Real-time line loss analysis method and system based on cross-modal measurement data fusion

Abstract

The invention relates to a real-time line loss analysis method and system based on cross-modal measurement data fusion, and the method comprises the following steps: S1, collecting multi-source heterogeneous data, carrying out the preliminary quality screening of the collected original data through employing an improved isolation forest algorithm, and obtaining an original data set and a corresponding data quality preliminary evaluation label; s2, preprocessing the original data set according to the data quality preliminary evaluation label to obtain preprocessed data; s3, according to the preprocessed data, constructing a comprehensive feature vector and an importance score matrix of each feature; s4, deeply combining the physical law of the power system with artificial intelligence to construct a hybrid analysis model, and performing hybrid analysis based on the comprehensive feature vector and the importance scoring matrix to obtain an analysis result; and S5, the analysis result is converted into visual and visual display and operable decision suggestions. According to the invention, high-precision real-time analysis of power distribution network line loss is realized, and intelligence and accuracy of line loss management are effectively improved.

Description

Technical Field

[0001] The present invention relates to the field of line loss analysis, and in particular to a real-time line loss analysis method and system based on cross-modal measurement data fusion. Background Art

[0002] Distribution network line loss is a core indicator that impacts the economic efficiency and safety of power systems. With the widespread adoption of smart meters, sensor terminals, and various monitoring devices, all aspects of the power system generate massive, heterogeneous, and real-time measurement data, including multimodal information such as current, voltage, power, temperature, and fault alarms. This provides the data foundation for more refined and intelligent line loss analysis and management, but also brings the following challenges: First, multi-source data suffers from inconsistent formats, varying temporal and spatial resolutions, high levels of noise and anomalies, and uneven quality. Failure to effectively screen and integrate this heterogeneous data can easily lead to distorted line loss analysis results and even misleading decision-making. Second, traditional single algorithms often struggle to simultaneously account for the laws of power physics and the advantages of data-driven approaches, failing to effectively capture complex operating conditions and hidden causes of line losses. Furthermore, innovative solutions are urgently needed to intuitively present analysis results and effectively guide actual operations, maintenance, scheduling, and management. Summary of the Invention

[0003] In order to solve the above problems, the purpose of the present invention is to provide a real-time line loss analysis method and system based on cross-modal measurement data fusion, so as to realize high-precision real-time analysis of distribution network line losses and effectively improve the intelligence and accuracy of line loss management.

[0004] To achieve the above object, the present invention adopts the following technical solutions: A real-time line loss analysis method based on cross-modal measurement data fusion includes the following steps: S1: Collect multi-source heterogeneous data and use the improved isolation forest algorithm to perform preliminary quality screening on the collected raw data to obtain the raw data set and the corresponding preliminary data quality assessment labels; S2: Preprocess the original data set based on the preliminary data quality assessment label to obtain preprocessed data; S3: Based on the preprocessed data, construct a comprehensive feature vector and an importance score matrix for each feature; S4: Deeply integrate the physical laws of the power system with artificial intelligence to build a hybrid analysis model, and perform hybrid analysis based on the comprehensive eigenvector and importance score matrix to obtain analysis results; S5: Convert analysis results into intuitive visualizations and actionable decision recommendations.

[0005] Furthermore, multi-source heterogeneous data, including electrical operation data of SCADA systemscada , high-frequency phasor data D of the PMU system pmu , user-side data D of the electricity consumption information collection system ami , infrared and visible light image data D img , environmental monitoring data D env And the static topological data D of the GIS system gis ; The electrical operation data includes the voltage U of each node of the power grid scada , current I scada , active power P scada , reactive power Q scada The high-frequency phasor data of the PMU system includes the voltage phasor V of the substation and the line pmu , current phasor I pmu , phase angle θ pmu , frequency f pmu ; The environmental monitoring data includes temperature T env 、Humidity H env , wind speed W env ; Static topological data of GIS system, namely power network topology structure Topo gis , including static information of nodes, lines, and equipment parameters; The above data is automatically synchronized and archived through the ETL interface, and the original data set D raw The structure is as follows: D raw ={D scada ,D pmu ,D ami ,D img ,D env ,D gis}.

[0006] Furthermore, the improved isolation forest algorithm is used to perform preliminary quality screening on the collected raw data, as follows: According to the original data set obtained , n is the number of samples, each x i is the multimodal feature concatenation at the data collection moment i, including D scada ,D pmu ,D ami ,D img ,D env ,D gis ; Each feature j has a corresponding data quality weight w j ; Weighting the features: ; in, Represents element-wise multiplication; represents the feature j data collected at time i; When selecting the cutting point of the improved isolation forest algorithm, if j is a time series feature, a time correlation constraint is introduced: suppose the current feature j corresponds to the time series t, and the generation of the cutting point p considers not only the feature value range but also the time neighborhood relationship: ; Where r(t) is the time autocorrelation function; α is the adjustment coefficient; 、 are the minimum and maximum values ​​of feature j in the sample set respectively; U represents uniform distribution; In the tree construction process of the improved isolation forest algorithm, each tree is generated as follows: Build Input: Sample Subset , is the jth data in the sample subset, n is the amount of data;, the current tree depth l , maximum depth l max ; Output: an isolation tree; Termination condition: If |X|≤1 or l ≥ l max, then returns the leaf node, which records the number of samples |X|; Select Features: By probability Randomly select feature j, where Avgweight j is the average weight of feature j on X; Select the cutting point: If j is a time series feature, p is selected according to the above time-aware cutting point generation method; otherwise, and Choose p uniformly randomly between Split the data and get the left subtree data X left , right subtree data X right : ; Recursively construct subtrees, including left and right subtrees: Using weighted path length: At node s, if the segmentation is based on feature j, the cumulative path length when passing through node s is , where w j (x′) is the weight of the current sample x′ on feature j; Weighted path length calculation: ; Among them, Path(x′) is the path of sample x′ in a single weighted isolation tree, that is, the set of all segmentation steps from the root node to its leaf node; h dw (x′) is the weighted path length; is the weight value of the path from j to s under sample x′; After generating a forest of t′ trees, calculate the weighted average path length for the sample x′ : ; in, is the weighted path length of sample x′ on the kth weighted isolation tree; Normalized Anomaly Score: ; Where c(n′′) is the normalization factor of node n′′; Get the abnormality score s for each sample dw , and then set the threshold τ to determine abnormality:

[0007] Among them, Q is the final abnormality discrimination label.

[0008] Further preprocessing is as follows: According to the anomaly score and anomaly discrimination label, all data samples marked as severe anomalies are directly removed. The processed data is D filtered : ; in, is the tolerable floating space, which is used to retain some boundary samples to enhance the robustness of the model; Q i For data x i Abnormal discrimination label of ; To D filtered Each data x in i And each feature x ij Missing data were detected and imputed using mean values; Only in the anomaly score s dw When it is in the adjustment interval, the abnormal value is replaced by the context-based mean; and normalization is performed to obtain the preprocessed data set D pre . Furthermore, based on the preprocessed data, a comprehensive feature vector and an importance score matrix of each feature are constructed, as follows: According to the processed data set D preFor time series numerical data, statistical features including mean, extreme value, variance, slope, and kurtosis are extracted. For image data, a convolutional neural network model is used to extract key visual features, including fault hotspots and structural change features. For environmental and geographic information, trend features are extracted through statistical windows. Static or categorical data is vectorized and all normalized data is merged into a single comprehensive feature vector in a splicing manner. After obtaining the comprehensive feature vector, the specific contribution of each type of feature to the subsequent line loss prediction of the power system is quantified.

[0009] Furthermore, the physical laws of the power system are deeply integrated with artificial intelligence to build a hybrid analysis model, as follows: Define the physical model module, including the first physical model and the second physical model. Specifically: First physical model: For the i′th sample, the theoretical line loss is calculated based on physical quantities. :

[0010] in, is the k′th phase current of the i′th sample, is the corresponding resistance; K is the number of phases; Second physical model: Define voltage and current over-limit state function to constrain subsequent AI branch outputs:

[0011] in, The voltage value measured for the i′th sample; are the normal upper and lower bounds of voltage under physical constraints; is the indicator function; It is a voltage state constraint function, which returns 0 when normal and 1 when out of limit, and is used for loss function or early warning output; Weight the comprehensive feature vector z by the importance vector m to obtain the enhanced feature : ; Among them, z i′ is the comprehensive feature vector of the i′th sample; m is the feature importance weight vector; ⊙ is the element-by-element multiplication; Then the physical quantity is added as a feature to form a mixed feature vector : ; in, represents vector concatenation; p i′ is the physical feature vector of the i′th sample; Build a hybrid neural network model, including a two-branch neural network with a physical branch and an AI branch: The physical branch inputs the physical feature p i′ , through a fully connected layer FC phys Generating physical latent representations :

[0012] The AI ​​branch inputs importance-weighted features , through the multi-layer perceptron MLP AI Generating AI Latent Representations :

[0013] Fusion layer: concatenates the outputs of the two branches and generates the final task-related representation through a joint fully connected layer : ; Among them, W f 、b f are the weight and bias of the fully connected layer respectively; ReLU is the activation function; The output layer obtains the final line loss prediction value , anomaly detection probability and voltage prediction value : ; ; ; Among them, W loss 、b loss are the weight vector and bias of the line loss prediction output layer respectively; W o 、b o are the weight and bias of the anomaly detection output layer respectively; W volt 、b volt are the weights and bias of the voltage prediction output layer respectively.

[0014] Furthermore, the loss function of the hybrid analysis model includes task loss, physical consistency loss, and physical constraint loss, as follows: Mission loss L task : ; Among them, y i′ is the true value of sample i′; n′ is the number of samples; Physical consistency loss L consist : ; Physical constraint loss L constraint : ⋅ Among them, λ voltage is the physical constraint weight; Final loss function: L=L task +λ1L consist +λ2L constraint ; Among them, λ1 and λ2 are adjustable hyperparameters that control the embedding strength of physical laws.

[0015] Furthermore, the analysis results are transformed into intuitive visualizations and actionable decision-making recommendations, as follows: Perform structured summarization and template-driven automated presentation of hybrid analysis model results, including line loss prediction, anomaly detection, and voltage over-limit. Using line loss prediction, abnormal probability, and voltage over-limit indicators as input, a multi-level heat map is dynamically generated and displayed overlaid on the power grid geographic information map, allowing for intuitive visualization of the risk distribution of different substations and lines. Deep reinforcement learning is introduced to automatically generate the optimal response strategy based on the analysis results of the hybrid analysis model, historical data and power network simulation.

[0016] Furthermore, deep reinforcement learning is introduced to automatically generate the optimal response strategy based on the analysis results of the hybrid analysis model, historical data, and power network simulation. The details are as follows: Based on historical data and grid topology, a simulation environment is built to simulate grid operation status; State transition: grid state s t In action a t Transfer to s t+1 The dynamic process is realized by the simulator: s t+1 =E(s t ,a t ); Where E is the simulation engine; The state vector is represented as: ; in, are the line loss prediction value, abnormality detection probability, and voltage prediction value at the current time t respectively; I t is the line current, P t is the active power, V t is the node voltage, λ t is the load rate; H loss,t 、H anom,tRespectively represent the current line loss thermal value and abnormal thermal value of each area; Action a t Defined as a multi-level optimized response strategy, including: Instant Action : Load adjustment, switch operation, capacitor switching; Short-term maintenance : Distribute inspection work orders and spot inspection equipment lists chec ; Long-term transformation :Apply for technical transformation budget and plan upgrade area Z upgrade ; The action space is a hierarchical structure: ; Define the reward function r t : ; Among them, α1, α2, α3 are weight coefficients; Cost(a t ) is the action cost; V ref is the voltage reference value; ΔV is the voltage tolerance range; is the predicted line loss value at time t; is the abnormal event discrimination output at time t; is the predicted voltage value at time t; β′ is the weight coefficient of the control action cost term; Adopting a hierarchical reinforcement learning framework combined with the Actor-Critic algorithm: Critic network: evaluate the state value V(s t ;θv) or action value Q(s t ,a t ;θ q ); where θv and θq are the parameters of the state value function V and the action value function Q respectively; Actor network: outputs a hierarchical action policy π(a t ∣s t ;θ π ); Optimization goal: maximize the cumulative discount reward :

[0017] Where γ is the discount factor; is the expected function; T is the total time; Update formula: ; in, is the gradient of the objective function J with respect to the policy parameter θ; is the logarithmic policy gradient, indicating that in state s t Next select action a t The gradient of the log probability of with respect to the parameter θ; A(s t ,a t ) is the advantage function, estimated by TD-error: ; Among them, V(s t ) is the current state s t The value function estimate of .

[0018] A real-time line loss analysis system based on cross-modal measurement data fusion includes a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, it specifically performs the steps of the real-time line loss analysis method based on cross-modal measurement data fusion as described above.

[0019] The present invention has the following beneficial effects: 1. This invention innovatively introduces modal differentiation, feature weighting, and data quality weighting into the multi-source heterogeneous data structure of the power system, performs hierarchical standardization on data with different sampling frequencies and heterogeneous features, and effectively suppresses the interference of low-quality data in actual operation through weight allocation and weighted path length calculation, significantly reducing false detections and missed detections. 2. This invention explicitly embeds the physical laws of the power system, such as line loss theory calculation and state constraint criteria, into the AI ​​modeling process. This not only ensures the credibility and security of the model output at the physical level and avoids abnormal results that have no physical meaning, but also achieves a high degree of integration between data-driven and knowledge-driven models, making full use of domain prior knowledge. Through feature importance weighting and hybrid physical feature splicing, the model can adaptively highlight the information most relevant to business value. The weighted comprehensive feature vector enhances the model's ability to perceive anomalies and key business features, effectively improving the accuracy and robustness of the analysis. 3. The present invention not only presents information in an intuitive and visual form, including multi-dimensional interactive heat maps and large screens of core business indicators, but also generates actionable suggestion lists through automated text, providing all-round, hierarchical and personalized decision support for operation and maintenance personnel, and responding to potential risks in real time, proposing immediate, short-term and long-term optimization suggestions, which greatly improves the scientific and intelligent level of distribution network line loss management and operation and maintenance decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 Flow chart of the method of the present invention. DETAILED DESCRIPTION

[0021] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments: refer to Figure 1 In this embodiment, a real-time line loss analysis method based on cross-modal measurement data fusion is provided, comprising the following steps: S1: Collect multi-source heterogeneous data and use the improved isolation forest algorithm to perform preliminary quality screening on the collected raw data to obtain the raw data set and the corresponding preliminary data quality assessment labels; S2: Preprocess the original data set based on the preliminary data quality assessment label to obtain preprocessed data; S3: Based on the preprocessed data, construct a comprehensive feature vector and an importance score matrix for each feature; S4: Deeply integrate the physical laws of the power system with artificial intelligence to build a hybrid analysis model, and perform hybrid analysis based on the comprehensive eigenvector and importance score matrix to obtain analysis results; S5: Convert analysis results into intuitive visualizations and actionable decision recommendations.

[0022] In this embodiment, multi-source heterogeneous data, including electrical operation data D of the SCADA system scada (voltage, current, power, etc., sampling period 5-15 seconds), high-frequency phasor data D of the PMU system pmu (millisecond sampling), user-side data D of the electricity consumption information collection system ami (15 minutes sampling), infrared and visible light image data D img (hourly or daily sampling), environmental monitoring data D env (temperature, humidity, wind speed, etc., minute-level sampling) and static topological data of the GIS system D gis ; The electrical operation data includes the voltage U of each node of the power grid scada , current I scada , active power P scada , reactive power Q scada (Sampling period 5-15 seconds); The high-frequency phasor data of the PMU system includes the voltage phasor V of the substation and the line pmu , current phasor I pmu , phase angle θ pmu , frequency f pmu ; The environmental monitoring data includes temperature T env 、Humidity H env , wind speed W env ; Static topological data of GIS system, namely power network topology structure Topo gis, including static information of nodes, lines, and equipment parameters; The above data is automatically synchronized and archived through the ETL interface, and the original data set D raw The structure is as follows: D raw ={D scada ,D pmu ,D ami ,D img ,D env ,D gis}.

[0023] In this embodiment, the improved isolation forest algorithm is used to perform preliminary quality screening on the collected raw data, as follows: According to the original data set obtained , n is the number of samples, each x i is the multimodal feature concatenation at the data collection moment i, including D scada ,D pmu ,D ami ,D img ,D env ,D gis ; Each feature j has a corresponding data quality weight w j ; Weighting the features: ; in, Represents element-wise multiplication; represents the feature j data collected at time i; When selecting the cutting point of the improved isolation forest algorithm, if j is a time series feature, a time correlation constraint is introduced: suppose the current feature j corresponds to the time series t, and the generation of the cutting point p considers not only the feature value range but also the time neighborhood relationship: ; Where r(t) is the time autocorrelation function (e.g. ), t′ is the current time, β is the time correlation attenuation coefficient); α is the adjustment coefficient (e.g., α = 0.1); 、 are the minimum and maximum values ​​of feature j in the sample set respectively; U represents uniform distribution; In the tree construction process of the improved isolation forest algorithm, each tree is generated as follows: Build Input: Sample Subset , is the jth data in the sample subset, n is the amount of data; the current tree depth l , maximum depth l max ; Output: an isolation tree; Termination condition: If |X|≤1 or l ≥ l max, then returns the leaf node, which records the number of samples |X|; Select Features: By probability Randomly select feature j, where Avgweight j is the average weight of the feature on X; Select the cutting point: If j is a time series feature (such as voltage or current), p is selected according to the above time-aware cutting point generation method; otherwise, and Choose p uniformly randomly between Split the data and get the left subtree data X left , right subtree data X right : ; Recursively construct subtrees, including left and right subtrees: The path length h(x′) in the traditional isolation forest is defined as the number of edges from the root node to the leaf node where the sample is located.

[0024] This solution uses weighted path length: At node s, if the segmentation is based on feature j, the value of the cumulative path length when passing through the node is , where w j (x′) is the weight of the current sample x′ on feature j; Weighted path length calculation: ; Among them, Path(x′) is the path of sample x′ in a single weighted isolation tree, that is, the set of all segmentation steps from the root node to its leaf node; h dw (x′) is the weighted path length; is the weight value of the path from j to s under sample x′; After generating a forest of t′ trees, calculate the weighted average path length for the sample x′ : ; in, is the weighted path length of sample x′ on the kth weighted isolation tree; Normalized Anomaly Score: ; Where c(n′′) is the normalization factor of node n′′; Get the abnormality score s for each sample dw, and then set the threshold τ (τ=0.6) to determine the abnormality:

[0025] Among them, Q is the final abnormality discrimination label.

[0026] In this embodiment, the preprocessing is as follows: According to the anomaly score and anomaly discrimination label, all data samples marked as severely abnormal are directly removed to avoid extreme values ​​affecting subsequent modeling. The processed data is D filtered : ; in, is the tolerable floating space, which is used to retain some boundary samples to enhance the robustness of the model; Q i For data x i Abnormal discrimination label of ; To D filtered Each data x in i And each feature x ij Detect missing values ​​(such as NaN, null values, invalid codes) and use mean interpolation; Only in the anomaly score s dw When it is in the adjustment interval, the abnormal value is replaced by the context-based mean; and normalization is performed to obtain the preprocessed data set D pre . In this embodiment, a comprehensive feature vector and an importance score matrix of each feature are constructed based on the preprocessed data, as follows: According to the processed data set D pre For time-series numerical data such as SCADA and PMU, statistical features including mean, extreme value, variance, slope, and kurtosis are extracted. For image data such as infrared and visible light, convolutional neural network models are used to extract key visual features, including fault hotspots and structural change features. For environmental and geographic information, trend features are extracted through statistical windows. Static or categorical data is vectorized (such as One-Hot), and all normalized data is merged into a single comprehensive feature vector through splicing. After obtaining the comprehensive feature vector, the specific contribution of each type of feature to the subsequent line loss prediction of the power system is quantified, and a tree model (such as random forest, XGBoost, etc.) is used to automatically output the feature splitting gain index to form a feature importance ranking.

[0027] In this embodiment, the physical laws of the power system are deeply combined with artificial intelligence to build a hybrid analysis model, as follows: Define the physical model module, including the first physical model and the second physical model. Specifically: First physical model (physical calculation of line loss): For the i′th sample (such as a certain line / station), the theoretical line loss is calculated based on physical quantities (current, resistance, voltage, etc.) :

[0028] in, is the k′th phase current (or total current) of the sample, is the corresponding resistance; K is the number of phases; Second physical model (state constraint function): defines voltage over-limit and current over-limit state quantity functions to constrain subsequent AI branch outputs:

[0029] in, The voltage value measured for the i′th sample; are the normal upper and lower bounds of voltage under physical constraints; is the indicator function; It is a voltage state constraint function, which returns 0 if normal and 1 if out of limit. It can be used for loss function or early warning output. Weight the comprehensive feature vector z by the importance vector m to obtain the enhanced feature : ; Among them, z i′ is the comprehensive feature vector of the i′th sample (features from S3, multimodal concatenation, dimension d); m is the feature importance weight vector (dimension d, reflecting the importance of each dimension of features in subsequent tasks); ⊙ is element-by-element multiplication; Then the physical quantity is added as a feature to form a mixed feature vector : ; in, represents vector concatenation; p i′ is the physical feature vector of the i′th sample (e.g. 、V i′ wait); Build a hybrid neural network model, including a two-branch neural network with a physical branch and an AI branch: The physical branch inputs the physical feature p i′ , through a fully connected layer FC phys Generating physical latent representations :

[0030] The AI ​​branch inputs importance-weighted features , through the multi-layer perceptron MLPAI Generating AI Latent Representations :

[0031] Fusion layer: concatenates the outputs of the two branches and generates the final task-related representation through a joint fully connected layer : ; Among them, W f 、b f are the weight and bias of the fully connected layer respectively; ReLU is the activation function; The output layer obtains the final line loss prediction value , anomaly detection probability and voltage prediction value : ; ; ; Among them, W loss 、b loss are the weight vector and bias of the line loss prediction output layer respectively; W o 、b o are the weight and bias of the anomaly detection output layer respectively; W volt 、b volt are the weights and bias of the voltage prediction output layer respectively.

[0032] In this embodiment, the loss function of the hybrid analysis model includes task loss, physical consistency loss, and physical constraint loss, as follows: Mission loss L task : ; Among them, y i′ is the true value of sample i′; n′ is the number of samples; Physical consistency loss L consist : ; Physical constraint loss L constraint : ⋅ Among them, λ voltage is the physical constraint weight; The final loss function L: L=L task +λ1L consist +λ2L constraint ; Among them, λ1 and λ2 are adjustable hyperparameters that control the embedding strength of physical laws.

[0033] In this embodiment, the analysis results are converted into intuitive visualization and actionable decision suggestions, as follows: The hybrid analysis model's analysis results, including line loss prediction, anomaly detection, and voltage over-limit, are structured and summarized, and template-driven and automatically expressed. This converts complex time series predictions, anomaly distribution, and attribution information into text analysis reports that are easy for business personnel to understand. Using line loss prediction, abnormal probability, and voltage over-limit indicators as input, a multi-level heat map is dynamically generated and displayed overlaid on the power grid geographic information map, allowing for intuitive visualization of the risk distribution of different substations and lines. Deep reinforcement learning is introduced to automatically generate the optimal response strategy based on the analysis results of the hybrid analysis model, historical data and power network simulation.

[0034] In this embodiment, deep reinforcement learning is introduced to automatically generate the optimal response strategy based on the analysis results of the hybrid analysis model, historical data, and power network simulation. The specific details are as follows: Based on historical data and grid topology, build a simulation environment (such as a power system simulator based on OpenDSS or MATLAB / Simulink) to simulate grid operation status; State transition: grid state s t In action a t Transfer to s t+1 The dynamic process is realized by the simulator: s t+1 =E(s t ,a t ); Where E is the simulation engine; Status t Integrate the following information: Model analysis results: current line loss prediction value, anomaly detection probability, and voltage prediction value; Physical state: line current It, active power Pt, node voltage Vt, load factor λt; Historical state: historical data within a sliding window (such as the state sequence at several moments in the past); Heat map aggregation index: current line loss heat value H in each area loss,t , abnormal thermal value H anom,t wait.

[0035] The state vector is represented as: ; in, are the line loss prediction value, abnormality detection probability, and voltage prediction value at the current time t respectively; I t is the line current, P t is the active power, V t is the node voltage, λ t is the load rate; H loss,t 、H anom,t Respectively represent the current line loss thermal value and abnormal thermal value of each area; Action a t Defined as a multi-level optimized response strategy, including: Instant Action : Load adjustment (adjustment ratio δ), switch operation (open / close), capacitor switching; Short-term maintenance : Dispatching inspection work orders (object list Linsp), inspection equipment list L chec ; Long-term transformation :Apply for technical transformation budget and plan upgrade area Z upgrade ; The action space is a hierarchical structure: ; Define the reward function r t : ; Among them, α1, α2, and α3 are weight coefficients; Cost(a t ) is the action cost; V ref is the voltage reference value; ΔV is the voltage tolerance range; is the predicted line loss value at time t; is the abnormal event discrimination output at time t; is the predicted voltage value at time t; β′ is the weight coefficient of the control action cost term; Use a hierarchical reinforcement learning framework (such as Option-Critic or MAXQ) combined with the Actor-Critic algorithm: Critic network: evaluate the state value V(s t ;θv) or action value Q(s t ,a t ;θ q ); where θv and θq are the parameters of the state value function V and the action value function Q respectively; Actor network: outputs a hierarchical action policy π(a t ∣s t ;θ π ); Optimization goal: maximize the cumulative discount reward :

[0036] Where γ is the discount factor; is the expected function; T is the total time; Update formula: ; in, is the gradient of the objective function J with respect to the policy parameter θ; is the logarithmic policy gradient, indicating that in state s t Next select action a t The gradient of the log probability of with respect to the parameter θ; A(s t ,a t ) is the advantage function, estimated by TD-error: ; Among them, V(s t ) is the current state s t The value function estimate of .

[0037] A real-time line loss analysis system based on cross-modal measurement data fusion, comprising a processor, a memory, and a computer program stored in the memory. When the processor executes the computer program, the system specifically performs the steps of the real-time line loss analysis method based on cross-modal measurement data fusion as described above. Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0038] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowcharts and / or block diagrams, as well as combinations of processes and / or blocks in the flowcharts and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowcharts and / or block diagrams. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0039] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0040] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0041] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any other manner. Any person skilled in the art may utilize the above-disclosed technical content to modify or modify the present invention into equivalent embodiments. However, any simple modifications, equivalent variations, and modifications to the above embodiments that do not depart from the technical content of the present invention and are based on the technical essence of the present invention remain within the scope of protection of the present invention.

Claims

1. A real-time line loss analysis method based on cross-modal measurement data fusion, characterized in that: The following steps are involved: S1: Collect multi-source heterogeneous data and use the improved isolation forest algorithm to perform preliminary quality screening on the collected raw data to obtain the raw data set and the corresponding preliminary data quality assessment labels; S2: Preprocess the original data set based on the preliminary data quality assessment label to obtain preprocessed data; S3: Based on the preprocessed data, construct a comprehensive feature vector and an importance score matrix for each feature; S4: Deeply integrate the physical laws of the power system with artificial intelligence to build a hybrid analysis model, and perform hybrid analysis based on the comprehensive eigenvector and importance score matrix to obtain analysis results; S5: Convert analysis results into intuitive visualizations and actionable decision recommendations.

2. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 1 is characterized in that: The multi-source heterogeneous data includes the electrical operation data of the SCADA system. scada , high-frequency phasor data D of the PMU system pmu , user-side data D of the electricity consumption information collection system ami , infrared and visible light image data D img , environmental monitoring data D env And the static topological data D of the GIS system gis ; The electrical operation data includes the voltage U of each node of the power grid scada , current I scada , active power P scada , reactive power Q scada The high-frequency phasor data of the PMU system includes the voltage phasor V of the substation and the line pmu , current phasor I pmu , phase angle θ pmu , frequency f pmu ; The environmental monitoring data includes temperature T env 、Humidity H env , wind speed W env ; Static topological data of GIS system, namely power network topology structure Topo gis , including static information of nodes, lines, and equipment parameters; The above data is automatically synchronized and archived through the ETL interface, and the original data set D raw The structure is as follows: D raw ={D scada ,D pmu ,D ami ,D img ,D env ,D gis }。 3. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 2 is characterized in that: The improved isolation forest algorithm is used to perform preliminary quality screening on the collected raw data, as follows: According to the original data set obtained , n is the number of samples, each x i is the multimodal feature concatenation at the data collection moment i, including D scada ,D pmu ,D ami ,D img ,D env ,D gis ; Each feature j has a corresponding data quality weight w j ; Weighting the features: ； in, Represents element-wise multiplication; represents the feature j data collected at time i; When selecting the cutting point of the improved isolation forest algorithm, if j is a time series feature, a time correlation constraint is introduced: suppose the current feature j corresponds to the time series t, and the generation of the cutting point p considers not only the feature value range but also the time neighborhood relationship: ； Where r(t) is the time autocorrelation function; α is the adjustment coefficient; 、 are the minimum and maximum values ​​of feature j in the sample set respectively; U represents uniform distribution; In the tree construction process of the improved isolation forest algorithm, each tree is generated as follows: Build Input: Sample Subset , is the jth data in the sample subset, n is the amount of data; the current tree depth l , maximum depth l max ; Output: an isolation tree; Termination condition: If |X|≤1 or l ≥ l max, then returns the leaf node, which records the number of samples |X|; Select Features: By probability Randomly select feature j, where Avgweight j is the average weight of feature j on X; Select the cutting point: If j is a time series feature, p is selected according to the above time-aware cutting point generation method; otherwise, and Choose p uniformly randomly between Split the data and get the left subtree data X left , right subtree data X right : ； Recursively construct subtrees, including left and right subtrees: Using weighted path length: At node s, if the segmentation is based on feature j, the cumulative path length when passing through node s is , where w j (x′) is the weight of the current sample x′ on feature j; Weighted path length calculation: ； Among them, Path(x′) is the path of sample x′ in a single weighted isolation tree, that is, the set of all segmentation steps from the root node to its leaf node; h dw (x′) is the weighted path length; is the weight value of the path from j to s under sample x′; After generating a forest of t′ trees, calculate the weighted average path length for the sample x′ : ； in, is the weighted path length of sample x′ on the kth weighted isolation tree; Normalized Anomaly Score: ； Where c(n′′) is the normalization factor of node n′′; Get the abnormality score s for each sample dw , and then set the threshold τ to determine abnormality: ； Among them, Q is the final abnormality discrimination label.

4. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 3 is characterized in that: The preprocessing is as follows: According to the anomaly score and anomaly discrimination label, all data samples marked as severe anomalies are directly removed. The processed data is D filtered : ； in, is the tolerable floating space, which is used to retain some boundary samples to enhance the robustness of the model; Q i For data x i Abnormal discrimination label of ; To D filtered Each data x in i And each feature x ij Missing data were detected and imputed using mean values; Only in the anomaly score s dw When it is in the adjustment interval, the abnormal value is replaced by the context-based mean; and normalization is performed to obtain the preprocessed data set D pre。 5. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 4 is characterized in that: Based on the preprocessed data, a comprehensive feature vector and an importance score matrix of each feature are constructed, as follows: According to the processed data set D pre For time series numerical data, statistical features including mean, extreme value, variance, slope, and kurtosis are extracted; for image data, a convolutional neural network model is used to extract key visual features, including fault hotspots and structural change features; for environmental and geographic information, trend features are extracted through statistical windows; Vectorize static or categorical data and merge all normalized data into a single comprehensive feature vector in a splicing manner; After obtaining the comprehensive feature vector, the specific contribution of each type of feature to the subsequent line loss prediction of the power system is quantified.

6. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 1 is characterized in that: The hybrid analysis model is constructed by deeply combining the physical laws of the power system with artificial intelligence, as follows: Define the physical model module, including the first physical model and the second physical model. Specifically: First physical model: For the i′th sample, the theoretical line loss is calculated based on physical quantities. : ； in, is the k′th phase current of the i′th sample, is the corresponding resistance; K is the number of phases; Second physical model: Define voltage and current over-limit state function to constrain subsequent AI branch outputs: ； in, The voltage value measured for the i′th sample; are the normal upper and lower bounds of voltage under physical constraints; is the indicator function; It is a voltage state constraint function, which returns 0 when normal and 1 when out of limit, and is used for loss function or early warning output; Weight the comprehensive feature vector z by the importance vector m to obtain the enhanced feature : ； Among them, z i′ is the comprehensive feature vector of the i′th sample; m is the feature importance weight vector; ⊙ is the element-by-element multiplication; Then the physical quantity is added as a feature to form a mixed feature vector : ； in, represents vector concatenation; p i′ is the physical feature vector of the i′th sample; Build a hybrid neural network model, including a two-branch neural network with a physical branch and an AI branch: The physical branch inputs the physical feature p i′ , through a fully connected layer FC phys Generating physical latent representations : ； The AI ​​branch inputs importance-weighted features , through the multi-layer perceptron MLP AI Generating AI Latent Representations : ； Fusion layer: concatenates the outputs of the two branches and generates the final task-related representation through a joint fully connected layer : ； Among them, W f 、b f are the weight and bias of the fully connected layer respectively; ReLU is the activation function; The output layer obtains the final line loss prediction value , anomaly detection probability and voltage prediction value : ； ； ； Among them, W loss 、b loss are the weight vector and bias of the line loss prediction output layer respectively; W o 、b o are the weight and bias of the anomaly detection output layer respectively; W volt 、b volt are the weights and bias of the voltage prediction output layer respectively.

7. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 6 is characterized in that: The loss function of the hybrid analysis model includes task loss, physical consistency loss, and physical constraint loss, as follows: Mission loss L task : ； Among them, y i′ is the true value of sample i′; n′ is the number of samples; Physical consistency loss L consist : ； Physical constraint loss L constraint : ⋅ Among them, λ voltage is the physical constraint weight; Final loss function: L=L task +λ1L consist +λ2L constraint ; Among them, λ1 and λ2 are adjustable hyperparameters that control the embedding strength of physical laws.

8. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 6, characterized in that: The analysis results are converted into intuitive visualization and actionable decision-making recommendations as follows: Perform structured summarization and template-driven automated presentation of hybrid analysis model results, including line loss prediction, anomaly detection, and voltage over-limit. Using line loss prediction, abnormal probability, and voltage over-limit indicators as input, a multi-level heat map is dynamically generated and displayed overlaid on the power grid geographic information map, allowing for intuitive visualization of the risk distribution of different substations and lines. Deep reinforcement learning is introduced to automatically generate the optimal response strategy based on the analysis results of the hybrid analysis model, historical data and power network simulation.

9. The real-time line loss analysis method based on cross-modal measurement data fusion according to claim 8, characterized in that: The introduction of deep reinforcement learning automatically generates the optimal response strategy based on the analysis results of the hybrid analysis model, historical data and power network simulation, as follows: Based on historical data and grid topology, a simulation environment is built to simulate grid operation status; State transition: grid state s t In action a t Transfer to s t+1 The dynamic process is realized by the simulator: s t+1 =E(s t ,a t ); Where E is the simulation engine; The state vector is represented as: ； in, are the line loss prediction value, abnormality detection probability, and voltage prediction value at the current time t respectively; I t is the line current, P t is the active power, V t is the node voltage, λ t is the load rate; H loss,t 、H anom,t Respectively represent the current line loss thermal value and abnormal thermal value of each area; Action a t Defined as a multi-level optimized response strategy, including: Instant Action : Load adjustment, switch operation, capacitor switching; Short-term maintenance : Distribute inspection work orders and spot inspection equipment lists chec ; Long-term transformation :Apply for technical transformation budget and plan upgrade area Z upgrade ; The action space is a hierarchical structure: ； Define the reward function r t : ； Among them, α1, α2, α3 are weight coefficients; Cost(a t ) is the action cost; V ref is the voltage reference value; ΔV is the voltage tolerance range; is the predicted line loss value at time t; is the abnormal event discrimination output at time t; is the predicted voltage value at time t; β′ is the weight coefficient of the control action cost term; Adopting a hierarchical reinforcement learning framework combined with the Actor-Critic algorithm: Critic network: evaluate the state value V(s t ;θv) or action value Q(s t ,a t ;θ q ); where θv and θq are the parameters of the state value function V and the action value function Q respectively; Actor network: outputs a hierarchical action policy π(a t ∣s t ;θ π ); Optimization goal: maximize the cumulative discount reward : ； Where γ is the discount factor; is the expected function; T is the total time; Update formula: ； in, is the gradient of the objective function J with respect to the policy parameter θ; is the logarithmic policy gradient, indicating that in state s t Next select action a t The gradient of the log probability of with respect to the parameter θ; A(s t ,a t ) is the advantage function, estimated by TD-error: ； Among them, V(s t ) is the current state s t The value function estimate of .

10. A real-time line loss analysis system based on cross-modal measurement data fusion, characterized in that: The invention comprises a processor, a memory and a computer program stored in the memory. When the processor executes the computer program, the method specifically performs the steps of a real-time line loss analysis method based on cross-modal measurement data fusion as described in any one of claims 1 to 9.

Images