Energy heat supply pipe network flow adjusting method based on machine learning

Abstract

The invention discloses an energy heat supply pipe network flow adjusting method based on machine learning. The method comprises the steps that S1, a standardized dynamic feature data set is obtained; s2, constructing a Fourier neural operator load prediction model by using the standardized dynamic feature data set, completing training to obtain a node thermal load prediction model, and outputting a node thermal load prediction result in real time; s3, generating a local load mapping matrix according to a node thermal load prediction result; s4, constructing a multi-objective optimization function according to the local load mapping matrix; s5, initializing an improved starvation game search algorithm, and searching to obtain an optimal flow distribution scheme; s6, real-time flow self-adaptive adjustment of each node is completed; and S7, dynamic optimization control over the local flow of the micro-grid energy heat supply pipe network is achieved. The intelligent level of flow regulation of the micro-grid energy heat supply pipe network in the multi-node dynamic load environment is remarkably improved.

Description

Technical Field

[0001] The present invention relates to the field of energy heating technology, and in particular to a method for regulating flow in an energy heating pipe network based on machine learning. Background Art

[0002] With the rapid development of distributed energy systems and the energy internet, microgrid technology has garnered widespread attention for improving local energy efficiency, ensuring energy supply security, and enhancing control flexibility. As a crucial component of microgrids, microgrid energy heating systems play a key role in winter heating, efficient waste heat utilization, and industrial waste energy recovery. Given the operational characteristics of heating networks, local flow regulation strategies have become a key control element, impacting system stability and energy efficiency.

[0003] Existing technologies generally rely on time series or statistical models, such as ARIMA and support vector regression, to predict the load of microgrid heating networks. These models are prone to large prediction errors when dealing with nonlinear and sudden heat load demands, and are unable to accurately characterize high-frequency dynamic disturbances and multi-node coupling mechanisms. Furthermore, existing optimization methods, often based on traditional numerical optimization algorithms or standard intelligent algorithms, possess certain global search capabilities. However, they suffer from slow computational speeds, unstable convergence, or coarse adjustment granularity when dealing with complex coupling constraints, multi-objective decision-making, and real-time feedback responses. These methods struggle to meet the high-precision, robustness, and real-time requirements of microgrid systems.

[0004] In addition, during actual operation, due to the large number of pipeline nodes, uneven load distribution, and significant differences in response characteristics, traditional algorithms are usually unable to dynamically adjust the optimization process based on forecast information. They ignore the sensitive impact of load forecast errors on the quality of optimization decisions and lack an adaptive correction mechanism, which leads to deviations between the optimization adjustment results and actual operation, resulting in adjustment lag and local over-adjustment problems. At the same time, the current flow regulation strategy lacks fine-grained linkage design with the actual execution unit in terms of flow instruction generation and control implementation. It fails to reflect load priority, adjustment amplitude constraints and safety redundancy control at the control instruction level, affecting the execution reliability and adjustment response speed of the overall system.

[0005] In summary, it is urgent to propose a new flow regulation method that integrates prediction and optimization for high dynamic load response and adaptive decision-making to meet the high intelligence and high responsiveness operation requirements of modern microgrid heating systems. Summary of the Invention

[0006] One purpose of the present invention is to propose a flow regulation method for an energy heating pipeline network based on machine learning. The present invention significantly improves the intelligent level of flow regulation of a microgrid energy heating pipeline network in a multi-node dynamic load environment.

[0007] According to an embodiment of the present invention, a method for regulating flow in an energy heating network based on machine learning includes the following steps: S1. Collecting real-time operating parameters of each node in the microgrid energy heating system, integrating the real-time operating parameters into a multi-node dynamic feature dataset, and preprocessing the multi-node dynamic feature dataset to obtain a standardized dynamic feature dataset; S2. Build a Fourier neural operator load forecasting model using a standardized dynamic feature dataset and complete training to obtain a node thermal load forecasting model. This node thermal load forecasting model is deployed in the microgrid control center and outputs node thermal load forecast results in real time based on the latest standardized dynamic feature dataset. S3. Generate a local load mapping matrix based on the node heat load forecast results. The local load mapping matrix is ​​used to quantify the heat load demand of each node within the short-term forecast period. S4. Construct a multi-objective optimization function based on the local load mapping matrix, including energy efficiency, heat balance, and regulation response speed, and set corresponding constraints. S5. Using the multi-objective optimization function as the evaluation criterion, initialize the population parameters, feasible flow solution space, and iterative control parameters of the improved Hunger Game search algorithm, execute the iterative calculation process of the improved Hunger Game search algorithm, and use the node heat load prediction results and local load mapping matrix to evaluate individual fitness during the iteration process to search for the optimal flow allocation solution; S6. Convert the optimal flow distribution plan into a flow regulation instruction and send it to the heating network execution unit to complete the real-time flow adaptive regulation of each node; S7. Continuously monitor the operating status of the pipeline network after adaptive flow adjustment to achieve dynamic optimization control of the local flow of the microgrid energy heating pipeline network.

[0008] Optionally, the real-time operating parameters include node temperature, node flow rate, node pressure and historical heat load change records, and the preprocessing includes denoising processing, feature scaling processing and dimensional unification processing.

[0009] Optionally, S2 includes the following steps: S21. Based on standardized dynamic feature dataset Get the frequency domain characteristic components : ; in, represents the standardized dynamic feature dataset, Represents a single sample input vector at time t, which contains the set of operating status parameters of all nodes at time t. Indicates the node temperature of each node at time t, reflecting the heat transfer status of the unit node in the microgrid heating system. It represents the node flow velocity of each node at time t, which is the rate at which the heat carrier flows through the node per unit time. Indicates the node pressure of each node at time t, which is used to measure the pressure state of the heating medium at the node. Represents the historical heat load change record at time t, which is the extracted value of the historical time series heat load state of each node. Represents the total number of samples in the standardized dynamic feature data set; the standardized dynamic feature data set is decomposed into frequency domain features using the Fourier transform operator. For the moment Time The frequency domain coefficients of the node real-time operation parameters corresponding to the Fourier frequency components; S22. Calculate frequency domain sensitivity factor , evaluate the contribution of different Fourier frequency components to the node heat load prediction results: ; in, Indicates the The frequency domain sensitivity factor of the Fourier frequency component to the node heat load prediction result, Indicates that at the prediction time The following only uses Nodal heat load prediction results calculated using Fourier frequency components; S23. Based on frequency domain sensitivity factor Generate a frequency-domain adaptive cropping mask: ; in, For the The frequency-domain adaptive clipping mask value of the Fourier frequency components, The frequency domain adaptive clipping threshold is determined through the joint constraint optimization of the overall energy utilization efficiency target and the local flow response speed target of the microgrid energy heating pipeline network. The clipped Fourier frequency domain convolutional network adaptively highlights the frequency components whose contribution to the local flow regulation of the microgrid energy heating pipeline network is greater than the threshold. S24. Use frequency-domain adaptive cropping mask values ​​to construct an improved Fourier neural operator network. The improved Fourier neural operator network includes an input layer, an adaptive cropping Fourier frequency-domain convolution layer, a nonlinear activation function layer, and an output layer. The characteristic output expression of the adaptive cropping Fourier frequency-domain convolution layer is as follows: ; in, For the Layer Fourier frequency domain feature output, For the Layer Fourier frequency domain feature input, 、 Respectively The weight parameters and bias parameters of the layer Fourier frequency domain convolutional network, is a nonlinear activation function, is the Hadamard product operation; S25. The final output features of the adaptively clipped Fourier frequency domain convolution layer are used as the final layer representation of the neural network forward propagation and input into the regression mapping function to generate the node heat load prediction results. A composite optimization loss function is constructed based on the node heat load prediction results. : ; in, represents the composite optimization loss for joint optimization of local flow prediction and response performance of microgrid energy heating network, For the The node heat load prediction results of samples are: For the The actual results of node heat load of samples, Based on the The evaluation index of the flow regulation response speed of the microgrid energy heating network calculated by samples is The weight coefficient is adaptively determined based on the real-time operating status of the microgrid energy heating network, which is used to balance the node heat load prediction accuracy and the network flow regulation response speed. M represents the total number of training samples involved in the calculation of the composite optimization loss function. S26. Using composite optimization loss function As the objective function, the weight parameters of the improved Fourier neural operator network are adjusted using the back propagation algorithm. , bias parameters The node heat load prediction model is deployed in the microgrid control center to output the real-time node heat load prediction results. : ; in, To predict the time The node heat load prediction result set of Indicates the Nodes at time The predicted heat load value, is the total number of nodes in the microgrid energy heating network.

[0010] Optionally, S3 includes the following steps: S31. Construct a local load mapping matrix based on the node heat load prediction results : ; in, Indicates that at the prediction time , No. Node pair The load impact factor of each node is established based on the thermal load coupling relationship and thermal transmission path between nodes, and meets the following requirements: ; in, Indicates the Node and The thermal coupling weight between nodes has a value range of [0,1] and is used to quantify the intensity of the heating interaction between nodes. Represents the total number of nodes in the microgrid energy heating network, including all heating nodes distributed in the heating network with heat load regulation or monitoring capabilities; S32. Calculate the local load intensity index of each node based on the local load mapping matrix The local load intensity index is obtained by weighted accumulation of all load impact factors in the i-th row of the local load mapping matrix. The weighted accumulation operation is used to aggregate the total thermal coupling impact of the i-th node on all other nodes in the prediction period. The larger the local load intensity index, the stronger the traffic response regulation demand faced by the node in the current period: ; in, Indicates the Nodes at the prediction time The local load intensity is used to measure the heat load concentration and regulation priority of each node in the microgrid energy heating network within a short-term forecast period; S34. Integrate the local load intensity index into the local load mapping matrix to form an enhanced local load mapping matrix structure The enhanced local load mapping matrix structure is used to simultaneously characterize the thermal load coupling relationship between nodes in the microgrid energy heating network and the local load concentration and regulation priority of each node in the short-term forecast period. The enhanced local load mapping matrix structure consists of two parts: the original local load mapping matrix and the local load intensity index. The original local load mapping matrix is ​​used to express the load influence relationship between each node in the forecast period. The local load intensity index set is used to represent the local load concentration of each node itself in the forecast period. The local load intensity index set is obtained by weighted accumulation of all elements in each row of the original local load mapping matrix. Each row corresponds to a node, which represents the total load coupling between the node and all other nodes. The larger the total load coupling value, the greater the load response pressure of the node in the current forecast period.

[0011] Optionally, the S4 includes the following steps: S41. Constructing a multi-objective optimization function for the microgrid energy heating network based on the enhanced local load mapping matrix structure , the multi-objective optimization function includes energy utilization efficiency target, heat balance target and regulation response speed target: ; in, is the comprehensive evaluation value of the multi-objective optimization function, 、 、 are the weight coefficients of energy efficiency target, heat balance target and regulation response speed target, respectively. It is the weighted unit energy consumption loss index after flow adjustment of all nodes in the microgrid energy heating network. is the heat balance objective function, To adjust the response speed objective function; S42. Set constraints in the multi-objective optimization function, where the constraints include node flow boundary constraints, total heat supply balance constraints, and flow regulation smoothness constraints.

[0012] Optionally, the weighted unit energy loss index is introduced into the local load intensity : ; in, is the weighted unit energy consumption loss after flow adjustment of all nodes in the microgrid energy heating network. For the The energy consumption sensitivity coefficient of each node is For the Nodes at the prediction time The local load strength, To adjust the output flow, is the optimal target flow value calculated based on the heat load prediction and coupling relationship; Constructing the heat balance objective function , evaluate the heating coordination under the coupling conditions between nodes: ; in, is the local load mapping matrix No. Rank Column elements; Constructing an objective function for regulating response speed , used to measure the dynamic change rate of traffic adjustment: ; in, For the Dynamic adjustment sensitivity coefficient of each node, For the Nodes at time The flow rate change rate.

[0013] Optionally, the node flow boundary constraint is used to limit the minimum and maximum value ranges of each node's adjusted output flow value. The node flow boundary constraint is based on the node flow lower limit value. and the node traffic upper limit Together, the node adjusts the output flow value The value of must be between the lower limit and the upper limit of the node flow rate to ensure the physical feasibility of the flow regulation process and the safety of equipment operation; The total heat balance constraint is used to limit the difference between the sum of the output flow values ​​of each node in the entire network and the sum of the optimal target flow values ​​of each node. The total heat balance constraint is determined by the maximum allowable heat balance deviation threshold. Control, the difference between the sum of the adjusted output flow values ​​of all nodes and the sum of the optimal target flow values ​​of all nodes must not exceed the maximum thermal balance deviation threshold, which is used to ensure the global supply and demand dynamic balance of the entire microgrid heating system; The flow regulation smoothness constraint is used to limit the change range of the regulated output flow value of any node between two consecutive moments. The flow regulation smoothness constraint is determined by the maximum flow change threshold. Control, the absolute value of the difference between the regulated output flow values ​​of each node between the current moment and the next moment must be less than or equal to the maximum flow change amplitude threshold, which is used to ensure the continuity and regulation stability of the control behavior during the flow regulation process.

[0014] Optionally, the S5 includes the following steps: S51. Multi-objective optimization function As the fitness evaluation criterion, the population parameters of the improved Hunger Game search algorithm with local traffic adaptive adjustment characteristics are initialized. The population parameters include the population size , initial individual location information and initial hunger state factor , where population size Represents the total number of individuals in the population used to search for the optimal flow distribution scheme for the microgrid energy heating network, and the initial individual position information Indicates the The initial corresponding microgrid energy heating network node flow distribution scheme for each individual: ; in, Indicates the Individuals initially correspond to Traffic distribution value of each node; Initial hunger state factor It is used to determine the local and global search balance of the optimization process of the control algorithm in the process of adaptive regulation of local flow in the microgrid energy heating network. The initial value depends on the node heat load prediction error and the local load intensity, and the value range is [0,1]. S52. Constructing the feasible flow solution space of microgrid energy heating network based on node flow boundary constraints The feasible flow solution space is used to limit the legal value range of the node flow allocation value corresponding to each individual in the improved hunger game search algorithm. The feasible flow solution space is defined by setting the node flow lower limit value of each node. and the node traffic upper limit Determine that the flow distribution scheme corresponding to each individual in the improved Hunger Game search algorithm is represented by the position vector of the individual in the population. The element represents the individual corresponding to the The flow distribution value of each node; if the flow distribution value of any node in the position vector of an individual during the search process is less than the lower limit of the node flow corresponding to the node , it will be corrected to the lower limit of the node flow; if the flow distribution value is greater than the upper limit of the node flow corresponding to the node , then correct it to the upper limit of node flow; Ensure that the flow distribution value of each individual in the algorithm search process is always within the legal physical boundaries defined by the node flow lower limit and the node flow upper limit, thereby forming a feasible flow solution space that meets the actual operation constraints of the microgrid energy heating pipeline network; S53. Initialize the iterative control parameters of the improved Hunger Game search algorithm, which include the maximum number of iterations and the dynamic hunger factor attenuation coefficient. , dynamic hunger factor attenuation coefficient The calculation introduces node heat load prediction error and local load intensity index , used to dynamically adjust the exploration and exploitation efforts of the search process: ; in, Indicates the node heat load prediction results The difference between the actual result of node heat load and the actual result of node heat load at time The prediction error, represents the allowed node heat load prediction error threshold, represents the prediction error sensitivity coefficient; S54. Execute the iterative calculation process of the improved Hunger Game search algorithm, in the In the iteration process, the local load mapping matrix between nodes is used Individual position adaptive update driven jointly with hunger state factor: ; in, For the The first iteration Individuals correspond to The updated value of traffic distribution of nodes, For the The first iteration Individuals correspond to The current traffic distribution value of each node, For the The best individual in the current population at the iteration corresponds to The traffic distribution value of each node, is the local load mapping matrix Middle Rank Column elements, For the The first iteration The hunger state factor of each individual, The optimization initiative weight coefficient is determined dynamically by combining the individual historical optimal position and the local load intensity index; S55. Call the node heat load prediction results in each iteration process And the enhanced local load mapping matrix structure , evaluate the fitness function value of each individual : ; in, Indicates the The first iteration The fitness function value of each individual is used to evaluate the performance of the node traffic allocation scheme; S56. When the number of iterations reaches the maximum number of iterations or the population fitness convergence condition is met, the iteration is terminated and the optimal individual position in the current population is output. As a microgrid energy heating network at the predicted time The optimal traffic distribution scheme: ; in, It represents the optimal flow distribution scheme for each node of the microgrid energy heating network obtained by optimizing the improved Hunger Game search algorithm, which is used to guide the real-time adaptive and precise adjustment of the flow of nodes in the microgrid energy heating network.

[0015] Optionally, the S6 includes the following steps: S61. The optimal traffic allocation solution obtained by optimizing the improved Hunger Game search algorithm As the target regulation benchmark, the flow regulation instruction set containing the flow control parameters of each node is parsed and generated. The flow regulation instruction set includes the target flow value, regulation rate reference value and execution timestamp of each node, which is used to guide the execution unit of the heating pipe network to perform real-time flow control adjustment; S62. Each flow regulation instruction structure definition includes a node number parameter , target flow value , Current adjustment reference value , Allows adjustment of the maximum stride And the release time stamp:

[0016] Node number parameter Used to uniquely identify the first A regulation execution node; Target flow value Taken from the optimal flow distribution scheme Traffic distribution value of each node Current adjustment reference value Obtain the actual flow operation status of the current node based on the real-time monitoring system of the heating pipe network; Allows adjustment of maximum stride length The upper limit of traffic regulation is set based on the node regulation capability and device operation constraints; Sending time stamp To adjust the system time when the instruction takes effect, and to synchronize the flow control behavior of multiple nodes; S63. Compare target flow value With the current adjustment reference value , judge whether the node meets the adjustment execution conditions, if , then the node is marked as needing adjustment, where is the minimum effective adjustment threshold; S64. Execute the phased adjustment plan for the nodes marked as requiring adjustment, according to the local load intensity of the nodes. Determine the priority of the adjustment step, give priority to assigning large adjustment steps to nodes with high load intensity, and form a hierarchical adjustment strategy, so that the response of key nodes is prioritized and the adjustment of non-key nodes is restricted; S65. Convert the structured flow control instruction set into a data format recognizable by the heating network execution unit. Verify the reliability and integrity of the instruction data through the communication module of the microgrid control center before issuing it. This ensures that the instruction data is securely transmitted and accurately delivered to the node control layer. S66. The heating network execution unit receives and interprets the corresponding node flow regulation instruction. Based on the control characteristics of the node regulation valve, electric actuator, or variable frequency pump hardware equipment, it drives the hardware module to perform physical regulation operations on the flow state of the target node until the actual flow value gradually converges to the target flow value. , completing the local flow adaptive regulation of the microgrid energy heating network.

[0017] Optionally, the dynamic optimization control of the local flow of the microgrid energy heating pipeline includes real-time collection of adjusted real-time operating parameters and appending them to the multi-node dynamic feature data set, and based on the updated multi-node dynamic feature data set, performing online incremental learning and updating of the node heat load prediction model, and dynamically adjusting the multi-objective optimization function weight parameters and the improved Hunger Game search algorithm control parameters according to the real-time prediction error and system energy consumption indicators, so that the prediction-optimization-regulation closed loop continues to adaptively iterate.

[0018] The beneficial effects of the present invention are: 1. The present invention constructs an adaptive Fourier neural operator based on a frequency domain sensitivity clipping mechanism to improve the accuracy and generalization ability of heat load prediction. A frequency domain sensitivity factor analysis mechanism is introduced into the Fourier neural operator structure. According to the degree of influence of different Fourier frequency components on the node heat load prediction results, a frequency domain adaptive clipping mask is generated to achieve dynamic filtering of frequency redundant components, and a clipping-enhanced Fourier neural operator structure that is highly adapted to the local load response characteristics of the microgrid is constructed. Compared with the traditional full-frequency Fourier neural network, it can greatly reduce network redundant calculations while maintaining prediction accuracy, improve the model's modeling ability and real-time responsiveness to local heat loads in a highly dynamic multi-node microgrid environment, avoid prediction lag problems, and ensure the input data quality of the optimization algorithm.

[0019] 2. The present invention proposes a structure-driven Hunger Game search algorithm that integrates local load mapping to implement a global-local collaborative traffic optimization strategy. The inter-node thermal load influence factor matrix and local load intensity index are introduced into the individual update mechanism of the Hunger Game search algorithm to construct a locally guided dynamic perturbation model to improve the adaptability of individual search paths to node adjustment priorities and traffic imbalances. At the same time, an adaptive adjustment factor is constructed based on the thermal load prediction error to dynamically modulate the search behavior, achieving a balance between the individual's extensive exploration in the early stage of the search and local convergence in the later stage. This effectively overcomes the problems of insufficient search accuracy and slow response speed of traditional intelligent optimization algorithms when facing multi-objective and multi-constraint traffic adjustment problems, and improves the adaptability of the adjustment algorithm to sudden load changes.

[0020] 3. Based on the optimal flow distribution result, the present invention designs a structured flow regulation instruction format including key control parameters such as node number, target flow value, current flow status, maximum regulation step and execution timestamp, and constructs regulation priority based on local load intensity to form a hierarchical regulation strategy to ensure the practicality and goal orientation of the regulation action. The regulation instruction mechanism can achieve fine linkage with the execution units (smart valves, variable frequency pumps) in the actual heating pipeline network, effectively ensuring the continuity, response speed and stability of the control behavior, solving the problems of high degree of abstraction and weak implementation of control data in traditional solutions, and realizing a closed-loop adaptive operation mechanism of prediction, optimization and control. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings: Figure 1 This is a flow chart of a method for regulating flow in an energy heating network based on machine learning proposed by the present invention; Figure 2 This is a structural diagram of an adaptive Fourier neural operator load prediction model in a method for regulating energy heating pipe network flow based on machine learning proposed in the present invention; Figure 3 This is a schematic diagram of the dynamic adjustment of individual flow position update process and structural parameters in the improved Hunger Game search algorithm in the energy heating pipeline flow regulation method based on machine learning proposed by the present invention. DETAILED DESCRIPTION

[0022] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0023] refer to Figure 1-Figure 3 , a method for regulating flow in an energy heating network based on machine learning, comprising the following steps: S1. Collecting real-time operating parameters of each node in the microgrid energy heating system, integrating the real-time operating parameters into a multi-node dynamic feature dataset, and preprocessing the multi-node dynamic feature dataset to obtain a standardized dynamic feature dataset; S2. Build a Fourier neural operator load forecasting model using a standardized dynamic feature dataset and complete training to obtain a node thermal load forecasting model. This node thermal load forecasting model is deployed in the microgrid control center and outputs node thermal load forecast results in real time based on the latest standardized dynamic feature dataset. S3. Generate a local load mapping matrix based on the node heat load forecast results. The local load mapping matrix is ​​used to quantify the heat load demand of each node within the short-term forecast period. S4. Construct a multi-objective optimization function based on the local load mapping matrix, including energy efficiency, heat balance, and regulation response speed, and set corresponding constraints. S5. Using the multi-objective optimization function as the evaluation criterion, initialize the population parameters, feasible flow solution space, and iterative control parameters of the improved Hunger Game search algorithm, execute the iterative calculation process of the improved Hunger Game search algorithm, and use the node heat load prediction results and local load mapping matrix to evaluate individual fitness during the iteration process to search for the optimal flow allocation solution; S6. Convert the optimal flow distribution plan into a flow regulation instruction and send it to the heating network execution unit to complete the real-time flow adaptive regulation of each node; S7. Continuously monitor the operating status of the pipeline network after adaptive flow adjustment to achieve dynamic optimization control of the local flow of the microgrid energy heating pipeline network.

[0024] In this embodiment, the real-time operating parameters include node temperature, node flow rate, node pressure and historical heat load change records, and the preprocessing includes denoising, feature scaling and dimension unification.

[0025] In this embodiment, S2 includes the following steps: S21. Based on standardized dynamic feature dataset Get the frequency domain characteristic components : ; in, represents the standardized dynamic feature dataset, Represents a single sample input vector at time t, which contains the set of operating status parameters of all nodes at time t. Indicates the node temperature of each node at time t, reflecting the heat transfer status of the unit node in the microgrid heating system. It represents the node flow velocity of each node at time t, which is the rate at which the heat carrier flows through the node per unit time. Indicates the node pressure of each node at time t, which is used to measure the pressure state of the heating medium at the node. Represents the historical heat load change record at time t, which is the extracted value of the historical time series heat load state of each node. Represents the total number of samples in the standardized dynamic feature data set; the standardized dynamic feature data set is decomposed into frequency domain features using the Fourier transform operator. For the moment Time The frequency domain coefficients of the node real-time operation parameters corresponding to the Fourier frequency components; The standardized dynamic feature data set is processed by the Fourier transform operator to convert the node temperature, node flow velocity, node pressure and heat load change parameters in the time domain into corresponding frequency domain expressions, and the frequency domain feature components reflecting the periodicity and fluctuation characteristics of the node operation are extracted.

[0026] S22. Calculate frequency domain sensitivity factor , evaluate the contribution of different Fourier frequency components to the node heat load prediction results: ; in, Indicates the The frequency domain sensitivity factor of the Fourier frequency component to the node heat load prediction result, Indicates that at the prediction time The following only uses Nodal heat load prediction results calculated using Fourier frequency components; S23. Based on frequency domain sensitivity factor Generate a frequency-domain adaptive cropping mask: ; in, For the The frequency-domain adaptive clipping mask value of the Fourier frequency components, The frequency domain adaptive clipping threshold is determined through the joint constraint optimization of the overall energy utilization efficiency target and the local flow response speed target of the microgrid energy heating pipeline network. The clipped Fourier frequency domain convolutional network adaptively highlights the frequency components whose contribution to the local flow regulation of the microgrid energy heating pipeline network is greater than the threshold. S24. Use frequency-domain adaptive cropping mask values ​​to construct an improved Fourier neural operator network. The improved Fourier neural operator network includes an input layer, an adaptive cropping Fourier frequency-domain convolution layer, a nonlinear activation function layer, and an output layer. The characteristic output expression of the adaptive cropping Fourier frequency-domain convolution layer is as follows: ; in, For the Layer Fourier frequency domain feature output, For the Layer Fourier frequency domain feature input, 、 Respectively The weight parameters and bias parameters of the layer Fourier frequency domain convolutional network, is a nonlinear activation function, is the Hadamard product operation; The formula introduced in S24 will adaptively clip the mask in the frequency domain By integrating the Hadamard product method into the Fourier frequency domain convolution process, a structural breakthrough is achieved based on the traditional Fourier neural operator structure. The effect is that in the microgrid energy heating network, different nodes have different sensitive frequency distributions for regulating flow, and direct full-frequency convolution calculation will cause a lot of computational redundancy and inefficient response. The tailoring mechanism effectively eliminates the frequency characteristic components that have a low contribution to the local heat load prediction accuracy, retains the high-contribution and high-coupling frequency area characteristics, and significantly improves the prediction model's response capability to highly dynamic nodes and the calculation convergence speed.

[0027] S25. The final output features of the adaptively clipped Fourier frequency domain convolution layer are used as the final layer representation of the neural network forward propagation and input into the regression mapping function to generate the node heat load prediction results. A composite optimization loss function is constructed based on the node heat load prediction results. : ; in, represents the composite optimization loss for joint optimization of local flow prediction and response performance of microgrid energy heating network, For the The node heat load prediction results of samples are: For the The actual results of node heat load of samples, Based on the The evaluation index of the flow regulation response speed of the microgrid energy heating network calculated by samples is The weight coefficient is adaptively determined based on the real-time operating status of the microgrid energy heating network, which is used to balance the node heat load prediction accuracy and the network flow regulation response speed. M represents the total number of training samples involved in the calculation of the composite optimization loss function. In the embodiment, the regression mapping function is constructed based on the final features output by the frequency domain convolution layer of the Fourier neural operator, and a low-rank linear mapping structure is adopted to map the high-dimensional frequency domain feature tensor to the heat load prediction value corresponding to each node. The regression mapping function is composed of a trainable weight matrix and a bias vector, and is jointly trained with the entire network under the guidance of a composite loss function, thereby achieving accurate fitting of the node heat load. Its construction not only considers the minimization of prediction error, but also integrates the flow regulation response speed constraint, so that the output result has physical feasibility and regulation adaptability during the optimization process, thereby meeting the real-time and high-precision load prediction requirements in the microgrid energy system.

[0028] Composite loss function in S25 By combining the heat load prediction error with the node adjustment response speed index The combined incorporation into the same optimization objective realizes the fusion training of prediction performance and regulation execution performance. Unlike the traditional loss function that only considers prediction error, the present invention not only focuses on accurate prediction during training, but also simultaneously optimizes fast regulation. It is suitable for application environments that require fast closed-loop control response in microgrid scenarios. Under conditions where dynamic load changes drastically, it can significantly reduce response lag and regulation disturbances.

[0029] S26. Using composite optimization loss function As the objective function, the weight parameters of the improved Fourier neural operator network are adjusted using the back propagation algorithm. , bias parameters The node heat load prediction model is deployed in the microgrid control center to output the real-time node heat load prediction results. : ; in, To predict the time The node heat load prediction result set of Indicates the Nodes at time The predicted heat load value, is the total number of nodes in the microgrid energy heating network.

[0030] In this embodiment, S3 includes the following steps: S31. Construct a local load mapping matrix based on the node heat load prediction results : ; in, Indicates that at the prediction time , No. Node pair The load impact factor of each node is established based on the thermal load coupling relationship and thermal transmission path between nodes, and meets the following requirements: ; in, Indicates the Node and The thermal coupling weight between nodes has a value range of [0,1] and is used to quantify the intensity of the heating interaction between nodes. Represents the total number of nodes in the microgrid energy heating network, including all heating nodes distributed in the heating network with heat load regulation or monitoring capabilities; S32. Calculate the local load intensity index of each node based on the local load mapping matrix The local load intensity index is obtained by weighted accumulation of all load impact factors in the i-th row of the local load mapping matrix. The weighted accumulation operation is used to aggregate the total thermal coupling impact of the i-th node on all other nodes in the prediction period. The larger the local load intensity index, the stronger the traffic response regulation demand faced by the node in the current period: ; in, Indicates the Nodes at the prediction time The local load intensity is used to measure the heat load concentration and regulation priority of each node in the microgrid energy heating network within a short-term forecast period; S34. Integrate the local load intensity index into the local load mapping matrix to form an enhanced local load mapping matrix structure The enhanced local load mapping matrix structure is used to simultaneously characterize the thermal load coupling relationship between nodes in the microgrid energy heating network and the local load concentration and regulation priority of each node in the short-term forecast period. The enhanced local load mapping matrix structure consists of two parts: the original local load mapping matrix and the local load intensity index. The original local load mapping matrix is ​​used to express the load influence relationship between each node in the forecast period. The local load intensity index set is used to represent the local load concentration of each node itself in the forecast period. The local load intensity index set is obtained by weighted accumulation of all elements in each row of the original local load mapping matrix. Each row corresponds to a node, which represents the total load coupling between the node and all other nodes. The larger the total load coupling value, the greater the load response pressure of the node in the current forecast period.

[0031] In this embodiment, S4 includes the following steps: S41. Constructing a multi-objective optimization function for the microgrid energy heating network based on the enhanced local load mapping matrix structure , the multi-objective optimization function includes energy utilization efficiency target, heat balance target and regulation response speed target: ; in, is the comprehensive evaluation value of the multi-objective optimization function, 、 、 are the weight coefficients of energy efficiency target, heat balance target and regulation response speed target, respectively. It is the weighted unit energy consumption loss index after flow adjustment of all nodes in the microgrid energy heating network. is the heat balance objective function, To adjust the response speed objective function; S42. Set constraints in the multi-objective optimization function, where the constraints include node flow boundary constraints, total heat supply balance constraints, and flow regulation smoothness constraints.

[0032] In this embodiment, the local load intensity is introduced into the process of weighting the unit energy consumption loss index. : ; in, is the weighted unit energy consumption loss after flow adjustment of all nodes in the microgrid energy heating network. For the The energy consumption sensitivity coefficient of each node is For the Nodes at the prediction time The local load strength, To adjust the output flow, is the optimal target flow value calculated based on the heat load prediction and coupling relationship; Weighted unit energy consumption loss Introduced node local load intensity As a dynamic weighting factor, energy consumption optimization is not only related to the absolute flow deviation, but also to the regulation criticality of the node in the overall thermal network. This solves the defect that the average weight in traditional energy consumption optimization cannot distinguish high-priority nodes. When the local load pressure is high, it can automatically allocate regulation resources to key nodes first, thereby optimizing the overall regulation cost and energy efficiency ratio of the system.

[0033] Constructing the heat balance objective function , evaluate the heating coordination under the coupling conditions between nodes: ; in, is the local load mapping matrix No. Rank Column elements; Heat balance objective function Incorporating load coupling matrix The difference in flow deviation is used to construct a coupling imbalance measure among multiple nodes through the influence propagation path between nodes, breaking the traditional single-node benchmarking balance strategy and establishing a node-interference heating coordination objective function for the heat load network structure, so that the optimization results can better reflect the operational stability of the entire heat network system and avoid the disadvantages of single-point optimality and system imbalance.

[0034] Constructing an objective function for regulating response speed , used to measure the dynamic change rate of traffic adjustment: ; in, For the Dynamic adjustment sensitivity coefficient of each node, For the Nodes at time The flow rate change rate.

[0035] Response speed objective function Introducing flow rate change and node sensitivity coefficient , constructing a truly observable dynamic cost evaluation metric for regulation. This structure enables the algorithm to perceive regulation smoothness during the optimization process, avoiding water hammer effects caused by drastic fluctuations in flow between nodes or frequent starts and stops of heat exchangers, thereby improving the controllability of regulation behavior and equipment protection capabilities.

[0036] In this embodiment, the node flow boundary constraint is used to limit the minimum and maximum value range of each node's output flow value. The node flow boundary constraint is based on the node flow lower limit. and the node traffic upper limit Together, the node adjusts the output flow value The value of must be between the lower limit and the upper limit of the node flow rate to ensure the physical feasibility of the flow regulation process and the safety of equipment operation; The total heat balance constraint is used to limit the difference between the sum of the output flow values ​​of each node in the entire network and the sum of the optimal target flow values ​​of each node. The total heat balance constraint is determined by the maximum allowable heat balance deviation threshold. Control, the difference between the sum of the adjusted output flow values ​​of all nodes and the sum of the optimal target flow values ​​of all nodes must not exceed the maximum thermal balance deviation threshold, which is used to ensure the global supply and demand dynamic balance of the entire microgrid heating system; The flow regulation smoothness constraint is used to limit the change range of the regulated output flow value of any node between two consecutive moments. The flow regulation smoothness constraint is determined by the maximum flow change threshold. Control, the absolute value of the difference between the regulated output flow values ​​of each node between the current moment and the next moment must be less than or equal to the maximum flow change amplitude threshold, which is used to ensure the continuity and regulation stability of the control behavior during the flow regulation process.

[0037] In this embodiment, S5 includes the following steps: S51. Multi-objective optimization function As the fitness evaluation criterion, the population parameters of the improved Hunger Game search algorithm with local traffic adaptive adjustment characteristics are initialized. The population parameters include the population size , initial individual location information and initial hunger state factor , where population size Represents the total number of individuals in the population used to search for the optimal flow distribution scheme for the microgrid energy heating network, and the initial individual position information Indicates the The initial corresponding microgrid energy heating network node flow distribution scheme for each individual: ; in, Indicates the Individuals initially correspond to Traffic distribution value of each node; Initial hunger state factor It is used to determine the local and global search balance of the optimization process of the control algorithm in the process of adaptive regulation of local flow in the microgrid energy heating network. The initial value depends on the node heat load prediction error and the local load intensity, and the value range is [0,1]. S52. Constructing the feasible flow solution space of microgrid energy heating network based on node flow boundary constraints The feasible flow solution space is used to limit the legal value range of the node flow allocation value corresponding to each individual in the improved hunger game search algorithm. The feasible flow solution space is defined by setting the node flow lower limit value of each node. and the node traffic upper limit Determine that the flow distribution scheme corresponding to each individual in the improved Hunger Game search algorithm is represented by the position vector of the individual in the population. The element represents the individual corresponding to the The flow distribution value of each node; if the flow distribution value of any node in the position vector of an individual during the search process is less than the lower limit of the node flow corresponding to the node , it will be corrected to the lower limit of the node flow; if the flow distribution value is greater than the upper limit of the node flow corresponding to the node , then correct it to the upper limit of node flow; Ensure that the flow distribution value of each individual in the algorithm search process is always within the legal physical boundaries defined by the node flow lower limit and the node flow upper limit, thereby forming a feasible flow solution space that meets the actual operation constraints of the microgrid energy heating pipeline network; S53. Initialize the iterative control parameters of the improved Hunger Game search algorithm, which include the maximum number of iterations and the dynamic hunger factor attenuation coefficient. , dynamic hunger factor attenuation coefficient The calculation introduces node heat load prediction error and local load intensity index , used to dynamically adjust the exploration and exploitation efforts of the search process: ;

[0038] in, Indicates the node heat load prediction results The difference between the actual result of node heat load and the actual result of node heat load at time The prediction error, represents the allowed node heat load prediction error threshold, represents the prediction error sensitivity coefficient; The node prediction error and local load intensity Coupling is used to dynamically adjust the search activity in the Hunger Games search algorithm. The formula is designed so that individuals maintain higher exploration capabilities when prediction uncertainty is high and the adjustment task is heavy; and when the system tends to be stable, it automatically suppresses large perturbations in the search. This adaptive adjustment mechanism based on real-time operating status is significantly better than traditional fixed step size or static convergence functions, reflecting stronger algorithm intelligence and controllability of the search process.

[0039] S54. Execute the iterative calculation process of the improved Hunger Game search algorithm, in the In the iteration process, the local load mapping matrix between nodes is used Individual position adaptive update driven jointly with hunger state factor: ; in, For the The first iteration Individuals correspond to The updated value of traffic distribution of nodes, For the The first iteration Individuals correspond to The current traffic distribution value of each node, For the The best individual in the current population at the iteration corresponds to The traffic distribution value of each node, is the local load mapping matrix Middle Rank Column elements, For the The first iteration The hunger state factor of the individual is obtained by updating the hunger state factor of the pth individual according to the dynamic hunger factor attenuation coefficient. The optimization initiative weight coefficient is determined dynamically by combining the individual historical optimal position and the local load intensity index; Hunger status factor It is the control The search intensity and search direction of each individual in the improved Hunger Game search algorithm are important dynamic parameters. The design purpose is to simulate the mechanism in which individuals in real hunger behavior enhance their desire to explore due to resource shortages, and to make adaptive adjustments based on the actual operating status of the microgrid energy heating pipeline network, thereby improving search efficiency and local regulation accuracy.

[0040] The calculation of the hunger state factor adopts an update model based on a dynamic attenuation mechanism, which specifically includes the following two parts: Initial definition: At the iteration, the hunger state factor of each individual Can be initialized to a The random value in the interval indicates that the individual has a high initial exploration enthusiasm, which avoids falling into the local optimum in the early stage of the search.

[0041] Dynamic update mechanism: At the tth iteration, the hunger state factor of the pth individual According to the hunger state factor of the previous iteration , current forecast error level and the individual local load influence weight The combined calculation yields: ; in, For the The hunger state factor of each individual in the previous iteration, For the The local load intensity of the node currently associated with each individual, is the average local load intensity of all nodes in the current population, It is the local load enhancement factor (the recommended value range is 0.1-0.3), which is used to amplify the regulation priority of high-load nodes.

[0042] When the system load forecast error is large (i.e. Large) or local load intensity at a node Significantly above average When the individual's hunger status factor is automatically increased , driving it closer to the optimal traffic solution and increasing search aggressiveness; If the current prediction error is small or the node load associated with the individual is low, then It will gradually weaken to avoid excessive disturbance to the current optimal solution, thereby improving the convergence stability of the system.

[0043] The hunger state factor model fully integrates the three key indicators of prediction, load and control in the microgrid heating scenario, forming a dynamic, node-adaptive search-driven mechanism. It significantly enhances the processing ability of the hunger game search algorithm in the present invention for complex, multi-objective optimization problems, and is suitable for flow optimization and regulation in scenarios with drastic fluctuations in node load or local sudden heat demand.

[0044] The position update formula in S54 not only refers to the optimal individual, but also introduces the load coupling matrix A thermal coupling-driven group collaborative perturbation mechanism was established, which enables each individual in the population to have the dual driving forces of global guidance (convergence to the optimal) + local structure guidance (adjustment according to thermal network coupling) during updating, effectively improving the information dissemination efficiency and multi-peak search capabilities during the search process, and avoiding falling into local optimality.

[0045] S55. Call the node heat load prediction results in each iteration process And the enhanced local load mapping matrix structure , evaluate the fitness function value of each individual : ; in, Indicates the The first iteration The fitness function value of each individual is used to evaluate the performance of the node traffic allocation scheme; The fitness function construction in S55 explicitly introduces three key input parameters: the current individual flow solution , Heat load prediction using Fourier neural operator output , enhanced load mapping structure The coupled input structure ensures that the fitness evaluation not only considers the rationality of the solution itself, but also reflects the solution's external prediction support and the coordination ability of the entire network. It is a systematic evaluation mechanism for the prediction-optimization-control closed loop, which effectively supports the full-process convergence control of the adaptive adjustment model.

[0046] S56. When the number of iterations reaches the maximum number of iterations or the population fitness convergence condition is met, the iteration is terminated and the optimal individual position in the current population is output. As a microgrid energy heating network at the predicted time The optimal traffic distribution scheme: ;

[0047] in, It represents the optimal flow distribution scheme for each node of the microgrid energy heating network obtained by optimizing the improved Hunger Game search algorithm, which is used to guide the real-time adaptive and precise adjustment of the flow of nodes in the microgrid energy heating network.

[0048] In this embodiment, S6 includes the following steps: S61. The optimal traffic allocation solution obtained by optimizing the improved Hunger Game search algorithm As the target regulation benchmark, the flow regulation instruction set containing the flow control parameters of each node is parsed and generated. The flow regulation instruction set includes the target flow value, regulation rate reference value and execution timestamp of each node, which is used to guide the execution unit of the heating pipe network to perform real-time flow control adjustment; S62. Each flow regulation instruction structure definition includes a node number parameter , target flow value , Current adjustment reference value , Allows adjustment of the maximum stride Time stamp for issuing : Node number parameter Used to uniquely identify the first A regulation execution node; Target flow value Taken from the optimal flow distribution scheme Traffic distribution value of each node Current adjustment reference value Obtain the actual flow operation status of the current node based on the real-time monitoring system of the heating pipe network; Allows adjustment of maximum stride length The upper limit of traffic regulation is set based on the node regulation capability and device operation constraints; Sending time stamp To adjust the system time when the instruction takes effect, and to synchronize the flow control behavior of multiple nodes; S63. Compare target flow value With the current adjustment reference value , judge whether the node meets the adjustment execution conditions, if , then the node is marked as needing adjustment, where is the minimum effective adjustment threshold; S64. Execute the phased adjustment plan for the nodes marked as requiring adjustment, according to the local load intensity of the nodes. Determine the priority of the adjustment step, give priority to assigning large adjustment steps to nodes with high load intensity, and form a hierarchical adjustment strategy, so that the response of key nodes is prioritized and the adjustment of non-key nodes is restricted; S65. Convert the structured flow control instruction set into a data format recognizable by the heating network execution unit. Verify the reliability and integrity of the instruction data through the communication module of the microgrid control center before issuing it. This ensures that the instruction data is securely transmitted and accurately delivered to the node control layer. S66. The heating network execution unit receives and interprets the corresponding node flow regulation instruction. Based on the control characteristics of the node regulation valve, electric actuator, or variable frequency pump hardware equipment, it drives the hardware module to perform physical regulation operations on the flow state of the target node until the actual flow value gradually converges to the target flow value. , completing the local flow adaptive regulation of the microgrid energy heating network.

[0049] In this embodiment, the dynamic optimization control of the local flow of the microgrid energy heating pipeline network includes real-time collection of adjusted real-time operating parameters and appending them to the multi-node dynamic feature data set. Based on the updated multi-node dynamic feature data set, the node heat load prediction model is updated through online incremental learning. The weight parameters of the multi-objective optimization function and the control parameters of the improved Hunger Game search algorithm are dynamically adjusted according to the real-time prediction error and the system energy consumption index, so that the prediction-optimization-regulation closed loop continues to adaptively iterate.

[0050] Example 1: At 1:45 PM on December 22, 2024, the smart energy management center of an industrial park received an alert from the microgrid control platform stating that "load forecast deviations in node Group-B have increased for five consecutive minutes, indicating a surge in local heat load." The park's microgrid system, controlled by the master control node MC-01, consists of 36 heating nodes and three gas-fired substations, distributed in a ring topology. The total heating area is approximately 237,000 square meters.

[0051] At that time, the control system recorded a historical load value of 98.6 kW for node "Node-17." However, the Fourier neural network operator model deployed on the edge control unit predicted a value of 112.3 kW for the next 10 minutes, with a prediction error of 13.7%. Further observation revealed that the outdoor temperature at this node dropped sharply between 12:45 and 1:45 PM (from -8.1°C to -13.3°C), and the indoor load demand increased simultaneously, indicating a significant "prediction lag risk."

[0052] The system then invoked the proposed method, with the control center automatically invoking the Fourier neural operator prediction module to perform frequency-domain modeling of the dynamic heat load for Nodes 17 to 20 (Group B). Using an adaptive frequency clipping mechanism, the model identified high-impact frequencies concentrated between 0.08 Hz and 0.12 Hz. With the sensitivity mask activated in real time, the model completed heat load reconstruction and short-term prediction in less than nine seconds. The new predicted value for Node 17 was revised to 114.1 kW, with the error reduced to 2.1%.

[0053] Then the corresponding module generates a local load mapping matrix. The system identifies that the coupling influence values ​​of Node-17 on Node-18 and Node-19 are 0.29 and 0.21 respectively. At the same time, the local load intensity of Node-17 is The calculated value is 1.36, the largest among all nodes. The system marks it as a "high priority adjustment node" and generates the following code structure mapping result: Node-17: LocalLoadStrength:1.36; CouplingtoNode-18:0.29; CouplingtoNode-19:0.21; Based on the results, the system calls the multi-objective optimization module and sets the energy consumption sensitivity coefficient is 1.12, adjusting the response sensitivity coefficient The optimization objective function is constructed as 0.93. Due to the high stability of the historical adjustment response of this node, the upper limit of the system adjustment step is It is set to 0.5 kg / s, and the target flow rate is adjusted to a target value of 6.4 kg / s.

[0054] At 2:03 PM, the improved Hunger Game search algorithm initiated flow regulation optimization. In the 14th iteration, the algorithm locked onto the optimal individual flow configuration, increasing the target flow rate at Node-17 from the current 5.8 kg / s to 6.37 kg / s. This simultaneously reduced the flow rate at Node-18 by 0.3 kg / s and at Node-19 by 0.2 kg / s to balance the local heating load. The total optimization algorithm execution time was 38.2 seconds, nearly three times faster than the traditional genetic algorithm.

[0055] Then, the system generates a structured flow regulation instruction based on the result according to claim 6. The instruction content is as follows: Node-17 → Flow rate target: 6.37 kg / s; Current value: 5.80 kg / s; Maximum step: 0.5 kg / s; Status: Needs adjustment; Node-18 → Flow target: 5.62 kg / s; Current value: 5.92 kg / s; Status: Participating in regulation; Node-19 → Traffic target: 5.70 kg / s; Current value: 5.90 kg / s; Status: Participating in regulation After the command was issued, the electric valve of Node-17 adjusted the flow rate step by step within 17 seconds, with each step being 0.1 kg / s, and finally reached the target flow rate at 14:05:31. The control center received the adjustment feedback confirmation: Node-17 → Flow rate has stabilized to 6.37 kg / s; response time: 72 seconds; predicted load error reduced to 0.9 kW; Over the next 15 minutes, the indoor temperature in Group-B rose by an average of 0.6°C, user complaints decreased significantly, and the heating trunk pressure remained stable. The system recorded the data before and after the adjustment, as shown in Table 1: Table 1 System record data before and after adjustment

[0056] The comparison experiment with the traditional method is shown in Table 2 below: Table 2 Comparative data between the present invention and the traditional method

[0057] Through the deployment and verification of the method of the present invention, the system has a significant ability to predict and anticipate node thermal load fluctuations at critical moments, and quickly forms precise control results through structural optimization and dynamic adjustment mechanisms. Compared with traditional methods, it has greatly improved in adjustment accuracy, response time, and energy consumption optimization, which fully proves that the present invention has engineering usability and promotion value.

[0058] The present invention constructs an adaptive Fourier neural operator based on the frequency domain sensitivity clipping mechanism to improve the heat load prediction accuracy and generalization ability, introduces the frequency domain sensitivity factor analysis mechanism into the Fourier neural operator structure, generates a frequency domain adaptive clipping mask according to the degree of influence of different Fourier frequency components on the node heat load prediction results, realizes dynamic filtering of frequency redundant components, and constructs a clipping-enhanced Fourier neural operator structure that is highly adapted to the local load response characteristics of the microgrid. Compared with the traditional full-frequency Fourier neural network, it can greatly reduce network redundant calculations while maintaining prediction accuracy, improve the model's modeling ability and real-time responsiveness for local heat loads in a highly dynamic multi-node microgrid environment, avoid prediction lag problems, and ensure the input data quality of the optimization algorithm.

[0059] The present invention proposes a structure-driven Hunger Game search algorithm that integrates local load mapping to implement a global-local collaborative traffic optimization strategy. The inter-node thermal load influence factor matrix and local load intensity index are introduced into the individual update mechanism of the Hunger Game search algorithm to construct a locally guided dynamic perturbation model, thereby improving the adaptability of the individual search path to node regulation priority and traffic imbalance. At the same time, an adaptive regulation factor is constructed based on the thermal load prediction error to dynamically modulate the search behavior, achieving a balance between the individual's extensive exploration in the early stage of the search and local convergence in the later stage. This effectively overcomes the problems of insufficient search accuracy and slow response speed of traditional intelligent optimization algorithms when facing multi-objective and multi-constraint traffic regulation problems, and improves the adaptability of the regulation algorithm to sudden load changes.

[0060] Based on the optimal flow distribution result, the present invention designs a structured flow regulation instruction format including key control parameters such as node number, target flow value, current flow status, maximum regulation step and execution timestamp, and constructs regulation priority based on local load intensity to form a hierarchical regulation strategy to ensure the practicality and goal orientation of the regulation action. The regulation instruction mechanism can achieve fine linkage with the execution unit in the actual heating pipeline network, effectively ensuring the continuity, response speed and stability of the control behavior, solving the problems of high degree of abstraction and weak implementation of control data in traditional solutions, and realizing a closed-loop adaptive operation mechanism of prediction, optimization and control.

[0061] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for regulating flow in an energy heating network based on machine learning, characterized in that: The steps include: S1. Collect the real-time operating parameters of each node in the microgrid energy heating system, integrate the real-time operating parameters into a multi-node dynamic feature dataset, and preprocess them to obtain a standardized dynamic feature dataset; S2. Build a Fourier neural operator load prediction model using a standardized dynamic feature dataset, complete training to obtain a node thermal load prediction model, and output node thermal load prediction results in real time. S3. Generate a local load mapping matrix based on the node heat load prediction results; S4. Construct a multi-objective optimization function based on the local load mapping matrix and set corresponding constraints; S5. Using the multi-objective optimization function as the evaluation criterion, initialize the improved Hunger Game search algorithm and execute the improved Hunger Game search algorithm iterative calculation process. During the iterative process, the node heat load prediction results and the local load mapping matrix are used to evaluate individual fitness and search for the optimal traffic allocation solution. S6. Convert the optimal flow distribution plan into a flow regulation instruction and send it to the heating network execution unit to complete the real-time flow adaptive regulation of each node; S7. Continuously monitor the operating status of the pipeline network after adaptive flow adjustment to achieve dynamic optimization control of the local flow of the microgrid energy heating pipeline network.

2. The method for regulating flow of energy heating pipe network based on machine learning according to claim 1, characterized in that: The real-time operating parameters include node temperature, node flow rate, node pressure and historical heat load change records, and the preprocessing includes denoising processing, feature scaling processing and dimension unification processing.

3. The method for regulating flow of energy heating pipe network based on machine learning according to claim 1, characterized in that: The S2 comprises the following steps: S21. Based on standardized dynamic feature dataset Get the frequency domain characteristic components ; S22. According to the frequency domain characteristic components Calculating frequency domain sensitivity factors , evaluate the contribution of different Fourier frequency components to the node heat load prediction results: ； in, Indicates the The frequency domain sensitivity factor of the Fourier frequency component to the node heat load prediction result, Indicates that at the prediction time The following only uses Nodal heat load prediction results calculated using Fourier frequency components; S23. Based on frequency domain sensitivity factor Generate frequency-domain adaptive clipping masks , the frequency domain sensitivity factor is greater than the frequency domain adaptive clipping threshold, the frequency domain adaptive clipping mask is 1, otherwise it is 0; S24. Use frequency-domain adaptive cropping mask values ​​to construct an improved Fourier neural operator network. The improved Fourier neural operator network includes an input layer, an adaptive cropping Fourier frequency-domain convolution layer, a nonlinear activation function layer, and an output layer. The characteristic output expression of the adaptive cropping Fourier frequency-domain convolution layer is as follows: ； in, For the Layer Fourier frequency domain feature output, For the Layer Fourier frequency domain feature input, 、 Respectively The weight parameters and bias parameters of the layer Fourier frequency domain convolutional network, is a nonlinear activation function, is the Hadamard product operation, Represents the total number of layers of the improved Fourier neural operator network; S25. The final output features of the adaptively clipped Fourier frequency domain convolution layer are used as the final layer representation of the neural network forward propagation and input into the regression mapping function to generate the node heat load prediction results. A composite optimization loss function is constructed based on the node heat load prediction results. : ； in, represents the composite optimization loss for joint optimization of local flow prediction and response performance of microgrid energy heating network, For the The node heat load prediction results of samples are: For the The actual results of node heat load of samples, Based on the The evaluation index of the flow regulation response speed of the microgrid energy heating network calculated by samples is The weight coefficient is determined adaptively based on the real-time operating status of the microgrid energy heating network, and M represents the total number of training samples involved in the calculation of the composite optimization loss function; S26. Using composite optimization loss function As the objective function, the weight parameters of the improved Fourier neural operator network are adjusted using the back propagation algorithm. , bias parameters The node heat load prediction model is deployed in the microgrid control center to output the real-time node heat load prediction results. .

4. The method for regulating flow of energy heating pipe network based on machine learning according to claim 3 is characterized in that: The S3 includes the following steps: S31. Construct a local load mapping matrix based on the node heat load prediction results : ； in, Indicates that at the prediction time , No. Node pair The load impact factor of each node, Represents the total number of nodes in the microgrid energy heating network; S32. Calculate the local load intensity index of each node based on the local load mapping matrix , the local load intensity index is obtained by weighted accumulation of all load influencing factors in the i-th row in the local load mapping matrix; S34. Integrate the local load intensity index into the local load mapping matrix to form an enhanced local load mapping matrix structure ,The enhanced local load mapping matrix structure is used to simultaneously characterize the ,heat load coupling relationship between nodes in the microgrid energy ,heating network and the local load concentration and regulation priority ,of each node in the short-term prediction period.

5. The method for regulating flow of energy heating pipe network based on machine learning according to claim 4 is characterized in that: The S4 comprises the following steps: S41. Constructing a multi-objective optimization function for the microgrid energy heating network based on the enhanced local load mapping matrix structure , multi-objective optimization function It is obtained by weighting the weighted unit energy consumption loss index, the heat balance objective function and the regulation response speed objective function after the flow of all nodes in the microgrid energy heating network is adjusted; S42. Set constraints in the multi-objective optimization function, where the constraints include node flow boundary constraints, total heat supply balance constraints, and flow regulation smoothness constraints.

6. The method for regulating flow of energy heating pipe network based on machine learning according to claim 5 is characterized in that: The weighted unit energy loss index is introduced into the process of local load intensity : ； in, is the weighted unit energy consumption loss after flow adjustment of all nodes in the microgrid energy heating network. For the The energy consumption sensitivity coefficient of each node is To adjust the output flow, is the optimal target flow value calculated based on the heat load prediction and coupling relationship; Constructing the heat balance objective function , evaluate the heating coordination under the coupling conditions between nodes: ； in, is the local load mapping matrix No. Rank Column elements; Constructing an objective function for regulating response speed , used to measure the dynamic change rate of traffic adjustment: ； in, For the Dynamic adjustment sensitivity coefficient of each node, For the Nodes at time The flow rate change rate.

7. The method for regulating flow of energy heating pipe network based on machine learning according to claim 5, characterized in that: The node flow boundary constraint is used to limit the minimum and maximum value ranges of each node's adjusted output flow value. The node flow boundary constraint is based on the node flow lower limit. and the node traffic upper limit together constitute; The total heat balance constraint is controlled by the maximum allowable heat balance deviation threshold. The difference between the sum of the regulated output flow values ​​of all nodes and the sum of the optimal target flow values ​​of all nodes must not exceed the maximum heat balance deviation threshold. The flow regulation smoothness constraint is controlled by the maximum flow change amplitude threshold. The absolute value of the difference between the regulated output flow values ​​of each node between the current moment and the next moment must be less than or equal to the maximum flow change amplitude threshold.

8. The method for regulating flow of energy heating pipe network based on machine learning according to claim 5, characterized in that: The S5 comprises the following steps: S51. Multi-objective optimization function As the fitness evaluation criterion, the population parameters of the improved Hunger Game search algorithm with local traffic adaptive adjustment characteristics are initialized. The population parameters include the population size , initial individual location information and initial hunger state factor , where population size Represents the total number of individuals in the population used to search for the optimal flow distribution scheme for the microgrid energy heating network, and the initial individual position information middle Indicates the Individuals initially correspond to Traffic distribution value of each node; S52. Constructing the feasible flow solution space of microgrid energy heating network based on node flow boundary constraints , the flow distribution scheme corresponding to each individual in the improved Hunger Game search algorithm is represented by the position vector of the individual in the population. The element represents the individual corresponding to the The flow distribution value of each node; if the flow distribution value of any node in the position vector of an individual during the search process is less than the lower limit of the node flow corresponding to the node , it will be corrected to the lower limit of the node flow; if the flow distribution value is greater than the upper limit of the node flow corresponding to the node , then correct it to the upper limit of node flow; S53. Initialize the iterative control parameters of the improved Hunger Game search algorithm, which include the maximum number of iterations and the dynamic hunger factor attenuation coefficient. : ； in, Indicates the node heat load prediction results The difference between the actual result of node heat load and the actual result of node heat load at time The prediction error, represents the allowed node heat load prediction error threshold, represents the prediction error sensitivity coefficient; S54. Execute the iterative calculation process of the improved Hunger Game search algorithm, in the In the iteration process, the local load mapping matrix between nodes is used Individual position adaptive update driven jointly with hunger state factor: ； in, For the The first iteration Individuals correspond to The updated value of traffic distribution of nodes, For the The first iteration Individuals correspond to The current traffic distribution value of each node, For the The best individual in the current population at the iteration corresponds to The traffic distribution value of each node, For the The first iteration The hunger state factor of the individual is obtained by updating the hunger state factor of the pth individual according to the dynamic hunger factor attenuation coefficient. The optimization initiative weight coefficient is determined dynamically by combining the individual historical optimal position and the local load intensity index; S55. Call the node heat load prediction results in each iteration process And the enhanced local load mapping matrix structure , evaluate the fitness function value of each individual : ； in, Indicates the The first iteration The fitness function value of each individual is used to evaluate the performance of the node traffic allocation scheme; S56. When the number of iterations reaches the maximum number of iterations or the population fitness convergence condition is met, the iteration is terminated and the optimal individual position in the current population is output. As a microgrid energy heating network at the predicted time The optimal traffic distribution scheme .

9. The method for regulating flow of energy heating pipe network based on machine learning according to claim 5, characterized in that: The S6 comprises the following steps: S61. The optimal traffic allocation solution obtained by optimizing the improved Hunger Game search algorithm As the target regulation benchmark, the flow regulation instruction set containing the flow control parameters of each node is parsed and generated. The flow regulation instruction set includes the target flow value, regulation rate reference value and execution timestamp of each node; S62. Each flow regulation instruction structure definition includes a node number parameter , target flow value , Current adjustment reference value , Allows adjustment of the maximum stride Time stamp for issuing ; S63. Compare target flow value With the current adjustment reference value , judge whether the node meets the adjustment execution conditions, if , then the node is marked as needing adjustment, where is the minimum effective adjustment threshold; S64. Execute the phased adjustment plan for the nodes marked as requiring adjustment, according to the local load intensity of the nodes. Determine the priority of the adjustment step, give priority to assigning large adjustment steps to nodes with high load intensity, and form a hierarchical adjustment strategy, so that the response of key nodes is prioritized and the adjustment of non-key nodes is restricted; S65. The heating network execution unit receives and interprets the corresponding node flow regulation instruction. Based on the control characteristics of the node regulation valve, electric actuator, or variable frequency pump hardware equipment, it drives the hardware module to perform physical regulation operations on the flow state of the target node until the actual flow value gradually converges to the target flow value. , completing the local flow adaptive regulation of the microgrid energy heating network.

10. The method for regulating flow of energy heating pipe network based on machine learning according to claim 1, characterized in that: The dynamic optimization control of the local flow of the microgrid energy heating pipeline network includes dynamically adjusting the weight parameters of the multi-objective optimization function and the control parameters of the improved Hunger Game search algorithm according to the real-time prediction error and the system energy consumption index, so that the prediction-optimization-regulation closed loop continues to adaptively iterate.

Images