Intelligent equipment health monitoring method and system for power grid production technical transformation project

Abstract

The application discloses a power grid production technical transformation project intelligent equipment health monitoring method and system, relates to the power equipment state monitoring technical field, and comprises the following steps: based on equipment historical data, constructing a personalized health assessment model; analyzing the real-time operation data of the equipment through the personalized health assessment model, calculating and outputting the health state quantitative value; judging the health grade according to the health state quantitative value, and generating the equipment maintenance operation decision; and optimizing and updating the personalized health assessment model based on the execution result of the operation decision. The application effectively solves the problems of high misjudgment rate, inability to accurately reflect individual equipment differences and fine tracking of long-term performance degradation process caused by the static unified threshold of the prior art.

Description

Technical Field

[0001] This invention relates to the field of power equipment condition monitoring technology, and more specifically, to an intelligent equipment health monitoring method and system for power grid production technology upgrading projects. Background Technology

[0002] Power grid production technology upgrade projects typically involve equipment replacement and technological transformation to meet the high-efficiency operation requirements of modern power systems. With the development of intelligent and digital technologies, traditional equipment maintenance models are gradually being replaced by health management models based on data monitoring and intelligent analysis. Equipment failures often have a significant impact on the stability and security of power systems; therefore, timely and accurate assessment of equipment health is crucial for improving the safety and efficiency of power grid operation.

[0003] Existing systems typically define health status (normal / abnormal) in a static, binary manner, with thresholds based on industry standards or broad-area statistical values, ignoring the individual differences of each device, such as manufacturer, commissioning time, operating history, and microenvironment. This often leads to misjudgments when using a uniform standard. Furthermore, devices exist in a long-term "performance degradation" or "sub-healthy" state between "completely normal" and "completely failed," which current technologies struggle to quantify and track with precision. Summary of the Invention

[0004] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide an intelligent equipment health monitoring method and system for power grid production technical transformation projects. By constructing a personalized health assessment model, the system dynamically analyzes real-time operating data and outputs continuously quantified health status values, thereby generating differentiated maintenance decisions. Based on the execution results, the model is optimized and updated in a closed loop. This solves the problems of high misjudgment rate, inability to accurately reflect individual equipment differences, and inability to finely track the long-term performance degradation process caused by the use of static uniform thresholds in the prior art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] The intelligent equipment health monitoring method for power grid production technology upgrading projects includes the following steps: constructing a personalized health assessment model based on historical equipment data; analyzing real-time operating data of the equipment through the personalized health assessment model, calculating and outputting a quantitative value of health status; determining the health level based on the quantitative value of health status, and generating equipment maintenance operation decisions; and optimizing and updating the personalized health assessment model based on the execution results of the operation decisions.

[0007] In a preferred embodiment, the construction of the personalized health assessment model specifically involves: acquiring historical multi-source time-series data of the target device during its healthy operation period, and dividing it into multiple operational micro-state vectors according to a preset time window; constructing a phase space based on the historical multi-source time-series data, mapping the operational micro-state vectors into the phase space, and using a density clustering algorithm to identify one or more health attractors to form a health attraction domain; analyzing the transition probabilities of the micro-state vectors between health attractors and constructing a micro-state transition diagram; constructing a health entropy calculation function based on the health attraction domain and the micro-state transition diagram; and encapsulating the health attraction domain, the micro-state transition diagram, and the health entropy calculation function to form a personalized health assessment model.

[0008] In a preferred embodiment, the construction of the health entropy calculation function specifically involves: for a state point in phase space, calculating its Mahalanobis distance to each healthy attractor in the healthy attraction domain; based on the Mahalanobis distance, calculating the membership degree of the state point to each healthy attractor using Gaussian radial basis functions; based on the state transition probability matrix of the microstate transition graph, calculating the local transition entropy of each state node; and combining the membership degree and the local transition entropy to construct the health entropy calculation function.

[0009] In a preferred embodiment, the calculation and output of the health status quantification value specifically involves: mapping the real-time operating data vector of the target device to the health status ground state in a high-dimensional Hilbert space using a preset nonlinear encoder; constructing a Hamiltonian operator representing the state transition law based on the health attraction domain and historical health data; performing a virtual time-dependent evolution of the health status ground state using the Hamiltonian operator to obtain the health status wavefunction; calculating the probability amplitude of the health status wavefunction in each intrinsic health state defined by the health attractor; sampling the squared modulus of the probability amplitude as the probability that the device is currently in each intrinsic health state, and outputting the center value of the intrinsic health state corresponding to the highest probability as the real-time health status quantification value.

[0010] In a preferred embodiment, the health level determination specifically involves: establishing a three-dimensional mapping matrix of health status quantification value, core component aging coefficient, and real-time load fluctuation coefficient; using an improved radial basis function neural network to obtain a dynamic health level threshold surface from the three-dimensional mapping matrix, and dividing the device state using the surface; constructing three-dimensional coordinate points based on the collected data and projecting them onto the dynamic level threshold surface to determine the initial health region of the device; calculating the disorder of the state trajectory based on the device's historical state data and micro-state transition diagram, and correcting the initial health region of the device; and outputting the health level of the device based on the corrected health region.

[0011] In a preferred embodiment, the equipment maintenance operation decision is specifically as follows: if the health level is in the safe domain, perform a first type of operation, record status data and continue monitoring; if the health level is in the early warning domain, perform a second type of operation, generate early warning information and send it to the remote monitoring center; if the health level is in the alarm domain or emergency domain, trigger an advanced decision optimization process.

[0012] In a preferred embodiment, the advanced decision optimization process specifically includes: generating a topological complex representing the state-operation mapping relationship based on historical data using the Vietoris-Rips complex construction method; performing continuous homology analysis on the topological complex to extract topological features; mapping the current health state and the target health state of the device to two points in the topological complex respectively; calculating all homotopic equivalent path classes connecting the two points, and selecting the optimal homotopic path according to a preset criterion; and inverting the optimal homotopic path into a specific executable operation instruction sequence.

[0013] In a preferred embodiment, the step of selecting the optimal homotopy path according to preset criteria specifically involves: selecting the optimal homotopy path based on geometric length and topological durability criteria; for each homotopy path class, calculating the geometric length of all paths in that class, and selecting the path with the shortest geometric length as the representative of that class; calculating the total durability of all representative paths, and selecting the representative path with the highest total durability as the optimal homotopy path.

[0014] In a preferred embodiment, the optimization update specifically involves: within a preset time period after the execution of the operation decision, collecting a sequence of quantitative health status values ​​of the device, extracting its statistical features and combining them with the operation type code to form an operation feedback vector; calculating the deformation parameters of the affected health attractor based on the operation feedback vector, including the adjustment amount of the center displacement vector and the covariance matrix; smoothing the health attractor based on the deformation parameters, including center displacement adjustment and shape covariance adjustment; reweighting and updating the transition probabilities in the microstate transition diagram based on the performance of the device status in the actual transition path; and updating the personalized health assessment model with the updated health attraction domain and the microstate transition diagram.

[0015] The system for intelligent equipment health monitoring in power grid production technology upgrading projects includes: a model building module, which builds a personalized health assessment model based on historical equipment data; a result output module, which analyzes the real-time operating data of the equipment through the personalized health assessment model, calculates and outputs a quantitative value of the health status; a judgment module, which judges the health level based on the quantitative value of the health status and generates equipment maintenance operation decisions; and an optimization module, which optimizes and updates the personalized health assessment model based on the execution results of the operation decisions.

[0016] The technical effects and advantages of the intelligent equipment health monitoring method and system for power grid production upgrading projects of this invention are as follows:

[0017] 1. This invention solves the problem of misjudgment caused by the use of a uniform static threshold in existing technologies by constructing a personalized health assessment model based on the historical data of individual devices and using dynamic features such as health attraction domain and micro-state transition diagram for quantitative analysis, thereby achieving accurate, continuous and personalized assessment of the health status of each device.

[0018] 2. This invention solves the problem that existing technologies cannot effectively warn of "sub-health" states and provide forward-looking decision-making by integrating health quantification values ​​with multi-dimensional parameters and introducing a decision optimization process based on topological data analysis. At the same time, through a closed-loop optimization mechanism, the model has self-learning capabilities, realizing intelligent operation and maintenance from passive response to proactive warning. Attached Figure Description

[0019] Figure 1 This is a schematic diagram of the intelligent equipment health monitoring method for power grid production technology upgrading projects of the present invention.

[0020] Figure 2 This is a schematic diagram of the system structure of the intelligent equipment health monitoring method for power grid production technology upgrading projects of the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0022] Example 1, Figure 1 This invention provides an intelligent equipment health monitoring method for power grid production technology upgrading projects, comprising the following steps:

[0023] S1, based on historical device data, builds a personalized health assessment model;

[0024] In this embodiment, the construction of a personalized health assessment model based on historical device data specifically includes:

[0025] The target equipment is a 110kV oil-immersed transformer. Assuming a healthy operating period of 6 hours, within this timeframe, data collection, including winding temperature, will be performed using existing instruments. Core vibration Load current and insulating oil dielectric loss Multi-source time-series data is collected once per minute to obtain a four-dimensional data point. Based on the preset time window of one hour, the collected multi-source time-series data is divided according to the preset time window to obtain multiple running micro-state vectors. Each time window's microstate vector will contain 60 data points.

[0026] Select winding temperature Core vibration Load current and insulating oil dielectric loss The phase space is constructed using four parameters. The construction method involves embedding historical data with a time delay to ensure temporal correlation within the phase space. The preset time window is 30 minutes. The data is divided into multiple micro-state vectors, each represented as a point in the phase space. For each time interval... The method for mapping the microstate vector to the phase space is as follows:

[0027]

[0028] in, Indicates the device at time A state point is the overall health status of a device at a certain moment.

[0029] Each running microstate vector is mapped to the phase space through time-delay embedding. All microstate vectors are represented as multiple points in the phase space, forming a distribution.

[0030] In this four-dimensional phase space, density clustering algorithms (such as DBSCAN) are used to cluster the micro-state vectors in the phase space to identify healthy attractors. The purpose of density clustering algorithms is to find dense regions in the phase space, which represent the stable states of the device under healthy operating conditions. Setting the radius threshold and minimum number of points for the clustering algorithm affects the clustering results. The radius threshold is determined by analyzing the k-nearest neighbor distances of data points; the radius threshold determines the distance between adjacent micro-state vectors, and the minimum number of points determines the size of the dense region, generally set to twice the data dimension, or selected through rules of thumb and experimentation.

[0031] Clustering algorithms identify one or more health attractors, which represent typical states of the device in a healthy operating state. Health attractors are dense regions in phase space, and the operating state of the device fluctuates stably within these regions. Connecting multiple health attractor regions forms a health attraction domain, which represents the overall range of the device's health state.

[0032] Based on the historical microstate vectors of the equipment during its healthy operation period, the transition of these microstate vectors between different healthy attractors is analyzed. In the time series, the transition probability of a microstate vector from one healthy attractor to another is calculated. Specifically, for a given microstate vector... Calculate its health attractor To health attractants Transition probability:

[0033]

[0034] in, To attract health To health attractants Number of transfers, For health attractants Total number of transfers.

[0035] The calculated transition probabilities are constructed into a state transition probability matrix, which is then visualized as a directed weighted graph, i.e., a microstate transition graph. Nodes in the graph represent health attractors, the direction of the edges indicates the transition direction, and the weights represent the transition probabilities.

[0036] The health entropy calculation function, based on the health attraction domain and micro-state transition diagram, is constructed as follows:

[0037] Mahalanobis distance is used to measure the similarity between a state and a healthy attractor, considering the correlation between various dimensions. This step is to quantify the position of each state relative to the healthy attractor. For any state in phase space... and health attractants The formula for calculating Mahalanobis distance is:

[0038]

[0039] in, It is a vector of state points (such as winding temperature, core vibration, etc.). For health attractants The mean vector (i.e., the center of the attractor). For health attractants The covariance matrix reflects the distribution and correlation of data points within the attractor region.

[0040] Mahalanobis distance can be used to calculate the distance from each state point to each healthy attractor, thereby measuring how close the point is to different attractors.

[0041] Gaussian radial basis functions are used to calculate the membership degree of a state point within a healthy attractor region. A higher membership degree indicates that the state point is closer to the attractor. For a state point... and health attractants Mahalanobis distance The membership degree is calculated using the Gaussian radial basis function, and the formula is as follows:

[0042]

[0043] in, State point Belongs to health attractor membership degree It is a health attractant The standard deviation of the membership degree determines the range of its expansion; a larger standard deviation indicates a higher membership degree. This will make the change in membership degree more slow.

[0044] Membership degree Reflects the state point In health attractant The degree of membership in a region. The higher the degree of membership, the higher the state point. The more likely it is to belong to the health attractor .

[0045] Extracting the transition probability matrix from the microstate transition diagram ,in Indicates from state Transition to state The probability of a healthy attractor Local transfer entropy Measured from health attractants The uncertainty of the departure transition. The calculation formula is:

[0046]

[0047] The larger the local transfer entropy, the stronger the transfer from the healthy attractor. The more uncertain the starting point, the more complex the dynamic behavior of the system.

[0048] A health entropy calculation function is constructed by combining membership degree and local transition entropy. Health entropy reflects the current health status of the equipment and the uncertainty of the system's dynamic behavior. The calculation formula is:

[0049]

[0050] This function combines membership degree (describing the similarity between a state point and a health attractor) and local transition entropy (describing the dynamic uncertainty of the system), providing a quantitative assessment of the device's health status. A higher entropy value indicates a more uncertain health status, potentially requiring more monitoring and maintenance.

[0051] S2 analyzes the real-time operating data of the equipment through a personalized health assessment model, calculates and outputs a quantitative value of the health status;

[0052] In this embodiment, the step of analyzing the real-time operating data of the device through a personalized health assessment model, calculating and outputting a quantitative value of the health status, specifically involves:

[0053] The real-time operating data vector is input into a pre-trained feature encoder, which converts the specific device operating parameters into an abstract feature vector in a high-dimensional Hilbert space. This feature vector represents the current operating state of the device, referred to as the health state ground state. The feature encoder can be an autoencoder based on a neural network, which is trained on a large amount of historical health data and can extract key features of the device state.

[0054] By analyzing historical health data of equipment, we can construct patterns of equipment state transitions and understand the transition modes between different health states. For example, the transition of equipment from a normal state to a minor fault, and then to a severe fault.

[0055] Based on the state transition rules, construct the Hamiltonian operator for the device health state. It describes how the health status of the equipment changes over time, specifically:

[0056]

[0057] in, and These are health attractors and health attractants quantum state, The device is from a health attractor Transferred to The transition amplitude is obtained by statistically analyzing the frequency of past state transitions. , It is a constant factor that controls the scale of the transfer amplitude.

[0058] The health state ground state of the device driven by the Hamiltonian operator This involves performing a virtual time-dependent evolution. Time-dependent evolution represents the change of a system over time, similar to the wave function evolution in quantum mechanics. The health state wave function of the device... It evolves over time according to the following equation:

[0059]

[0060] in, The imaginary unit, i.e. .

[0061] Calculate the wave function of health status using the matrix exponentiation method or other numerical methods. Evolution:

[0062]

[0063] This step uses numerical solutions to obtain the device's value at a future moment. The health state wave function.

[0064] health state wave function Intrinsic health states can be defined by the health attractor. Expand above:

[0065]

[0066] in, Is the wave function in an intrinsic healthy state? The probability amplitude.

[0067] The probability amplitude of each intrinsic health state Indicates the current status of the device in the health attractor. The probability of [something]. The probability amplitude is obtained by calculating the inner product:

[0068]

[0069] probability amplitude Modulus square This represents the probability that the device is currently in that intrinsic health state. For each intrinsic health state... Calculate its corresponding probability amplitude The square of the modulus gives the probability that the device is currently in each intrinsic health state:

[0070]

[0071] in, This indicates that the device is in an intrinsically healthy state. The probability of.

[0072] From the probabilities of all intrinsic health states, select the central value corresponding to the intrinsic health state with the highest probability. As a quantitative value of the device's real-time health status, i.e. This health value reflects the health status of the device at the current moment.

[0073] S3, determine the health level based on the quantified health status value, and generate equipment maintenance operation decisions;

[0074] In this embodiment, the step of determining the health level based on the quantified health status value and generating equipment maintenance operation decisions specifically involves:

[0075] Health status quantification value The aging coefficient of the core components has been calculated through the previous steps. This is an indicator reflecting the inherent aging degree of the equipment, which can be estimated through measured data of the degree of polymerization (DP value) of the insulating paper or based on a model based on the number of years of operation. Real-time load fluctuation coefficient. It is an indicator reflecting the severity of the current operating conditions, obtained by calculating the ratio of the standard deviation of the load current to its mean within the current time window. The larger the value, the more severe the load fluctuation and the greater the impact stress on the equipment. The aging coefficient of core components and the real-time load fluctuation coefficient can be obtained using existing technologies, and will not be described in detail here.

[0076] The above three variables , , As input, construct a three-dimensional mapping matrix:

[0077]

[0078] This matrix represents the relationships between different variables during the health assessment of a device.

[0079] Using a large amount of historical data (including normal operation data and data prior to known failures) and their corresponding expert-annotated health levels (safe, warning, alarm, emergency) as training samples, a modified radial basis function neural network (RBFNN) is trained under supervision. After training, the neural network internally defines a dynamic health level threshold surface. This surface is a complex boundary in three-dimensional space that divides different health level regions. For any input three-dimensional coordinate point, the network can output its corresponding health region. The network training process optimizes weights and parameters using historical health data from multiple devices (including health status quantification values, aging coefficients, and load fluctuation coefficients) to ensure that the network can accurately predict the health level of devices.

[0080] If all three parameters are in the healthy range, it is in the safe range. If the health status quantification value decreases, but the aging coefficient and load fluctuation coefficient are relatively stable, it is in the warning range. If the health status quantification value decreases significantly, and the aging coefficient and load fluctuation coefficient fluctuate greatly, it is in the alarm range. If the health status deteriorates severely, and the aging coefficient and load fluctuation coefficient are both in the high-risk range, the equipment may be about to fail, which is the emergency range.

[0081] Collect the three current variables of the target device , , A three-dimensional coordinate point is generated and projected onto the previously generated dynamic health level threshold surface to determine the current health level region of the device. Based on the projection result, the initial health region of the device is determined. For example, if the three-dimensional coordinate point is projected onto the warning domain, the device is currently in a warning state.

[0082] The initial judgment is based on the current instantaneous state. To further improve reliability, a time-series-based correction is introduced. Historical health status data (e.g., health status quantification, aging coefficient, load fluctuation coefficient, etc.) of the device over the past hour is collected to form a historical state sequence. Based on the historical micro-state data of the device over the past hour, a corresponding micro-state transition graph is constructed. The nodes of the transition graph represent different health states, and the edge weights represent the probability of transitioning from one health state to another. Disorder refers to the uncertainty or complexity of the device's state trajectory, representing the degree of "disorder" in the device's state transitions. It reflects the fluctuation range of the device's health status; the greater the disorder, the greater the fluctuation of the device's state. The disorder of the state trajectory is calculated based on the device's historical state trajectory and the micro-state transition graph. Specifically, this can be achieved through the following methods:

[0083]

[0084] If the degree of disorder A high level indicates that the device's status is unstable and the initial health zone may need to be adjusted. For example, if the device is in the safe zone but has a high level of disorder, it may need to be adjusted to the warning zone or alarm zone to detect potential problems earlier. If the device is already in the alarm zone or emergency zone and has a high level of disorder, special attention needs to be paid to the device's changing trend. The device's health level is output based on the corrected health zone (safe zone, warning zone, alarm zone, emergency zone).

[0085] If the health level is in the safe domain, perform the first type of operation, record the status data and continue monitoring;

[0086] If the health level is in the warning zone, perform the second type of operation, generate warning information, and send it to the remote monitoring center;

[0087] If the health level is in the alarm or emergency domain, the advanced decision optimization process will be triggered.

[0088] The advanced decision optimization process is as follows:

[0089] Obtain device operation instructions and status response data Equipment status response data includes multi-dimensional vectors such as equipment health status quantification values, aging coefficients, and load fluctuation coefficients, while operation commands include events such as equipment start-up, stop-up, and load adjustment.

[0090] Each "state-operation-new state" record is mapped to a high-dimensional data point. All these data points constitute a point cloud in a high-dimensional space. Dot cloud It is a high-dimensional space containing the relationships between states, operations, and new states, describing the state transformations of the device under different operations. A topological complex representing the state-operation mapping is generated using the Vietoris-Rips complex construction method. Specifically, for point clouds... Each point in the data is represented by a scale parameter. Draw a high-dimensional sphere with a radius. If there is an intersection between the high-dimensional spheres of two points, connect these points to form a simplex (such as a line segment, triangle, etc.), and fill the space with these simplexes. The set of all simplexes constitutes a topological complex. The Vietoris-Rips complex provides a topological framework for the relationship between state and operation, characterizing the changes in device state under different operating modes.

[0091] Topological complexes were constructed Then, by gradually increasing the scale parameter Continuous homology analysis was performed on the topological complex to calculate homology groups at different scales, especially 0-dimensional and 1-dimensional homology groups. The 0-dimensional homology group represents the connectivity between device states, while the 1-dimensional homology group represents the ring structure in the device states. The generators of the homology group at different scales... The "birth" and "death" moments can be used to create barcodes:

[0092]

[0093] Barcode length This reflects the stability of the topological feature; longer barcodes indicate stable topological features, while shorter barcodes are usually noise.

[0094] The current health status of the device and expected target health status Add point cloud Update to new point cloud :

[0095]

[0096] This will further determine the current health status. and expected target health status Mapped to topological complexes respectively The current state point and the target state point, in the topological complex In this process, we find all paths from the current state to the target state. These paths are divided into several homotopy equivalence classes based on their topology, meaning paths that can be transformed into each other through continuous deformation. For each homotopy class, we calculate the geometric length of the path. That is, the sum of the weights of each edge in the path:

[0097]

[0098] in, The weight of each segment on the path represents the "cost" or "risk" of transitioning from one state to another. This represents the number of edges in the path.

[0099] Based on the geometric length of the path, the shortest path is selected as the representative path. Simultaneously, considering topological durability, the durability of the simplexes traversed by the path is calculated, and the path with the highest total durability is selected.

[0100]

[0101] in This represents all the simplexes traversed along the path.

[0102] Extracting the geometric center point from the optimal path This represents the central state of the path. These geometric center points reflect the critical states in the transition process from the current state to the target state of the device, based on the geometric center points. This is matched against the "status-action" records in the historical data to deduce the specific operation instructions. Each center point corresponds to an operation sequence. For example, if the path starts from... → → → The system will search the historical records and find the most similar state. Execute operation The state most often reached later This process continues to generate a sequence of operation instructions. Based on the inverted operation sequence, a topology-optimal set of operation instructions is generated. Each operation instruction is followed by a wait time and a status check, for example: execute operation. Wait 10 minutes to execute the operation. Wait 15 minutes and check if the equipment is close to the target state. .

[0103] S4. The personalized health assessment model is optimized and updated based on the results of the operational decisions.

[0104] In this embodiment, the optimization and updating of the personalized health assessment model based on the execution results of operational decisions specifically involves:

[0105] Within a preset time period after the system generates and executes an operational decision (whether a simple warning or an advanced topology optimization command), the system initiates a model optimization and update process to ensure that the evaluation model can adapt to changes in equipment status. The system collects a real-time sequence of quantitative health status values ​​of the target equipment through sensors and monitoring devices. Extract key statistical features (such as mean, standard deviation, maximum, minimum, etc.), and assign a specific operation type code to each executed operation decision. The statistical characteristics of the quantified health status value are combined with the operation type code to form an operation feedback vector. To analyze the impact of the operation on the equipment's health attractor, the deformation parameters of the health attractor need to be calculated. The deformation parameters include the center displacement vector and the covariance matrix adjustment. By calculating the feedback vector and combining it with the changes in the quantified health status value of the equipment after different operations, the center displacement vector and the covariance matrix adjustment are obtained. The health attractor is then smoothed according to the deformation parameters, including center displacement adjustment and shape covariance adjustment. The center point position of the attractor is adjusted according to the center displacement vector, and the shape of the health attractor is adjusted according to the covariance matrix adjustment. After smoothing adjustment, the health attractor represents the "new health attraction domain" of the equipment's health status after the operation, that is, the stable region of the equipment in the new health state.

[0106] Based on the operational feedback vector and the device's performance in the actual transition path, the transition probabilities in the microstate transition graph are reweighted. Assuming the device's health status transition path differs from expectations after the operation, the transition probabilities in the microstate transition graph are reweighted and updated, adjusting the edge weights (i.e., transition probabilities) to more accurately reflect changes in the device's actual health status. The updated health attraction domain and microstate transition graph are then used to update the personalized health assessment model, providing more precise maintenance decision support.

[0107] This optimization and update process is based on existing technology and will not be described in detail.

[0108] Example 2, Figure 2 The present invention provides a system for intelligent equipment health monitoring in power grid production technology upgrading projects, comprising:

[0109] The model building module constructs personalized health assessment models based on historical device data.

[0110] The results output module analyzes the real-time operating data of the equipment through a personalized health assessment model, calculates and outputs a quantitative value of the health status;

[0111] The judgment module determines the health level based on the quantitative value of the health status and generates equipment maintenance operation decisions.

[0112] The optimization module optimizes and updates the personalized health assessment model based on the execution results of operational decisions.

[0113] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0114] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0115] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0116] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0117] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0118] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent equipment health monitoring of power grid production technical transformation projects, characterized in that, The method comprises the following steps: Based on the device historical data, a personalized health assessment model is constructed, specifically: obtaining historical multi-source time series data of the device in the health running period, and dividing it into multiple running micro-state vectors according to a preset time window; based on the historical multi-source time series data, a phase space is constructed, the running micro-state vector is mapped into the phase space, and one or more health attractors are identified by using a density clustering algorithm to form a health attractor domain; Analyze the transition probability of the micro-state vector between the health attractors, and construct a micro-state transition graph; Based on the health attractor domain and the micro-state transition graph, a health entropy calculation function is constructed; based on the health attractor domain, the micro-state transition graph and the health entropy calculation function, a personalized health assessment model is formed; The health entropy calculation function is constructed, specifically: for the state point in the phase space, the Mahalanobis distance from each health attractor is calculated; based on the Mahalanobis distance, the membership degree of the state point to each health attractor is calculated by using a Gaussian radial basis function; Based on the state transition probability matrix of the micro-state transition graph, the local transition entropy of each state node is calculated; the membership degree and the local transition entropy are integrated to construct the health entropy calculation function; The real-time running data of the device is analyzed by the personalized health assessment model, and a health state quantitative value is calculated and output; According to the health state quantitative value, the health level is judged, and the device maintenance operation decision is generated; Based on the execution result of the operation decision, the personalized health assessment model is optimized and updated. 2.The smart device health monitoring method for power grid production technical transformation project of claim 1, characterized in that, The health state quantitative value is calculated and output, specifically: The real-time running data vector of the device is mapped into the health state ground state in the high-dimensional Hilbert space by using a preset nonlinear encoder; Based on the health attractor domain and the historical health data, a Hamiltonian operator representing the state transition rule is constructed; The health state ground state is virtually evolved by using the Hamiltonian operator to obtain a health state wave function; The probability amplitude of the health state wave function on each eigen-health state defined by the health attractor is calculated; The modulus square of the probability amplitude is sampled as the probability that the device is currently in each eigen-health state, and the center value of the eigen-health state corresponding to the highest probability is output as the real-time health state quantitative value. 3.The method of claim 2, wherein, The health level judgment is specifically: A three-dimensional mapping matrix of the health state quantitative value, the core component aging coefficient and the real-time load fluctuation coefficient is established; The three-dimensional mapping matrix is obtained through an improved radial basis function neural network to obtain a dynamic health level threshold surface, and the device state is divided by the surface; Based on the collected data, three-dimensional coordinate points are constructed and projected onto the dynamic level threshold surface to determine the initial health region of the device; Based on the device historical state data and the micro-state transition graph, the chaos degree of the state trajectory is calculated to correct the initial health region of the device; According to the corrected health region, the health level of the device is output.

4. The intelligent equipment health monitoring method for power grid production technical transformation projects according to claim 3, characterized in that, The device maintenance operation decision is specifically: If the health level is in the safe domain, execute the first type of operation, record the state data and continue to monitor; If the health level is in the warning domain, execute the second type of operation, generate warning information and send it to the remote monitoring center; If the health level is in the alarm domain or the emergency domain, trigger the advanced decision optimization process.

5. The intelligent equipment health monitoring method for power grid production technical transformation projects according to claim 4, characterized in that, The advanced decision optimization process is specifically: The Vietoris-Rips complex construction method is used to generate a topological complex representing the state-operation mapping relationship based on historical data; Persistent homology analysis is performed on the topological complex to extract topological features; The current health state and the target health state of the device are mapped to two points in the topological complex, respectively; All homotopy equivalent path classes connecting the two points are calculated, and the optimal homotopy path is selected according to a preset criterion; The optimal homotopy path is inverted to a specific executable operation instruction sequence.

6. The intelligent equipment health monitoring method for power grid production technical transformation projects according to claim 5, characterized in that, The optimal homotopy path is selected according to the preset criterion, specifically: The optimal homotopy path is selected according to the geometric length and topological persistence criterion; For each homotopy path class, the geometric length of all paths in the class is calculated, and the path with the shortest geometric length is selected as the representative of the class; The total persistence of all representative paths is calculated, and the representative path with the highest total persistence is selected as the optimal homotopy path.

7. The intelligent equipment health monitoring method for power grid production technical transformation projects according to claim 6, characterized in that, The optimization update, specifically: Within a preset time period after the operation decision is executed, the health state quantitative value sequence of the device is collected, the statistical features are extracted and combined with the operation type code to form an operation feedback vector; The deformation parameters of the affected health attractor are calculated based on the operation feedback vector, including the center displacement vector and the covariance matrix adjustment amount; The health attractor is smoothly adjusted based on the deformation parameters, including center displacement adjustment and shape covariance adjustment; Based on the performance of the device state in the actual transition path, the transition probability in the micro-state transition graph is re-weighted and updated; The personalized health assessment model is updated based on the updated health attractor domain and micro-state transition graph.

8. The system for intelligent equipment health monitoring method of power grid production technological transformation project according to claims 1-7, characterized in that, It includes: A model construction module constructs a personalized health assessment model based on device historical data; A result output module analyzes real-time running data of the device through the personalized health assessment model, calculates and outputs health state quantitative values; A judgment module judges the health level according to the health state quantitative values and generates device maintenance operation decisions; An optimization module optimizes and updates the personalized health assessment model based on the execution results of the operation decisions.

Images