real time the dynamic load disturbances, changes in pipeline heat loss and system operation characteristics under multi-source data coupling. Especially in complex thermal topology and multi-condition switching scenarios, the evaluation results are delayed and the accuracy is limited. It is urgent to establish a more intelligent, dynamic, scenario-oriented and adaptable operation efficiency prediction and analysis mechanism.

[0003] In one existing technology, patent CN114693069A proposes a "central heating system stability assessment method and system", which evaluates the system safety and stability through simulation and operating parameters. The advantage of this method is that it can warn of abnormal risks, but it relies on simulation model and control strategy verification, and the assessment process is relatively complicated; in another existing technology, patent CN113007784B proposes a "large-scale heating pipeline network comprehensive evaluation method", which constructs five dimensions including heating quality, energy consumption, equipment efficiency, parameter control and intelligence, and uses scoring and weighting methods to give a comprehensive assessment, but the evaluation indicators are relatively macro, and it is difficult to reflect the differences in operating efficiency at the node level.

[0004] The existing technology generally has the problems of relatively simple model structure and insufficient collaborative modeling capabilities between variables; the current data-driven optimization mechanism fails to deeply model the nonlinear coupling relationship and disturbance propagation behavior between different sensor variables, making it difficult to effectively perceive and characterize the structural evolution characteristics between nodes in the heating pipeline network, thereby limiting the accuracy of system operation status assessment and the dynamic identification capability of efficiency deviation. Summary of the Invention

[0005] The present invention proposes a method for evaluating the operating efficiency of a heating network based on data prediction, aiming to achieve accurate prediction of heating node flow and comprehensive analysis of operating efficiency. The method first acquires and preprocesses historical operating data to construct an original time series input matrix; then introduces a coupling attention factor and a disturbance spectrum distribution projection factor, constructs a multi-channel transformation kernel and performs structural transformation to generate a structural transformation result matrix; then, calculates the structural response characteristics and constructs a graph adjacency matrix, introduces a disturbance amplitude vector, a stability factor and a transition caustic factor, constructs a disturbance-driven graph propagation mechanism, completes feature propagation, and generates a structural disturbance response representation matrix; then, by introducing a kurtosis disturbance factor, identifies structural behavior patterns, and selects corresponding predictors for node flow prediction; finally, compares and analyzes the prediction results with the actual operating data to determine the degree of fit between the operating status of the heating node and the target demand, thereby completing the evaluation and analysis of the operating efficiency of the heating network.

[0006] In a first aspect, in order to solve the above technical problems, the present invention provides a method for evaluating the operating efficiency of a heating network based on data prediction, comprising:

[0007] S1. Acquire and preprocess historical operating data in the heating network to construct an original time series input matrix. The original time series input matrix includes the flow rate, temperature, pressure, supply and return water temperature difference of the heating node, the flow rate of the heating pipe section, and the time series characteristic value of the outside temperature in a continuous time window;

[0008] S2. Introducing a coupling attention factor and a disturbance spectrum distribution projection factor to construct a multi-channel transformation kernel structure, and performing a structural transformation operation on the original time series input matrix based on the multi-channel transformation kernel structure to obtain a structural transformation result matrix;

[0009] S3. Based on the structural transformation result matrix, the response path common track coupling degree, the structural gradient amplitude difference, the trajectory change staggered directionality and the response time offset synchronization are calculated to construct a graph adjacency matrix;

[0010] S4, introducing a disturbance amplitude vector, a response stability factor, and a transition caustic factor, and combining them with the graph adjacency matrix to construct a disturbance-driven graph propagation modulation mechanism, and using the disturbance-driven graph propagation modulation mechanism to perform feature propagation processing on the structural transformation result matrix to obtain a structural disturbance response representation matrix;

[0011] S5. Introducing a kurtosis disturbance factor, constructing a multi-mode response structure discrimination mechanism, and performing structural behavior identification on the structural disturbance response representation matrix based on the discrimination mechanism to obtain a structural behavior identification result;

[0012] S6. Select a corresponding basic predictor for prediction based on the structural behavior identification result to obtain a predicted value of the heating node flow, wherein the basic predictor includes a trend predictor, a hybrid predictor, and a jump predictor;

[0013] S7. Compare and analyze the predicted value with the actual operation record of the heating network, combine it with the historical reference benchmark, judge the degree of fit between the node operation status and the target heating demand, and output the evaluation conclusion of the heating network operation efficiency.

[0014] Preferably, in step S1, historical data of the heating network covering a typical operating cycle is first collected, including parameters such as flow, pressure, supply and return water temperature, pipe section flow rate and external ambient temperature at key nodes; then, the data is time-aligned, anomalies are eliminated and missing data are completed to obtain continuous and complete operating data; then, the data is normalized, and the sliding window and step size are set according to the thermal characteristics to divide the original sequence into multiple time series segments; finally, the multivariate data is organized in chronological order to construct the original time series input matrix in a unified format to provide data support for subsequent modeling and regulation.

[0015] Preferably, the historical operating data in the heating pipeline network usually has the characteristics of strong multi-variable coupling, uneven propagation of state disturbances between nodes, and implicit structural response characteristics. Especially in the case of dynamic working condition changes or sudden load disturbances, there are differences in the response degrees of different physical variables in the spatial structure, which makes it difficult for the original time series input matrix to directly reveal the collaborative behavior pattern and structural driving mechanism between nodes. The physical variables include flow, pressure, and temperature difference. To address the above problems, the present invention introduces coupling attention factors and disturbance spectrum distribution projection factors, and realizes the adaptive characterization of the mutual influence relationship and structural perception weight between each variable in the input matrix by constructing a transformation kernel module with multi-channel structural modeling capabilities. In the process of structural transformation, the multi-channel transformation kernel can fuse the node spatial distribution and variable response intensity, realize structural deconstruction and reconstruction operations under disturbance drive, and finally generate a structural transformation result matrix with structural differentiation and disturbance perception, providing a structural embedding expression basis for subsequent graph modeling and control strategy design.

[0016] Furthermore, in step S2, a coupling attention factor and a disturbance spectrum distribution projection factor are introduced to construct a multi-channel transformation kernel structure. A structural transformation operation is performed on the original time series input matrix based on the multi-channel transformation kernel structure. The process of obtaining the structural transformation result matrix includes:

[0017] S21, based on the original time series input matrix, extract the local neighborhood time series difference characteristics and global statistical characteristics of the variable, construct a structural tension entropy function, and determine the coupling attention factor according to the structural tension entropy function ;

[0018] S22. Construct a structural gradient term based on the original time series input matrix. The calculation formula is:

[0019] ,

[0020] in, For As the center, is the local neighborhood with the neighborhood radius, For input variables In the neighborhood The local mean within , is the structural gradient, which indicates the deviation of the point from its local structure;

[0021] S23. Determine the perturbation spectrum distribution projection factor based on the structural gradient term using a channel perturbation strategy. The calculation formula is:

[0022] ,

[0023] in, is the time window length, is the number of input variables, is the channel index, The constructed sinusoidal modulation spectrum basis, For channel The projection factor of the perturbation spectrum distribution;

[0024] S24. Construct a multi-channel heterogeneous transformation kernel structure based on the coupling attention factor and the disturbance spectrum distribution projection factor. The calculation formula is:

[0025] ,

[0026] in, For channel At the point The transformed kernel response on It is an oscillating frequency capture core. is a Gaussian smoothing filter;

[0027] S25, performing a channel response mapping operation on the original time series input matrix based on the multi-channel heterogeneous transformation kernel structure to obtain a structure transformation result matrix .

[0028] Furthermore, in step S2, based on the original time series input matrix, the local temporal differences and global statistical characteristics of the variables are extracted, the structural tension entropy function is constructed, and the coupling attention factor is determined to regulate the structural fluctuation sensitivity of the variables; then a structural gradient term is constructed to describe the microscale change trend of the variables, and based on this term, a disturbance spectrum distribution projection factor is generated to extract the frequency domain features in the channel domain; the coupling attention factor is fused with the disturbance spectrum projection factor to construct a multi-channel heterogeneous transformation kernel to enhance the expression of channel response differences; finally, the structural transformation of the original input is completed, and the structural transformation result matrix is ​​generated to provide multi-scale feature support for subsequent mapping and recognition.

[0029] Preferably, the historical data of the heating pipe network flow system often exhibits complex characteristics such as weak response path coupling, uneven structural fluctuations, staggered disturbance directions and time sequence offsets. Especially when the load fluctuates violently or the pipe network structure is complex, it is difficult for the variables between nodes to change synchronously, which affects the identification of collaborative relationships. To solve the above problems, the present invention introduces four types of structural characteristic indicators based on the structural transformation results: common rail coupling degree, gradient difference, staggered directionality and time synchronization, and comprehensively generates an adjacency weight matrix through multi-dimensional modeling to enhance the system structure expression and disturbance response modeling capabilities, thereby improving the prediction accuracy and control effect of the regulation model.

[0030] Furthermore, in step S3, based on the structural transformation result matrix, the response path common track coupling degree, the structural gradient amplitude difference, the trajectory change staggered directionality and the response time offset synchronization are calculated. The process of constructing the graph adjacency matrix includes:

[0031] S31, based on the structural transformation result matrix, extract the structural change trend characteristics of each variable in the time dimension, and calculate the variable pair The response path common rail coupling degree between them is calculated as follows:

[0032] ,

[0033] in, For variables At the time point The unit direction change rate of the variable The structural change trend characteristics in the time dimension, For variable pairs The response path common rail coupling degree, The closer the value of the measure is to 1, the better the variable With variables The more consistent the response direction is, the stronger the structural common rail property is;

[0034] S32. According to the structural transformation result matrix, construct the structural gradient amplitude index of each variable in the time dimension, and the calculation formula is:

[0035] ,

[0036] in, For variables The structural gradient amplitude index reflects the variable The average intensity of the response change per unit time over the entire time period, The larger the value, the more intense the variable response. The smaller the value, the smoother the variable response;

[0037] S33, based on the structural gradient amplitude index, combined with the response difference relationship between the variable pairs, determine the structural gradient amplitude modulation factor ;

[0038] S34. Based on the structural change trend characteristics of each variable in the time dimension, determine whether there is frequent intersection in the response direction of the variable pair in the time dimension, and construct a trajectory change intersection directionality index. The calculation formula is:

[0039] ,

[0040] in, For variable pairs The trajectory changes staggered direction index, when When it approaches 0, it means and The response directions of the two are consistent most of the time. When it approaches 1, it indicates and The two directions are frequently opposite, and the structures are significantly intertwined;

[0041] S35. Based on the structural change trend characteristics, evaluate the degree of response time offset of the variable on the time structure and construct a response time offset synchronization index. ;

[0042] S36. Based on the response path co-track coupling degree, the structural gradient amplitude modulation factor, the trajectory change interlaced directionality index and the response time offset synchronization, a final graph adjacency matrix is ​​constructed. The calculation formula is:

[0043] ,

[0044] in, For variable pairs The adjacency matrix of 、 、 Construct hyperparameters that influence weights for controlling graph structure.

[0045] Furthermore, in step S3, based on the structural transformation result matrix, the change trend of each variable in the time dimension is extracted, and the co-track coupling degree of the response path is calculated to measure the coordinated change relationship between variables and identify potential dynamic associations; at the same time, a structural gradient amplitude index is constructed to reflect the degree of change in the local response intensity, and a structural gradient amplitude modulation factor is generated based on this to highlight the structural jump area; combined with the interleaving of the response directions between variables, a trajectory change interleaving directionality index is constructed to capture the instability of the disturbance path; further, the synchronization of the variable response moments is evaluated, and a response time offset synchronization index is generated to characterize asynchronous behavior; finally, the above four types of indicators are integrated to construct a graph adjacency matrix to comprehensively express the structural response and disturbance propagation characteristics, providing a basis for subsequent graph model analysis and regulation optimization.

[0046] Preferably, the heating pipeline network often exhibits complex characteristics such as uneven disturbance, asymmetric coupling and nonlinear response during dynamic operation. Especially when the cold and hot loads alternate or the load is uneven, the node response is prone to sudden diffusion, and it is difficult for traditional static graphs to accurately characterize its propagation path; for this reason, the present invention introduces the disturbance amplitude vector, the disturbance difference adjustment index and the transition dispersion factor to construct a disturbance-driven graph propagation modulation mechanism, realize the direction-sensitive propagation and adaptive control of the disturbance behavior, effectively improve the model's characterization accuracy of disturbance coupling and flow changes, and enhance the robustness and response sensitivity of the control system.

[0047] Furthermore, in step S4, a disturbance amplitude vector, a disturbance difference adjustment index, and a transition caustic factor are introduced, and combined with the graph adjacency matrix, a disturbance-driven graph propagation modulation mechanism is constructed. The structural transformation result matrix is ​​subjected to feature propagation processing using the disturbance-driven graph propagation modulation mechanism to obtain a structural disturbance response representation matrix. The process includes:

[0048] S41. Introducing the perturbation amplitude vector, the perturbation difference adjustment index, and the transition caustic factor, and combining them with the graph adjacency matrix, a perturbation-driven graph propagation modulation mechanism is constructed.

[0049] S42, according to the graph propagation modulation mechanism, performing disturbance response association processing on the structure transformation result matrix to construct the difference propagation term ;

[0050] S43, constructing a disturbance auto-amplitude difference function according to the difference propagation term, and the calculation formula is:

[0051] ,

[0052] in, is the disturbance amplitude difference function. When the disturbance amplitude is small, it is approximately a linear response. When the disturbance amplitude is large, it automatically suppresses the change. is the input vector of the perturbation auto-amplitude difference function, i.e., the difference propagation term constructed in step S42, The square norm is obtained by accumulating the squares of the difference propagation terms in all hidden layers;

[0053] S44. Introduce a residual retention factor, combine it with the disturbance amplitude difference function, and construct a disturbance-driven response characteristic mapping mechanism. Based on the response characteristic mapping mechanism, perform nonlinear modulation and dynamic update on the structural state of the current node to obtain a structural disturbance response representation matrix composed of the structural disturbance response tensors of all nodes.

[0054] Furthermore, in step S41, the perturbation amplitude vector, the perturbation difference adjustment index, and the transition caustic factor are introduced, and combined with the graph adjacency matrix, the process of constructing the perturbation-driven graph propagation modulation mechanism includes:

[0055] S411, performing feature extraction processing on the input response sequence of each variable in the structural transformation result matrix to determine the disturbance amplitude vector;

[0056] S412, performing amplitude analysis processing on the disturbance amplitude vector to construct a response stability factor;

[0057] S413. Constructing a disturbance difference adjustment index based on the response stability factor;

[0058] S414, performing structural fluctuation feature processing on the response sequence of each variable connection pair in the graph adjacency matrix to construct a transition caustic index;

[0059] S415. Based on the disturbance amplitude vector, the response stability factor and the transition caustics index, perform matrix fusion calculation to construct a disturbance-driven graph propagation modulation mechanism.

[0060] Furthermore, in step S4, in order to characterize the diffusion characteristics of disturbances between nodes, the response sequence of each node is first extracted to construct a disturbance amplitude vector, which is used to measure its memory of historical disturbances; then the amplitude change is analyzed to obtain the response stability factor, which improves the ability to identify areas with severe fluctuations; the disturbance difference adjustment index is constructed in combination with the stability factor to achieve the regulation of the response intensity of key nodes; at the same time, the structural fluctuation characteristics of the variable pairs are extracted, and the transition dispersion factor is constructed to reflect the consistency and conflict degree of the disturbance propagation direction; the above three types of factors are integrated to establish a disturbance-driven graph propagation modulation mechanism, and based on this, difference propagation and nonlinear mapping are performed to form a structural disturbance response representation; this mechanism enhances the model's perception of local mutations and multi-layer propagation characteristics, providing structural precision support for subsequent regulation.

[0061] Preferably, the heating pipeline network often experiences nonlinear changes in structural response during the regulation process, especially in the load mutation or regulation switching stage, the node variables may show sudden jumps or slow changes, and traditional static methods are difficult to identify their complex structural behaviors; for this reason, the present invention introduces a kurtosis disturbance factor, and establishes a structural curvature and peak evaluation model by constructing disturbance difference, extracting jump intensity and extreme value ratio and other indicators, so as to achieve accurate identification of trend-type, jump-type and mixed-type structures, which helps to improve the key node identification ability and the classification accuracy of regulation behavior.

[0062] Furthermore, in step S5, a kurtosis disturbance factor is introduced to construct a multi-mode response structure discrimination mechanism, and structural behavior recognition is performed on the structural disturbance response representation matrix based on the discrimination mechanism. The process of obtaining the structural behavior recognition result includes:

[0063] S51, extracting the amplitude jump vector of the structural response representation matrix and constructing the structural disturbance difference ;

[0064] S52: performing a change trend extraction operation on the structural disturbance difference, constructing a structural curvature characterization method, and obtaining the overall structural curvature strength based on the structural curvature characterization method. ;

[0065] S53, performing a mutation concentration evaluation operation on the structural disturbance difference, building a multi-mode response structure discrimination mechanism in combination with the overall structural curvature strength, and determining a kurtosis disturbance factor based on the multi-mode response structure discrimination mechanism, the calculation formula is:

[0066] ,

[0067] in, For input variables The maximum amplitude mutation in the structural channel dimension, For input variables The average jump level in the structural channel dimension, that is, the background perturbation intensity, For input variables The corresponding kurtosis disturbance factor;

[0068] S54, when , indicating the input variable It is relatively stable and belongs to trend type variable. , indicating the input variable There is a slight disturbance but no sudden jump, which is a mixed variable. , indicating the input variable There is a significant boundary jump, which is a jump type variable.

[0069] Furthermore, in step S5, in order to accurately identify the flow disturbance behavior of the heating network, the amplitude jump vector of each variable is first extracted, and the structural disturbance difference is constructed to capture its instantaneous mutation characteristics; then the change trend is extracted, the structural curvature intensity is calculated, and the nonlinear turning degree of the variable response is measured; combined with the distribution concentration of the mutation points, a mutation concentration index is formed to describe the aggregation characteristics of the disturbance behavior; finally, the two are integrated to construct a multi-mode response structure discrimination mechanism, generate a kurtosis disturbance factor, and realize the adaptive classification of trend-type, jump-type and mixed-type structural behaviors, which helps to enhance the model's recognition ability and control accuracy for complex disturbances.

[0070] Preferably, the flow of the heating pipe network exhibits diverse structural behaviors, and different nodes may show stable trends or jump mutations, which makes it difficult for a unified predictor to take into account the characteristics of various variables and easily cause prediction errors; for this reason, the present invention constructs a multi-predictor system based on the structural identification results, and adopts trend predictors, hybrid predictors and jump predictors to adapt to different structural types, so as to achieve accurate modeling of steady changes, local disturbances and mutation responses, thereby improving the overall prediction accuracy.

[0071] Furthermore, in step S6, according to the structural behavior identification result, a corresponding basic predictor is selected for prediction to obtain a predicted value of the heating node flow, which includes:

[0072] S61, performing specific site response extraction and weighted combination operations on the structural perturbation response representation matrix to construct a trend predictor, the calculation formula is:

[0073] ,

[0074] in, is the unit axis vector, is the predicted value of the heating node flow;

[0075] S62: Perform a dimension inversion operation on the structural disturbance response representation matrix to obtain a mirror inversion result of the structural response vector, perform response enhancement processing on the structural disturbance response representation matrix and the mirror inversion result, and construct a hybrid predictor. The calculation formula is:

[0076] ,

[0077] where {H}^{*}_{i}[j]={H}_{i}[d-j+1] is the mirror-inverted result of the structural response vector;

[0078] S63, performing boundary position response extraction and combination operations on the structural disturbance response representation matrix to construct a jump predictor, the calculation formula is:

[0079] ,

[0080] in, is the unit terminal axis vector;

[0081] S64. Set the preset structural response mode threshold group and specify the specific value of each threshold. According to the interval position of the kurtosis disturbance factor, select the corresponding basic predictor from the trend predictor, hybrid predictor and jump predictor to predict the flow of the heating node. .

[0082] Preferably, in step S7, the predicted value is compared and analyzed with the actual operation record of the heating network, and the degree of fit between the node operation status and the target heating demand is determined in combination with the historical reference benchmark, and the process of outputting the evaluation conclusion of the heating network operation efficiency includes:

[0083] S71. Predicted flow values ​​of each node in the matrix according to the structural disturbance response , combined with the heat supply node flow statistics in the historical time window, extract the minimum value of the heat supply node in the sliding historical window With the maximum value , construct the reasonable operation flow range of the node [{q}^{min}_{i},{q}^{max}_{i}] Serves as a reference boundary for evaluation;

[0084] S72, when there are 70% of the predicted values ​​in the current cycle In [{q}^{min}_{i},{q}^{max}_{i}] If the node is within the specified range and does not show a systematic deviation trend, it is judged that the node is operating in a stable state and its efficiency is normal;

[0085] S73, when there are 30% of the predicted values ​​in the current cycle Above the upper limit When , the node is evaluated to have heat redundancy or oversupply trend and marked as "efficiency overflow" state;

[0086] S74, when there are 30% of the predicted values ​​in the current cycle Below the lower limit When , the node is assessed to have insufficient response or undersupply risk and is marked as “efficiency leakage” state.

[0087] In the second aspect, the present invention provides a device for evaluating the operation efficiency of a heating network based on data prediction, including: a data acquisition module for collecting and preprocessing data such as flow, temperature, pressure, supply and return water temperature difference, pipe section flow rate and air temperature of a heating node, and constructing an original time series input matrix; a structural transformation module for introducing a coupling attention factor and a disturbance spectrum projection factor, constructing a multi-channel transformation kernel, performing structural transformation on the original data, and outputting a structural transformation result matrix; a graph modeling module for calculating structural feature indicators based on the structural transformation results, constructing a graph adjacency matrix, and reflecting the association relationship between nodes; a graph propagation module, Combining the disturbance amplitude, stability and dispersion factor, a disturbance-driven graph propagation mechanism is constructed to achieve feature enhancement and obtain a structural disturbance response representation; the structure recognition module introduces the kurtosis disturbance factor based on the disturbance response representation to identify the node structural behavior pattern; the prediction output module selects a trend, hybrid or jump predictor based on the recognition result and outputs the node flow prediction value; the efficiency evaluation module compares and analyzes the predicted value with the actual operation record of the heating pipeline network, combines the historical reference benchmark to judge the degree of fit between the node operation status and the target heating demand, and outputs the evaluation conclusion of the heating pipeline operation efficiency.

[0088] Compared with the prior art, the present invention has the following beneficial effects:

[0089] The present invention proposes a method for evaluating the operating efficiency of a heating network based on data prediction, aiming to achieve accurate prediction of heating node flow and efficiency analysis of operating status. The method first acquires and preprocesses historical operating data to construct an original time series input matrix. Then, a coupling attention factor and a disturbance spectrum distribution projection factor are introduced to construct a multi-channel transformation kernel and perform structural transformation to generate a structural transformation result matrix. Next, the structural response characteristics are calculated and a graph adjacency matrix is ​​constructed. The disturbance amplitude vector, stability factor and transition caustic factor are introduced to construct a disturbance-driven graph propagation mechanism, complete feature propagation and generate a structural disturbance response representation matrix. Then, by introducing a kurtosis disturbance factor, the structural behavior pattern is identified and the corresponding predictor is selected to predict the node flow. Finally, the prediction results are compared and analyzed with the actual operating records. Combined with the historical reference benchmark, the degree of fit between the node operating status and the target heating demand is judged, and an evaluation conclusion of the operating efficiency of the heating network is output. BRIEF DESCRIPTION OF THE DRAWINGS

[0090] Figure 1 It is a step-by-step diagram of a method for evaluating the operation efficiency of a heating network based on data prediction;

[0091] Figure 2 It is a schematic diagram of the structure of a heating network operation efficiency evaluation method based on data prediction;

[0092] Figure 3 It is the structural diagram of the structural transformation module;

[0093] Figure 4 It is a structural diagram of the graph modeling module;

[0094] Figure 5 This is the structural diagram of the graph propagation module;

[0095] Figure 6 It is the structural diagram of the structure recognition module;

[0096] Figure 7 It is the structure diagram of the prediction output module;

[0097] Figure 8 It is the structure diagram of the efficiency evaluation module;

[0098] Figure 9 It is a comparison chart between predicted value and actual flow;

[0099] Figure 10 It is the time evolution curve of the prediction error. DETAILED DESCRIPTION

[0100] The present invention proposes a method for evaluating the operating efficiency of a heating network based on data prediction, aiming to achieve accurate prediction of heating node flow and comprehensive analysis of operating efficiency. The method first acquires and preprocesses historical operating data to construct an original time series input matrix; then introduces a coupling attention factor and a disturbance spectrum distribution projection factor, constructs a multi-channel transformation kernel and performs structural transformation to generate a structural transformation result matrix; then, calculates the structural response characteristics and constructs a graph adjacency matrix, introduces a disturbance amplitude vector, a stability factor and a transition caustic factor, constructs a disturbance-driven graph propagation mechanism, completes feature propagation, and generates a structural disturbance response representation matrix; then, by introducing a kurtosis disturbance factor, identifies structural behavior patterns, and selects corresponding predictors for node flow prediction; finally, compares and analyzes the prediction results with the actual operating data to determine the degree of fit between the operating status of the heating node and the target demand, thereby completing the evaluation and analysis of the operating efficiency of the heating network.

[0101] Reference Figures 1 to 8The first embodiment of the present invention provides a method for evaluating the operating efficiency of a heating network based on data prediction. The method first collects and preprocesses the historical operating data of the heating network through a data acquisition module to construct an original time series matrix containing information such as flow, temperature, pressure, supply and return water temperature difference, flow rate and air temperature; then, the structure transformation module introduces coupling attention and disturbance spectrum factors, performs multi-channel transformation to extract high-order coupling and nonlinear interaction features between variables, and generates a structure transformation result matrix; on this basis, the graph modeling module calculates the coupling degree, gradient difference, staggered directionality and time synchronization between nodes to construct a graph adjacency matrix; the graph propagation module combines the disturbance amplitude, stability and caustic factor to perform feature propagation and enhancement to obtain a structural disturbance response representation; the structure identification module uses the kurtosis disturbance factor based on the response representation matrix to identify the node behavior type; the prediction output module selects the corresponding trend, hybrid or jump predictor to predict the flow according to the identification result; finally, the efficiency evaluation module compares and analyzes the predicted value with the actual operation record of the heating network, and judges the degree of fit between the node operation status and the target heating demand in combination with the historical reference benchmark, and outputs the evaluation conclusion of the heating network operation efficiency. The complete process is as follows: Figure 1 shown.

[0102] Construct a heating network operation efficiency evaluation model based on data prediction, the structure of which is as follows Figure 2 As shown, the specific steps are:

[0103] S1. Acquire and preprocess historical operating data in the heating network to construct an original time series input matrix. The original time series input matrix includes the flow, temperature, pressure, supply and return water temperature difference of the heating node, the flow rate of the heating pipe section, and the time series characteristic values ​​of the outside air temperature in a continuous time window.

[0104] Furthermore, in step S1, the operating data of each heating node in the heating system is collected, and the operating data includes the instantaneous flow value of the node, the supply water temperature, the return water temperature, the node pressure, the supply and return water temperature difference, the flow velocity of the corresponding pipe section, and the external air temperature data at the corresponding moment; according to the set time window length and sliding step size, the operating data are time-aligned and serialized to construct a multivariate time series data set; the time series data in the multivariate time series data set are pre-processed by filling missing values, eliminating outliers and normalization to obtain a pre-processed multivariate time series data set; based on the pre-processed multivariate time series data set, the original time series input matrix is ​​constructed in chronological order, and each row of the matrix corresponds to a multivariate joint observation value under a time step, which is used for subsequent structural modeling and predictive analysis.

[0105] It should be explained that the mathematical model of the original time series input matrix is ​​X={[{x}_{n,1},...,{x}_{n,j},...,{x}_{n,t}]}^{N\times T}\in {R}^{N\times T} ,in, is the original multivariate time series input matrix, For the The value of the input variable at the first time point, For the The input variable in The value at a time point, For the The input variable in The value at a time point, is the time window length, is the number of input variables.

[0106] Furthermore, the time window length and sliding step set in step S1 are set according to the response periodicity and prediction timeliness requirements of the operating state of the heating system, specifically including: selecting the time window length based on the delay characteristics and control cycle of heat transfer in the heating system, usually set to cover 1 to 3 typical thermal control cycles to ensure that the input data can fully reflect the dynamic change trend of the system. In this embodiment, the time window length is set to 96; the sliding step is determined according to the prediction update frequency and the system response speed, usually set to 1 to 5 sampling intervals to balance data redundancy and real-time requirements, and ensure the stability and generalization ability of the model on the continuous time axis. In this embodiment, the sliding step is set to 4; after setting the time window length and sliding step, the original operating data is serially sliced ​​according to the sliding window mechanism to generate a multivariate time series data set covering the target time period, which is used to construct the input matrix required for structural modeling.

[0107] S2. Introduce a coupling attention factor and a disturbance spectrum distribution projection factor to construct a multi-channel transformation kernel structure, perform a structural transformation operation on the original time series input matrix based on the multi-channel transformation kernel structure, and obtain a structural transformation result matrix.

[0108] Furthermore, the structure of the structure transformation module in step S2 is as follows: Figure 3 As shown, firstly, based on the original time series input matrix, the local neighborhood time series difference characteristics and global statistical characteristics of the variables are extracted, and the structural tension entropy function is constructed. According to the structural tension entropy function, the coupling attention factor is determined. , then construct the structural gradient term based on the original time series input matrix Then, based on the structural gradient term, the channel perturbation strategy is used to determine the perturbation spectrum distribution projection factor. According to the coupling attention factor and the perturbation spectrum distribution projection factor, a multi-channel heterogeneous transformation kernel structure is constructed. The calculation formula is: ,

[0109] in, For channel At the point The transformed kernel response on It is an oscillating frequency capture core. is a Gaussian smoothing filter; based on the multi-channel heterogeneous transformation kernel structure, a channel response mapping operation is performed on the original time series input matrix to obtain a structural transformation result matrix .

[0110] Furthermore, in step S2, based on the original time series input matrix, the local neighborhood time series difference characteristics and global statistical characteristics of the variables are extracted, and a structural tension entropy function is constructed. The process of determining the coupling attention factor based on the structural tension entropy function includes: based on the original time series input matrix, extracting the local neighborhood data of each variable at each time point, and calculating its local residual. The calculation formula is: ,

[0111] in, For the The input variable in The value at a time point, For A local neighborhood centered on , For variables In the neighborhood The local mean within , For the The variable in The local residual at each time point is the difference between the current value and the mean of its neighborhood, which is used to characterize short-term fluctuation behavior. The global mean and standard deviation of each input variable in the entire time window are used as important metrics to measure the stability and volatility of the input variable. The calculation formula is: ,

[0112] in, is the input variable Throughout the time window The global mean of For the input variables in the entire time window Based on the local residual and the standard deviation, the normalized mutation intensity of each input variable at each time point is calculated, and the calculation formula is: ,

[0113] in, For the The variable in The normalized mutation intensity of each time point is used to measure whether the variable has a significant change at that moment; the local residual value of each time point in the local residual is normalized to construct the local residual normalization probability, and the local entropy value is calculated based on the local residual normalization probability. The calculation formula is: , ,

[0114] in, Neighborhood The time index collection in the current time point All time indices in the local neighborhood centered at For the Input variables in the neighborhood The local residual normalized probability within , Neighborhood The local entropy value of the normalized probability of the local residual of the input variable can be used to measure the local structural complexity at that moment; the structural entropy deviation of each input variable is calculated in the time dimension to determine whether each input variable is in an atypical structural state. The calculation formula is: ,

[0115] in, For time point Input variables The structural entropy deviation of the input variable can be used to determine Whether it is in an atypical change state; combining the normalized mutation intensity and the structural entropy deviation, constructing a structural tension entropy function, the calculation formula is: ,

[0116] in, is the value of the structural tension entropy function, which is used to comprehensively characterize the dynamic significance and structural variation degree of each variable at the current moment. According to the calculation result of the structural tension entropy function, the original coupling attention factor is obtained by reverse mapping, and the factor is normalized to obtain the normalized coupling attention factor. The calculation formula is: , ,

[0117] in, is the original coupling attention factor, It is the normalized coupling attention factor, which is used for channel weight regulation in subsequent multi-channel structural transformation modeling.

[0118] S3. Based on the structural transformation result matrix, the response path co-track coupling degree, the structural gradient amplitude difference, the trajectory change staggered directionality and the response time offset synchronization are calculated to construct a graph adjacency matrix.

[0119] Furthermore, the structure diagram of the graph modeling module in step S3 is as follows: Figure 4 As shown, firstly, based on the structural transformation result matrix, the structural change trend characteristics of each variable in the time dimension are extracted, and the variable pair The response path common rail coupling degree between them is calculated as follows: ,

[0120] in, For variables At the time point The unit direction change rate of the variable The structural change trend characteristics in the time dimension, For variable pairs The response path common rail coupling degree, The closer the value of the measure is to 1, the better the variable With variables The more consistent the response direction is, the stronger the structural co-orbitality is. Then, according to the structural transformation result matrix, the structural gradient amplitude index of each variable in the time dimension is constructed, and the calculation formula is: ,

[0121] in, For variables The structural gradient amplitude index reflects the variable The average intensity of the response change per unit time over the entire time period, The larger the value, the more intense the variable response. The smaller the value, the flatter the variable response. Then, based on the structural gradient amplitude index and the response difference relationship between the variable pairs, the structural gradient amplitude modulation factor is determined. , the calculation formula is: ,

[0122] in, is the structural gradient amplitude modulation factor, which can measure the variable With variables Based on the structural change trend characteristics of each variable in the time dimension, it is judged whether there is frequent staggering in the response direction of the variable in the time dimension, and the trajectory change staggering directionality index is constructed. The calculation formula is: ,

[0123] in, For variable pairs The trajectory changes staggered direction index, when When it approaches 0, it means and The response directions of the two are consistent most of the time. When it approaches 1, it indicates and The two directions are frequently opposite, and the structures are significantly intertwined. Based on the structural change trend characteristics, the response time offset degree of the variable pair in the time structure is evaluated, and the response time offset synchronization index is constructed. Based on the response path co-track coupling degree, the structural gradient amplitude modulation factor, the trajectory change interleaving directionality index and the response time offset synchronization, the final graph adjacency matrix is ​​constructed, and the calculation formula is: ,

[0124] in, For variable pairs The adjacency matrix of 、 、 Construct hyperparameters that influence weights for controlling graph structure.

[0125] Furthermore, the mathematical model of the structural change trend characteristics of each variable in the time dimension in step S3 is: ,

[0126] in, For variables At the time point The unit directional rate of change, For variables The difference in unit direction change rate between two adjacent time points, For variables The structural change strength modulus length.

[0127] It should be noted that the process of evaluating the response time offset degree of the variable pair in the time structure based on the structural change trend characteristics and constructing the response time offset synchronization index in step S3 includes: calculating the response center value of each variable in the time dimension based on the structural change trend characteristics. The response center value represents the weighted distribution center of the variable response intensity in the time window, and the calculation formula is: ,

[0128] in, For variables At the time point The structural transformation result at Representing variables The main response time center, that is, the response time centroid, emphasizes the time period in which the variable response energy is most concentrated. The greater the response intensity at the location, the greater the impact on the response time centroid, and the more it can reflect its activated temporal position. The closer the response centroid times of the variables are, the more consistent the "potential phases" stimulated by the disturbance are, and the stronger the structural dependence is. Based on the main response time centers of the two variables, the response time offset synchronization index is constructed to reflect the degree of coordination of the two variables in the time structure. The calculation formula is: ,

[0129] in, For variable pairs The response time offset synchronization index reflects the degree of coordinated response of the variables in the time structure. When it approaches 1, the variable With variables The main response time center moment is almost the same when When it approaches 0, it indicates that the variable With variables The response shift between them is quite significant.

[0130] It should be explained that the response time offset synchronization index in step S3 is used to identify whether the variables are activated synchronously on the time axis. If the variables have common response peaks in similar time windows, it means that they may be driven by the same disturbance or transmitted by the same structural path. This synchronization reflects a kind of "potential phase consistency within the structure", which can be used as a dependency enhancement item in graph modeling and can be used as a multiplicative enhancement item in the subsequent adjacency matrix construction to improve the accuracy perception ability of structural coupling.

[0131] Furthermore, the control graph structure in step S3 constructs the hyperparameters that affect the weights 、 、 The design process includes: The response path common rail coupling degree and response time offset synchronization index are used to construct the synchronization bias potential energy ratio, which is calculated as follows: ,

[0132] in, is the synchronous bias potential energy ratio, It is a very small positive number, so as to avoid the denominator being zero, we usually take ; Based on variable pairs The trajectory change interlaced directional index and the structural gradient amplitude modulation factor are used to construct the directional conflict potential energy ratio, which is calculated as follows: ,

[0133] in, is the directional conflict potential energy ratio; based on the variable The degree of cooperative coupling and the structural perturbation characteristics are used to construct the coupling consistency potential energy ratio, which is calculated as follows: ,

[0134] in, is the coupling consistency potential energy ratio; based on the synchronization bias potential energy ratio , all variables are The absolute value of the corresponding value of the target set , and arrange the set in ascending order to obtain the absolute value sequence of synchronous bias potential energy. The calculation formula is: ,

[0135] in, is the absolute value sequence of the synchronous bias potential energy in ascending order. Item; when the number of input variables When it is an odd number, the synchronization control factor for The item , when the number of input variables When it is an even number, the synchronization control factor for The Item and The average value of the item; the operations in step S364 to step S365 are performed on the direction conflict potential energy ratio and the coupling consistency potential energy ratio respectively, to obtain the consistency control factor Directional regulatory factors Based on the synchronization control factor, the consistency control factor and the directionality control factor, combined with the response path co-track coupling degree, the structural gradient amplitude modulation factor, the trajectory change interlaced directionality index and the response time offset synchronization, a graph adjacency matrix is ​​constructed to obtain each variable pair The connection strength is calculated as follows: ,

[0136] in, For variable pairs The adjacency matrix of .

[0137] S4. Introduce the disturbance amplitude vector, response stability factor and transition dispersion factor, combine them with the graph adjacency matrix, construct a disturbance-driven graph propagation modulation mechanism, use the disturbance-driven graph propagation modulation mechanism to perform feature propagation processing on the structural transformation result matrix, and obtain the structural disturbance response representation matrix.

[0138] Furthermore, the structure diagram of the graph modeling module in step S4 is as follows: Figure 5As shown, first, feature extraction processing is performed on the input response sequence of each node in the structural transformation result matrix to determine the disturbance amplitude vector, amplitude analysis processing is performed on the node disturbance amplitude vector, a response stability factor is constructed, structural fluctuation feature processing is performed on the response sequence of each node connection pair in the graph adjacency matrix, a transition caustic index is constructed, matrix fusion calculation is performed based on the disturbance amplitude vector, the response stability factor and the transition caustic index, a disturbance-driven graph propagation modulation mechanism is constructed, and then, according to the graph propagation modulation mechanism, disturbance response association processing is performed on the structural transformation result matrix to construct a difference propagation term , the calculation formula is: ,

[0139] in, For the The node in The structural perturbation response of the layer is expressed as follows: is the difference propagation term from the neighbors after modulation, reflecting the “disturbance diffusion” on the structure; the disturbance auto-amplitude difference function is constructed based on the difference propagation term, and the calculation formula is: ,

[0140] in, is the disturbance amplitude difference function. When the disturbance amplitude is small, it is approximately a linear response. When the disturbance amplitude is large, it automatically suppresses the change. is the input vector of the perturbation amplitude difference function, i.e., the difference propagation term constructed in step S42, The square norm is obtained by accumulating the squares of the difference propagation terms in all hidden layers. The residual retention factor is introduced and combined with the disturbance amplitude difference function to construct a disturbance-driven response feature mapping mechanism. Based on the response feature mapping mechanism, the structural state of the current node is nonlinearly modulated and dynamically updated to obtain a structural disturbance response representation matrix composed of the structural disturbance response tensors of all nodes. The calculation formula is: ,

[0141] in, is the disturbance amplitude difference function, is the residual retention factor, which can control the current layer feature retention ratio. For the The node in The structural perturbation response of the layer.

[0142] Furthermore, in step S4, the disturbance amplitude vector, the disturbance difference adjustment index and the transition caustic factor are introduced, and the process of constructing the disturbance-driven graph propagation modulation mechanism in combination with the graph adjacency matrix includes: performing feature extraction processing on the input response sequence of each variable in the structural transformation result matrix to determine the disturbance amplitude vector, and the calculation formula is: ,

[0143] in, For variables At the time point The structural transformation result at is the average structural response, For nodes The total disturbance amplitude reflects the degree of deviation of its response from the mean value and can be regarded as "response memory". The amplitude analysis process is performed on the disturbance amplitude vector to construct the response stability factor. The calculation formula is: ,

[0144] in, For variables The response stability factor in the structural transformation channel measures the stability of its frequency disturbance. The larger the value, the more stable the structural response and the smaller the disturbance. Based on the response stability factor, a disturbance difference adjustment index is constructed, and the calculation formula is: ,

[0145] in, It is a symmetric regulation index composed of the stability of variable responses. When both variables are in a highly stable state, the larger the regulation factor value, the more effective it is in suppressing the unnecessary spread of disturbance differences and preventing the imbalance of propagation due to “over-suppression” or “over-amplification”. The structural fluctuation feature processing is performed on the response sequence of each variable connection pair in the graph adjacency matrix to construct the transition caustic index. The calculation formula is: ,

[0146] in, is the transition caustic index, when the variable With variables When the same direction or opposite direction is strong, the index value is higher. Directional cooperation or conflict both indicate important structural features, and the propagation intensity should be increased. Based on the perturbation amplitude vector, the response stability factor, and the transition caustic index, a matrix fusion calculation is performed to construct a perturbation-driven graph propagation modulation mechanism. The calculation formula is: ,

[0147] in, For variable pairs The adjacency matrix of is the difference in the magnitude vector of the variable perturbation, {\nu}_{ij}\in [0,{A}_{ij}] is the transition modulation weight matrix, i.e., the perturbation-driven graph propagation modulation mechanism.

[0148] Furthermore, the design process of the residual preservation factor in step S4 includes: constructing a perturbation symmetry ratio of the variable according to the transition modulation weight matrix, and the calculation formula is: ,

[0149] in, is the perturbation symmetry ratio, is the transition modulation weight matrix, It is a very small positive number, so as to avoid the denominator being zero, we usually take Based on the degree of symmetric deviation of the disturbance symmetry ratio, the residual preservation factor is defined, and the calculation formula is: ,

[0150] in, is the residual preservation factor, which is used to control the degree of preservation of the residual of the previous layer in the next layer. , indicating that all nodes have balanced input and output disturbances, then Approaching 1, the residual is highly preserved. If the overall disturbance is seriously unbalanced, then Automatically shrink to prevent error status transmission.

[0151] It should be noted that by performing the operations in step S3 to step S4 on the structural transformation result matrix of all input variables, the structural disturbance response representation matrix of all input variables can be obtained. The mathematical model is: H={[{H}^{(k+1)}_{1},...,{H}^{(k+1)}_{N}]}^{T}\in {R}^{N\times d} ,in, is the number of input variables, is the dimension of the perturbation response tensor, that is, the dimension of the structural channel generated after each variable is propagated in the structural graph, For the Variables are updated in graph reasoning The output of the layer, The matrix represents the structural perturbation response for all input variables.

[0152] It should be explained that in step S4, according to the graph propagation modulation mechanism, the disturbance response association processing is performed on the structural transformation result matrix, the difference propagation term is constructed, the disturbance self-variation amplitude difference function is constructed according to the difference propagation term, the residual retention factor is introduced, and the disturbance self-variation amplitude difference function is combined to construct a disturbance-driven response characteristic mapping mechanism. Based on the response characteristic mapping mechanism, the structural state of the current node is nonlinearly modulated and dynamically updated. The detailed process of obtaining the structural disturbance response representation is as follows: first, the initial structural state tensor of the 0th layer is set to the structural transformation result itself, and the mathematical model is: , ,in, , Respectively The first and The structural transformation result of each node is the initial embedded state of the graph propagation. According to the transition modulation weight matrix, the difference between each node and its neighbors is summed to construct the difference propagation term of the 0th layer. The calculation formula is: , based on the difference propagation term at layer 0 Construct a nonlinear suppression function to adaptively adjust the disturbance propagation intensity: ,in, is a regulating factor used to dynamically control the nonlinearity of the response; combined with the residual retention factor , update the structural status of layer 1: , which means that the structural response of the current node is injected with structural perturbation difference modulation information on the basis of maintaining the residual feature ratio to form the initial perturbation response feature; for subsequent layers, the following steps are repeated:

[0153] 1. Calculation of difference propagation term: ,

[0154] 2. Auto-amplitude difference function modulation: ,

[0155] 3. Feature updates: ,

[0156] Therefore, in actual implementation, the perturbation-driven graph propagation modulation mechanism does not constitute a closed-loop dependency structure, but directly provides the initial structural state tensor from the structural transformation result. , and taking this as a starting point, by constructing the difference propagation term, the self-variable amplitude difference function and the disturbance modulation update process layer by layer, the generation of the structural disturbance response representation is gradually completed, avoiding the problem of broken model structure dependency chain or self-reference, and ensuring the interpretability and coherence of the disturbance response mechanism in the entire graph.

[0157] S5. Introduce a kurtosis disturbance factor, construct a multi-mode response structure discrimination mechanism, perform structural behavior identification on the structural disturbance response representation matrix based on the discrimination mechanism, and obtain a structural behavior identification result.

[0158] Furthermore, the structure diagram of the structure recognition module in step S5 is as follows: Figure 6 As shown, first extract the amplitude jump vector of the structural response representation matrix and construct the structural perturbation difference The calculation formula is: {J}_{i}=[\left | {{H}_{i,2}-{H}_{i,1}} \right |,\left | {{H}_{i,3}-{H}_{i,2}} \right |,...,\left | {{H}_{i,d}-{H}_{i,d-1}} \right |]\in {R}^{d-1} ,

[0159] Among them, {J}_{i}[j] For input variables The corresponding The disturbance difference of the bit reflects the jump rate of the structural disturbance sequence. For input variables In the structural response The dimensional component, For input variables In the Peace and The amplitude jump between dimensions is the amplitude difference of the structural response. This difference constitutes Each item in reflects the jump / mutation degree of the response signal in the channel dimension; the change trend extraction operation is performed on the structural perturbation difference, and a structural curvature characterization method is constructed. Based on the structural curvature characterization method, the overall structural curvature intensity is obtained. , the calculation formula is: ,

[0160] in, For input variables The corresponding overall structural curvature intensity represents the nonlinear complexity of the disturbance trend; a mutation concentration evaluation operation is performed on the structural disturbance difference, and a multi-mode response structure discrimination mechanism is constructed in combination with the overall structural curvature intensity. The kurtosis disturbance factor is determined based on the multi-mode response structure discrimination mechanism, and the calculation formula is: ,

[0161] in, For input variables The maximum amplitude mutation in the structural channel dimension, For input variables The average jump level in the structural channel dimension, that is, the background perturbation intensity, For input variables The corresponding kurtosis disturbance factor; when , indicating the input variable It is relatively stable and belongs to trend type variable. , indicating the input variable There is a slight disturbance but no sudden jump, which is a mixed variable. , indicating the input variable There is a significant boundary jump, which is a jump type variable.

[0162] S6. According to the structural behavior identification result, a corresponding basic predictor is selected for prediction to obtain a predicted value of the heating node flow, wherein the basic predictor includes a trend predictor, a hybrid predictor, and a jump predictor.

[0163] Furthermore, the structure diagram of the prediction output module in step S6 is as follows: Figure 7 As shown, first, the specific site response extraction and weighted combination operations are performed on the structural perturbation response representation matrix to construct a trend predictor. The calculation formula is: ,

[0164] in, is the unit axis vector, is the predicted value of the heating node flow; the dimension inversion operation is performed on the structural disturbance response representation matrix to obtain the mirror inversion result of the structural response vector, the structural disturbance response representation matrix and the mirror inversion result are subjected to response enhancement processing to construct a hybrid predictor, and the calculation formula is: ,

[0165] where {H}^{*}_{i}[j]={H}_{i}[d-j+1] is the mirror-inverted result of the structural response vector; the boundary position response extraction and combination operations are performed on the structural disturbance response representation matrix to construct a jump predictor, and the calculation formula is: ,

[0166] in, is the unit terminal axis vector; a preset structural response mode threshold group is set, and the specific value of each threshold is specified. According to the interval position of the kurtosis disturbance factor, the corresponding basic predictor is selected from the trend predictor, hybrid predictor and jump predictor for prediction to obtain the predicted value of the heating node flow.

[0167] It should be explained that in step S64, the structural response mode threshold group includes the dividing point between trend type variables and mixed type variables. and the dividing point between mixed variables and jump variables , the dividing point between the trend variable and the mixed variable The value range is , in this embodiment, the value is 1.2, the dividing point between the mixed variable and the jump variable The value range is , in this embodiment, the value is 2.5. When , it means that the structural response mode of the input variable is stable and the disturbance is gentle, which is a trend variable, and the trend predictor is selected; when When , it means that the input variable has some disturbance but no sharp jump occurs, which is a mixed variable, and the mixed predictor is selected; when When , it indicates that the input variable has significant mutation or boundary activation behavior, which is a jump type variable, and the jump predictor is selected; the structural response mode threshold value group in step S64 、 It is an empirical setting and can be fine-tuned according to the distribution characteristics of the structural disturbance response data in the actual system, but it must meet the following requirements: reasonably separate the three structural response modes of trend, mixing and jump; ensure the sensitivity and stability of the predictor selection; avoid misclassification of boundary disturbance variables and improve the robustness of the prediction model.

[0168] Furthermore, the contents of steps S5 to S7 are executed on the structural disturbance response representation matrix of all input variables to obtain the predicted flow value matrix of each node. The mathematical model is: ,in, is the predicted flow value matrix of each node, (Right now ) indicates the The predicted value of each node.

[0169] S7. Compare and analyze the predicted value with the actual operation record of the heating network, combine it with the historical reference benchmark, judge the degree of fit between the node operation status and the target heating demand, and output the evaluation conclusion of the heating network operation efficiency.

[0170] Furthermore, in step S7, the predicted value is compared with the actual operation record of the heating network, and the degree of fit between the node operation status and the target heating demand is judged in combination with the historical reference benchmark, and the evaluation conclusion of the heating network operation efficiency is output. The process includes: the predicted flow value of each node in the structural disturbance response representation matrix is ​​expressed as follows: , combined with the heat supply node flow statistics in the historical time window, extract the minimum value of the heat supply node in the sliding historical window With the maximum value , construct the reasonable operation flow range of the node [{q}^{min}_{i},{q}^{max}_{i}] As a reference boundary for evaluation; when there are 70% of the predicted values ​​in the current period In [{q}^{min}_{i},{q}^{max}_{i}] If the node is stable and has no systematic deviation trend, the node is judged to be in a stable state and has normal efficiency. Above the upper limit When the node is evaluated to have heat redundancy or oversupply trend, it is marked as "efficiency overflow" state; when there are 30% of the predicted values ​​in the current cycle Below the lower limit When , the node is assessed to have insufficient response or undersupply risk and is marked as “efficiency leakage” state.

[0171] It should be noted that the algorithm implementation and testing of this embodiment are carried out on a portable computer platform equipped with Windows 11 Home Chinese Edition (version number 24H2) operating system, AMD Ryzen 9 7945HX processor, 16GB memory, NVIDIA GeForce RTX 4060 graphics card, and the operating environment is MATLAB 2023b. In the specific implementation process, it is divided into multiple modules according to function, and the orderly calling of each step is realized through the main program main.m: the feature extraction and structural transformation functions related to steps S1 to S2 are integrated in the structure_transform.m module, the graph construction and disturbance propagation modeling functions of steps S3 to S4 are encapsulated in the graph_construct.m module, the structural behavior discrimination strategy of step S5 is implemented by structure_identify.m, and the predictor execution and efficiency evaluation of steps S6 and S7 are completed by the predictor.m module. The configuration of related factors such as the time window length of 96, the sliding step size of 4, the number of propagation layers set in step S4 of 5, and the prediction deviation evaluation parameters in step S7 are all stored in the config.m configuration file. The code is read by main.m at runtime and the functions of each module are called in turn to complete the structural modeling and operation efficiency evaluation process of the heating pipe network flow data; further, the comparison chart between the predicted value and the actual flow in the heating pipe network operation efficiency evaluation method based on data prediction is as follows: Figure 9 As shown in the figure, the solid line represents the actual flow of a node in the pipe network, and the dotted line represents the flow prediction value obtained based on the prediction module of the present invention. It can be observed from the figure that the overall trend of the prediction curve is basically consistent with the actual flow, which can more accurately reflect the time series response characteristics of the heating node, indicating that the designed structure-aware prediction method has good fitting ability and generalization stability, providing a high-quality data basis for subsequent operation efficiency analysis; the prediction error evolution diagram is shown in Figure 10 As shown in the figure, the changing trend of the prediction error of the node in continuous time steps is shown. It can be seen from the figure that the error fluctuation is generally within a reasonable range, but there are sudden increases in some sections, indicating that there is a deviation between the operating status of the heating system and the predicted behavior during this period. This figure can be used as a key reference in the process of operating efficiency evaluation to assist in judging whether the system has potential risks such as supply and demand mismatch, dynamic load anomaly, etc.

[0172] The above are only preferred embodiments of the present invention. It should be pointed out that those skilled in the art can make several modifications and improvements without departing from the inventive concept of the present invention, and these all fall within the scope of protection of the present invention.

Claims

1. A method for evaluating the operation efficiency of a heating network based on data prediction, characterized in that: include: S1. Acquire and preprocess historical operation data in the heating network to construct an original time series input matrix. The original time series input matrix includes the flow rate, temperature, pressure, supply and return water temperature difference of the heating node, the flow rate of the heating pipe section, and the time series characteristic values ​​of the outside temperature in a continuous time window, specifically including: S11. Collecting operating data of each heating node in the heating system, the operating data including the instantaneous flow value, supply water temperature, return water temperature, node pressure, supply and return water temperature difference, flow velocity of the corresponding pipe section, and external temperature data at the corresponding time; S12. Performing time alignment and serialization processing on the operating data according to the set time window length and sliding step size to construct a multivariate time series dataset; S13, performing missing value filling, outlier removal, and normalization preprocessing on the time series data in the multivariate time series dataset to obtain a preprocessed multivariate time series dataset; S14. Based on the pre-processed multivariate time series data set, construct an original time series input matrix in chronological order. The mathematical model is: , in, is the original multivariate time series input matrix, where each row of the matrix corresponds to a multivariate joint observation at a time step, For the The value of the input variable at the first time point, For the The input variable in The value at a time point, For the The input variable in The value at each time point; is the time window length, is the number of input variables; S2. Introducing a coupling attention factor and a disturbance spectrum distribution projection factor to construct a multi-channel transformation kernel structure, performing a structural transformation operation on the original time series input matrix based on the multi-channel transformation kernel structure to obtain a structural transformation result matrix, specifically including: S21, based on the original time series input matrix, extract the local neighborhood time series difference characteristics and global statistical characteristics of the variable, construct a structural tension entropy function, and determine the coupling attention factor according to the structural tension entropy function ; S22. Construct a structural gradient term based on the original time series input matrix. The calculation formula is: , in, For As the center, is the local neighborhood of the neighborhood radius, For input variables In the neighborhood The local mean within , is the structural gradient, indicating the magnitude of the deviation from its local structure; S23. Determine the perturbation spectrum distribution projection factor based on the structural gradient term using a channel perturbation strategy. The calculation formula is: , in, is the time window length, is the number of input variables, is the channel index, For channel The constructed sinusoidal modulation spectrum basis, For channel The projection factor of the perturbation spectrum distribution; S24. Construct a multi-channel heterogeneous transformation kernel structure based on the coupling attention factor and the disturbance spectrum distribution projection factor. The calculation formula is: , in, For channel At the point The transformed kernel response on It is an oscillating frequency capture core. ; S25, performing a channel response mapping operation on the original time series input matrix based on the multi-channel heterogeneous transformation kernel structure to obtain a structure transformation result matrix ; S3. Based on the structural transformation result matrix, the response path common track coupling degree, structural gradient amplitude difference, trajectory change staggered directionality and response time offset synchronization are calculated to construct a graph adjacency matrix, specifically including: S31, based on the structural transformation result matrix, extract the structural change trend characteristics of each variable in the time dimension, and calculate the variable pair The response path common rail coupling degree between them is calculated as follows: , in, For variables At the time point The unit direction rate of change, that is, the variable The structural change trend characteristics in the time dimension, For variable pairs The response path common rail coupling degree, The closer the value of the measure is to 1, the better the variable With variables The more consistent the response direction is, the stronger the structural common rail property is; S32. According to the structural transformation result matrix, construct the structural gradient amplitude index of each variable in the time dimension, and the calculation formula is: , in, For variables The structural gradient amplitude index reflects the variable The average intensity of the response change per unit time over the entire time period, The larger the value, the more intense the variable response. The smaller the value, the smoother the variable response; S33, based on the structural gradient amplitude index, combined with the response difference relationship between the variable pairs, determine the structural gradient amplitude modulation factor ; S34. Based on the structural change trend characteristics of each variable in the time dimension, determine whether there is frequent intersection in the response direction of the variable pair in the time dimension, and construct a trajectory change intersection directionality index. The calculation formula is: ， in, For variable pairs The trajectory changes staggered direction index, when When it approaches 0, it means and The response directions of the two are consistent most of the time. When it approaches 1, it indicates and The two directions are frequently opposite, and the structures are significantly intertwined; S35. Based on the structural change trend characteristics, evaluate the degree of response time offset of the variable on the time structure and construct a response time offset synchronization index. ; S36. Based on the response path common track coupling degree, the structural gradient amplitude modulation factor, the trajectory change interlaced directionality index and the response time offset synchronization, a final graph adjacency matrix is ​​constructed. The calculation formula is: ， in, For variable pairs The adjacency matrix of 、 、 Construct hyperparameters that influence weights to control graph structure; S4. Introducing the disturbance amplitude vector, the disturbance difference adjustment index, and the transition caustic factor, and combining them with the graph adjacency matrix, constructing a disturbance-driven graph propagation modulation mechanism, and using the disturbance-driven graph propagation modulation mechanism to perform feature propagation processing on the structural transformation result matrix to obtain a structural disturbance response representation matrix, specifically including: S41. Introducing the perturbation amplitude vector, the perturbation difference adjustment index, and the transition caustic factor, combined with the graph adjacency matrix, to construct a perturbation-driven graph propagation modulation mechanism, specifically including: S411, performing feature extraction processing on the input response sequence of each node in the structural transformation result matrix to determine the disturbance amplitude vector; S412, performing amplitude analysis processing on the disturbance amplitude vector to construct a response stability factor; S413. Constructing a disturbance difference adjustment index based on the response stability factor; S414, performing structural fluctuation feature processing on the response sequence of each variable connection pair in the graph adjacency matrix to construct a transition caustic index; S415: performing matrix fusion calculation based on the disturbance amplitude vector, the disturbance difference adjustment index, and the transition caustics index to construct a disturbance-driven graph propagation modulation mechanism; S42, according to the graph propagation modulation mechanism, performing disturbance response association processing on the structure transformation result matrix to construct the difference propagation term ; S43, constructing a disturbance auto-amplitude difference function according to the difference propagation term, and the calculation formula is: ， in, The disturbance amplitude difference function is approximately linear when the disturbance amplitude is small, and automatically suppresses changes when the disturbance amplitude is large. is the input vector of the perturbation auto-amplitude difference function, i.e., the difference propagation term constructed in step S42, The square norm is obtained by accumulating the squares of the difference propagation terms in all hidden layers; S44. Introducing a residual retention factor and combining it with the disturbance auto-variation difference function to construct a disturbance-driven response characteristic mapping mechanism, performing nonlinear modulation and dynamic update on the structural state of the current node based on the response characteristic mapping mechanism, and obtaining a structural disturbance response representation matrix composed of structural disturbance response tensors of all nodes; S5. Introducing a kurtosis disturbance factor, constructing a multi-mode response structure discrimination mechanism, and performing structural behavior identification on the structural disturbance response representation matrix based on the discrimination mechanism to obtain a structural behavior identification result, specifically including: S51, extracting the amplitude jump vector of the structural disturbance response representation matrix and constructing the structural disturbance difference ; S52: performing a change trend extraction operation on the structural disturbance difference, constructing a structural curvature characterization method, and obtaining the overall structural curvature strength based on the structural curvature characterization method. ; S53, performing a mutation concentration evaluation operation on the structural disturbance difference, building a multi-mode response structure discrimination mechanism in combination with the overall structural curvature strength, and determining a kurtosis disturbance factor based on the multi-mode response structure discrimination mechanism, the calculation formula is: , in, For input variables The corresponding overall structural curvature strength, For input variables The maximum amplitude mutation in the structural channel dimension, For input variables The average jump level in the structural channel dimension, that is, the background perturbation intensity, For input variables The corresponding kurtosis disturbance factor; S54, when , indicating the input variable It is relatively stable and belongs to trend type variable. , indicating the input variable There is a slight disturbance but no sudden jump, which is a mixed variable. , indicating the input variable There is a significant boundary jump, which is a jump variable; S6. Select a corresponding basic predictor for prediction based on the structural behavior identification result to obtain a predicted value of the heating node flow, wherein the basic predictor includes a trend predictor, a hybrid predictor, and a jump predictor; S7. Compare and analyze the predicted value with the actual operation record of the heating network, combine it with the historical reference benchmark, judge the degree of fit between the node operation status and the target heating demand, and output the evaluation conclusion of the heating network operation efficiency.

2. A method for evaluating the operation efficiency of a heating network based on data prediction according to claim 1, characterized in that: According to the structural behavior identification results, the corresponding basic predictor is selected for prediction to obtain the predicted value of the heating node flow, which includes: S61, performing specific site response extraction and weighted combination operations on the structural perturbation response representation matrix to construct a trend predictor, the calculation formula is: , in, is the unit axis vector, is the predicted value of the heating node flow; S62: Perform a dimension inversion operation on the structural disturbance response representation matrix to obtain a mirror inversion result of the structural response vector, perform response enhancement processing on the structural disturbance response representation matrix and the mirror inversion result, and construct a hybrid predictor. The calculation formula is: , in, is the mirror-inverted result of the structural response vector; S63, performing boundary position response extraction and combination operations on the structural disturbance response representation matrix to construct a jump predictor, the calculation formula is: , in, is the unit terminal axis vector; S64. Set a preset structural response mode threshold group and specify the specific value of each threshold. According to the interval position of the kurtosis disturbance factor, select the corresponding basic predictor from the trend predictor, hybrid predictor and jump predictor to perform prediction, and obtain the predicted value of the heating node flow.

Images