A distributed energy metering method and system based on multi-energy flow fusion

Abstract

The application belongs to the technical field of distributed energy management, and provides a distributed energy metering method and system based on multi-energy flow fusion, which comprises the following steps: deploying a sensor network to collect energy flow data sequences of each energy node in a multi-energy flow distributed energy network; dividing the multi-energy flow distributed energy network into sub-networks; performing fluctuation analysis on the energy nodes; dividing the energy nodes into fluctuation energy nodes and stable energy nodes; for the fluctuation energy nodes, constructing and training an LSTM prediction model, predicting the energy flow data vector of the fluctuation energy nodes at the current time node, calculating the prediction error, and performing dynamic compensation by using an adaptive dynamic compensation algorithm to obtain the to-be-fused energy flow data vector of each energy node; constructing a sub-network energy flow collaborative fusion model; fusing the to-be-fused energy flow data vector in the sub-network; modifying the fusion result based on global energy balance; and calculating the metering index of the multi-energy flow distributed energy network.

Description

Technical Field

[0001] This invention belongs to the field of distributed energy management technology, specifically a distributed energy metering method and system based on multi-energy flow fusion. Background Technology

[0002] With the advancement of the global energy transition, distributed energy networks have become a research hotspot in the energy field due to their high efficiency, flexibility, and low carbon characteristics. These networks typically include energy nodes that input, convert, and store multiple energy forms such as electricity, heat, hydrogen, and wind power, forming a complex system with multiple energy flows coupled together. Against this backdrop, accurately measuring the energy flow and conversion efficiency of multiple energy flows has become a core requirement for optimizing system operation and improving energy utilization.

[0003] In existing technologies, multi-energy flow data acquisition typically employs independently deployed sensor networks. Sensors of different energy types lack a collaborative mechanism, resulting in insufficient handling of data heterogeneity. The units and acquisition frequencies of data such as electrical energy and thermal energy are not consistent, and the data of multiple types are not integrated in a time-series synchronization manner, making it impossible to capture the dynamic correlations in the energy conversion process.

[0004] Traditional distributed energy networks are often divided into subnetworks based on topological connections, neglecting spatial location correlations. The physical distance between energy nodes directly affects energy transmission loss and coupling strength, but existing division methods do not consider Euclidean spatial distances, resulting in weak physical correlations between energy nodes within subnetworks. The volatility quantification methods are also crude, and the judgment of energy node data volatility often uses fixed thresholds or simple statistics without considering the relative degree of data volatility, making it difficult to adapt to the characteristic differences of different types of energy nodes.

[0005] On the other hand, for fluctuating energy nodes, such as wind energy nodes and photovoltaic energy nodes, existing technologies lack effective means to handle dynamic drift. Traditional prediction models have poor adaptability, and models such as linear regression and ARIMA are difficult to capture the nonlinear dynamic characteristics of energy flow data. In strong fluctuation scenarios, the prediction error is large, the static compensation mechanism is lagging, and the method of using fixed compensation parameters cannot track the data drift trend in real time, resulting in the measurement results deviating from the actual value.

[0006] In terms of multi-energy flow fusion metering, existing technologies suffer from insufficient modeling of energy coupling relationships. Traditional numerical averaging or weighted fusion methods do not consider the topological connections and energy transmission paths between energy nodes, fail to capture the coupling characteristics between energy nodes, lack global energy conservation, and lack a systematic balance correction mechanism.

[0007] To address the above problems, this invention proposes a distributed energy metering method and system based on multi-energy flow fusion. Summary of the Invention

[0008] In order to overcome the shortcomings of the prior art, at least one technical problem raised in the background art is solved.

[0009] The technical solution adopted by this invention to solve the technical problem is: a distributed energy metering method based on multi-energy flow fusion, comprising:

[0010] Deploy a sensor network to collect energy flow data vectors from each energy node in a multi-energy flow distributed energy network, and integrate them according to time sequence to obtain the energy flow data sequence of the energy node.

[0011] The multi-energy flow distributed energy network is divided into sub-networks. Based on the energy flow data sequence of energy nodes in the sub-networks, the volatility of energy nodes is analyzed, and the energy nodes are divided into fluctuating energy nodes and stable energy nodes. For fluctuating energy nodes, dynamic drift analysis is triggered.

[0012] If dynamic drift analysis is triggered, an LSTM prediction model is constructed and trained to predict the energy flow data vector of the fluctuating energy node at the current time node. The prediction error is calculated and dynamically compensated using an adaptive dynamic compensation algorithm to obtain the energy flow data vector to be merged for each energy node.

[0013] A sub-network energy flow collaborative fusion model is constructed to fuse the energy flow data vectors to be fused within the sub-network, and the fusion result is corrected based on global energy balance. The measurement indicators of the multi-energy flow distributed energy network are then calculated.

[0014] The sub-network is divided as follows:

[0015] A cohesive hierarchical clustering algorithm is adopted to treat each energy node in the multi-energy flow distributed energy network as a subnetwork. Connected energy node pairs are obtained and filtered according to the energy node connection matrix. The connected energy node pairs with the smallest spatial distance are obtained. The two subnetworks containing connected energy node pairs are merged into one subnetwork. The distance between the subnetworks is recalculated and merged.

[0016] For any subnetwork, if the number of energy nodes in the subnetwork reaches the preset upper limit, or the spatial distance between the subnetwork and all other subnetworks is greater than the preset distance threshold, it is determined that the subnetwork has met the partitioning conditions. In each iteration, only the subnetworks that do not meet the partitioning conditions are iteratively clustered until all subnetworks meet the partitioning conditions.

[0017] The energy node connection matrix is ​​obtained as follows:

[0018] Based on the topological connection relationship of the multi-energy flow distributed energy network, an energy node connection matrix is ​​constructed. If two energy nodes are directly connected, they are determined to be a connected energy node pair, and the corresponding value of the connected energy node pair in the energy node connection matrix is ​​1. Otherwise, the corresponding value of the two energy nodes in the energy node connection matrix is ​​0, and the energy nodes themselves are not connected.

[0019] The method for acquiring the fluctuating energy node is as follows:

[0020] The energy flow data vector includes energy input, output effective energy, and recovered energy. For any energy node, the energy input of the energy node is acquired, and the coefficient of variation of the energy input during the data acquisition period is calculated. If the coefficient of variation is greater than the preset coefficient of variation threshold, the energy node is classified as a fluctuating energy node.

[0021] The prediction error is obtained as follows:

[0022] For fluctuating energy nodes, the energy flow data sequence of the fluctuating energy nodes during the data acquisition period is obtained. The part of the energy flow data sequence that does not include the current time node is extracted and normalized to obtain the normalized training sequence. In the normalized training sequence, the normalized historical window is extracted.

[0023] An LSTM prediction model is constructed and trained using a normalized training sequence. The normalized historical window is input into the trained LSTM prediction model, and the output results are denormalized to obtain the predicted energy flow data vector. The energy flow data vector of the energy node at the current time node is obtained and the predicted energy flow data vector is compared with the predicted energy flow data vector, and the deviation is calculated to obtain the prediction error.

[0024] The method for obtaining the energy flow data vector to be fused is as follows:

[0025] For fluctuating energy nodes, an adaptive dynamic compensation algorithm is adopted. Adaptive compensation parameters are designed, a sliding window is selected, and the standard deviation of the prediction error of the energy nodes within the sliding window is obtained and calculated. The adaptive compensation parameters are updated based on the standard deviation, and the predicted energy flow data vector is compensated using the adaptive compensation parameters to obtain the corrected predicted energy flow data vector. The energy flow data vector to be fused for the fluctuating energy nodes is the corrected predicted energy flow data vector.

[0026] For a stable energy node, the energy flow data vector to be merged for the stable energy node is the energy flow data vector of the current time node.

[0027] The calculation method for the aforementioned measurement indicators is as follows:

[0028] Metrics for multi-energy flow distributed energy networks include energy node-level metrics and sub-network-level metrics.

[0029] For energy node-level indicators, the standard energy flow data vector of each network energy node is obtained and processed to obtain the real-time conversion efficiency of each energy node. For sub-network-level indicators, the corrected sub-network energy flow feature vector of each network is obtained and processed to obtain the energy utilization rate of each sub-network.

[0030] The standard energy flow data vector is obtained as follows:

[0031] The energy flow data vectors to be merged at each energy node at the current time node are standardized by energy units, and different forms of energy are uniformly converted into standard coal equivalent. The conversion coefficient is set according to the general rules for comprehensive energy consumption calculation, and the standard energy flow data vector after energy unit standardization is obtained.

[0032] The modified sub-network energy flow feature vector is obtained as follows:

[0033] The energy node connection matrix of each energy node in the sub-network is obtained to determine the energy transmission path. A sub-network energy flow collaborative fusion model is constructed. The sub-network energy flow collaborative fusion model adopts a graph convolutional network. The standard energy flow data vector is forward propagated through a multi-layer GCN network to fuse the standard energy flow data vectors in the sub-network. The fused sub-network energy flow feature vector is output. The fusion results of all sub-networks are globally integrated. Energy conservation constraints are introduced to construct a global energy balance. If there is an error in the global energy balance, an adaptive weight allocation algorithm is used to correct the error of the sub-network energy flow feature vector of each sub-network to obtain the corrected sub-network energy flow feature vector.

[0034] A distributed energy metering system based on multi-energy flow fusion includes the following modules:

[0035] Acquisition Module: Deploys a sensor network to collect energy flow data vectors from each energy node in the multi-energy flow distributed energy network, and integrates them according to time sequence to obtain the energy flow data sequence of the energy node.

[0036] Node partitioning module: Divides the multi-energy flow distributed energy network into sub-networks. Based on the energy flow data sequence of energy nodes in the sub-networks, it performs volatility analysis on the energy nodes and classifies them into fluctuating energy nodes and stable energy nodes. For fluctuating energy nodes, it triggers dynamic drift analysis.

[0037] Dynamic compensation module: If dynamic drift analysis is triggered, an LSTM prediction model is constructed and trained to predict the energy flow data vector of the fluctuating energy node at the current time node, the prediction error is calculated and an adaptive dynamic compensation algorithm is used for dynamic compensation to obtain the energy flow data vector to be merged for each energy node.

[0038] Metering module: Constructs a sub-network energy flow collaborative fusion model, fuses the energy flow data vectors to be fused within the sub-network, corrects the fusion results based on global energy balance, and calculates the metering indicators of the multi-energy flow distributed energy network.

[0039] The beneficial effects of this invention are as follows:

[0040] 1. This invention comprehensively collects energy flow data vectors from each energy node in a multi-energy flow distributed energy network by deploying a sensor network, and integrates them into a sequence according to time sequence, laying a solid data foundation for accurate metering. Detailed fluctuation analysis of energy nodes can accurately distinguish between fluctuating and stable energy nodes. Dynamic drift analysis is carried out for fluctuating energy nodes, and LSTM prediction models and adaptive dynamic compensation algorithms are used to effectively correct data deviations, greatly improving the accuracy of energy flow data from each energy node. This makes the overall distributed energy metering results more accurate and reliable, providing a strong basis for energy management decisions.

[0041] 2. This invention constructs a sub-network energy flow collaborative fusion model based on the energy flow data vectors to be fused from energy nodes within the sub-network. This model fully considers the complex relationships between different energy forms and achieves the scientific fusion of standard energy flow data vectors. The metering indicators of the multi-energy flow distributed energy network calculated based on the fusion results can accurately reflect the operating status of the energy system, such as energy efficiency and losses. This helps energy managers accurately grasp energy usage, optimize energy allocation, reduce energy losses, and improve energy utilization efficiency. Simultaneously, it enhances the stability and reliability of the multi-energy flow distributed energy network, promoting the sustainable development of the energy system. Attached Figure Description

[0042] The invention will now be further described with reference to the accompanying drawings.

[0043] Figure 1 This is a flowchart illustrating the steps of a distributed energy metering method based on multi-energy flow fusion as described in an embodiment of the present invention.

[0044] Figure 2 This is a system module architecture diagram of a distributed energy metering system based on multi-energy flow fusion, as described in an embodiment of the present invention. Detailed Implementation

[0045] To make the technical means, creative features, objectives and effects of this invention easier to understand, the invention will be further described below in conjunction with specific embodiments.

[0046] Example 1

[0047] Please see Figure 1 As shown in the embodiment of the present invention, a distributed energy metering method based on multi-energy flow fusion includes the following steps:

[0048] S1: Deploy a sensor network to collect energy flow data vectors from each energy node in the multi-energy flow distributed energy network, and integrate them according to the time sequence to obtain the energy flow data sequence of the energy node.

[0049] In multi-energy flow distributed energy networks, corresponding sensors are deployed for different types of energy nodes to construct a sensor network. Energy nodes represent the basic units of energy input and conversion, including electrical energy nodes, thermal energy nodes, hydrogen energy nodes, wind energy nodes, etc. Through the constructed sensor network, energy flow data from each energy node is collected at preset fixed time intervals during the data acquisition period, and the data is integrated to obtain the energy flow data vector of the energy node. , where t represents the time node and i represents the energy node number.

[0050] ;

[0051] The data collection period refers to a fixed duration that ends at the current time point, and the data collection period moves over time.

[0052] in, The energy input energy of energy node i at time node t is obtained by collecting and calculating data from sensors deployed at the energy node's inlet. For example, electrical energy input is measured by a power sensor, and thermal energy input is measured and calculated by a combination of a temperature sensor and a flow sensor. The role of energy input energy is to reflect the energy input status of the energy node at the current time.

[0053] in, This represents the effective energy output of energy node i at time node t, which is obtained by collecting and calculating data from sensors deployed at the output port of the energy node. This represents the recovered energy of energy node i at time node t, which is measured by sensors deployed on the recovery device.

[0054] During the data acquisition period, for each energy node numbered i, all acquired energy flow data vectors will be... The energy flow data sequence is obtained by integrating the data in chronological order.

[0055] It should be noted that the purpose of this step is to deploy a sensor network to collect energy flow data from each energy node in the multi-energy flow distributed energy network in real time, and integrate it into a data sequence according to the time series. This provides basic time series data support for subsequent analysis, covers multiple types of energy nodes, realizes unified collection of all types of energy data, and ensures the temporal continuity of the data through time series integration, providing a time dimension basis for dynamic analysis.

[0056] S2: Divide the multi-energy flow distributed energy network into sub-networks. Based on the energy flow data sequence of energy nodes in the sub-networks, perform volatility analysis on the energy nodes and divide the energy nodes into fluctuating energy nodes and stable energy nodes. For fluctuating energy nodes, trigger dynamic drift analysis.

[0057] In a multi-energy flow distributed energy network, the network is divided into several sub-networks based on the network connection and spatial location relationships of the energy nodes.

[0058] Specifically, regarding the network connectivity of energy nodes, by analyzing the topology of the multi-energy flow distributed energy network, it is determined whether there are direct energy transmission lines or communication links between the energy nodes, thus obtaining the topological connectivity of the multi-energy flow distributed energy network.

[0059] Based on the topological connections of multi-energy flow distributed energy networks, an energy node connection matrix is ​​constructed. If energy node i is directly connected to energy node j, then the two energy nodes are considered a connected energy node pair.

[0060] ;

[0061] otherwise:

[0062] ;

[0063] Since the energy node itself has no connections, when i=j:

[0064] ;

[0065] Based on the spatial relationship of the energy nodes, the spatial coordinates of two energy nodes are obtained, and the spatial distance between the energy nodes is calculated using the Euclidean distance formula. The spatial distance between energy node i and energy node j is... .

[0066] A cohesive hierarchical clustering algorithm is used to partition the multi-energy flow distributed energy network into subnetworks. Initially, each energy node in the multi-energy flow distributed energy network is considered an independent subnetwork based on the energy node connection matrix. Filter out connected energy node pairs and obtain the spatial distance between energy node i and energy node j in all connected energy node pairs. The nodes are sorted from smallest to largest, and the two connected energy nodes contained in the closest spatial distance are obtained. The subnetworks containing the two energy nodes are merged into one subnetwork, and the distance between the subnetworks is recalculated and merged.

[0067] For any subnetwork, if the number of energy nodes in the subnetwork reaches the preset upper limit, or the spatial distance between the subnetwork and all other subnetworks is greater than the preset distance threshold, it is determined that the subnetwork has met the partitioning conditions. In the next iteration, subnetworks that meet the partitioning conditions will no longer be considered, and iterative clustering will continue only in subnetworks that do not meet the partitioning conditions until all subnetworks meet the partitioning conditions, at which point the subnetwork partitioning is determined to be complete.

[0068] After completing the subnetwork partitioning, based on any subnetwork, a volatility analysis is performed on each energy node within the subnetwork, and the energy nodes are divided into fluctuating energy nodes and stable energy nodes.

[0069] Specifically, for any energy node numbered i, obtain the energy input energy of energy node numbered i. Calculate the coefficient of variation of energy input during the data acquisition period. The formula is:

[0070] ;

[0071] Where n represents the number of time points within the data collection period. This represents the average energy input energy of energy node i during the data collection period.

[0072] It should be noted that the coefficient of variation is used to directly reflect the relative fluctuation of the energy input of the energy node during the data acquisition period.

[0073] If the coefficient of variation of an energy node at the current time point is greater than the preset coefficient of variation threshold, the energy input of the energy node is judged to have a high degree of relative fluctuation during the data collection period, and the energy node is a fluctuating energy node; otherwise, the energy node is judged to be a stable energy node.

[0074] If there are fluctuating energy nodes in the subnetwork, dynamic drift analysis is triggered for these fluctuating energy nodes.

[0075] It should be noted that in multi-energy flow distributed energy networks, the energy input of fluctuating energy nodes has a high degree of relative fluctuation during the data acquisition period. Therefore, fluctuating energy nodes are prone to dynamic drift of energy flow data, which affects the energy metering of fluctuating energy nodes.

[0076] It should be noted that this step divides the multi-energy flow distributed energy network into subnetworks based on topological connectivity and spatial location. By analyzing the volatility of energy nodes using the coefficient of variation, it identifies volatile energy nodes that require key monitoring, triggers dynamic drift analysis, reduces network complexity, focuses on local energy coupling relationships, and improves analysis efficiency. Volatility analysis accurately locates easily drifting energy nodes, avoids redundant calculations for stable energy nodes, optimizes resource allocation, and integrates agglomerative hierarchical clustering algorithm that combines the topological connectivity matrix and Euclidean distance. It introduces spatial location relationships into the energy network sub-partitioning, which is more consistent with physical transmission characteristics than traditional topological partitioning. The use of the coefficient of variation to quantify the degree of volatility replaces traditional threshold judgment, improving the objectivity of energy node classification.

[0077] S3: If dynamic drift analysis is triggered, construct and train an LSTM prediction model to predict the energy flow data vector of the fluctuating energy node at the current time node, calculate the prediction error and use an adaptive dynamic compensation algorithm to perform dynamic compensation, and obtain the energy flow data vector to be merged for each energy node.

[0078] If dynamic drift analysis is triggered, for fluctuating energy nodes, an LSTM prediction model is constructed and trained to predict the energy flow data vector of the fluctuating energy node at the current time node.

[0079] Specifically, an LSTM network architecture is constructed for the LSTM prediction model, including an input layer, an LSTM layer, a fully connected layer, and an output layer. The input layer has a dimension of 3 and is responsible for receiving the preprocessed energy flow data sequence. The LSTM layer, as the core of the entire LSTM prediction model, outputs a feature vector and has 128 hidden units. The feature vector output by the LSTM layer is mapped to a 3D space through the fully connected layer. The fully connected layer connects the output of the LSTM layer with the energy nodes of the output layer through a weight matrix to achieve further feature transformation. The input layer has a dimension of 3 and is responsible for outputting the predicted energy flow data vector. The mean squared error (MSE) is used as the loss function.

[0080] The energy flow data sequence of the fluctuating energy node during the data acquisition period is obtained. The part of the energy flow data sequence excluding the current time node is extracted and marked as the training sequence. Within the training sequence, normalization is performed on each component of the energy flow data vector to obtain the normalized training sequence. The normalized training sequence is divided into a training set and a validation set in an 8:2 ratio. The training set is used to learn the parameters of the LSTM prediction model, and the validation set is used to monitor the performance of the LSTM prediction model and adjust hyperparameters during the training process to avoid overfitting.

[0081] The LSTM prediction model is trained using a training set, and the Adam optimizer is used to update the parameters of the LSTM prediction model. An initial learning rate is set, and the learning rate is dynamically adjusted according to the loss function of the validation set during training. To prevent the LSTM prediction model from overfitting during training, an early stopping mechanism is introduced. If the loss function of the validation set does not decrease in consecutive iterations or reaches the preset number of iterations, the LSTM prediction model is judged to have completed prediction.

[0082] In the normalized training sequence, the energy flow data vectors of the m time nodes before the current time node are extracted and integrated to obtain the normalized historical window.

[0083] Based on the trained LSTM prediction model, a normalized historical window is input into the trained LSTM prediction model, and the output is denormalized to obtain the predicted energy flow data vector for the current time point. Where i represents the number of the fluctuating energy node in the energy node list. This indicates the current time point;

[0084] Define prediction error ;

[0085] ;

[0086] Prediction error represents the deviation between the acquired energy flow data vector and the predicted energy flow data vector, indirectly reflecting the dynamic drift of the energy flow data vector of the fluctuating energy node. An adaptive dynamic compensation algorithm is used to dynamically compensate the predicted energy flow data vector.

[0087] Specifically, design adaptive compensation parameters. Extract a sliding window of length w, ending at the previous time node. Obtain the prediction errors of the energy nodes at the w most recent time nodes within the sliding window, and calculate the standard deviation of the prediction errors within the sliding window. ;

[0088] The formula for updating the adaptive compensation parameters is:

[0089] ;

[0090] in, This indicates the set adaptive learning rate. This represents the preset reference error standard deviation.

[0091] Using adaptive compensation parameters For predicted energy flow data vectors Compensation is performed to obtain the corrected energy flow data vector. ;

[0092] ;

[0093] The energy flow data vector to be fused from the fluctuating energy node is obtained as the corrected energy flow data vector:

[0094] ;

[0095] Since the energy flow data of stable energy nodes is relatively stable, the energy flow data vector to be fused for stable energy nodes is the energy flow data vector collected at the current time node:

[0096] ;

[0097] It should be noted that the purpose of this step is to construct an LSTM model to predict energy flow data for fluctuating energy nodes, and to use an adaptive compensation algorithm to correct the data to ensure measurement accuracy. The LSTM model captures the time series dependence of energy flow data, improving the prediction accuracy in dynamic scenarios. The adaptive compensation parameters are dynamically adjusted according to historical errors to adapt to data drift characteristics in real time and reduce measurement deviations. The LSTM neural network is introduced into the prediction of energy node data drift. Combined with the early stopping mechanism and Adam optimizer, the problem of insufficient adaptability of traditional statistical models to nonlinear drift is solved. An adaptive compensation algorithm based on the standard deviation of sliding window error is proposed to achieve real-time error correction in dynamic scenarios.

[0098] S4: Construct a sub-network energy flow collaborative fusion model, fuse the energy flow data vectors to be fused within the sub-network, correct the fusion results based on global energy balance, and calculate the measurement indicators of the multi-energy flow distributed energy network.

[0099] For each sub-network, the energy flow data vector to be merged at the current time node of each energy node is standardized by energy unit, and different forms of energy such as electrical energy, thermal energy, and hydrogen energy are uniformly converted into standard coal equivalent. The conversion coefficient is set according to the general rules for comprehensive energy consumption calculation, and the standard energy flow data vector after energy unit standardization is obtained.

[0100] ;

[0101] ;

[0102] Among them, the energy node numbered i is the r-th energy node in the subnetwork s. This represents the standard energy flow data vector of the r-th energy node in subnetwork s. This represents the conversion coefficient corresponding to energy node number i. , , These represent standard energy flow data vectors. The standard energy input, standard output effective energy, and standard recovered energy are defined in this context.

[0103] The energy node connection matrix of each energy node in the sub-network is obtained to determine the energy transmission path. The sub-network energy flow collaborative fusion model is constructed. The sub-network energy flow collaborative fusion model uses a graph convolutional network (GCN) to fuse the standard energy flow data vectors in the sub-network.

[0104] Specifically, a graph convolutional network (GCN) is used to perform neighborhood aggregation on the standard energy flow data vector to capture the energy coupling characteristics between energy nodes. The formula is as follows:

[0105] ;

[0106] in, This represents the adjacency matrix with self-loops. Represents the identity matrix. Degree matrix, Indicates the first The energy flow eigenvector of the layer, Initially, it is a standard energy flow data vector. This represents a trainable weight matrix. This represents the ReLU activation function.

[0107] The standard energy flow data vector is forward-propagated through a multi-layer GCN network, outputting the fused sub-network energy flow feature vectors. The fusion results of all sub-networks are then globally integrated, and an energy conservation constraint is introduced to construct a global energy balance. If there are errors in the global energy balance, an adaptive weight allocation algorithm is used to correct the errors in the sub-network energy flow feature vectors of each sub-network, resulting in the corrected sub-network energy flow feature vectors. .

[0108] ;

[0109] in, , , These represent the total input energy, total output energy, and total recovery energy of the subnetwork, respectively.

[0110] Based on the standard energy flow data vectors of each network energy node and the corrected sub-network energy flow feature vectors, measurement indicators for multi-energy flow distributed energy networks are generated, including energy node-level indicators and sub-network-level indicators.

[0111] Among them, the energy node-level index is the real-time conversion efficiency of a single energy node. For the r-th energy node in the subnetwork s, the real-time conversion efficiency is calculated. The formula is:

[0112] ;

[0113] Among them, the sub-network level index is the energy utilization rate of a single sub-network. For sub-network s, the energy utilization rate is calculated. The formula is:

[0114] ;

[0115] By calculating the metering indicators of multi-energy flow distributed energy networks, distributed energy metering based on multi-energy flow fusion is realized.

[0116] It should be noted that the purpose of this step is to integrate the standard energy flow data of the sub-network using GCN, introduce energy conservation constraints to correct the global balance, generate energy node-level indicators and sub-network-level indicators, realize multi-energy flow fusion measurement, standardize energy units and unify multi-energy flow measurement dimensions to facilitate cross-type energy comparison, capture the energy coupling characteristics between energy nodes using graph convolutional networks, and combine them with global energy conservation constraints to ensure the physical consistency of the fusion results. Multi-dimensional measurement indicators provide quantitative basis for system optimization. By applying graph convolutional networks to multi-energy flow fusion and using network topology to enhance the modeling of energy coupling relationships, the limitations of traditional numerical fusion methods are overcome, and a two-layer measurement framework of local fusion and global conservation is constructed to solve the consistency problem between local data and global balance in distributed energy networks.

[0117] The technical solution of this invention is as follows: A sensor network is deployed to collect energy flow data vectors from each energy node in a multi-energy flow distributed energy network. These data vectors are then integrated according to time series to obtain energy flow data sequences for the energy nodes. The multi-energy flow distributed energy network is divided into sub-networks. Based on the energy flow data sequences of the energy nodes in the sub-networks, volatility analysis is performed on the energy nodes, classifying them into fluctuating energy nodes and stable energy nodes. For fluctuating energy nodes, dynamic drift analysis is triggered. If dynamic drift analysis is triggered, an LSTM prediction model is constructed and trained to predict the energy flow data vectors of fluctuating energy nodes at the current time node. The prediction error is calculated and dynamically compensated using an adaptive dynamic compensation algorithm to obtain the energy flow data vectors to be fused for each energy node. A sub-network energy flow collaborative fusion model is constructed to fuse the energy flow data vectors to be fused within the sub-network. The fusion result is corrected based on global energy balance, and the measurement indicators of the multi-energy flow distributed energy network are calculated.

[0118] Example 2

[0119] like Figure 2 As shown in the embodiment of the present invention, a distributed energy metering system based on multi-energy flow fusion includes the following modules:

[0120] Acquisition Module: Deploys a sensor network to collect energy flow data vectors from each energy node in the multi-energy flow distributed energy network, and integrates them according to time sequence to obtain the energy flow data sequence of the energy node.

[0121] Node partitioning module: Divides the multi-energy flow distributed energy network into sub-networks. Based on the energy flow data sequence of energy nodes in the sub-networks, it performs volatility analysis on the energy nodes and classifies them into fluctuating energy nodes and stable energy nodes. For fluctuating energy nodes, it triggers dynamic drift analysis.

[0122] Dynamic compensation module: If dynamic drift analysis is triggered, an LSTM prediction model is constructed and trained to predict the energy flow data vector of the fluctuating energy node at the current time node, the prediction error is calculated and an adaptive dynamic compensation algorithm is used for dynamic compensation to obtain the energy flow data vector to be merged for each energy node.

[0123] Metering module: Constructs a sub-network energy flow collaborative fusion model, fuses the energy flow data vectors to be fused within the sub-network, corrects the fusion results based on global energy balance, and calculates the metering indicators of the multi-energy flow distributed energy network.

[0124] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A distributed energy metering method based on multi-energy flow fusion, characterized in that: include: Deploy a sensor network to collect energy flow data vectors from each energy node in a multi-energy flow distributed energy network, and integrate them according to time sequence to obtain the energy flow data sequence of the energy node; The multi-energy flow distributed energy network is divided into sub-networks. Based on the energy flow data sequence of energy nodes in the sub-networks, the volatility of energy nodes is analyzed. Energy nodes are divided into fluctuating energy nodes and stable energy nodes. For fluctuating energy nodes, dynamic drift analysis is triggered. If dynamic drift analysis is triggered, an LSTM prediction model is constructed and trained to predict the energy flow data vector of the fluctuating energy node at the current time node. The prediction error is calculated and an adaptive dynamic compensation algorithm is used for dynamic compensation to obtain the energy flow data vector to be fused for each energy node. A sub-network energy flow collaborative fusion model is constructed to fuse the energy flow data vectors to be fused within the sub-network, and the fusion result is corrected based on the global energy balance to calculate the measurement indicators of the multi-energy flow distributed energy network. The sub-network is divided as follows: A cohesive hierarchical clustering algorithm is used to treat each energy node in the multi-energy flow distributed energy network as a subnetwork. Connected energy node pairs are obtained and filtered according to the energy node connection matrix. The connected energy node pairs with the smallest spatial distance are obtained. Two subnetworks containing connected energy node pairs are merged into one subnetwork. The distance between the subnetworks is recalculated and merged. For any subnetwork, if the number of energy nodes in the subnetwork reaches the preset limit, or the spatial distance between the subnetwork and all other subnetworks is greater than the preset distance threshold, it is determined that the subnetwork has met the partitioning conditions. In each iteration, only the subnetworks that do not meet the partitioning conditions are iteratively clustered until all subnetworks meet the partitioning conditions. The calculation method for the aforementioned measurement indicators is as follows: Metrics for multi-energy flow distributed energy networks include energy node-level metrics and sub-network-level metrics. For energy node-level indicators, the standard energy flow data vector of each network energy node is obtained and processed to obtain the real-time conversion efficiency of each energy node. For sub-network-level indicators, the corrected sub-network energy flow feature vector of each sub-network is obtained and processed to obtain the energy utilization rate of each sub-network. The modified subnetwork energy flow feature vector is obtained as follows: The energy node connection matrix of each energy node in the sub-network is obtained to determine the energy transmission path. A sub-network energy flow collaborative fusion model is constructed. The sub-network energy flow collaborative fusion model adopts a graph convolutional network. The standard energy flow data vector is forward propagated through a multi-layer GCN network to fuse the standard energy flow data vectors in the sub-network. The fused sub-network energy flow feature vector is output. The fusion results of all sub-networks are globally integrated. Energy conservation constraints are introduced to construct a global energy balance. If there is an error in the global energy balance, an adaptive weight allocation algorithm is used to correct the error of the sub-network energy flow feature vector of each sub-network to obtain the corrected sub-network energy flow feature vector.

2. The distributed energy metering method based on multi-energy flow fusion according to claim 1, characterized in that: The energy node connection matrix is ​​obtained as follows: Based on the topological connection relationship of the multi-energy flow distributed energy network, an energy node connection matrix is ​​constructed. If two energy nodes are directly connected, they are determined to be a connected energy node pair, and the corresponding value of the connected energy node pair in the energy node connection matrix is ​​1. Otherwise, the corresponding value of the two energy nodes in the energy node connection matrix is ​​0, and the energy nodes themselves are not connected.

3. The distributed energy metering method based on multi-energy flow fusion according to claim 1, characterized in that: The method for acquiring the fluctuating energy node is as follows: The energy flow data vector includes energy input, output effective energy, and recovered energy. For any energy node, the energy input of the energy node is acquired, and the coefficient of variation of the energy input during the data acquisition period is calculated. If the coefficient of variation is greater than the preset coefficient of variation threshold, the energy node is classified as a fluctuating energy node.

4. The distributed energy metering method based on multi-energy flow fusion according to claim 1, characterized in that: The prediction error is obtained as follows: For fluctuating energy nodes, the energy flow data sequence of the fluctuating energy nodes during the data acquisition period is obtained. The part of the energy flow data sequence that does not include the current time node is extracted and normalized to obtain the normalized training sequence. In the normalized training sequence, the normalized historical window is extracted. An LSTM prediction model is constructed and trained using a normalized training sequence. The normalized historical window is input into the trained LSTM prediction model, and the output results are denormalized to obtain the predicted energy flow data vector. The energy flow data vector of the energy node at the current time node is obtained and the predicted energy flow data vector is compared with the predicted energy flow data vector, and the deviation is calculated to obtain the prediction error.

5. A distributed energy metering method based on multi-energy flow fusion according to claim 4, characterized in that: The method for obtaining the energy flow data vector to be fused is as follows: For fluctuating energy nodes, an adaptive dynamic compensation algorithm is adopted. Adaptive compensation parameters are designed, a sliding window is extracted, the standard deviation of the prediction error of the energy nodes within the sliding window is obtained and calculated, the adaptive compensation parameters are updated based on the standard deviation, and the predicted energy flow data vector is compensated using the adaptive compensation parameters to obtain the corrected predicted energy flow data vector. The energy flow data vector to be fused for the fluctuating energy nodes is the corrected predicted energy flow data vector. For a stable energy node, the energy flow data vector to be merged for the stable energy node is the energy flow data vector of the current time node.

6. The distributed energy metering method based on multi-energy flow fusion according to claim 1, characterized in that: The standard energy flow data vector is obtained as follows: The energy flow data vectors to be merged at each energy node at the current time node are standardized by energy units, and different forms of energy are uniformly converted into standard coal equivalent. The conversion coefficient is set according to the general rules for comprehensive energy consumption calculation, and the standard energy flow data vector after energy unit standardization is obtained.

7. A distributed energy metering system based on multi-energy flow fusion, characterized in that: Includes the following modules: Acquisition module: Deploys a sensor network to acquire energy flow data vectors from each energy node in the multi-energy flow distributed energy network, and integrates them according to time series to obtain the energy flow data sequence of the energy node; Node partitioning module: Divides the multi-energy flow distributed energy network into sub-networks. Based on the energy flow data sequence of energy nodes in the sub-networks, it performs volatility analysis on the energy nodes and classifies them into fluctuating energy nodes and stable energy nodes. For fluctuating energy nodes, it triggers dynamic drift analysis. Dynamic compensation module: If dynamic drift analysis is triggered, an LSTM prediction model is constructed and trained to predict the energy flow data vector of the fluctuating energy node at the current time node, the prediction error is calculated and an adaptive dynamic compensation algorithm is used for dynamic compensation to obtain the energy flow data vector to be merged for each energy node. Metering module: Constructs a sub-network energy flow collaborative fusion model, fuses the energy flow data vectors to be fused within the sub-network, corrects the fusion results based on global energy balance, and calculates the metering indicators of the multi-energy flow distributed energy network; The sub-network is divided as follows: A cohesive hierarchical clustering algorithm is used to treat each energy node in the multi-energy flow distributed energy network as a subnetwork. Connected energy node pairs are obtained and filtered according to the energy node connection matrix. The connected energy node pairs with the smallest spatial distance are obtained. Two subnetworks containing connected energy node pairs are merged into one subnetwork. The distance between the subnetworks is recalculated and merged. For any subnetwork, if the number of energy nodes in the subnetwork reaches the preset limit, or the spatial distance between the subnetwork and all other subnetworks is greater than the preset distance threshold, it is determined that the subnetwork has met the partitioning conditions. In each iteration, only the subnetworks that do not meet the partitioning conditions are iteratively clustered until all subnetworks meet the partitioning conditions. The calculation method for the aforementioned measurement indicators is as follows: Metrics for multi-energy flow distributed energy networks include energy node-level metrics and sub-network-level metrics. For energy node-level indicators, the standard energy flow data vector of each network energy node is obtained and processed to obtain the real-time conversion efficiency of each energy node. For sub-network-level indicators, the corrected sub-network energy flow feature vector of each sub-network is obtained and processed to obtain the energy utilization rate of each sub-network. The modified subnetwork energy flow feature vector is obtained as follows: The energy node connection matrix of each energy node in the sub-network is obtained to determine the energy transmission path. A sub-network energy flow collaborative fusion model is constructed. The sub-network energy flow collaborative fusion model adopts a graph convolutional network. The standard energy flow data vector is forward propagated through a multi-layer GCN network to fuse the standard energy flow data vectors in the sub-network. The fused sub-network energy flow feature vector is output. The fusion results of all sub-networks are globally integrated. Energy conservation constraints are introduced to construct a global energy balance. If there is an error in the global energy balance, an adaptive weight allocation algorithm is used to correct the error of the sub-network energy flow feature vector of each sub-network to obtain the corrected sub-network energy flow feature vector.

Images