Intelligent operation and maintenance method and device for highway strong current system

Abstract

The invention belongs to the technical field of expressway operation and maintenance, and relates to an intelligent operation and maintenance method and device for an expressway strong current system, and the method comprises the following steps: collecting and preprocessing multi-source data; constructing a state space model and defining a system state vector; obtaining an estimated state value by adopting a state estimation algorithm; performing fault anomaly detection based on the estimated state value and the multi-source data; extracting a device feature vector and calculating a health index; predicting the remaining service life of the equipment and judging the potential fault risk, and generating a maintenance work order and determining the priority when the threshold value is exceeded. According to the technical scheme of the invention, work orders are generated through multi-source data acquisition, state modeling and estimation, anomaly detection, health assessment and life prediction, so that the whole-process intelligent operation and maintenance of the highway strong current system is realized, and the real-time sensing and diagnosis capability of the operation state can be improved. And the requirements of operation and maintenance real-time performance improvement, data unified management, intelligent diagnosis enhancement and emergency disposal high efficiency are met.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of highway operation and maintenance, and particularly relates to an intelligent operation and maintenance method and device for a highway strong current system. BACKGROUND

[0002] In the prior art, the operation and maintenance of the highway strong current system usually relies on manual inspection and regular maintenance, and the safety operation requirements of the equipment are met through manual experience and periodic maintenance. However, the existing operation and maintenance mode has some obvious deficiencies. Due to the wide coverage and complex operating environment of the highway strong current system, the manual method often fails to find potential hidden dangers in time, resulting in frequent post-maintenance situations.

[0003] At the same time, the existing operation and maintenance system still has the problems of dispersion and isolation in data management. Different types of strong current equipment have different sources, non-uniform monitoring protocols and collection methods, and the operation and maintenance personnel cannot uniformly process and integrate the collected operation information, so that the system operation state cannot be fully reflected, and the accurate analysis of the operation dynamics under complex working conditions is also limited.

[0004] In addition, the functions of the existing monitoring system mostly stop at the stage of parameter recording and alarm prompting, and cannot effectively combine multi-source data to carry out state estimation, anomaly detection and health assessment. In actual operation, the system cannot predict potential failures in time, the equipment reliability management level is insufficient, and the operation risk accumulation and passive maintenance arrangement are easy to occur.

[0005] Therefore, the existing technology often has the problems of lagging operation and maintenance mode, scattered and isolated data, insufficient diagnosis capability and untimely emergency response. This is the deficiency of the prior art.

[0006] Therefore, the present application provides an intelligent operation and maintenance method and device for a highway strong current system to solve the above-mentioned defects in the prior art, which is very necessary. SUMMARY

[0007] The purpose of the present application is to provide an intelligent operation and maintenance method and device for a highway strong current system to solve the above-mentioned technical problems.

[0008] To achieve the above-mentioned purpose, the present application provides the following technical solution: An intelligent operation and maintenance method for a highway strong current system, comprising the following steps: Collecting multi-source data and performing data preprocessing, the multi-source data including at least one of voltage, current, temperature, vibration and switch action signal, and the data preprocessing including at least one of denoising, cleaning and normalization; Based on the multi-source data, a state space model is constructed, and a system state vector is defined in the state space model to represent the dynamic relationship between the system running state and the external input; A state estimation algorithm is used to estimate the system state vector to obtain an estimated state value; Based on the estimated state value and the multi-source data, fault anomaly detection is performed, which includes mutation detection based on cumulative sum and change point detection, fault detection based on residual statistical test, and anomaly detection based on machine learning; Based on the multi-source data, an equipment feature vector is extracted, and a health index is calculated; According to the health index, the remaining useful life of the equipment is predicted, and the potential failure risk is determined. When the potential failure risk exceeds a preset risk threshold, a corresponding maintenance work order is generated, and the priority of the maintenance work order is determined.

[0009] By using the above technical solutions, through collecting multi-source data, constructing a state space model, state estimation and fault anomaly detection, extracting equipment features and calculating a health index, and generating a maintenance work order combined with life prediction, the whole process intelligent operation and maintenance of the highway strong power system is realized, which can improve the real-time perception and diagnosis capability of the running state, and meet the needs of real-time improvement of operation and maintenance, unified management of data, enhanced intelligent diagnosis, and efficient emergency disposal.

[0010] Among them, by collecting multi-source data such as voltage, current, temperature, vibration and switch action signal, and combining with denoising, cleaning, normalization and other preprocessing operations, the standardized processing of different source running information is realized, thereby avoiding the problem of scattered and isolated data, and providing a unified input basis for subsequent modeling and analysis; Based on multi-source data, a state space model is constructed, and a system state vector is defined therein, so that the dynamic relationship between the system running state and the external input can be accurately represented; Combined with the state estimation algorithm, the system state is estimated to obtain the estimated state value, which can reflect the real running condition of the equipment, overcoming the limitations of traditional manual inspection mode lag and information lag; Based on the estimated state value and the multi-source data, abnormality detection is carried out, combined with mutation detection, residual statistical test and machine learning algorithm, a multi-level fault diagnosis mechanism is realized, which can timely discover abnormal conditions in operation, improve the intelligent identification ability of equipment failure, and thus improve the problem of insufficient diagnosis capability of the existing technology; By extracting the equipment feature vector and calculating the health index, the remaining useful life is predicted and the potential failure risk is determined, and when the risk exceeds the threshold, a maintenance work order is automatically generated and the priority is determined. This process realizes the transformation from "after-the-fact maintenance" to "predictive maintenance", and can directly associate the risk result with the work order scheduling, so as to respond in advance before the fault occurs, and meet the timeliness demand of emergency response in highway strong power operation.

[0011] As preferred, the multi-source data is weighted least square fused, and the fused multi-source data is taken as the observation input of the state estimation algorithm.

[0012] In the technical solution, the multi-source data is uniformly processed by using the weighted least square method, and the following technical effects can be achieved: First, by introducing the weighting strategy, the role of high confidence data can be highlighted, and the interference of unstable data can be weakened in the face of different signal source collection accuracies and large noise level differences, so that the fusion result is closer to the true state of the system, thereby significantly improving the reliability and stability of the input information in the operation and maintenance process. Second, by completing error correction and redundancy utilization in advance before the data is input to the state estimation algorithm, the convergence speed of the algorithm can be faster, and the estimation process can be smoother, thereby avoiding estimation bias and misjudgment risk caused by single data anomaly, and providing a more solid foundation for subsequent system modeling and state quantity calculation. Third, by globally processing signals from voltage, current, temperature, vibration and action, a unified data fusion interface can be formed in a large-scale highway strong electric system, which not only improves the centralization level of data management, but also ensures the input consistency of subsequent state estimation and diagnosis process, thereby meeting the efficiency and real-time demand of operation and maintenance information processing in the highway scene. Fourth, by continuously accumulating noise characteristics and credibility information of different sensors in the long-term operation process, the weighting parameters can be dynamically adjusted to realize adaptive optimization, so that the fusion model can maintain precision and stability as the actual operating conditions change, ensuring that the entire operation and maintenance system always has good data support capability and long-term reliability.

[0013] As preferred, when a device fails, the time difference of arrival method is used for fault location based on the signals collected by the sensors, and the position of the fault section is determined in combination with the topology information of the strong electric system.

[0014] In the technical solution, the signals collected by the sensors are processed in combination with the time difference of arrival method, and the following technical effects can be achieved: First, by using the multi-point distributed sensors to capture the surge or sudden change signal generated by the electrical fault, the propagation path can be calculated according to the time difference of the signal arriving at different measuring points, so that the possible fault range can be quickly narrowed, and compared with the traditional manual checking method, the positioning speed is significantly improved, and the loss caused by long-term shutdown is reduced. Secondly, combining the preliminary fault location calculated by the time difference with the topology information of the highway strong power system can determine the fault section in a complex line branch structure, thereby avoiding the ambiguity caused by relying solely on time difference calculation, achieving more accurate positioning effect, and ensuring that the emergency repair team can arrive at the correct location in the first time; Thirdly, in a wide range of power equipment arrangement environment, the method can ensure the time difference calculation accuracy through synchronous sampling and unified clock mechanism, thereby ensuring that the positioning process is not disturbed by the distance of the sensor or the sampling delay, and improving the stability and reliability of the overall system in the actual environment. Fourthly, by completing the positioning and generating the location result within a short time after the fault occurs, accurate basis can be provided for the subsequent issuance of maintenance instructions, thereby enabling the operation and maintenance system to have rapid reaction capability in the event of a sudden electrical accident, meeting the high requirements of highway operation on safety and continuity.

[0015] As a preferred, the waveform of voltage or current is subjected to discrete Fourier transform, the total harmonic distortion rate is calculated, and the equipment feature vector is extracted based on multi-source data and the total harmonic distortion rate.

[0016] In the technical solution, the discrete Fourier transform is performed and the total harmonic distortion rate is calculated, and the equipment feature vector is extracted by combining multi-source data, which can achieve the following technical effects: Firstly, converting the time domain waveform into frequency domain information and quantifying the harmonic component can reveal the change of power quality of the equipment during operation, thereby identifying potential harmonic pollution and power abnormalities, ensuring that the feature vector not only contains basic operation parameters but also reflects power fluctuation characteristics, providing more comprehensive feature support for equipment state analysis; Secondly, by calculating the total harmonic distortion rate and including it in the feature vector construction together with conventional voltage, current, temperature, vibration and other multi-source data, the importance of frequency domain indicators can be highlighted while maintaining data consistency, thereby enhancing the sensitivity of the feature to equipment performance degradation and making the generation of health indicators more representative and accurate; Thirdly, the combination of frequency domain and time domain data enables them to maintain high diagnostic discrimination in different operating scenarios, thereby avoiding the limitations of single time domain parameters that cannot accurately reflect fault signs, ensuring that the abnormal detection and life prediction links have a more stable and reliable input basis; Fourthly, by performing Fourier transform on the waveform and extracting the total harmonic distortion rate as the core indicator, the feature vector can cover basic electrical parameters, dynamic fluctuation information and harmonic quality indicators, thereby providing higher-dimensional support for subsequent health assessment and risk judgment, meeting the needs of comprehensive monitoring and in-depth analysis of the highway strong power system.

[0017] As preferred, the health indicator is generated by a linear model or a nonlinear model, the linear model being a weighted sum of the device feature vector according to feature weights, and the nonlinear model being a deep neural network including a neural network weight matrix, a bias, an activation function and a scaling function.

[0018] In the technical solution, the health indicator is generated by using a linear model or a nonlinear model, which can achieve the following technical effects: First, the linear model can highlight the influence of key features on the device state by weighting and summing each index of the feature vector according to the weight, thereby quickly forming an intuitive and easily interpretable health score during operation and maintenance, thereby meeting the efficiency requirements of batch evaluation of large-scale equipment while ensuring the transparency and traceability of the results. Second, the nonlinear model can depict complex nonlinear relationships between device operation data by introducing a weight matrix, a bias and an activation function through a deep neural network structure, thereby maintaining high fitting accuracy in the face of large fluctuations in working conditions and multiple feature dimensions, thereby avoiding the shortcomings of traditional linear methods that are difficult to capture potential degradation patterns, and making predictive analysis more consistent with actual operation rules. Third, the combination of linear and nonlinear models can provide flexible options for different application scenarios. In simple scenarios, a linear weighting model is used for quick deployment, and in complex working conditions, a deep neural network is used to obtain higher accuracy, thereby ensuring that the health indicator generation is both universal and adaptive, ensuring stable operation of the system in various environments. Fourth, as a core factor in device lifecycle management, the health indicator generated by the above model can provide high-quality input for subsequent life prediction and risk assessment, enabling maintenance strategies to be based on scientific quantification, thereby forming a closed loop in the overall operation and maintenance system and meeting the actual needs of highway power system upgrades in safety, reliability and intelligence.

[0019] As preferred, the remaining useful life or potential failure risk is fed back to the energy efficiency scheduling and the power distribution is adjusted when generating a maintenance work order.

[0020] In the technical solution, the energy efficiency scheduling feedback and power distribution adjustment can achieve the following technical effects: First, using the results of life prediction and risk assessment directly on energy efficiency scheduling can optimize power distribution according to the health status of the device during operation, which not only ensures stable power supply for key devices, but also allocates standby resources in advance in the case of increased potential risk, thereby improving the overall safety redundancy of the system. Secondly, by combining work order generation with energy efficiency scheduling, the drawbacks of the separation of maintenance and energy management in the traditional mode can be avoided, so that the maintenance decision is no longer isolated, but interacts with the power distribution, thereby reducing the possibility of high-risk equipment accelerating damage due to continuous high-load operation, and ensuring the coordination of the operation and maintenance process; Thirdly, under this condition, the energy efficiency scheduling no longer relies only on load fluctuation and power demand, but considers the health status and remaining life of the equipment, so that a more scientific energy optimization strategy can be realized, which can reduce energy waste and maintenance cost, and meet the dual demands of economy and efficiency of the highway strong power system; Fourthly, by feeding the risk information to the energy efficiency scheduling system in real time, the power supply strategy can be quickly adjusted in emergency situations, so that the power can be prioritized and redistributed among different areas and equipment, thereby maintaining power supply continuity and safety when abnormal events occur, and meeting the requirements of high reliability and rapid response of the operation and maintenance system.

[0021] As a preferred, the priority of the maintenance work order is mainly based on the weighted calculation of the health index, the criticality of the equipment and the downtime loss, and the result of the weighted calculation is used for priority sorting of multiple work orders.

[0022] In the technical scheme, by introducing weighted calculation in the priority sorting of maintenance work orders, the following technical effects can be achieved: Firstly, by including the equipment health index in the priority score, the system can identify equipment with serious deterioration in running state in time, so that maintenance resources can be prioritized to solve objects with high risk degree, ensuring that the overall operation and maintenance decision matches the actual running state of the equipment, and avoiding delays and deviations that may be caused by relying solely on artificial experience; Secondly, by combining the criticality of the equipment in the system for weighting, the equipment that has the greatest impact on the overall operation safety and continuity of the highway can be highlighted, thereby realizing the alignment of resource input and system importance, so as to ensure that limited maintenance forces prioritize key links, and avoid the multiplication of operation risks and economic losses caused by the downtime of key equipment; Thirdly, under this condition, downtime loss is considered as an evaluation factor, which can convert the potential economic impact into a quantitative index for priority calculation, so as to not only focus on safety risks, but also consider the economic benefits of operation and maintenance, making the decision more comprehensive and scientific, and meeting the lean management needs of the intelligent highway operation and maintenance system; Fourthly, the weighted fusion result of the health index, criticality and downtime loss is used for sorting multiple work orders, which can quickly form a clear execution sequence under the condition of multiple tasks and limited resources, so that the maintenance plan has stronger operability and execution force, ensuring that the entire operation and maintenance process is efficient and orderly, and meeting the comprehensive requirements of safety, economy and timeliness of the highway strong power system.

[0023] Further, the application also provides a highway strong power system intelligent operation and maintenance device, comprising: A data acquisition module is configured to monitor equipment in the highway strong power system and collect multi-source data, wherein the multi-source data comprises at least one of voltage, current, temperature, vibration and switch action signal; An edge computing module is configured to receive the multi-source data and perform data preprocessing, wherein the data preprocessing comprises at least one of denoising, cleaning and normalization; An analysis and diagnosis module is configured to perform fault anomaly detection based on the multi-source data after data preprocessing, calculate health indicators of the equipment and predict the remaining useful life; A maintenance instruction module is configured to generate maintenance work orders according to the health indicators or the remaining useful life, and sort and distribute the maintenance work orders according to priority scores.

[0024] By adopting the above technical solutions, the data acquisition module, the edge computing module, the analysis and diagnosis module and the maintenance instruction module are arranged in the system, thereby realizing intelligent operation and maintenance of the whole process of the highway strong power system, and improving the integrity of information perception, the accuracy of operation diagnosis and the efficiency of maintenance scheduling.

[0025] The data acquisition module monitors and collects multi-source information such as voltage, current, temperature, vibration and switch action signal, thereby realizing comprehensive coverage of operation data and providing continuous and reliable raw input for system operation state; the edge computing module performs preprocessing such as denoising, cleaning and normalization on the received data, thereby realizing standardized processing of multi-source heterogeneous data, effectively eliminating noise interference and maintaining data consistency, and thereby laying a foundation for subsequent modeling and analysis; the analysis and diagnosis module performs fault anomaly detection on the preprocessed data, and generates health indicators and life prediction in combination with feature calculation, thereby realizing dynamic evaluation of equipment operation state, discovering potential risks in time and quantifying equipment reliability, and providing a scientific basis for maintenance planning; the maintenance instruction module sorts and reasonably distributes the generated maintenance work orders according to priority, thereby realizing efficient use of resources and orderly advancement of task scheduling, ensuring that key equipment is given priority, and thereby improving the timeliness of emergency disposal and the safety of overall system operation.

[0026] Preferably, the health indicators are generated by a linear weighting model or a nonlinear model, and the nonlinear model is a deep neural network model, which comprises a neural network weight matrix, a bias, an activation function and a scaling function.

[0027] In the analysis and diagnosis module, the feature information is comprehensively processed by using the weighting model and the deep network, thereby achieving the following technical effects: First, by proportionally integrating various operating characteristics according to their importance through linear weighting, an intuitive health degree evaluation result can be quickly obtained, which not only ensures the simplicity of calculation, but also facilitates the understanding and tracing of the model basis for the operation and maintenance personnel, and provides an efficient and feasible tool for batch analysis of large-scale equipment. Second, when facing complex operating environments and large data dimensions, the introduction of deep networks for nonlinear mapping of feature relationships can capture the potential patterns and degradation rules hidden behind the data, thereby improving the sensitivity of health indicators to early abnormalities and enhancing the early warning capability of device state changes. Third, linear and nonlinear methods complement each other in the same system, allowing flexible switching and combination in different application scenarios. In simple working conditions, linear integration is relied on for rapid deployment, while in complex working conditions, deep networks are relied on for higher accuracy, thereby ensuring the adaptability and long-term stability of the health indicator generation process.

[0028] As a preferred embodiment, the remaining useful life is predicted by a survival analysis method, which models the device failure time through a survival function, a hazard function, or a cumulative hazard function.

[0029] The survival analysis method in the analysis and diagnosis module models and predicts the device life, which can achieve the following technical effects: First, by quantifying the probability of the device remaining in normal operation within different time intervals through the survival function, an intuitive reference based on the time dimension can be provided to the operation and maintenance personnel. This not only allows the evaluation of the survival probability of the device at a certain time in the future, but also helps to develop a reasonable inspection and maintenance cycle, avoiding excessive maintenance or excessive reliance on experience. Second, by depicting the trend of failure risk over time through the hazard function, the law of gradual accumulation of risk in the later stage of device operation can be revealed, thereby allowing for early intervention measures at the stage of rapid risk increase, thereby reducing the probability of sudden failure and providing data basis for the priority allocation of operation and maintenance resources, making the entire operation and maintenance system more proactive and proactive. Third, by modeling the cumulative effect of risk in the time dimension using the cumulative hazard function, the full-cycle risk accumulation of the device from initial operation to potential failure can be exhibited. This allows the operation and maintenance department to grasp the health change process of the device throughout its life cycle, ensuring timely adjustment of maintenance plans at critical stages and improving the safety and reliability of the system.

[0030] The beneficial effects of this invention are that by generating work orders through multi-source data acquisition, state modeling and estimation, anomaly detection, health assessment and life prediction, it realizes intelligent operation and maintenance of the entire process of highway power systems, which can improve the real-time perception and diagnosis capabilities of the operating status and meet the needs of improving the real-time performance of operation and maintenance, unified data management, enhanced intelligent diagnosis and efficient emergency response.

[0031] Furthermore, the design principle of this invention is reliable, the structure is simple, and it has a very wide range of application prospects.

[0032] Therefore, it is evident that the present invention has outstanding substantive features and significant progress compared with the prior art, and the beneficial effects of its implementation are also obvious. Attached Figure Description

[0033] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0034] Figure 1 This is a flowchart of an intelligent operation and maintenance method for a highway power system provided by the present invention; Figure 2 This is a schematic diagram of an intelligent operation and maintenance device for a highway power system provided by the present invention.

[0035] The module consists of 1. a data acquisition module, 2. an edge computing module, 3. an analysis and diagnosis module, and 4. a maintenance instruction module. Detailed Implementation

[0036] The present invention will now be described in detail with reference to the accompanying drawings and specific embodiments. The following embodiments are explanations of the present invention, but the present invention is not limited to the following implementation methods.

[0037] Example 1: like Figure 1 As shown in the figure, this embodiment provides an intelligent operation and maintenance method for a highway power system, which includes the following steps: Step S1: Collect multi-source data and perform data preprocessing. The multi-source data includes at least one of voltage, current, temperature, vibration and switching action signals. The data preprocessing includes at least one of noise reduction, cleaning and normalization. Step S2: Based on multi-source data, construct a state-space model and define a system state vector in the state-space model to represent the dynamic relationship between the system's operating state and external inputs; Step S3: estimating the system state vector by using a state estimation algorithm to obtain an estimated state value; Step S4: performing fault anomaly detection based on the estimated state value and the multi-source data, the fault anomaly detection including mutation detection based on cumulative sum and change point detection, fault detection based on residual statistical test, and anomaly detection based on machine learning; Step S5: extracting a device feature vector based on the multi-source data and calculating a health index; Step S6: predicting the remaining useful life of the device according to the health index and determining a potential fault risk, and when the potential fault risk exceeds a preset risk threshold, generating a corresponding maintenance work order and determining the priority of the maintenance work order.

[0038] By using the above technical solutions, the multi-source data is collected, the state space model is constructed, the state estimation and fault anomaly detection are performed, the device feature is extracted and the health index is calculated, and the maintenance work order is generated in combination with the life prediction, thereby realizing the whole-process intelligent operation and maintenance of the highway strong power system, improving the real-time perception and diagnosis capability of the running state, and meeting the needs of real-time improvement of operation and maintenance, unified management of data, enhanced intelligent diagnosis, and efficient emergency disposal.

[0039] Specifically, by collecting multi-source data such as voltage, current, temperature, vibration, and switch action signals, and combining with preprocessing operations such as denoising, cleaning, and normalization, the standardized processing of different sources of running information is realized, thereby avoiding the problem of scattered and isolated data, and providing a unified input basis for subsequent modeling and analysis; the state space model is constructed based on the multi-source data, and the system state vector is defined therein, so that the dynamic relationship between the system running state and the external input can be accurately characterized; then the system state is estimated by using a state estimation algorithm, and the estimated state value obtained can reflect the real running condition of the device, overcoming the limitations of the traditional manual inspection mode lag and information lag; based on the estimated state value and the multi-source data, anomaly detection is carried out, and in combination with the mutation detection, residual statistical test, and machine learning algorithm, a multi-level fault diagnosis mechanism is realized, which can timely discover abnormal conditions in operation and improve the intelligent identification capability of device faults, thereby improving the problem of insufficient diagnosis capability of the prior art; by extracting the device feature vector and calculating the health index, the remaining useful life is predicted and the potential fault risk is determined, and when the risk exceeds the threshold, the maintenance work order is automatically generated and the priority is determined, and this process realizes the change from "after-event maintenance" to "predictive maintenance", and can directly associate the risk result with the work order scheduling, thereby responding in advance before the fault occurs, and meeting the needs of timeliness of emergency response in the operation and maintenance of the highway strong power.

[0040] Hereinafter, according to the embodiments of the present application, the above steps S1 to S6 are specifically described.

[0041] In step S1, it is first necessary to collect multi-source data on the operating environment of the expressway strong electric system. Since the expressway involves power stations, power distribution rooms, tunnels, toll stations and service areas along the line, and various types of electrical equipment are deployed, the collection objects include not only basic electrical quantities such as voltage and current, but also temperature, mechanical vibration, and action signals of switch devices such as circuit breakers and contactors.

[0042] In order to ensure the comprehensiveness and accuracy of the collection, multiple types of sensors are installed at each key device, such as voltage and current transformers for electrical quantity acquisition, thermocouples or thermistors for temperature monitoring, acceleration sensors for vibration characteristic measurement, and auxiliary switch contacts or intelligent collection units for recording switch actions. All sensors are connected to the edge computing node through a unified collection network to realize real-time aggregation of raw signals. Exemplarily, these sensors can use M-type sensors.

[0043] Raw data cannot be directly entered into the analysis link, so data preprocessing is required. In the preprocessing process, first, the noise in the signal needs to be suppressed. Taking voltage and current waveforms as examples, due to strong electromagnetic interference and switching harmonics of power electronic equipment in the field, the original sampling waveform will contain a large number of high-frequency components. By setting a digital filter (such as a finite impulse response filter or a wavelet transform denoising method), high-frequency interference can be effectively removed while retaining the main characteristics of the fundamental wave and low-order harmonics, thereby ensuring that the signal is clean and representative; in temperature and vibration signal processing, sliding average or adaptive filtering can also be used to smooth the sampling curve, so that subsequent analysis is based on stable data.

[0044] After denoising, data cleaning needs to be addressed. Some sensors may produce incomplete or abnormal values due to sampling packet loss, communication interruption or hardware failure, such as current sampling points that exceed the physical reasonable range or temperature values that change abruptly. For such cases, an abnormal value identification algorithm can be configured to find and process error data through threshold discrimination or residual inspection based on historical trends. For small-scale missing points, linear interpolation or spline interpolation can be used for repair; for large areas of invalid data, they are directly excluded to avoid affecting the stability of the overall modeling.

[0045] The last step is normalization. Due to the obvious difference in the dimension and order of magnitude of different physical quantities, for example, the voltage may be in the kilovolt level, and the temperature is only in the tens of degrees Celsius. If directly input into the modeling or diagnosis algorithm, it is easy to cause imbalance between the characteristic quantities. Therefore, it is necessary to standardize the mapping of all collected data. Common methods include using a zero-mean unit-variance standardization method, or using interval scaling to map the data to the [0, 1] range. In this way, the voltage, current, temperature, vibration and switch signal are uniformly converted into dimensionless and comparable input quantities, so as to ensure that various features can play a role on the same scale in subsequent state modeling and algorithm processing.

[0046] Through the above process, step S1 can realize the conversion of the multi-source heterogeneous signals obtained from the complex field environment to the standardized data that can be directly modeled and utilized, ensuring that the state estimation and abnormal detection of the highway strong electric system are based on accurate, stable and consistent data input.

[0047] In step S2, based on the preprocessed multi-source data obtained in step S1, a discrete-time state space model is established to describe the dynamic relationship between the running state of the highway strong electric system and the external input in a unified mathematical framework.

[0048] Specifically, in the embodiments of the present application, the sampling period is taken as the discrete time The system evolution is represented by The system state vector with a dimension of is selected as the internal representation of the system. This vector can include but is not limited to the linearized variables of the voltage amplitude and phase angle of the bus or key node, the equivalent thermal state of the transformer core / winding, the SOC (State of Charge) and equivalent internal resistance parameter of the energy storage device, the equivalent thermal-electric coupling state of the long-distance cable, and the equivalent position / wear of the switching element, which are used to completely cover the main electric-thermal dynamics of the strong electric equipment in a mathematical sense; meanwhile, the exogenous / control input is defined, which can be the tap voltage regulation instruction, the reactive power compensation switching command, the energy storage charging and discharging power setting, the ring network or bypass switching instruction, and the active / reactive power target from the upper layer dispatching, etc. These quantities represent the observable or recordable external action exerted on the system; in order to accommodate the non-stationarity of the field operation and the model simplification error, the process disturbance is explicitly introduced in the state evolution equation to absorb the influence of unmodeled dynamics and disturbances.

[0049] Based on the above setting, the system evolution can be expressed as a nonlinear or weakly nonlinear discrete state equation:

[0050] where, is a slowly time-varying or condition-varying state transition map, is a zero-mean stochastic process with covariance .

[0051] When linearizable around the operating point, a linear time-varying approximation can be obtained:

[0052] where, is a state transition matrix, reflecting the relationship of the system state itself evolving over time, obtained from the linearized Jacobian ; is an input matrix, reflecting the effect of external inputs on the state evolution, obtained from the linearized Jacobian ; in the scenario of network topology changing due to switching, and are piecewise time-varying with the change of topology mode, so that the structural changes of the power distribution network are reflected in the state dynamics in the form of parameter time-varying, realizing the unified description of the electro-thermal coupling and topology switching dynamics of the strong electric system.

[0053] is to establish a connection between the abstract state quantity and the field instrument reading, at each sensor / measuring point , a dimensional observation vector at time is defined, whose elements can correspond to the instantaneous or RMS (Root Mean Square) value of phase voltage / phase current, device temperature, vibration index, and power quality characteristics derived from voltage or current waveform (such as THD (Total Harmonic Distortion) commonly used in subsequent analysis, but not limited to, which corresponds to the nonlinear mapping of the observation function to the state output at the measurement level).

[0054] The measurement relationship adopts a general multi-sensor measurement model:

[0055] where, is the observation function matched with the th sensor type and installation location, which can be linear or nonlinear; represents a zero-mean observation noise with covariance (which can vary over time).

[0056] For the convenience of overall modeling, the observations of each measuring point can be spliced into an aggregated observation with a dimension of ​ The corresponding observation mapping is .

[0057] In the embodiments of the present application, if linearization is performed on the operating point , the following is obtained:

[0058] wherein, is the observation matrix, which is obtained from the linearized Jacobian ; is the observation noise, the covariance of which is composed of the noise covariance of each measurement point according to the block diagonal or correlation structure.

[0059] Through this "state-observation" dual structure, a differentiable mapping between the internal mechanism of the system and the external measurable quantity is completed, so that measurements of different sources and different physical quantities are consistently expressed in a unified mathematical space.

[0060] Further, in the embodiments of the present application, in order to ensure the implementability of the model in engineering, the initial conditions and the time-varying nature of the parameters need to be determined during the modeling stage. Illustratively, the initial state may be approximately obtained from the historical steady-state solution, the power-on self-test reading or the statistics of a time window, and the covariance reflects the uncertainty; the system parameters (such as the equivalent admittance of the line, the load characteristic parameters, the heat capacity and the heat resistance, etc.) slowly drift or change with the environment and the season in long-term operation, and these changes are reflected in the time subscript of or the time-varying form of , , in the linear model, thereby ensuring the adaptability and robustness of the model to the long-term and cross-scene operation of the strong power system; for scenes containing phasor measurements and three-phase unbalanced working conditions, the sub-state can be expressed in the phasor domain, the symmetrical component domain or the synchronous rotating coordinate system (dq0) during modeling to reduce the dimension and enhance the observability, but regardless of the coordinate representation, the transformation between it and the physical measurement is included in the definition of or to maintain the consistency of the observation mapping and the field measurement.

[0061] In combination with the complexity of the highway strong power system, the state space model in the above step S2 covers not only the linearized power flow and power balance relationship of the electrical side of the distribution network, but also the slow states related to the heat-life of the key equipment, and through the time-varying description of the input and the topology mode, the model can adapt to different operation scenes and operation strategies without changing the unified mathematical structure.

[0062] In step S3, a state estimation algorithm needs to be used to calculate the system state vector based on the state space model established in step S2, so as to obtain an estimated state value that can truly reflect the running state of the device.

[0063] However, in the strong electric system of the expressway, the multi-source data such as voltage, current, temperature, vibration and switching signal have complex sources and different measurement accuracies. If not reasonably fused and estimated, the error of a single channel may be directly amplified, resulting in distorted analysis conclusion.

[0064] Therefore, in the embodiments of the present application, in the state estimation link, a weighted least squares (WLS) method is first introduced to fuse the multi-source data, and then the fusion result is taken as the input of a state estimation algorithm such as Kalman filter (KF) or recursive least squares (RLS), so as to ensure the stability and accuracy of the estimation result.

[0065] Specifically, first, a linear observation model of the system at time is considered, which can be expressed as:

[0066] wherein is an observation vector, which is spliced by the measurement values of multiple sensors, including electrical quantities (such as phase voltage and phase current), environmental quantities (such as temperature) and mechanical quantities (such as vibration acceleration); is an observation matrix, which is determined by the device layout and the measurement point configuration, and is spliced by each in the vertical direction; is a state vector, which is a quantity to be estimated; is an observation noise, which is usually assumed to be a random variable with a mean of zero.

[0067] Since the noise characteristics of different sensors are quite different, for example, the noise of the current transformer increases significantly when affected by electromagnetic interference, while the random error of the temperature sensor is relatively stable, therefore, a weighting matrix needs to be constructed, wherein is the covariance matrix (diagonal or block diagonal) of the overall observation noise, based on which, the weight of the channel with small noise will be larger, while the weight of the channel with large noise will be smaller, and thus the weighted least squares estimation formula is:

[0068] The implementation of the formula in engineering is as follows: first, the noise variances of each measurement point are obtained by using the calibration data and the historical running statistics of the sensors, and the covariance matrix is constructed.Secondly, the weight matrix is obtained by matrix inversion Finally, the fused estimation value is calculated according to the formula.

[0069] The method has the advantages that it can obtain a globally consistent estimation under the condition of multi-sensor redundant observation, and even if an individual measuring point temporarily fails or noise increases, the overall estimation result can still remain stable. For example, in a tunnel ventilation scene, if a temperature sensor drifts due to environmental humidity, the algorithm will automatically reduce the weight of the measuring point in estimation, avoiding the transmission of false information to the state estimation result.

[0070] In this step, the multi-source data is uniformly processed by using the weighted least squares method, which can achieve the following technical effects: First, by introducing a weighting strategy, the role of high-confidence data can be highlighted, and the interference of unstable data can be weakened, so that the fusion result is closer to the true state of the system, thereby significantly improving the reliability and stability of the input information in the operation and maintenance process; Second, by completing error correction and redundancy utilization in advance before the data is input into the state estimation algorithm, the convergence speed of the algorithm can be faster, and the estimation process can be smoother, thereby avoiding estimation bias and misjudgment risk caused by single data anomaly, and providing a more solid foundation for subsequent system modeling and state quantity calculation; Third, by globally processing data from voltage, current, temperature, vibration and action signals, a unified data fusion interface can be formed in a large-scale highway strong power system, which not only improves the centralization level of data management, but also ensures the input consistency of the subsequent state estimation and diagnosis process, thereby meeting the efficiency and real-time demand of operation and maintenance information processing in the highway scene; Fourth, by continuously accumulating noise characteristics and credibility information of different sensors in the long-term operation process, the weighting parameters can be dynamically adjusted to realize adaptive optimization, so that the fusion model can maintain precision and stability as the actual operating conditions change, ensuring that the entire operation and maintenance system always has good data support capability and long-term reliability.

[0071] After completing the weighted fusion, the state estimation algorithm part is entered. For the scene where the approximation is linear and the noise follows a Gaussian distribution, the Kalman filter method can be used for dynamic recursion.

[0072] Specifically, the Kalman filter is divided into prediction and update processes, wherein the prediction (time update) formula is:

[0073]

[0074] When the observation At time of arrival, the update (observation correction) formula is used:

[0075]

[0076]

[0077] wherein, represents the state at time State prediction before observation arrives; represents the posterior estimation at time ; , represents the corresponding prediction error covariance matrix; is the process noise covariance, in actual engineering, it can be set by statistics of equipment disturbance, for example, the variance of bus voltage fluctuation is modeled after long-term monitoring; is the unit matrix; is the Kalman gain, used to balance the contribution of prediction results and actual observations, when the observation noise is small, the gain increases, and the measurement result is more dependent; when the observation noise is large, the model prediction is more dependent. This dynamic weighting mechanism ensures stable state estimation values under different working conditions. For example, at night on the highway under low load, the current sensor signal noise is relatively obvious, and the algorithm will automatically reduce its weight to avoid false observations dominating the results.

[0078] Further, in some actual working conditions, the system not only contains state quantities, but also contains unknown parameters, for example, the resistance and inductance of the power transmission line change with temperature rise, and the capacity of the battery gradually decreases with the number of cycles. Therefore, in the embodiments of the present application, the RLS method is introduced to identify the parameters online.

[0079] Specifically, assuming that the state or observation model contains an unknown parameter vector , the regression feature vector is , and the observation is , the recursive formula is:

[0080]

[0081]

[0082] wherein is the parameter covariance matrix, is the forgetting factor, when When approaching 1, the algorithm gives less weight to new data, and the parameter updates slowly, which is suitable for the scenario where the parameter is long-term stable; when is small, it can quickly track mutations, and is suitable for the scenario where the device parameter changes rapidly due to faults or external environmental changes, such as in the strong power system of the expressway, the cable insulation resistance rapidly decreases due to humidity changes, at this time, the value of can be reduced to achieve rapid updates.

[0083] It should be noted that the entire state estimation process not only outputs the state vector but also outputs the error covariance matrix , which provides a confidence interval for the estimation result. In actual operation and maintenance, this information is extremely important, and operation and maintenance personnel can determine whether the estimation result is reliable according to this information, for example, if the confidence interval of the temperature estimation is too wide, it means that the sensor data is insufficient or the noise is too large, and the system can automatically trigger the calling of redundant measuring points or prompt manual review.

[0084] In the above step S3, by combining the fusion of multi-source data and the state estimation algorithm, the state space model defined in the state space model during the operation of the strong power system of the expressway can be recursively calculated, so as to obtain the estimated state value at the corresponding time, which is used as the basic input for subsequent steps of fault and anomaly detection and device health index analysis.

[0085] In step S4, based on the estimated state value obtained in the foregoing steps and the actually collected multi-source data, fault and anomaly detection is performed.

[0086] Because the running environment of the strong power system of the expressway is complex, sudden abnormalities of electrical quantities may occur, slow-accumulating hidden dangers may occur, and even nonlinear abnormal patterns between multiple variables exist. Therefore, in the embodiments of the present application, a multi-level parallel method is adopted in the fault and anomaly detection mechanism, including mutation detection based on CUSUM (Cumulative Sum), statistical test based on residual, and anomaly recognition detection based on a machine learning model, and when a fault occurs, TDOA (Time Difference Of Arrival) method is combined for fault positioning, and the determination of the fault section is realized in cooperation with the topology structure of the system.

[0087] In mutation detection, the CUSUM method is used to identify the change point of the monitored quantity. Specifically, the expected value of a certain monitored index in the normal state is , the tolerance drift constant (usually set by performance requirements) is , and the observation value is , the cumulative sum statistic is defined as , and when Exceeding the preset threshold When an abnormal change occurs in the monitored quantity, it is immediately determined. This method is particularly effective in detecting step changes in current or voltage waveforms; for example, when a circuit breaker fails to operate, causing a sharp rise in current, CUSUM can trigger an alarm within a short time. (Threshold) The settings can be achieved by combining the protection action level and actual operation statistics, thereby striking a balance between detection sensitivity and false alarm rate.

[0088] In the detection of incremental or latent faults, statistical tests based on residuals can be used. Residuals refer to the difference between the actual observed value and the estimated value predicted by the state-space model. Further define the standardized residual statistic: ,in, Let be the residual covariance matrix. An anomaly is considered to exist if the chi-square distribution threshold, determined based on degrees of freedom and confidence levels, is exceeded. Unlike simple threshold monitoring, this method uses model predictions as a reference, enabling it to identify potential inconsistencies in the system when observations appear normal. For example, in the initial degradation stage of cable insulation, voltage fluctuations may still be within acceptable limits, but the difference from model predictions may gradually accumulate. In this case, residual statistics will reflect the problem in advance.

[0089] Under complex operating conditions, traditional statistical methods may be insufficient to identify anomalies involving nonlinearity and multidimensional coupling. Therefore, in this embodiment, a machine learning-based anomaly detection method is also introduced. Specifically, the autoencoder model, trained on normal data, enables the network to reconstruct the input. For new input samples, if the reconstruction error... If the threshold is exceeded, it is judged as abnormal; this method can discover complex pattern anomalies that traditional feature indicators cannot show.

[0090] Alternatively, the Isolation Forest (IF) method can be used. This method constructs multiple random trees to segment the samples. When the average path length of a sample point within a tree is significantly shorter than that of normal points, it indicates that the point is an anomaly. This type of method does not require an explicit physical model and can maintain its detection capability even under unknown conditions. It is suitable for atypical anomalies in highway power systems under different climatic, seasonal, or load conditions.

[0091] Once a fault is detected, it is necessary to further determine its spatial location. In some embodiments of this application, the TDOA method combined with system topology information can be used for location. Specifically, it is assumed that sensors deployed at different locations are at time... , The fault wavefront signal (e.g., a sudden change in current during a surge or arc) is captured. Assume the signal propagation speed is... , then the distance difference between the fault point and the two measuring points satisfies: , wherein, is the physical distance from the fault point to the sensor , and is the physical distance from the fault point to the sensor .

[0092] Further, by simultaneous collection of the multi-point sensors, a set of equations can be formed, and then the spatial position of the fault point is solved by using the least square method or other optimization methods. To ensure accuracy, all sensors need to be synchronized in a unified clock, and after obtaining the geometric solution of the fault position, the network topology of the highway strong power system is combined to map to a specific line section, to complete the accurate positioning of the fault section. For example, in a distribution line with multiple branches, if the calculation result shows that the fault point is located near the intersection of the main trunk and a branch, then the topology structure is combined to determine that the fault occurs in the specific branch, so as to guide the maintenance personnel to quickly reach the target position. Exemplarily, GPS (Global Positioning System) time service or IEEE 1588 precision clock synchronization protocol can be used for unified clock synchronization.

[0093] In this step, by using the signals collected by the sensors and combining the time difference of arrival method for processing, the following technical effects can be achieved: First, by means of the multi-point arranged sensors to capture the surge or sudden signal generated by the electrical fault, the propagation path can be calculated according to the time difference of arrival of the signal at different measuring points, so as to quickly narrow down the possible fault range, compared with the traditional manual checking method, the positioning speed is significantly improved, and the loss caused by long-time shutdown is reduced; Second, combining the preliminary fault position calculated by the time difference with the topology information of the highway strong power system, the fault section can be clearly attributed in the complex line branch structure, so as to avoid the ambiguity caused by simply relying on time difference calculation, realize more accurate positioning effect, and ensure that the emergency repair team can arrive at the correct position in the first time; Third, in the environment of large-scale power equipment arrangement, this method can ensure the time difference calculation accuracy through synchronous sampling and unified clock mechanism, so as to ensure that the positioning process is not disturbed by the distance nearness or sampling delay of the sensors, and improve the stability and reliability of the overall system in the actual environment; Fourth, by completing the positioning and generating the position result within a short time after the fault occurs, accurate basis can be provided for the subsequent issuance of maintenance instructions, so as to enable the operation and maintenance system to have rapid reaction capability under sudden electrical accidents, and meet the high requirements of highway operation on safety and continuity.

[0094] In the embodiments of the present application, in order to improve the robustness and accuracy of detection, in actual deployment, allow the mutation detection, residual statistical test and machine learning model to run at the same time, and fuse different detection results at the upper layer; when the three methods determine the abnormality at the same time, trigger high-level alarm; if only a single method detects the abnormality, enter the observation delay confirmation mode to avoid false alarm. Through this hierarchical fusion mechanism, both the sensitivity of detection can be guaranteed, and the false alarm probability caused by single-point abnormality in complex environment can be reduced.

[0095] In summary, in step S4, by combining the estimated state value with the multi-source observation data, fault and abnormality detection of the high-speed highway strong electric system operation state can be realized at different levels; the mutation detection can quickly identify the sharp deviation of the operation parameter, the residual statistical test can reveal the potential inconsistency, and the machine learning model can cover complex and atypical abnormal patterns; when a fault occurs, combined with the time difference method and system topology information, the specific line section where the fault is located can be further determined. The above detection and positioning mechanism provides necessary data and state support for subsequent construction of health indicators based on device characteristics and residual life prediction.

[0096] In step S5, based on the multi-source data and state estimation results obtained in the foregoing steps, the device feature vector is further extracted, and the health index is calculated based on this.

[0097] Specifically, the device feature vector, as a high-dimensional representation form of operation data, covers the characteristics of electrical quantities, environmental quantities and device working states, and is the basis for realizing device health quantification and subsequent life prediction.

[0098] In some embodiments of the present application, in order to ensure the sensitivity of the feature vector to device performance degradation and abnormality, on the basis of conventional voltage, current, temperature and vibration data, power quality indicators such as harmonic distortion are introduced, and a more comprehensive feature set is constructed combined with multi-source information.

[0099] In the feature construction process, first, the time domain waveform of voltage or current is subjected to Discrete Fourier Transform (DFT) to obtain its frequency domain components. Let the sampling sequence be , the number of sampling points be , and the m-th harmonic component be:

[0100] Among them, represents the amplitude of the harmonic, and THD can be defined as: ​

[0101] wherein, is the fundamental amplitude, is the amplitude of the th harmonic, is the selected maximum harmonic order. The obtained by the above calculation can quantitatively represent the power quality situation, and if the significantly increases, it indicates that there may be non-linear loads, increased harmonic sources or equipment degradation in the system.

[0102] After obtaining the , the is combined with voltage, current, temperature, vibration and switching operation frequency and other conventional monitoring quantities to construct the feature vector of the equipment. Specifically, for a certain equipment , the feature vector extracted at time is:

[0103] wherein, may represent the normalized voltage, the normalized current, the short-time change rate of the RMS current, the temperature change amplitude, the root mean square value of the vibration signal, the number of switching actions within a certain time window, the THD, etc. Other components can be extended according to the operating characteristics of the equipment, such as the load fluctuation rate, the phase imbalance coefficient, etc., wherein each element is normalized. The feature vector formed in this way not only contains the intuitive information of the basic physical quantities, but also introduces the frequency domain and statistical characteristics, and can depict the operating state of the equipment from multiple dimensions. Exemplarily, the normalized voltage , the normalized current , the short-time change rate of the RMS current , the number of switching actions within a certain time window , , and the like. .

[0104] In this step, the total harmonic distortion rate is calculated by discrete Fourier transform, and the feature vector of the equipment is extracted in combination with multi-source data, which can achieve the following technical effects: Firstly, the time-domain waveform is converted into frequency-domain information and the harmonic components are quantified, which can reveal the changes of the power quality of the equipment during operation, so as to identify potential harmonic pollution and power abnormalities, and ensure that the feature vector not only contains basic operating parameters, but also reflects the power fluctuation characteristics, thereby providing more comprehensive feature support for equipment state analysis. Second, by calculating the total harmonic distortion rate and incorporating it into the feature vector along with conventional multi-source data such as voltage, current, temperature, and vibration, the importance of frequency domain indicators can be highlighted while maintaining data consistency. This enhances the sensitivity of features to equipment performance degradation, making the generation of health indicators more representative and accurate. Third, the combined use of frequency domain and time domain data enables it to maintain a high diagnostic discrimination under different operating scenarios, thereby avoiding the limitation that a single time domain parameter cannot accurately reflect fault symptoms, and ensuring that the anomaly detection and life prediction links have a more stable and reliable input basis. Fourth, by performing Fourier transform on the waveform and extracting the total harmonic distortion rate as a core indicator, the feature vector can encompass basic electrical parameters, dynamic fluctuation information, and harmonic quality indicators. This provides higher-dimensional support for subsequent health assessments and risk determinations, meeting the needs of highway power systems for comprehensive monitoring and in-depth analysis.

[0105] After feature extraction is completed, the device's health index (HI) needs to be calculated. The health index is a quantitative measure of the device's status, and its value range is usually set in [0,1] or [0,100]. In the embodiments of this application, the lower the HI value, the closer the device status is to normal, and the higher the value, the greater the risk of degradation or potential failure.

[0106] In some embodiments of this application, to ensure that the health indicator is both intuitive and adaptable to complex scenarios, HI can be generated using a linear or nonlinear model. Specifically, in a linear model, the health indicator is calculated using a weighted summation: Let at time... Device feature vectors The corresponding weight is health indicators for:

[0107] in, For equipment At any moment The Features; weights The importance of features in health metrics can be obtained through statistical analysis of historical data, expert experience, or regression model training based on accident cases. For example, when research indicates that current overload and temperature rise are key precursors to transformer failure, the corresponding features will be assigned higher weights. Under linear models, the calculation of health indicators is transparent, facilitating manual interpretation and traceability, making it suitable for scenarios requiring rapid batch evaluation of a large number of devices.

[0108] In nonlinear models, deep neural networks are introduced to map features in order to capture the complex nonlinear relationships between features: Let the input layer be the device feature vector. A multi-layer structure is formed by linear mapping through weight matrices and bias terms, and by introducing nonlinear activation functions between layers. Its basic form is as follows:

[0109] in, For the first The output of the layer, This is the weight matrix. For bias vectors, For the activation function, you can choose ReLU (Rectified Linear Unit), Sigmoid (Sigmoid Function), or Tanh (Hyperbolic Tangent Function), etc. The final output layer is scaled by a scaling function. The results are mapped to a preset range of health indicators, for example, using the Sigmoid function or a maximum / minimum normalization mapping, mapping the output to [0,1]. For instance, when the second layer is the final output layer, the output of the first layer is... The output of the second layer is After mapping Through training a neural network, it can automatically learn the complex nonlinear coupling relationships between features, thus providing accurate health indicators even when facing atypical working conditions.

[0110] In engineering implementation, linear models are computationally simple and suitable for deployment on edge devices for real-time and rapid evaluation; nonlinear models are computationally complex but can run in the cloud or on a central server, and can be trained using a large number of historical samples before being used for online prediction. In order to ensure the applicability of the model in different scenarios, in the embodiments of this application, linear and nonlinear models are allowed to be used in parallel. For example, the linear model can be used for rapid screening first, and then the nonlinear model can be used for refined diagnosis of suspected abnormal devices.

[0111] This step, which uses linear or nonlinear models to generate health indicators, can achieve the following technical effects: First, the linear model can highlight the impact of key features on equipment status by weighting and summing the various indicators of the feature vector according to their weights. This allows for the rapid generation of intuitive and easily interpretable health scores during operation and maintenance, thus meeting the efficiency requirements for batch evaluation of large-scale equipment while ensuring the transparency and traceability of the results. Second, the nonlinear model introduces weight matrix, bias and activation function through deep neural network structure, which can depict the complex nonlinear relationship between device operation data, thereby maintaining high fitting accuracy in the face of large working condition fluctuations and multiple feature dimensions, thereby avoiding the shortcomings of traditional linear methods that are difficult to capture potential degradation patterns, and making the predictive analysis more in line with the actual operation law; Third, the combination of linear and nonlinear models under this condition can provide flexible choices for different application scenarios. In simple scenarios, linear weighted models are used for quick deployment, and deep neural networks are used in complex working conditions to obtain higher accuracy. In this way, it can ensure that the health indicators generated have both universality and adaptability, ensuring that the system can run stably in various environments; Fourth, as the core factor of device life cycle management, the health indicators generated by the above model can provide high-quality input for subsequent life prediction and risk assessment, so that the maintenance strategy is based on scientific quantification, thereby forming a closed loop in the overall operation system and meeting the actual needs of highway strong power system for safety, reliability and intelligent upgrading.

[0112] Further, in the process of feature vector construction and health indicator calculation, data normalization and dynamic adjustment are also needed. Due to the large difference in physical dimensions of different features, such as voltage measured in kV, temperature measured in Celsius, vibration measured in acceleration, and scaling function mapping, if not handled uniformly, it will lead to a certain feature occupying too much weight in the model.

[0113] Therefore, in the embodiments of the present application, the normalization method can use zero mean standardization or maximum and minimum value normalization, and the general formula is respectively:

[0114] wherein, is the feature vector after normalization, , , and are the mean, standard deviation, minimum value and maximum value of the feature respectively. Through the above method, all features are ensured to be in a unified scale, thereby avoiding the bias caused by numerical imbalance.

[0115] Further, to ensure the effectiveness of the health indicator in long-term operation, the feature weights or neural network parameters can also be updated based on new data: in linear models, the weights can be re-estimated by logistic regression or Lasso algorithm (Least Absolute Shrinkage and Selection Operator); in nonlinear models, new data can be absorbed without losing existing knowledge through incremental learning or transfer learning. This dynamic updating mechanism ensures that the sensitivity of the health indicator to the degradation characteristics of the equipment does not decrease over time.

[0116] Through the above method, step S5 can convert multi-source data into a feature vector and calculate the equipment health indicator based on it; the construction process of the feature vector introduces multiple features in time domain, frequency domain and statistical dimension, so that the health indicator not only depends on the basic operating parameters, but also integrates power quality and dynamic fluctuation information; the generation of the health indicator combines linear and nonlinear models to ensure calculation efficiency while enhancing adaptability to complex working conditions. This result will be used for life prediction and risk judgment in subsequent steps.

[0117] In step S6, the remaining useful life of the equipment is further predicted using the equipment health indicator obtained in step S5, and the potential failure risk is judged based on the life prediction result. When the potential failure risk level exceeds the preset risk threshold, a maintenance work order is automatically generated, and the resource is reasonably allocated and scheduled by comprehensively calculating the work order priority, thereby establishing a closed-loop mechanism from health assessment to maintenance execution.

[0118] Specifically, in the life prediction link, survival analysis method can be used to model the equipment operating life, and let the random variable represent the failure time of the equipment, the current time is , and the historical observation information is , then the remaining life of the equipment at time can be represented as , which represents the expected value of the future running time of the equipment based on the existing historical information under the condition that the equipment has not failed yet.

[0119] In the embodiments of the present application, the distribution of which can be characterized by the survival function , that is, , which represents the probability that the equipment can still operate normally after time . The corresponding hazard function reflects the instantaneous failure rate of the equipment when operating to time , which is defined as:

[0120] Its integral form is the cumulative hazard function , and satisfies the relationship: .

[0121] In practical engineering, parametric distribution is often used to model life, in the embodiments of the present application, Weibull distribution is used, and its probability density function is: , wherein, is a shape parameter, is a scale parameter, and the corresponding survival function is: When , the hazard function increases with time, which is suitable for describing the life distribution of aging equipment; when , the life obeys the exponential distribution, indicating that the failure rate is constant; when , the failure rate decreases with time, which is suitable for describing equipment with high early failure rate; through maximum likelihood estimation of the historical operation data and failure records of the equipment, the optimal estimation values of and can be obtained.

[0122] Further, based on the survival function, the failure probability of the equipment in a future period of time can be calculated, for example, the failure probability of the equipment in the next time is: When the probability exceeds the threshold , it is determined that the potential failure risk is over-limit, triggering the generation of a preventive maintenance work order; the threshold can be set according to the importance of the equipment, the safety level or the historical operation experience. Exemplarily, for a main transformer of a substation, the threshold can be set to 5%; for a general load-side distribution switch, the threshold can be set to 10%.

[0123] In some other embodiments of the present application, the health index can also be compared with its threshold , when , it is determined that the potential failure risk is over-limit, triggering the generation of a preventive maintenance work order.

[0124] In some embodiments of the present application, in the maintenance work order generation process, not only the work order itself is output, but also the remaining service life prediction result or potential failure risk is fed back to the energy efficiency scheduling module, so that the health state directly affects the power distribution strategy. For example, when the remaining life prediction of a certain section of cable is less than 30 days, the system will automatically reduce the power load it carries and transfer part of the load to the line with better health state through energy efficiency scheduling, thereby reducing the risk of overload operation of the cable; when the risk index of a certain energy storage battery cluster increases, the system will accordingly reduce its charge and discharge depth to avoid accelerated degradation due to overuse.

[0125] Through energy efficiency scheduling feedback and power distribution adjustment in this step, the following technical effects can be achieved: First, using the life prediction and risk assessment results to directly affect the energy efficiency scheduling can enable the system to optimize power distribution according to the health state of the equipment during operation, which not only ensures stable power supply for critical equipment, but also allocates standby resources in advance in case of increased potential risk, thereby improving the overall safety redundancy of the system; Second, combining work order generation with energy efficiency scheduling can avoid the drawbacks of the separation of maintenance and energy management in traditional modes, so that maintenance decisions are no longer isolated, but interact with power distribution, thereby reducing the possibility of accelerated damage of high-risk equipment due to continuous high-load operation and ensuring the coordination of the operation and maintenance process; Third, under this condition, the energy efficiency scheduling no longer relies only on load fluctuations and power demand, but also considers the health state and remaining life of the equipment, thereby enabling more scientific energy optimization strategies, which can reduce energy waste and maintenance costs and meet the dual demands of economic efficiency and high efficiency of the highway strong power system; Fourth, real-time feedback of risk information to the energy efficiency scheduling system can quickly adjust the power supply strategy in case of emergencies, prioritize the redistribution of power between different areas and equipment, and thus maintain power supply continuity and safety when abnormal events occur, meeting the requirements of the operation and maintenance system for high reliability and rapid response.

[0126] In some embodiments of the present application, in the work order priority sorting link, the health index, equipment criticality, and downtime loss are considered comprehensively, and the final priority score is obtained through weighted calculation.

[0127] Specifically, let the health index corresponding to a certain work order be , the equipment criticality be , and the downtime loss be , then the priority scoring function can be represented as: where , , are weight coefficients, satisfying In the formula, the higher the health index, the worse the equipment state, and the higher the corresponding score; the criticality of the equipment is determined by its position and importance in the system, for example, the criticality of the main power supply line is higher than that of the branch line; the shutdown loss is estimated by the direct economic loss and indirect traffic impact caused by the fault shutdown.

[0128] In this step, by introducing a weighted calculation in the maintenance work order priority sorting, the following technical effects can be achieved: First, by including the equipment health index in the priority score, the system can identify equipment with serious deterioration in running state in time, so that maintenance resources can be prioritized to address high-risk objects, ensuring that the overall operation and maintenance decision matches the actual operating condition of the equipment and avoids delays and deviations that may be caused by relying solely on human experience; Second, combined with the criticality of the equipment in the system, the equipment that has the greatest impact on the overall operation safety and continuity of the expressway can be highlighted, thereby aligning resource investment with system importance, so as to ensure that limited maintenance forces prioritize key links and avoid exponentially magnified operational risks and economic losses due to key equipment shutdown; Third, under this condition, the shutdown loss is considered as an evaluation factor, which can convert the potential economic impact into a quantitative index participating in the priority calculation, so as to not only focus on safety risks, but also consider the economic benefits of operation and maintenance, making the decision more comprehensive and scientific, and meeting the lean management needs of the intelligent expressway operation and maintenance system; Fourth, the weighted fusion result of the health index, criticality and shutdown loss is used to sort multiple work orders, which can quickly form a clear execution sequence in the case of many tasks and limited resources, so that the maintenance plan has stronger operability and execution force, ensuring that the entire operation and maintenance process is efficient and orderly, meeting the comprehensive requirements of safety, economy and timeliness of the expressway power system.

[0129] In some embodiments of the present application, the criticality of the equipment can be calculated by the topological centrality index in graph theory, for example, using betweenness centrality or degree centrality to measure the importance of the equipment in the power network, the higher the criticality of the node, the higher the maintenance priority; the shutdown loss can be evaluated by combining historical data and economic models, for example, the loss caused by the shutdown of expressway tunnel lighting mainly reflects the traffic safety risk and social impact, while the loss caused by the shutdown of toll station power supply directly reflects the reduction of economic benefits, the system can automatically call the corresponding loss model according to different scenarios to ensure that the priority calculation meets the actual operating conditions.

[0130] After the work order sorting is completed, the maintenance tasks are pushed in priority order for an operation and maintenance personnel or an automatic scheduling system to execute, and if multiple work orders are triggered at the same time, the work order with the highest score is executed first; in task allocation, the geographic location and available resources of the operation and maintenance personnel can also be combined for secondary optimization to reduce response time and cost.

[0131] Through the above method, step S6 realizes life prediction and risk determination driven by health indicators, generates a maintenance work order when the risk exceeds the limit, and feeds back the life and risk information to energy efficiency scheduling to ensure that the electric energy distribution is coordinated with the equipment state; the priority sorting mechanism ensures that limited resources can be efficiently allocated to the most critical link, so that the maintenance activities not only meet the safety requirements, but also take into account the economy. This process is connected with the foregoing data collection, state modeling, state estimation, anomaly detection and health evaluation steps to form a complete intelligent operation and maintenance chain.

[0132] In summary, the method realizes intelligent operation and maintenance of the highway strong power system through sequentially performing multi-source data collection and preprocessing, state space model construction, state estimation algorithm calculation, fault anomaly detection based on estimated state values and observations, device feature vector extraction and health indicator calculation, and life prediction and maintenance work order generation based on health indicators. Not only can the running state be uniformly modeled and dynamically estimated, but also faults can be identified in time and the health level of the equipment can be quantified under different working conditions, and then maintenance work orders are triggered through residual life prediction and risk determination, and resource optimization configuration is completed in combination with energy efficiency scheduling. The overall process forms a closed-loop system from data perception, modeling and estimation, anomaly identification to maintenance execution, which helps to improve the safety, reliability and scientificity of system operation and management.

[0133] Those skilled in the art can understand that, in the embodiment description of the present application, the symbolic representations of the inverse and transpose of a matrix, such as and actually have the same meaning, and are also two equivalent notations. The above different notations are only differences in mathematical expression forms and do not constitute any limitation on the essence of the technical solutions of the present application. In actual application, different symbolic representations can be selected according to personal habits or specific programming environments, as long as the mathematical operations expressed are consistent, which all belong to the protection scope of the present application.

[0134] It should be noted that, although the embodiments of the present application are described with reference to Figure 1 Steps S1 to S6 are introduced and described in sequence, but this does not mean that steps S1 to S6 must be executed in strict sequence. The embodiments of the present application are described in the order of Figure 1The order in which steps S1 to S6 are described is provided to facilitate understanding of the technical solutions of the embodiments of this application by those skilled in the art. In other words, in the embodiments of this application, the order of steps S1 to S6 can be appropriately adjusted according to actual needs.

[0135] Example 2: like Figure 2 As shown in the figure, this embodiment provides an intelligent operation and maintenance device for a highway power system, comprising: Data acquisition module 1 is used to monitor equipment in the high-voltage power system of highways and acquire multi-source data, including at least one of voltage, current, temperature, vibration and switch action signals; Edge computing module 2 is used to receive multi-source data and perform data preprocessing, including at least one of noise reduction, cleaning and normalization. Analysis and diagnosis module 3 is used to detect faults and anomalies, calculate equipment health indicators and predict remaining service life based on multi-source data that has undergone data preprocessing; Maintenance instruction module 4 is used to generate maintenance work orders based on health indicators or remaining service life, and to sort and assign maintenance work orders according to priority scores.

[0136] By adopting the above technical solution and setting up a data acquisition module 1, an edge computing module 2, an analysis and diagnosis module 3, and a maintenance instruction module 4 in the system, intelligent operation and maintenance of the entire process of the highway power system is realized, which can improve the integrity of information perception, the accuracy of operation diagnosis, and the efficiency of maintenance scheduling.

[0137] Specifically, the data acquisition module 1 monitors and collects multi-source information such as voltage, current, temperature, vibration, and switching action signals, achieving comprehensive coverage of operational data and providing continuous and reliable raw input for system operation status. The edge computing module 2 performs preprocessing such as denoising, cleaning, and normalization on the received data, achieving standardized processing of multi-source heterogeneous data, effectively eliminating noise interference and maintaining data consistency, thus laying the foundation for subsequent modeling and analysis. The analysis and diagnosis module 3 performs fault and anomaly detection on the preprocessed data and generates health indicators and life predictions by combining feature calculations, achieving dynamic evaluation of equipment operation status, timely detection of potential risks, and quantification of equipment reliability, providing a scientific basis for maintenance plans. The maintenance instruction module 4 prioritizes and rationally allocates the generated maintenance work orders, achieving efficient resource utilization and orderly task scheduling, ensuring that critical equipment is given priority, thereby improving the timeliness of emergency response and the overall safety of system operation.

[0138] In some embodiments of the present application, the health index is generated by a linear weighting model or a nonlinear model, and the nonlinear model is a deep neural network model including a neural network weight matrix, a bias, an activation function, and a scaling function.

[0139] The analysis and diagnosis module 3 can achieve the following technical effects by comprehensively processing the feature information by using the weighted model and the deep network: First, the linear weighting form is used to proportionally integrate various operating characteristics according to importance, so that an intuitive health degree evaluation result can be quickly obtained, which not only ensures the simplicity of calculation, but also facilitates the understanding and tracing of the model by the operation and maintenance personnel, thereby providing an efficient and feasible tool for batch analysis of large-scale equipment. Second, when facing a complex operating environment and a large number of data dimensions, the deep network is introduced to perform nonlinear mapping on the feature relationship, so that the potential patterns and degradation rules hidden in the data can be captured, thereby improving the sensitivity of the health index to early abnormalities and enhancing the early warning capability of the device state change. Third, the linear and nonlinear methods complement each other in the same system, which can be flexibly switched and combined in different application scenarios. In simple working conditions, linear integration is relied on to achieve rapid deployment, and in complex working conditions, a deep network is relied on to obtain higher precision, thereby ensuring the adaptability and long-term stability of the health index generation process.

[0140] In some embodiments of the present application, the remaining useful life is predicted by a survival analysis method, which models the device failure time by a survival function, a hazard function, or a cumulative hazard function.

[0141] The analysis and diagnosis module 3 can achieve the following technical effects by modeling and predicting the device life by the survival analysis method: First, the survival function is used to quantify the probability of the device maintaining normal operation in different time intervals, which can provide an intuitive reference based on the time dimension for the operation and maintenance personnel, so that not only the survival probability of the device at a certain time in the future can be evaluated, but also a reasonable inspection and maintenance cycle can be developed to avoid excessive repair or excessive reliance on experience. Second, the hazard function is used to describe the trend of the failure risk over time, which can reveal the law of gradual accumulation of risk in the later stage of the device operation, thereby enabling intervention measures to be taken in advance in the rapid risk rising stage, thereby reducing the probability of sudden failure and providing data basis for the priority allocation of operation and maintenance resources, making the entire operation and maintenance system more proactive and active. Third, the cumulative effect of risk in the time dimension is modeled by using the cumulative hazard function, which can show the risk accumulation of the equipment from the initial operation to the potential failure, and thus can help the operation and maintenance department to grasp the health change process of the whole life cycle of the equipment, ensure timely adjustment of the maintenance plan in the key stage, and improve the safety and reliability of the system.

[0142] In summary, the system forms an intelligent operation and maintenance platform covering the whole life cycle of the strong current system of the expressway by setting the data acquisition module 1, the edge computing module 2, the analysis and diagnosis module 3 and the maintenance instruction module 4, and realizes a complete closed loop from multi-source information acquisition, preprocessing, state analysis to maintenance execution through the linkage of data and functions among the modules. Among them, the data acquisition module 1 ensures the comprehensive coverage of the operation information, the edge computing module 2 realizes the standardized processing of multi-source heterogeneous data, the analysis and diagnosis module 3 completes the fault detection, health assessment and life prediction on this basis, and the maintenance instruction module 4 converts the evaluation results into specific maintenance tasks and performs priority scheduling. The system structure is clear, the functions are complementary, and the unified management and intelligent operation and maintenance of the strong current equipment can be realized under the complex operation environment.

[0143] The above disclosed is only the preferred embodiment of the present application, but the present application is not limited thereto, any non-creative change that any person skilled in the art can think of, and several improvements and refinements made without departing from the principles of the present application, shall fall within the protection scope of the present application.

Claims

1. A highway strong electric system intelligent operation and maintenance method, characterized in that, The method comprises the following steps: Collecting multi-source data and performing data preprocessing, the multi-source data comprising at least one of voltage, current, temperature, vibration and switch action signals, the data preprocessing comprising at least one of denoising, cleaning and normalization; Based on the multi-source data, a state space model is constructed, in which a system state vector is defined to represent the dynamic relationship between the system operating state and external input; A state estimation algorithm is used to estimate the system state vector to obtain an estimated state value; Based on the estimated state value and the multi-source data, fault anomaly detection is performed, which comprises mutation detection based on cumulative sum and change point detection, fault detection based on residual statistical test and anomaly detection based on machine learning; Based on the multi-source data, an equipment feature vector is extracted, and a health index is calculated; According to the health index, the remaining useful life of the equipment is predicted, and a potential failure risk is determined, and when the potential failure risk exceeds a preset risk threshold, a corresponding maintenance work order is generated and the priority of the maintenance work order is determined.

2. The intelligent operation and maintenance method for a highway strong current system according to claim 1, characterized in that, The multi-source data is fused by weighted least squares, and the fused multi-source data is used as the observation input of the state estimation algorithm.

3. The intelligent operation and maintenance method for a highway strong current system according to claim 1, characterized in that, When the equipment fails, based on the signals collected by the sensors, the time difference of arrival method is used for fault location, and the position of the fault section is determined in combination with the topology information of the strong electric system.

4. The intelligent operation and maintenance method for a highway strong current system according to claim 1, characterized in that, The waveforms of the voltage or the current are subjected to discrete Fourier transform, the total harmonic distortion rate is calculated, and the equipment feature vector is extracted based on the multi-source data and the total harmonic distortion rate.

5. The intelligent operation and maintenance method for a highway strong current system according to claim 1, characterized in that, The health index is generated by a linear model or a nonlinear model, the linear model performs weighted summation on the equipment feature vector according to feature weights, and the nonlinear model is a deep neural network, which comprises a neural network weight matrix, a bias, an activation function and a scaling function.

6. The intelligent operation and maintenance method for a highway strong current system according to claim 1, characterized in that, When the maintenance work order is generated, the remaining useful life or the potential failure risk is fed back to energy efficiency scheduling and the electric energy distribution is adjusted.

7. The intelligent operation and maintenance method for a highway strong current system according to claim 1, characterized in that, The priority of the maintenance work order is mainly based on the health index, equipment criticality and shutdown loss for weighted calculation, and the result of the weighted calculation is used for priority sorting of multiple work orders.

8. A highway strong current system intelligent operation and maintenance device, characterized in that, The method comprises: A data acquisition module for monitoring equipment in a highway strong electric system and collecting multi-source data, the multi-source data comprising at least one of voltage, current, temperature, vibration and switch action signals; An edge computing module for receiving the multi-source data and performing data preprocessing, the data preprocessing comprising at least one of denoising, cleaning and normalization; An analysis and diagnosis module for performing fault anomaly detection based on the multi-source data after the data preprocessing, calculating the health index of the equipment and predicting the remaining useful life; A maintenance instruction module for generating a maintenance work order according to the health index or the remaining useful life, and sorting and distributing the maintenance work order according to the priority score.

9. The intelligent operation and maintenance device for highway strong current system according to claim 8, characterized in that, The health indicator is generated by a linear weighting model or a non-linear model, the non-linear model being a deep neural network model comprising neural network weight matrices, biases, activation functions, and scaling functions.

10. The intelligent operation and maintenance device for highway strong current system according to claim 8, characterized in that, The remaining useful life is predicted by a survival analysis method modeling equipment failure times by a survival function, a hazard function, or a cumulative hazard function.

Images