Electric power system infrastructure operation state diagnosis method and device and medium

Abstract

The invention relates to the field of operation state diagnosis, in particular to an electric power system infrastructure operation state diagnosis method and device and a medium, and the method comprises the following steps: obtaining real-time data including electric power system infrastructure environment data and equipment data in real time; acquiring historical data including historical power system infrastructure environment data, equipment data and historical operation state scores; performing data preprocessing on the real-time data and the historical data, and constructing a multi-modal feature set; inputting the multi-modal feature set into a self-adaptive multi-modal interaction model, extracting a dynamic interaction relationship between the environment data and the equipment data by using a self-attention mechanism, obtaining interaction features, fusing the interaction features, and outputting fused features; and calculating an operation state score based on the fused features, and estimating the operation state of the power system infrastructure. Compared with the prior art, the method has the advantages of remarkably improving the accuracy and adaptability of operation state evaluation and the like.

Description

Technical Field

[0001] The present invention relates to the field of operation status diagnosis, and in particular to a method, device and medium for diagnosing the operation status of power system infrastructure. Background Art

[0002] With the continuous development of the power industry, the infrastructure of the power system is also increasing. The operating environment of these infrastructures is complex, and real-time and effective operating status monitoring is required to avoid safety problems. The switch station is an important facility in the power system, and its operating status directly affects the safety and reliability of the power system. However, the operating environment of the switch station is complex, and it often faces problems such as high temperature, high humidity, flooding, smoke, and equipment aging, which may cause equipment failure or even safety accidents. At present, traditional inspection methods and single parameter monitoring cannot achieve a comprehensive and accurate evaluation of the overall operating status of the station, especially under harsh environmental conditions, it is difficult to timely warn of potential risks.

[0003] Current technologies deploy a variety of sensor devices for condition monitoring, including partial discharge sensors, temperature and humidity sensors, vibration accelerometers, and water immersion detectors. For example, China Southern Power Grid proposed an IoT-based switch station environmental monitoring system, but it is mainly limited to data collection and lacks high-level intelligent diagnostic capabilities. Institutions such as the China Electric Power Research Institute have introduced machine learning technologies, such as support vector machines (SVMs) and random forests (RFs), to detect abnormal operating conditions. However, these methods mainly target single fault types (such as partial discharge anomalies) and do not achieve fusion analysis of multi-dimensional data.

[0004] The research focus abroad has gradually shifted to multimodal data fusion analysis. For example, companies such as ABB and Siemens have developed intelligent switch station monitoring systems that can combine vibration, temperature and partial discharge signals for anomaly detection, but these systems are mostly hardware-based and lack deep learning algorithms based on historical data. Japan has performed outstandingly in the field of dynamic risk assessment of power equipment, for example, using Bayesian networks and time series models to predict equipment failure risks and update risk distribution maps in real time. However, its model relies heavily on data preprocessing and is difficult to adapt to complex scenarios with drastic data fluctuations.

[0005] To this end, there is an urgent need for an intelligent diagnostic method based on multi-dimensional data that can adapt to complex scenarios with drastic data fluctuations and accurately diagnose and predict the operating status of switch station houses. Summary of the invention

[0006] The purpose of the present invention is to provide a method, device and medium for diagnosing the operating status of power system infrastructure, using multi-source data such as partial discharge, mechanical properties, cable temperature, ambient temperature, etc., combined with historical data and real-time monitoring results, to construct an operating status scoring model, and through dynamic weight adjustment, accurate diagnosis and risk prediction of the station building operating status can be achieved to ensure the safe and stable operation of the power system.

[0007] The purpose of the present invention can be achieved by the following technical solutions:

[0008] A method for diagnosing the operating status of an electric power system infrastructure comprises the following steps:

[0009] Acquire real-time data including power system infrastructure environmental data and equipment data, wherein the environmental data is data related to the operating environment of the power system infrastructure, and the equipment data is data related to the operating status of key equipment in the power system infrastructure;

[0010] Access historical data including historical power system infrastructure environmental data, equipment data, and historical operating status scores;

[0011] Performing data preprocessing on the real-time data and historical data to construct a multimodal feature set;

[0012] The multimodal feature set is input into the adaptive multimodal interaction model, and the dynamic interaction relationship between the environment data and the device data is extracted using the self-attention mechanism to obtain the interaction features and fuse them, and then output the fused features.

[0013] The operating status score is calculated based on the fusion features to obtain the operating status diagnosis results of the power system infrastructure.

[0014] The environmental data include environmental humidity, environmental temperature, water level height on the ground of the power system infrastructure and smoke concentration.

[0015] The equipment data includes the amplitude and main frequency of the equipment partial discharge signal, the RMS value of the mechanical vibration signal and the cable temperature.

[0016] The pre-processing is specifically as follows:

[0017] Data cleaning: For missing value problems, interpolation methods based on adjacent features are used for interpolation; for outlier problems, the 3σ principle or local outlier factor method is used for detection and elimination;

[0018] Time synchronization: align multi-source data based on timestamps;

[0019] Normalization: Use standardized formula Perform normalization processing;

[0020] Time alignment: Align historical data with real-time data in time to form a multimodal feature set within the time window W = [tk, t], where t is the time stamp of the real-time data and k is the length of time to trace back to the past.

[0021] The adaptive multimodal interaction model performs the following steps to output fusion features:

[0022] The environmental data X env and device data X device Mapped to high-dimensional feature space respectively:

[0023] Q env , K env , V env =X env W Q , X env W K , X env W V

[0024] Q device , K device , V device =X device W Q , X device W K , X device W V

[0025] Among them, Q, K, V are query, key and value matrices respectively, which are used for attention calculation, the subscript env represents the environment, and device represents the device; W Q , W K , W V is a learnable weight matrix;

[0026] The attention weight of the environment data on the device data and the attention weight of the device data on the environment data are calculated by using the cross-modal attention mechanism, where the attention weight of the environment data on the device data is A env→device The calculation method is:

[0027]

[0028] The attention weight of device data to environment data A device→env The calculation method is:

[0029]

[0030] Among them, dd k is the scaling factor of the feature dimension;

[0031] Calculate interaction features based on attention weights:

[0032] Z device =A env→device V device

[0033] Z env =A env→device V device

[0034] Among them, Z device , Z env are the interaction characteristics of the device and the environment, respectively;

[0035] After fusing the interactive features, output:

[0036] Z=[Z device , Z env ]

[0037] Among them, Z is the fusion feature.

[0038] The calculation of the running status score based on the fusion feature is specifically as follows:

[0039] Using fusion features, we build an operation status scoring model:

[0040]

[0041] Among them, S is the running status score; n is the number of features in the fusion feature time series; w i is the weight of the i-th feature, satisfying Dynamically calculated through the attention mechanism:

[0042]

[0043] Z i is the value of the i-th feature in the fusion feature; f i is the normalized score of the i-th feature, based on the normalization formula:

[0044]

[0045] Among them, x i is the original data value corresponding to the i-th feature, min(x) represents the minimum value of the original data corresponding to this feature, and max(x) represents the maximum value of the original data corresponding to this feature.

[0046] The power system infrastructure operation status diagnosis result is obtained by grading based on the operation status score, and the status is divided into four levels from high to low according to the operation status score, namely:

[0047] The equipment is running normally without any abnormality and the operating status is excellent;

[0048] There are slight abnormalities in the operation of the equipment, and the operating status is good;

[0049] There are obvious abnormalities in the operation of the equipment, and the operating status is medium;

[0050] There are serious problems with the device and the operating status is poor.

[0051] The method further includes: using a time series prediction model to predict the future operating state of the power system infrastructure:

[0052]

[0053] in, is the predicted future operating status score, t is the current time, and m is the number of predicted time steps;

[0054] The loss function of the time series prediction model is:

[0055] L=λ 1 L state +λ 2 L trend

[0056] Among them, λ 1 and λ 2 is the weight coefficient, L state The state score loss uses the mean square error to measure the difference between the predicted score and the true score:

[0057]

[0058] in, is the operational status score predicted by the model, is the real running status score, n is the number of features in the fused feature time series;

[0059] L trend The cross entropy loss is used to measure the trend classification accuracy:

[0060]

[0061] Among them, y i,c It is the real operating status level trend category, is the probability of the operating status level predicted by the model, and C is the number of divided operating status levels.

[0062] A device for diagnosing the operating status of an electric power system infrastructure comprises a memory, a processor, and a program stored in the memory. The processor implements the method described when executing the program.

[0063] A storage medium stores a program, which implements the method when executed.

[0064] Compared with the prior art, the present invention has the following beneficial effects:

[0065] a) Based on the deep coupling analysis method of environment and equipment data, the present invention effectively solves the limitation of separate analysis of environment and equipment status in traditional diagnostic methods through multimodal data fusion and dynamic interaction modeling. Through the cross-modal attention mechanism, the present invention can capture the complex relationship between environmental factors (such as temperature, humidity, water level, etc.) and equipment signals (such as partial discharge signals, mechanical vibration, cable temperature, etc.) in real time, thereby achieving accurate status diagnosis and comprehensive risk prediction. This innovative design significantly improves the accuracy of operating status evaluation and diagnostic efficiency.

[0066] b) The present invention realizes real-time scoring and trend prediction of the operating status of power system infrastructure, and builds an integrated intelligent diagnosis framework from current status evaluation to future risk warning. Real-time scoring provides quantitative analysis results of equipment and environmental status, and trend prediction identifies potential risk change directions in advance to help operation and maintenance personnel make scientific decisions. Through the dynamic weight allocation mechanism, the system can automatically adjust feature contributions in extreme environments (such as high humidity and high temperature) to ensure the robustness and adaptability of diagnosis, greatly enhancing the system's support for multiple operating scenarios.

[0067] c) The present invention has strong result interpretability and user-friendliness. Through the attention mechanism, the present invention not only provides a comprehensive score, but also clearly displays the contribution of each feature to the score, helping users understand the basis of the diagnosis results.

[0068] d) The present invention reduces the cost of station operation and maintenance and improves system reliability. Through real-time diagnosis and active early warning, the frequency of manual inspections is reduced, the allocation of maintenance resources is optimized, and the accumulation and sudden occurrence of equipment failures are avoided, thereby extending the service life of the equipment. The present invention is applicable to the intelligent operation and maintenance of various infrastructures such as switch stations, substations, and data centers, and can be extended to scenarios such as smart cities and industrial plants, with broad application prospects and economic value. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 The present invention is a flow chart of the method. DETAILED DESCRIPTION

[0070] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0071] This embodiment takes a switch station building as an example to provide a method for diagnosing the operating status of power system infrastructure. Figure 1 As shown, the following steps are included:

[0072] S1, real-time acquisition of real-time data including switch station environment data and equipment data.

[0073] (1) Environmental data

[0074] Environmental data refers to data related to the station house operating environment, describing the changes in the station house's external conditions and internal environment, including:

[0075] H env (t): Ambient humidity, indicating the humidity level of the air in the station building (unit: %).

[0076] T env (t): Ambient temperature, indicating the temperature of the air in the station building (unit: ℃).

[0077] H water (t): Water level, indicating the water level height on the station floor (unit: cm).

[0078] S smoke (t): Smoke density, indicating the smoke content in the air (unit: ppm).

[0079] (2) Equipment data

[0080] Equipment data refers to data related to the operating status of key equipment in the station house, including:

[0081] PD amp (t): Amplitude of the partial discharge signal (unit: pC).

[0082] PD freq (t): Main frequency of partial discharge signal (unit: Hz).

[0083] V rms (t): RMS value of the mechanical vibration signal (unit: m / s 2 ).

[0084] T cable (t): Cable temperature, indicating the cable shell or core temperature (unit: ℃).

[0085] S2, obtain historical data including historical switch station house environment data, equipment data and historical operation status scores.

[0086] S3, performs data preprocessing on real-time data and historical data to build a multimodal feature set.

[0087] S31, data cleaning: For the missing value problem, the interpolation method based on adjacent features is used for interpolation; for the outlier problem, the 3σ principle or local outlier factor (LOF) method is used for detection and elimination;

[0088] S32, time synchronization: align multi-source data based on timestamps to form a unified time series input matrix;

[0089] S33, normalization: using the standardized formula Perform normalization to ensure that different features are comparable;

[0090] S34, time alignment: align historical data with real-time data in time to form a multimodal feature set within the time window W = [tk, t], where t is the time stamp of the real-time data, meaning that the multimodal data with t as the reference point is being analyzed and processed, and k is the length of time looking back into the past, which determines how much historical data is included in the time window.

[0091] S4, inputs the multimodal feature set into the adaptive multimodal interaction model, uses the self-attention mechanism to extract the dynamic interaction relationship between environmental data and device data, obtains the interaction features, fuses them, and outputs the fused features.

[0092] The adaptive multimodal interaction model performs the following steps to output fused features:

[0093] S41, the environmental data X env and device data X device Mapped to high-dimensional feature space respectively:

[0094] Q env , K env , V env =X env W Q , X env W K , X env W V

[0095] Q device , K device , V device =X device W Q , X device W K , X device W V

[0096] Among them, Q, K, and V are query, key, and value matrices, respectively, which are used for attention calculation. The subscript env represents the environment, and device represents the device.Q , W K , W V is a learnable weight matrix.

[0097] S42, using a cross-modal attention mechanism to calculate the attention weight of the environment data on the device data and the attention weight of the device data on the environment data, where the attention weight A of the environment data on the device data is env→device The calculation method is:

[0098]

[0099] The attention weight of device data to environment data A device→env The calculation method is:

[0100]

[0101] Among them, dd k It is the scaling factor of the feature dimension to prevent the attention value from being too large.

[0102] S43, calculate the interaction features based on the attention weights:

[0103] Z device =A env→device V device

[0104] Z env =A env→device V device

[0105] Among them, Z device , Z env are the interaction characteristics of the device and the environment, respectively.

[0106] S44, fuses the interactive features and outputs:

[0107] Z=[Z device , Z env ]

[0108] Among them, Z is the fusion feature, which is used for status scoring and trend prediction.

[0109] S5, calculating the operation status score based on the fusion features to obtain the operation status diagnosis result of the power system infrastructure.

[0110] Using the fusion feature Z, we build an operation status scoring model:

[0111]

[0112] Among them, S is the running status score, ranging from 0 to 100; n is the number of features in the fused feature time series; w iis the weight of the i-th feature, satisfying Dynamically calculated through the attention mechanism:

[0113]

[0114] Z i is the value of the i-th feature in the fusion feature; f i is the normalized score of the i-th feature, based on the normalization formula:

[0115]

[0116] Among them, x i is the original data value corresponding to the i-th feature, min(x) represents the minimum value of the original data corresponding to this feature, and max(x) represents the maximum value of the original data corresponding to this feature.

[0117] The operating status diagnosis results of the switch station house are graded based on the operating status score. The status is divided into four levels from high to low according to the operating status score, as shown in Table 1 below.

[0118] Table 1

[0119] Rating range Status Level describe suggestion 90～100 excellent The equipment is running normally without any abnormality Routine inspection 70～89 good There is a slight abnormality, which needs to be monitored Enhance monitoring and focus on troubleshooting local problems 50～69 middle There are obvious abnormalities and potential risks Develop maintenance plans and repair in time 0～49 Difference There is a serious problem that needs to be addressed immediately Stop the equipment and conduct a comprehensive investigation of potential hazards

[0120] S6, using time series prediction models (such as LSTM) to predict the future operating status of power system infrastructure:

[0121]

[0122] in, is the predicted future operating status score, t is the current time, and m is the number of predicted time steps.

[0123] The loss function of the time series forecasting model is:

[0124] L=λ 1 L state +λ 2 L trend

[0125] Among them, λ 1 and λ 2 is the weight coefficient, L state The state score loss uses the mean square error to measure the difference between the predicted score and the true score:

[0126]

[0127] in, is the operational status score predicted by the model, is the real running status score, n is the number of features in the fused feature time series;

[0128] L trend For the trend prediction loss, the cross entropy loss is used to measure the trend classification accuracy:

[0129]

[0130] Among them, y i,c It is the real operating status level trend category, is the probability of the operating status level predicted by the model, and C is the number of divided operating status levels.

[0131] In addition to being applied to the operation status diagnosis of switch station rooms, the method of the present invention can also be applied to the environment and power supply system status monitoring of data center computer rooms, as well as the temperature, humidity, and power supply safety diagnosis of server cabinets. The application process is as follows:

[0132] 1) Collect data from data center power supply equipment (such as UPS, distribution cabinet) and environmental sensor data (such as room temperature, humidity, and smoke concentration).

[0133] 2) The adaptive multimodal interaction model and the operation status scoring model of the present invention are used to evaluate the operation status of the computer room, and the time series prediction model is used to predict the impact of future environmental condition changes on server operation.

[0134] 3) Optimize server load or adjust environmental control (such as cooling, dehumidification) in advance based on the evaluation results.

[0135] The above is an introduction to a method embodiment. The following is a further explanation of the solution of the present invention through an apparatus embodiment.

[0136] In a preferred embodiment, the power system infrastructure operation status diagnosis device includes:

[0137] Real-time data acquisition module: real-time acquisition of real-time data including power system infrastructure environmental data and equipment data, wherein the environmental data is data related to the operating environment of the power system infrastructure, and the equipment data is data related to the operating status of key equipment in the power system infrastructure;

[0138] Historical data acquisition module: acquires historical data including historical power system infrastructure environmental data, equipment data and historical operation status scores;

[0139] Data preprocessing module: performs data preprocessing on the real-time data and historical data to construct a multimodal feature set;

[0140] Feature extraction module: Input the multimodal feature set into the adaptive multimodal interaction model, use the self-attention mechanism to extract the dynamic interaction relationship between environmental data and device data, obtain the interaction features, fuse them, and output the fused features;

[0141] Operation status diagnosis module: Calculate the operation status score based on the fusion features to obtain the operation status diagnosis results of the power system infrastructure.

[0142] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the described module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0143] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program code.

[0144] The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning, or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims

1. A method for diagnosing the operating status of power system infrastructure, characterized in that: The following steps are involved: Acquire real-time data including power system infrastructure environmental data and equipment data, wherein the environmental data is data related to the operating environment of the power system infrastructure, and the equipment data is data related to the operating status of key equipment in the power system infrastructure; Access historical data including historical power system infrastructure environmental data, equipment data, and historical operating status scores; Performing data preprocessing on the real-time data and historical data to construct a multimodal feature set; The multimodal feature set is input into the adaptive multimodal interaction model, and the dynamic interaction relationship between the environment data and the device data is extracted using the self-attention mechanism to obtain the interaction features and fuse them, and then output the fused features. The operating status score is calculated based on the fusion features to obtain the operating status diagnosis results of the power system infrastructure.

2. A method for diagnosing the operating status of an electric power system infrastructure according to claim 1, characterized in that: The environmental data include environmental humidity, environmental temperature, water level height on the ground of the power system infrastructure and smoke concentration.

3. A method for diagnosing the operating status of an electric power system infrastructure according to claim 1, characterized in that: The equipment data includes the amplitude and main frequency of the equipment partial discharge signal, the RMS value of the mechanical vibration signal and the cable temperature.

4. A method for diagnosing the operating status of an electric power system infrastructure according to claim 1, characterized in that: The pre-processing is specifically as follows: Data cleaning: For missing value problems, interpolation methods based on adjacent features are used for interpolation; for outlier problems, the 3σ principle or local outlier factor method is used for detection and elimination; Time synchronization: align multi-source data based on timestamps; Normalization: Use standardized formula Perform normalization processing; Time alignment: Align historical data with real-time data in time to form a multimodal feature set within the time window W = [tk, t], where t is the time stamp of the real-time data and k is the length of time to trace back to the past.

5. A method for diagnosing the operating status of an electric power system infrastructure according to claim 1, characterized in that: The adaptive multimodal interaction model performs the following steps to output fusion features: The environmental data X env and device data X device Mapped to high-dimensional feature space respectively: Q env ,K env ,V env =X env W Q ,X env W K ,X env W V Q device ,K device ,V device =X device W Q ,X device W K ,X device W V Among them, Q, K, V are query, key and value matrices respectively, which are used for attention calculation, the subscript env represents the environment, and device represents the device; W Q , W K , W V is a learnable weight matrix; The attention weight of the environment data on the device data and the attention weight of the device data on the environment data are calculated by using the cross-modal attention mechanism, where the attention weight of the environment data on the device data is A env→device The calculation method is: The attention weight of device data to environment data A devicr→env The calculation method is: Among them, d k is the scaling factor of the feature dimension; Calculate interaction features based on attention weights: From device =A env→device In device From env =A env→device In device Among them, Z device , Z env are the interaction characteristics of the device and the environment, respectively; After fusing the interactive features, output: From=[From device ,WITH env ] Among them, Z is the fusion feature.

6. A method for diagnosing the operating status of an electric power system infrastructure according to claim 1, characterized in that: The calculation of the running status score based on the fusion feature is specifically as follows: Using fusion features, we build an operation status scoring model: Among them, S is the running status score; n is the number of features in the fusion feature time series; w i is the weight of the i-th feature, satisfying Dynamically calculated through the attention mechanism: Z i is the value of the i-th feature in the fusion feature; f i is the normalized score of the i-th feature, based on the normalization formula: Among them, x i is the original data value corresponding to the i-th feature, min(x) represents the minimum value of the original data corresponding to this feature, and max(x) represents the maximum value of the original data corresponding to this feature.

7. A method for diagnosing the operating status of an electric power system infrastructure according to claim 1, characterized in that: The power system infrastructure operation status diagnosis result is obtained by grading based on the operation status score, and the status is divided into four levels from high to low according to the operation status score, namely: The equipment is running normally without any abnormality and the operating status is excellent; There are slight abnormalities in the operation of the equipment, and the operating status is good; There are obvious abnormalities in the operation of the equipment, and the operating status is medium; There are serious problems with the device and the operating status is poor.

8. A method for diagnosing the operating status of an electric power system infrastructure according to claim 1, characterized in that: The method further includes: using a time series prediction model to predict the future operating state of the power system infrastructure: in, is the predicted future operating status score, t is the current time, and m is the number of predicted time steps; The loss function of the time series prediction model is: L=λ1L state +λ2L trend Among them, λ1 and λ2 are weight coefficients, L state The state score loss uses the mean square error to measure the difference between the predicted score and the true score: in, is the operational status score predicted by the model, is the real running status score, n is the number of features in the fused feature time series; L trend The cross entropy loss is used to measure the trend classification accuracy: Among them, y i,c It is the real operating status level trend category, is the probability of the operating status level predicted by the model, and C is the number of divided operating status levels.

9. A device for diagnosing the operating status of an electric power system infrastructure, comprising a memory, a processor, and a program stored in the memory, characterized in that: When the processor executes the program, the method according to any one of claims 1 to 8 is implemented.

10. A storage medium having a program stored thereon, characterized in that: When the program is executed, the method according to any one of claims 1 to 8 is implemented.

Images