Converter station equipment parameter analysis method and system based on multi-parameter fusion monitoring

Abstract

The invention relates to the field of converter station equipment parameter analysis, and provides a converter station equipment parameter analysis method and system based on multi-parameter fusion monitoring, and the method comprises the following steps: collecting multi-parameter characteristic data of converter station equipment; performing streaming processing and abnormal mode identification on the multi-parameter characteristic data to obtain equipment abnormity early warning information of the converter station equipment; constructing a time-varying graph neural network model based on the multi-parameter feature data, obtaining a space-time dependency relationship among parameters through a causal reasoning method, and dynamically adjusting the feature weight of the time-varying graph neural network model in combination with a reinforcement learning algorithm to obtain a running state evaluation model; and inputting the real-time state feature sequence into the operation state evaluation model to obtain an equipment operation state evaluation result, and visually displaying the equipment operation state evaluation result and the abnormal early warning information to obtain a comprehensive operation state evaluation report of the converter station equipment. According to the invention, real-time monitoring and evaluation of the operation state of the converter station equipment are realized.

Description

Technical Field

[0001] The present invention relates to the field of converter station equipment parameter analysis, and in particular to a converter station equipment parameter analysis method and system based on multi-parameter fusion monitoring. Background Art

[0002] Converter station equipment parameter analysis involves monitoring and analyzing various operating parameters of converter station equipment to assess its operating status, predict potential failures, and provide early warning information. Converter stations are key nodes in power systems, their primary function being to convert AC to DC and vice versa to enable long-distance power transmission and distribution. The operating status of converter station equipment directly impacts the safety and stability of the power system, making parameter analysis of converter station equipment crucial. However, converter station equipment parameter analysis involves a large amount of data, which cannot be displayed intuitively, resulting in poor real-time monitoring.

[0003] In the existing technology, feature information is extracted through time series characterization methods, the cloud similarity measurement method is used to calculate the similarity between time series data, and the probability output method is used to evaluate the device status. However, this method only focuses on the statistical characteristics of time series data and does not consider the causal relationship and spatiotemporal dependency between parameters. The evaluation model is relatively static and cannot be dynamically adjusted according to the device operating status. Summary of the Invention

[0004] In view of this, the present invention proposes a method and system for analyzing converter station equipment parameters based on multi-parameter fusion monitoring, which solves the problem that the existing technology only focuses on the statistical characteristics of time series data, does not consider the causal relationship and spatiotemporal dependence between parameters, and the evaluation model is relatively static and cannot be dynamically adjusted according to the equipment operating status.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a method for analyzing converter station equipment parameters based on multi-parameter fusion monitoring, comprising the following steps:

[0006] Collecting electrical parameter data, physical parameter data, and environmental parameter data of converter station equipment, performing time axis alignment and standardization preprocessing on the electrical parameter data, physical parameter data, and environmental parameter data, and integrating them through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data;

[0007] Stream processing is performed on the multi-parameter feature data using an edge computing algorithm to obtain a real-time state feature sequence, and abnormal pattern recognition is performed on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method to obtain equipment abnormality warning information of the converter station equipment;

[0008] A time-varying graphical neural network model is constructed based on the multi-parameter feature data, the spatiotemporal dependency relationship between the multiple parameters is obtained through a causal reasoning method, and the feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment;

[0009] The real-time status feature sequence is input into the operation status assessment model to obtain the equipment operation status assessment result, the equipment operation status assessment result and abnormal warning information are visualized to obtain a comprehensive operation status assessment report of the converter station equipment.

[0010] On the basis of the above technical solution, preferably, the electrical parameter data, physical parameter data and environmental parameter data of the converter station equipment are collected, the electrical parameter data, physical parameter data and environmental parameter data are time-axis aligned and standardized pre-processed, and integrated through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data, including:

[0011] Performing time series interpolation and dynamic time warping on the electrical parameter data, the physical parameter data, and the environmental parameter data, respectively, to obtain time-axis aligned monitoring data;

[0012] Wavelet transform is used to perform noise reduction and feature extraction on the time-axis aligned monitoring data to obtain electrical parameter data features, physical parameter data features and environmental parameter data features, and the features of different parameters are integrated through an adaptive fusion algorithm to obtain multi-parameter feature data.

[0013] Based on the above technical solution, preferably, the edge computing algorithm is used to perform streaming processing on the multi-parameter feature data to obtain a real-time state feature sequence, and abnormal pattern recognition is performed on the real-time state feature sequence based on dynamic threshold adjustment and sliding window method to obtain equipment abnormality warning information of the converter station equipment, including:

[0014] Inputting the multi-parameter feature data into an edge computing node, and performing real-time streaming processing on the multi-parameter feature data using an online sliding window algorithm to obtain the real-time state feature sequence;

[0015] An abnormality detection threshold is determined based on an adaptive dynamic threshold algorithm, and abnormal pattern recognition is performed on the real-time state feature sequence in combination with a multi-scale sliding window method to obtain abnormality warning information of the equipment.

[0016] On the basis of the above technical solution, preferably, the calculation formula of the online sliding window algorithm is:

[0017] ST(t)=β1·Wf(t)·F(t)+(1-β1)·MA(t);

[0018]

[0019] Among them, ST(t) is the real-time state feature sequence at time t, β1 is the smoothing factor, Wf(t) is the adaptive weight at time t, F(t) is the multi-parameter feature data at time t, MA(t) is the moving average in the sliding window at time t, α1 is the weight adjustment parameter, σ 2 is the variance parameter, F(t-1) is the multi-parameter feature data at time t-1, H(t) is the entropy weight factor at time t, and p i is the normalized probability value of the i-th feature, and n is the feature dimension;

[0020] The calculation formula of the adaptive dynamic threshold algorithm is:

[0021] TH(t)=μ(t)±k(t)·σ(t);

[0022]

[0023] Among them, TH(t) is the anomaly detection threshold at time t, μ(t) is the local mean of the real-time state feature sequence in the sliding window at time t, σ(t) is the standard deviation of the real-time state feature sequence in the sliding window at time t, k(t) is the adaptive adjustment coefficient at time t, k0 is the basic threshold coefficient, γ1 is the sensitivity parameter, and ΔF(t) is the rate of change of the multi-parameter feature data at time t.

[0024] On the basis of the above technical solution, preferably, a time-varying graphical neural network model is constructed based on the multi-parameter feature data, the spatiotemporal dependency relationship between multiple parameters is obtained through a causal reasoning method, and the feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status evaluation model for converter station equipment, including:

[0025] Constructing a parameter relationship graph based on the multi-parameter feature data, performing a double causal test on the parameter nodes based on time series Granger causality and transfer entropy to obtain the spatiotemporal dependency strength between the parameter nodes, and dynamically screening the spatiotemporal dependency strength based on an adaptive threshold to obtain the dynamic correlation matrix;

[0026] The dynamic association matrix is input into a time-varying graph neural network model, the policy network and the value network of the time-varying graph neural network model are separated through a dual neural network structure, and the feature weights are updated based on a temporal difference learning method to obtain the operation status evaluation model.

[0027] On the basis of the above technical solution, preferably, the calculation formula of the spatiotemporal dependency strength is:

[0028] C(a,b)=α2·G(a,b)+(1-α2)·T(a,b);

[0029]

[0030] Among them, C(a, b) is the spatiotemporal dependency strength between parameter nodes a and b, G(a, b) is the Granger causality index between parameter nodes a and b, T(a, b) is the transfer entropy index between parameter nodes a and b, α2 is the Granger causality balance factor, RSS full is the residual sum of squares of the autoregressive prediction model containing the historical information of parameter node b, RSS reduced is the residual sum of squares of the autoregressive prediction model containing only the historical information of parameter a, τ is the time lag factor, λ is the attenuation coefficient, Ω is the state space, and a T and b T are the state vectors of parameter nodes a and b at time T, respectively. T+1 is the state vector of parameter a at time T+1, and p(·) is the probability distribution function;

[0031] The calculation formula for updating the feature weight is:

[0032]

[0033] in, and They are Moment and The feature weight of the time-varying graph neural network model, η is the learning rate, for The instant reward at the moment, γ2 is the discount factor, and They are Moment and The state vector at time t, and They are Moment and The action vector at the moment, Q(·) is the action value function, is the gradient of the weight parameter w.

[0034] Based on the above technical solution, preferably, the real-time status feature sequence is input into the operation status assessment model to obtain the equipment operation status assessment result, and the equipment operation status assessment result and abnormal warning information are visualized to obtain a comprehensive operation status assessment report of the converter station equipment, including:

[0035] Inputting the real-time status feature sequence into the operating status evaluation model, and obtaining the real-time operating status evaluation score and key influencing factors of the equipment by fusing multi-level feature information and historical evaluation results;

[0036] A multi-level visualization interface is constructed, wherein the first layer of the multi-level visualization interface is used to display the real-time operation status evaluation score of the entire equipment, the second layer of the multi-level visualization interface is used to display the contribution and change trend of key influencing factors, and the third layer of the multi-level visualization interface is used to display detailed descriptions and processing suggestions of abnormal warning information, and generate a comprehensive operation status evaluation report, which includes all display information of the multi-level visualization interface.

[0037] On the basis of the above technical solution, preferably, it further includes:

[0038] Construct a functional dependency graph for converter station equipment and extract the functional coordination characteristics of the entire station using a graph embedding method. The graph embedding method includes the GraphSAGE algorithm. By constructing a power coupling relationship matrix between nodes and analyzing the power coupling relationship between converter valves and transformers, the topological characteristic data of the entire station are obtained.

[0039] A time-varying graph neural network model is constructed based on the multi-parameter feature data and the whole station topology feature data. The spatiotemporal dependency relationship between multiple parameters is obtained through a causal reasoning method. The feature weights of the time-varying graph neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status evaluation model for converter station equipment.

[0040] On the basis of the above technical solution, preferably, it further includes:

[0041] Identifying a current operating mode of the converter station in the multi-parameter characteristic data, including a constant voltage control mode, a constant current control mode, and a constant power control mode, and performing segmented processing on the multi-parameter characteristic data according to the identified operating mode;

[0042] A time-varying graphical neural network model is constructed based on the multi-parameter feature data, and sub-models are constructed for different operating modes. The spatiotemporal dependency of multiple parameters under each operating mode is obtained through a causal reasoning method. The feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status evaluation model for converter station equipment that takes operating mode switching into consideration.

[0043] In a second aspect, the present invention further provides a converter station equipment parameter analysis system based on multi-parameter fusion monitoring, the system comprising:

[0044] A data acquisition module is used to collect electrical parameter data, physical parameter data, and environmental parameter data of converter station equipment, perform time axis alignment and standardization preprocessing on the electrical parameter data, physical parameter data, and environmental parameter data, and integrate them through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data;

[0045] An anomaly identification module is used to perform streaming processing on the multi-parameter feature data using an edge computing algorithm to obtain a real-time state feature sequence, perform abnormal pattern recognition on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method, and obtain equipment abnormality warning information of the converter station equipment;

[0046] a parameter analysis module for constructing a time-varying graphical neural network model based on the multi-parameter feature data, obtaining the spatiotemporal dependencies between the multiple parameters through a causal reasoning method, and dynamically adjusting the feature weights of the time-varying graphical neural network model in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment;

[0047] The status assessment module is used to input the real-time status feature sequence into the operation status assessment model to obtain the equipment operation status assessment result, visualize the equipment operation status assessment result and abnormal warning information, and obtain a comprehensive operation status assessment report of the converter station equipment.

[0048] The converter station equipment parameter analysis method and system based on multi-parameter fusion monitoring of the present invention have the following beneficial effects compared with the prior art:

[0049] (1) By collecting and integrating the electrical, physical, and environmental parameter data of converter station equipment, using edge computing and streaming processing, and combining causal reasoning and reinforcement learning algorithms to build a time-varying graph neural network model for intelligent operation status assessment, and providing a comprehensive operation status assessment report through visual display, it can achieve comprehensive, real-time, and intelligent monitoring and evaluation of the operation status of converter station equipment, thereby improving the safety and stability of converter station equipment operation;

[0050] (2) By adopting an online sliding window algorithm and an adaptive dynamic threshold algorithm, multi-parameter feature data is processed in real time and anomaly detection is performed. The upper and lower thresholds of anomaly detection can be dynamically calculated, and the operating status of converter station equipment can be monitored in real time and anomaly warnings can be provided, thereby improving the accuracy and response speed of anomaly detection.

[0051] (3) By calculating the spatiotemporal dependency strength between parameter nodes and combining the Granger causality and transfer entropy indicators, important spatiotemporal dependencies are dynamically screened out, and the feature weights are updated using a dual neural network structure and temporal difference learning method, thereby improving the prediction accuracy and adaptability of the time-varying graphical neural network model, enabling a more accurate assessment of the operating status of converter station equipment and improving the reliability of equipment status assessment. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0053] Figure 1 This is a flow chart of a converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to the present invention;

[0054] Figure 2 This is a structural diagram of a converter station equipment parameter analysis system based on multi-parameter fusion monitoring of the present invention. DETAILED DESCRIPTION

[0055] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0056] See also Figure 1 The present invention provides a method for analyzing converter station equipment parameters based on multi-parameter fusion monitoring, comprising the following steps:

[0057] Collecting electrical parameter data, physical parameter data, and environmental parameter data of converter station equipment, performing time axis alignment and standardization preprocessing on the electrical parameter data, physical parameter data, and environmental parameter data, and integrating them through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data with unified representation;

[0058] Stream processing is performed on the multi-parameter feature data using an edge computing algorithm to obtain a real-time state feature sequence, and abnormal pattern recognition is performed on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method to obtain equipment abnormality warning information of the converter station equipment;

[0059] A time-varying graphical neural network model is constructed based on the multi-parameter feature data, the spatiotemporal dependency relationship between the multiple parameters is obtained through a causal reasoning method, and the feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment;

[0060] The real-time status feature sequence is input into the operation status assessment model to obtain the equipment operation status assessment result, the equipment operation status assessment result and abnormal warning information are visualized to obtain a comprehensive operation status assessment report of the converter station equipment.

[0061] Specifically, this embodiment collects and integrates the electrical parameter, physical parameter and environmental parameter data of the converter station equipment, uses edge computing and streaming processing, and combines causal reasoning and reinforcement learning algorithms to build a time-varying graphical neural network model for intelligent operation status assessment, and provides a comprehensive operation status assessment report through visual display, thereby realizing comprehensive, real-time and intelligent monitoring and evaluation of the operation status of the converter station equipment, and improving the safety and stability of the operation of the converter station equipment.

[0062] The electrical parameter data, physical parameter data and environmental parameter data of the converter station equipment are collected, the electrical parameter data, physical parameter data and environmental parameter data are time-axis aligned and standardized pre-processed, and the data are integrated through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data, including:

[0063] Performing time series interpolation and dynamic time warping on the electrical parameter data, the physical parameter data, and the environmental parameter data, respectively, to obtain time-axis aligned monitoring data;

[0064] Determine a unified time axis reference based on the sampling frequency of each parameter data, perform cubic spline interpolation on parameter data with a sampling frequency lower than the time axis reference, perform dynamic time warping on parameter data with a sampling frequency higher than the time axis reference, and align all parameter data to the unified time axis reference;

[0065] Using wavelet transform to perform noise reduction and feature extraction on the time-axis aligned monitoring data to obtain electrical parameter data features, physical parameter data features, and environmental parameter data features, and integrating the features of different parameters through an adaptive fusion algorithm to obtain multi-parameter feature data with unified representation;

[0066] Adaptive weight coefficients are set according to the importance of different parameter features, and the parameter features after wavelet transformation are fused using a weighted summation method, wherein the adaptive weight coefficients are dynamically calculated and updated through the information entropy of the parameter features.

[0067] Specifically, this embodiment ensures that all data are aligned on a unified time axis benchmark by performing time series interpolation and dynamic time warping on electrical parameter data, physical parameter data, and environmental parameter data, and uses wavelet transform for noise reduction and feature extraction. The features of different parameters are then integrated through an adaptive fusion algorithm to ultimately obtain multi-parameter feature data with a unified representation, thereby improving the accuracy and consistency of data processing and ensuring the comprehensiveness of the multi-parameter feature data.

[0068] In a specific embodiment, the electrical parameter data, physical parameter data, and environmental parameter data of the converter station equipment include:

[0069] The electrical parameter data includes electrical parameters, current parameters, power parameters, harmonic parameters, and control parameters;

[0070] Electrical parameters: valve-side voltage, grid-side voltage, and winding voltages of converter transformers; valve group voltage and unit voltage of thyristor converter valves; pole-to-ground voltage and pole-to-pole voltage of DC busbars; filter outlet voltage of AC filters; and DC-side filter voltage of DC filters.

[0071] Current parameters: valve group current and thyristor current distribution of thyristor converter valves, current of each phase of converter transformers, current of each filter group of AC filters, DC line current and grounding electrode current of DC field equipment, and through-current of smoothing reactors;

[0072] Power parameters: active power and reactive power of the converter transformer, overall transmission power and power factor of the converter station, reactive compensation power of the AC filter, and reactive regulation power of the static VAR compensation device;

[0073] Harmonic parameters: filtering efficiency and harmonic content of AC filters, harmonic content of DC filters, voltage / current harmonic distortion rate of converter transformers, and total harmonic distortion rate of the entire station electrical system;

[0074] Control parameters: trigger angle, commutation angle, overlap angle for the converter valve control system, control pulse accuracy and trigger timing for the valve control unit, pole current instruction and pole voltage instruction for the pole controller, power instruction and operation mode instruction for the station control system.

[0075] The physical parameter data include temperature parameters, vibration parameters, pressure parameters, flow parameters, and position parameters;

[0076] Temperature parameters: thyristor junction temperature, radiator temperature, and valve body temperature of thyristor converter valves; oil temperature, winding temperature, and core temperature of converter transformers; cooling water inlet and outlet temperatures and radiator temperature of valve cooling systems; oil temperature and coil temperature of smoothing reactors; and case temperature of DC capacitors;

[0077] Vibration parameters: transformer body vibration and bushing vibration of converter transformers, reactor body vibration of smoothing reactors, valve tower vibration and supporting structure vibration of thyristor converter valves, conductor vibration and support insulator vibration of DC field equipment;

[0078] Pressure parameters: SF6 gas pressure of GIS equipment, cooling water pressure and medium pressure of converter valve cooling system, oil conservator pressure and oil pressure of converter transformer, SF6 pressure and hydraulic operating mechanism pressure of circuit breaker / disconnector;

[0079] Flow parameters: valve cooling water flow and secondary refrigerant flow of valve cooling system, cooling oil flow and fan air volume of transformer cooling system, valve hall ventilation flow of air conditioning system, and gas flow monitoring of SF6 gas equipment;

[0080] Position parameters: three-phase switch position status of the disconnector, grounding knife position status of the grounding switch, inter-pole disconnector position of the DC circuit breaker, and converter transformer tap position of the tap changer.

[0081] The environmental parameter data include meteorological parameters, air quality parameters, electromagnetic environment parameters, and insulation parameters;

[0082] Meteorological parameters: valve hall temperature, humidity, and air pressure in the valve hall; outdoor temperature, humidity, wind speed, and wind direction in the outdoor equipment area; local temperature and humidity in the transformer area; and ambient temperature and humidity in the control room.

[0083] Air quality parameters: dust concentration and suspended particulate matter concentration in the valve hall, salt spray concentration and PM2.5 content in the outdoor equipment area, SF6 concentration and corrosive gas concentration in the GIS room, and oil and gas concentration in the transformer firewall area;

[0084] Electromagnetic environment parameters: background electromagnetic field strength and electromagnetic interference level in the control room, high-frequency electromagnetic field strength in the valve hall, power frequency electric field strength and power frequency magnetic field strength of the substation fence, shielding effect and leakage electromagnetic field level in the shielding room;

[0085] Insulation parameters: pollution degree and creepage distance of outdoor insulators, air insulation distance and phase-to-phase insulation level of valve halls, internal insulation status and insulating gas quality of GIS equipment, surface insulation performance and surface hydrophobicity of bushings.

[0086] The method of using an edge computing algorithm to perform streaming processing on the multi-parameter feature data to obtain a real-time state feature sequence, performing abnormal pattern recognition on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method to obtain equipment abnormality warning information of the converter station equipment includes:

[0087] Inputting the multi-parameter feature data into an edge computing node, and performing real-time streaming processing on the multi-parameter feature data using an online sliding window algorithm to obtain the real-time state feature sequence;

[0088] An abnormality detection threshold is determined based on an adaptive dynamic threshold algorithm, and abnormal pattern recognition is performed on the real-time state feature sequence in combination with a multi-scale sliding window method to obtain abnormality warning information of the equipment.

[0089] Specifically, this embodiment implements real-time monitoring of converter station equipment operating status and anomaly warnings by using an online sliding window algorithm at edge computing nodes to stream multi-parameter feature data in real time, combined with an adaptive dynamic threshold algorithm and a multi-scale sliding window method for anomaly pattern recognition. This embodiment leverages the advantages of edge computing to reduce data transmission latency, improve the real-time and accuracy of anomaly detection, and effectively ensure timely warnings of equipment operating status.

[0090] In a specific embodiment, the initial window length L and the sliding step size S of the online sliding window algorithm are set;

[0091] The calculation formula of the online sliding window algorithm is:

[0092] ST(t)=β1·Wf(t)·F(t)+(1-β1)·MA(t);

[0093]

[0094] Among them, ST(t) is the real-time state feature sequence at time t, β1 is the smoothing factor, Wf(t) is the adaptive weight at time t, F(t) is the multi-parameter feature data at time t, MA(t) is the moving average in the sliding window at time t, α1 is the weight adjustment parameter, σ 2 is the variance parameter, F(t-1) is the multi-parameter feature data at time t-1, H(t) is the entropy weight factor at time t, and p i is the normalized probability value of the i-th feature, and n is the feature dimension;

[0095] Dynamically calculate upper and lower thresholds for anomaly detection based on the statistical characteristics of historical data and current operating status;

[0096] The calculation formula of the adaptive dynamic threshold algorithm is:

[0097] TH(t)=μ(t)±k(t)·σ(t);

[0098]

[0099] Among them, TH(t) is the anomaly detection threshold at time t, μ(t) is the local mean of the real-time state feature sequence in the sliding window at time t, σ(t) is the standard deviation of the real-time state feature sequence in the sliding window at time t, k(t) is the adaptive adjustment coefficient at time t, k0 is the basic threshold coefficient, γ1 is the sensitivity parameter, and ΔF(t) is the rate of change of the multi-parameter feature data at time t.

[0100] Specifically, this embodiment adopts an online sliding window algorithm to process multi-parameter feature data in real time, and introduces an adaptive dynamic threshold algorithm to dynamically adjust the threshold range of anomaly detection.

[0101] The introduction of smoothing factors and adaptive weights in the calculation of real-time state feature sequences improves the stability of feature extraction. The adaptive weighting of different feature dimensions is achieved through the entropy weight factor and normalized probability value. The introduction of variance parameters enhances the algorithm's adaptability to data fluctuations.

[0102] The dynamic threshold calculation takes into account the local mean and standard deviation, making the threshold more in line with the actual operating status. The introduction of adaptive adjustment coefficients and sensitivity parameters improves the accuracy of anomaly detection. The dynamic adjustment of the threshold is achieved through the change rate of multi-parameter feature data.

[0103] This implementation uses an online sliding window algorithm and an adaptive dynamic threshold algorithm to perform real-time streaming processing and anomaly detection on multi-parameter feature data. It can dynamically calculate the upper and lower thresholds for anomaly detection, realize real-time monitoring of the operating status of converter station equipment and anomaly warning, and improve the accuracy and response speed of anomaly detection.

[0104] The time-varying graphical neural network model is constructed based on the multi-parameter feature data, the spatiotemporal dependency relationship between the multiple parameters is obtained through a causal reasoning method, and the feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment, including:

[0105] Constructing a parameter relationship graph based on the multi-parameter feature data, performing a double causal test on the parameter nodes based on time series Granger causality and transfer entropy to obtain the spatiotemporal dependency strength between the parameter nodes, and dynamically screening the spatiotemporal dependency strength based on an adaptive threshold to obtain the dynamic correlation matrix;

[0106] The dynamic association matrix is input into a time-varying graph neural network model, the policy network and the value network of the time-varying graph neural network model are separated through a dual neural network structure, and the feature weights are updated based on the temporal difference learning method, wherein the reward function is designed based on the model prediction accuracy and parameter importance to obtain the operation status evaluation model.

[0107] Specifically, this embodiment constructs a parameter relationship diagram, employs time-series Granger causality and transfer entropy for dual causal testing, dynamically screens the spatiotemporal dependencies between parameter nodes, and utilizes a dual neural network structure—a policy network and a value network—combined with a temporal difference learning method to update feature weights, achieving a precise assessment of the operating status of converter station equipment. This embodiment enhances the model's adaptability and prediction accuracy through the use of a dynamic correlation matrix and dual network structure.

[0108] The calculation formula for the spatiotemporal dependency strength is:

[0109] C(a,b)=α2·G(a,b)+(1-α2)·T(a,b);

[0110]

[0111] Among them, C(a, b) is the spatiotemporal dependency strength between parameter nodes a and b, G(a, b) is the Granger causality index between parameter nodes a and b, T(a, b) is the transfer entropy index between parameter nodes a and b, α2 is the Granger causality balance factor, RSS full is the residual sum of squares of the autoregressive prediction model containing the historical information of parameter node b, RSS reduced is the residual sum of squares of the autoregressive prediction model containing only the historical information of parameter a, τ is the time lag factor, λ is the attenuation coefficient, Ω is the state space, and a T and b T are the state vectors of parameter nodes a and b at time T, respectively. T+1 is the state vector of parameter a at time T+1, and p(·) is the probability distribution function;

[0112] The calculation formula for updating the feature weight is:

[0113]

[0114] in, and They are Moment and The feature weight of the time-varying graph neural network model, η is the learning rate, for The instant reward at the moment, γ2 is the discount factor, and They are Moment and The state vector at time t, and They are Moment and The action vector at the moment, Q(·) is the action value function, is the gradient of the weight parameter w.

[0115] Specifically, this embodiment introduces a dual causality test mechanism, using both Granger causality and transfer entropy indicators, and dynamically assigning weights between the two indicators through a balancing factor. It also considers the lag factor and attenuation coefficient to better capture the temporal correlation between parameters.

[0116] Residual Sum of Squares (RSS) full and RSS reduced The causal strength is evaluated by comparison, and the autoregressive prediction model is introduced to improve the accuracy of causal relationship judgment. The influence of historical information is considered through the time lag factor, making the model more consistent with the actual situation.

[0117] The information transfer amount is calculated based on the state space and probability distribution function, and the mutual information between the state vectors of the parameter nodes is considered. The prediction provides reference information for future status.

[0118] Adopting the temporal difference learning method, combining the immediate reward and discount factor, introducing the action value function to optimize the weight update strategy, through the gradient Achieve precise adjustment of weights.

[0119] This embodiment calculates the strength of the spatiotemporal dependency between parameter nodes, combines Granger causality and transfer entropy indicators, dynamically screens out important spatiotemporal dependencies, and uses a dual neural network structure and temporal difference learning method to update the feature weights, thereby improving the prediction accuracy and adaptability of the time-varying graphical neural network model, enabling more accurate assessment of the operating status of converter station equipment and improving the reliability of equipment status assessment.

[0120] The real-time status feature sequence is input into the operation status assessment model to obtain the equipment operation status assessment result. The equipment operation status assessment result and abnormal warning information are visualized to obtain a comprehensive operation status assessment report of the converter station equipment, including:

[0121] Inputting the real-time status feature sequence into the operating status evaluation model, and obtaining the real-time operating status evaluation score and key influencing factors of the equipment by fusing multi-level feature information and historical evaluation results;

[0122] The attention mechanism is used to weight the importance of the real-time status feature sequence. Combined with the temporal dependency of historical evaluation results, the real-time operation status evaluation score at the current moment is calculated. The key factors affecting the equipment operation status and their weights are identified based on the parameter contribution analysis method.

[0123] A multi-level visualization interface is constructed, wherein the first layer of the multi-level visualization interface is used to display the real-time operation status evaluation score of the entire equipment, the second layer of the multi-level visualization interface is used to display the contribution and change trend of key influencing factors, and the third layer of the multi-level visualization interface is used to display detailed descriptions and processing suggestions of abnormal warning information, and supports real-time linkage updates and interactive queries of information at each level to generate a comprehensive operation status evaluation report, which includes all display information of the multi-level visualization interface.

[0124] Specifically, this embodiment constructs a multi-level visualization interface, uses the attention mechanism to weight the importance of the real-time status feature sequence, calculates the operating status evaluation score based on the temporal dependency of historical evaluation results, and identifies key influencing factors based on parameter contribution analysis. It realizes a multi-level visualization display from overall evaluation score, key factor analysis to abnormal warning processing, supports real-time linkage updates and interactive queries, makes the operating status evaluation results of the converter station equipment more intuitive, comprehensive and easy to understand, and improves the interpretability of the status evaluation results.

[0125] In a specific embodiment, a functional dependency graph of converter station equipment is constructed, and the functional coordination characteristics of the entire station are extracted based on a graph embedding method. The graph embedding method includes a GraphSAGE algorithm. By constructing a power coupling relationship matrix between nodes and analyzing the power coupling relationship between converter valves and transformers, the topological characteristic data of the entire station are obtained.

[0126] A time-varying graph neural network model is constructed based on the multi-parameter feature data and the whole station topology feature data. The spatiotemporal dependency relationship between multiple parameters is obtained through a causal reasoning method. The feature weights of the time-varying graph neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status evaluation model for converter station equipment.

[0127] In a specific embodiment, the current operating mode of the converter station is identified in the multi-parameter characteristic data, including a constant voltage control mode, a constant current control mode, and a constant power control mode, and the multi-parameter characteristic data is segmented according to the identified operating mode;

[0128] A time-varying graphical neural network model is constructed based on the multi-parameter feature data, and sub-models are constructed for different operating modes. The spatiotemporal dependency of multiple parameters under each operating mode is obtained through a causal reasoning method. The feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status evaluation model for converter station equipment that takes operating mode switching into consideration.

[0129] See also Figure 2 The present invention also provides a converter station equipment parameter analysis system based on multi-parameter fusion monitoring, the system comprising:

[0130] A data acquisition module is used to collect electrical parameter data, physical parameter data, and environmental parameter data of converter station equipment, perform time axis alignment and standardization preprocessing on the electrical parameter data, physical parameter data, and environmental parameter data, and integrate them through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data;

[0131] An anomaly identification module is used to perform streaming processing on the multi-parameter feature data using an edge computing algorithm to obtain a real-time state feature sequence, perform abnormal pattern recognition on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method, and obtain equipment abnormality warning information of the converter station equipment;

[0132] a parameter analysis module for constructing a time-varying graphical neural network model based on the multi-parameter feature data, obtaining the spatiotemporal dependencies between the multiple parameters through a causal reasoning method, and dynamically adjusting the feature weights of the time-varying graphical neural network model in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment;

[0133] The status assessment module is used to input the real-time status feature sequence into the operation status assessment model to obtain the equipment operation status assessment result, visualize the equipment operation status assessment result and abnormal warning information, and obtain a comprehensive operation status assessment report of the converter station equipment.

[0134] Specifically, this embodiment of a converter station equipment parameter analysis system based on multi-parameter fusion monitoring implements multi-parameter data collection, preprocessing, fusion, real-time monitoring, anomaly identification, and status assessment for converter station equipment by constructing a system comprising a data acquisition module, an anomaly identification module, a parameter analysis module, and a status assessment module. Leveraging edge computing and a time-varying graph neural network model, the system can process and analyze multi-source heterogeneous data in real time, dynamically adjust feature weights, provide accurate equipment operating status assessments and anomaly warnings, and present comprehensive operating status assessment reports through a multi-level visual interface, thereby improving the safety, reliability, and intelligent management of converter station equipment.

[0135] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a converter station equipment parameter analysis method based on multi-parameter fusion monitoring.

[0136] The present invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of a method for analyzing converter station equipment parameters based on multi-parameter fusion monitoring, as described in an embodiment of the present invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0137] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A converter station equipment parameter analysis method based on multi-parameter fusion monitoring, characterized in that: The following steps are involved: Collecting electrical parameter data, physical parameter data, and environmental parameter data of converter station equipment, performing time axis alignment and standardization preprocessing on the electrical parameter data, physical parameter data, and environmental parameter data, and integrating them through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data; Stream processing is performed on the multi-parameter feature data using an edge computing algorithm to obtain a real-time state feature sequence, and abnormal pattern recognition is performed on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method to obtain equipment abnormality warning information of the converter station equipment; A time-varying graphical neural network model is constructed based on the multi-parameter feature data, the spatiotemporal dependency relationship between the multiple parameters is obtained through a causal reasoning method, and the feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment; The real-time status feature sequence is input into the operation status assessment model to obtain the equipment operation status assessment result, the equipment operation status assessment result and abnormal warning information are visualized to obtain a comprehensive operation status assessment report of the converter station equipment.

2. The converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to claim 1, characterized in that: The electrical parameter data, physical parameter data and environmental parameter data of the converter station equipment are collected, the electrical parameter data, physical parameter data and environmental parameter data are time-axis aligned and standardized pre-processed, and the data are integrated through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data, including: Performing time series interpolation and dynamic time warping on the electrical parameter data, the physical parameter data, and the environmental parameter data, respectively, to obtain time-axis aligned monitoring data; Wavelet transform is used to perform noise reduction and feature extraction on the time-axis aligned monitoring data to obtain electrical parameter data features, physical parameter data features and environmental parameter data features, and the features of different parameters are integrated through an adaptive fusion algorithm to obtain multi-parameter feature data.

3. The converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to claim 1, characterized in that: The method of using an edge computing algorithm to perform streaming processing on the multi-parameter feature data to obtain a real-time state feature sequence, performing abnormal pattern recognition on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method to obtain equipment abnormality warning information of the converter station equipment includes: Inputting the multi-parameter feature data into an edge computing node, and performing real-time streaming processing on the multi-parameter feature data using an online sliding window algorithm to obtain the real-time state feature sequence; An abnormality detection threshold is determined based on an adaptive dynamic threshold algorithm, and abnormal pattern recognition is performed on the real-time state feature sequence in combination with a multi-scale sliding window method to obtain abnormality warning information of the equipment.

4. The converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to claim 3 is characterized in that: The calculation formula of the online sliding window algorithm is: ST(t)=β1·Wf(t)·F(t)+(1-β1)·MA(t); Among them, ST(t) is the real-time state feature sequence at time t, β1 is the smoothing factor, Wf(t) is the adaptive weight at time t, F(t) is the multi-parameter feature data at time t, MA(t) is the moving average in the sliding window at time t, α1 is the weight adjustment parameter, σ 2 is the variance parameter, F(t-1) is the multi-parameter feature data at time t-1, H(t) is the entropy weight factor at time t, and p i is the normalized probability value of the i-th feature, and n is the feature dimension; The calculation formula of the adaptive dynamic threshold algorithm is: TH(t)=μ(t)±k(t)·σ(t); Among them, TH(t) is the anomaly detection threshold at time t, μ(t) is the local mean of the real-time state feature sequence in the sliding window at time t, σ(t) is the standard deviation of the real-time state feature sequence in the sliding window at time t, k(t) is the adaptive adjustment coefficient at time t, k0 is the basic threshold coefficient, γ1 is the sensitivity parameter, and ΔF(t) is the rate of change of the multi-parameter feature data at time t.

5. The converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to claim 1, characterized in that: The time-varying graphical neural network model is constructed based on the multi-parameter feature data, the spatiotemporal dependency relationship between the multiple parameters is obtained through a causal reasoning method, and the feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment, including: Constructing a parameter relationship graph based on the multi-parameter feature data, performing a double causal test on the parameter nodes based on time series Granger causality and transfer entropy to obtain the spatiotemporal dependency strength between the parameter nodes, and dynamically screening the spatiotemporal dependency strength based on an adaptive threshold to obtain the dynamic correlation matrix; The dynamic association matrix is input into a time-varying graph neural network model, the policy network and the value network of the time-varying graph neural network model are separated through a dual neural network structure, and the feature weights are updated based on a temporal difference learning method to obtain the operation status evaluation model.

6. The converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to claim 5, characterized in that: The calculation formula for the spatiotemporal dependency strength is: C(a,b)=α2·G(a,b)+(1-α2)·T(a,b); Among them, C(a,b) is the spatiotemporal dependency strength between parameter nodes a and b, G(a,b) is the Granger causality index between parameter nodes a and b, T(a,b) is the transfer entropy index between parameter nodes a and b, α2 is the Granger causality balance factor, RSS full is the residual sum of squares of the autoregressive prediction model containing the historical information of parameter node b, RSS reduced is the residual sum of squares of the autoregressive prediction model containing only the historical information of parameter a, τ is the time lag factor, λ is the attenuation coefficient, Ω is the state space, and a T and b T are the state vectors of parameter nodes a and b at time T, respectively. T+1 is the state vector of parameter a at time T+1, and p(·) is the probability distribution function; The calculation formula for updating the feature weight is: in, and They are Moment and The feature weight of the time-varying graph neural network model, η is the learning rate, for The instant reward at the moment, γ2 is the discount factor, and They are Moment and The state vector at time t, and They are Moment and The action vector at the moment, Q(·) is the action value function, is the gradient of the weight parameter w.

7. The method for analyzing converter station equipment parameters based on multi-parameter fusion monitoring according to claim 1, characterized in that: The real-time status feature sequence is input into the operation status assessment model to obtain the equipment operation status assessment result. The equipment operation status assessment result and abnormal warning information are visualized to obtain a comprehensive operation status assessment report of the converter station equipment, including: Inputting the real-time status feature sequence into the operating status evaluation model, and obtaining the real-time operating status evaluation score and key influencing factors of the equipment by fusing multi-level feature information and historical evaluation results; A multi-level visualization interface is constructed, wherein the first layer of the multi-level visualization interface is used to display the real-time operation status evaluation score of the entire equipment, the second layer of the multi-level visualization interface is used to display the contribution and change trend of key influencing factors, and the third layer of the multi-level visualization interface is used to display detailed descriptions and processing suggestions of abnormal warning information, and generate a comprehensive operation status evaluation report, which includes all display information of the multi-level visualization interface.

8. The converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to claim 1, characterized in that: Also includes: Construct a functional dependency graph for converter station equipment and extract the functional coordination characteristics of the entire station using a graph embedding method. The graph embedding method includes the GraphSAGE algorithm. By constructing a power coupling relationship matrix between nodes and analyzing the power coupling relationship between converter valves and transformers, the topological characteristic data of the entire station are obtained. A time-varying graph neural network model is constructed based on the multi-parameter feature data and the whole station topology feature data. The spatiotemporal dependency relationship between multiple parameters is obtained through a causal reasoning method. The feature weights of the time-varying graph neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status evaluation model for converter station equipment.

9. The converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to claim 1, characterized in that: Also includes: Identifying a current operating mode of the converter station in the multi-parameter characteristic data, including a constant voltage control mode, a constant current control mode, and a constant power control mode, and performing segmented processing on the multi-parameter characteristic data according to the identified operating mode; A time-varying graphical neural network model is constructed based on the multi-parameter feature data, and sub-models are constructed for different operating modes. The spatiotemporal dependency of multiple parameters under each operating mode is obtained through a causal reasoning method. The feature weights of the time-varying graphical neural network model are dynamically adjusted in combination with a reinforcement learning algorithm to obtain an operating status evaluation model for converter station equipment that takes operating mode switching into consideration.

10. A converter station equipment parameter analysis system based on multi-parameter fusion monitoring, used to execute a converter station equipment parameter analysis method based on multi-parameter fusion monitoring according to any one of claims 1 to 9, characterized in that: The system comprises: A data acquisition module is used to collect electrical parameter data, physical parameter data, and environmental parameter data of converter station equipment, perform time axis alignment and standardization preprocessing on the electrical parameter data, physical parameter data, and environmental parameter data, and integrate them through a multi-source heterogeneous data fusion algorithm to obtain multi-parameter feature data; An anomaly identification module is used to perform streaming processing on the multi-parameter feature data using an edge computing algorithm to obtain a real-time state feature sequence, perform abnormal pattern recognition on the real-time state feature sequence based on dynamic threshold adjustment and a sliding window method, and obtain equipment abnormality warning information of the converter station equipment; a parameter analysis module for constructing a time-varying graphical neural network model based on the multi-parameter feature data, obtaining the spatiotemporal dependencies between the multiple parameters through a causal reasoning method, and dynamically adjusting the feature weights of the time-varying graphical neural network model in combination with a reinforcement learning algorithm to obtain an operating status assessment model for converter station equipment; The status assessment module is used to input the real-time status feature sequence into the operation status assessment model to obtain the equipment operation status assessment result, visualize the equipment operation status assessment result and abnormal warning information, and obtain a comprehensive operation status assessment report of the converter station equipment.

Images