Power equipment intelligent fault early warning method, system and equipment based on health assessment model and medium

Abstract

The invention discloses an intelligent fault early warning method, system, equipment and medium for power equipment based on a health assessment model, and relates to the technical field of equipment fault early warning, and the method comprises the steps: collecting the characteristic parameters of the power equipment in real time through a distributed sensor network, and starting from the time dimension and the space dimension for the characteristic parameters; constructing a space-time correlation feature matrix; constructing a self-adaptive health assessment model, taking a coupling association relationship between the acquired characteristic parameters and the space-time association characteristic matrix as input for training, mapping the characteristic parameters into a health index HI, and outputting the health index HI; and constructing a fault prediction model, taking the HI and the characteristic parameters as input, outputting a prediction result, and generating graded early warning according to the prediction result. According to the method, the space-time correlation feature matrix is constructed, the change of the equipment operation state and the environment working condition can be comprehensively considered from the time dimension and the space dimension, and the comprehensive assessment capability of the equipment health state in the complex environment is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of equipment failure early warning, and in particular to an intelligent fault early warning method, system, equipment and medium for power equipment based on a health assessment model. Background Art

[0002] As modern power systems rapidly develop toward high voltage, large capacity, and intelligent systems, the scale of power equipment is growing exponentially. Their operating environments are generally facing severe challenges, including complex electromagnetic environments, extreme climatic conditions, and the coupling of multiple physical fields. Statistics show that over the past five years, 68% of power grid incidents in my country caused by power equipment failures were due to equipment degradation that was not promptly warned. The traditional "post-fault repair" model is no longer able to meet the stringent power supply reliability requirements of smart grids (above 99.99%). Building an accurate fault warning system has become a key technical bottleneck in ensuring the safe and economical operation of power systems.

[0003] The current mainstream power equipment fault warning technology has three core defects: First, the monitoring method based on threshold judgment (such as the state quantity over-limit alarm in the IEC 61850 standard) can only identify obvious faults, and the recognition of early potential faults is insufficient. The State Grid Corporation of China's 2022 operation report pointed out that this type of method has a missed reporting rate of 42% for progressive faults such as insulation degradation; second, traditional fault diagnosis models mostly rely on single parameter analysis (such as partial discharge, oil chromatography data), and lack of systematic integration of multi-dimensional characteristics of temperature field-electric field-mechanical stress. Laboratory tests of the Southern Power Grid show that the false alarm rate of single parameter models is as high as 35%; third, the existing health assessment system generally adopts a static weight distribution mechanism, which fails to establish an adaptive association between equipment status and operating conditions. The application case of a certain ultra-high voltage converter station shows that the assessment error of the fixed weight model can reach 28% when the load fluctuates by 30%.

[0004] In response to the above-mentioned technical bottlenecks, the present invention innovatively constructs an intelligent early warning system driven by dynamic health assessment. By introducing a multi-source heterogeneous data fusion mechanism, the equipment status parameters, environmental working conditions and historical operation and maintenance records are deeply integrated to establish a quantitative model of equipment health with self-evolution capabilities. A machine learning algorithm is used to dynamically optimize the feature weight distribution strategy to achieve an accurate portrait of the equipment health status. In particular, the system designs a fault evolution prediction module based on a deep belief network (DBN), combines the physical degradation mechanism of the equipment to build a digital twin, and advances the response time window of the traditional threshold alarm by 4-6 operation and maintenance cycles. Summary of the Invention

[0005] In view of the above-mentioned problems, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is that traditional fault diagnosis models mostly rely on single parameter analysis and lack a systematic integration of the multi-dimensional characteristics of temperature field, electric field and mechanical stress.

[0007] To solve the above technical problems, the present invention provides the following technical solutions: an intelligent fault warning method for power equipment based on a health assessment model, which includes: collecting characteristic parameters of power equipment in real time through a distributed sensor network, and preprocessing the characteristic parameters, constructing a spatiotemporal correlation feature matrix for the characteristic parameters from two dimensions: time dimension and space dimension; constructing an adaptive health assessment model, taking the coupled correlation relationship between the collected characteristic parameters and the spatiotemporal correlation feature matrix as input for training, mapping the characteristic parameters into a health index and outputting it; constructing a fault prediction model, taking the health index and characteristic parameters as input, outputting the prediction results, and generating a graded warning based on the prediction results.

[0008] As a preferred solution of the intelligent fault warning method for power equipment based on a health assessment model described in the present invention, the characteristic parameters of the power equipment include electrical parameters, mechanical parameters, chemical parameters and environmental parameters; the preprocessing of the characteristic parameters includes preprocessing the collected signals using a joint algorithm of wavelet packet transform and adaptive Kalman filtering.

[0009] As a preferred solution of the intelligent fault warning method for power equipment based on the health assessment model described in the present invention, wherein: the construction of the spatiotemporal correlation feature matrix includes extracting the trend term and the period term of each feature parameter from the time dimension; the extraction of the trend term includes obtaining the time series of the feature parameter from the time dimension as {x t |t=1,2,...,T z}, x t Represented as the data point of the characteristic parameter at time t, T z Expressed as the total time series, the triple exponential smoothing equation is constructed as:

[0010]

[0011] Among them, L t Expressed as the horizontal component at time t, T t Expressed as the trend component at time t, S t is represented as the seasonal component at time t, α, β, and γ are represented as smoothing coefficients, and m is represented as the length of the seasonal cycle. The extraction of the periodic term includes performing discrete Fourier transform on the time series and expressing it as follows:

[0012]

[0013] Among them, X kIt is represented by the kth frequency component, N is the number of sampling points, and i is the imaginary unit; for X k To extract the main frequency, the formula is:

[0014] and|X k |≥0.7max|X|)

[0015] Among them, f dominant Indicates the main frequency component, |X k | represents the amplitude of the kth frequency component, (|X|) represents the set of all frequency components; the characteristics of the trend term and the period term include the characteristic F corresponding to the trend term trend =[L t ,T t ], the characteristic F corresponding to the periodic term cycle =[f dominant ,A peak ], A peak Indicates the amplitude corresponding to the main frequency.

[0016] As a preferred solution of the intelligent fault warning method for power equipment based on the health assessment model described in the present invention, wherein: the construction of the spatiotemporal correlation feature matrix also includes quantifying the correlation strength between the equipment body parameters and the environmental conditions from the spatial dimension, and constructing the equipment and environment mutual information matrix M g,j Expressed as,

[0017]

[0018] Among them, MIC(.) represents the maximum information coefficient, d g Expressed as device parameters, e j It is represented as the environmental parameter, and I(.) is represented as the mutual information, which is used to measure d g With e j Correlation, G represents the grid division base, G x Represented as parameter d g The number of grid divisions, G y Expressed as parameter e j The number of grid divisions is increased; the trend characteristics, periodic characteristics and the mutual information matrix between equipment and environment are integrated to construct the spatiotemporal correlation matrix.

[0019] As a preferred solution of the intelligent fault warning method for power equipment based on a health assessment model described in the present invention, the construction of the adaptive health assessment model includes: constructing an input layer including 24-dimensional original features and 8-dimensional derived features; the 24-dimensional original features are the time domain and frequency domain features of the preprocessed feature parameters, and the 8-dimensional derived features are the environmental coupling features of the spatiotemporal correlation matrix; constructing the hidden layer includes adopting a bidirectional LSTM network structure and dynamically updating the weights using the particle swarm algorithm update rules; the output layer outputs the health index.

[0020] As a preferred solution of the intelligent fault early warning method for power equipment based on the health assessment model described in the present invention, the fault prediction model includes physical degradation equation correction and fault probability calculation; the output prediction result of the method using the health index and characteristic parameters as input includes embedding the health index value into the physical degradation equation, correcting the deviation of the pure data-driven model, and obtaining the remaining life RUL expressed as:

[0021]

[0022] Among them, E a Expressed as activation energy, used to reflect the material properties of the device, k B It is expressed as Boltzmann constant, T is the operating temperature of the equipment, which is obtained from the collected characteristic parameters, HI c It is expressed as the critical threshold of health index, HI(t) is expressed as the current health index, α and β are expressed as material aging coefficients, which are calibrated by historical data; the probability of failure P is calculated by the time series change rate of the health index f (t) is expressed as,

[0023]

[0024] Where λ is the baseline failure rate, which is related to the device type, and δ is the nonlinear correction factor.

[0025] As a preferred solution of the intelligent fault warning method for power equipment based on the health assessment model described in the present invention, the generation of graded warnings based on the prediction results includes graded warnings based on the health index, remaining life and probability of failure, specifically including the judgment conditions of the primary warning being:

[0026]

[0027] The judgment conditions for the intermediate warning are:

[0028] P f (t)≥0.8OR RUL≤t design ×15%

[0029] The judgment conditions for advanced warning are:

[0030] P f (t)≥0.95AND RUL≤72hours

[0031] in, It is expressed as the rate of change of health index over time, -0.05 / day means a decrease of 0.05 per day, t design Expressed as the design life of the equipment.

[0032] Another object of the present invention is to provide an intelligent fault warning system for power equipment based on a health assessment model, which can automatically provide early warning of equipment failures.

[0033] To solve the above technical problems, the present invention provides the following technical solutions: an intelligent fault warning system for power equipment based on a health assessment model, comprising: a data acquisition module, a spatiotemporal correlation module, a health assessment module and a fault prediction module; the data acquisition module collects characteristic parameters of power equipment in real time through a distributed sensor network and preprocesses the characteristic parameters; the spatiotemporal correlation module constructs a spatiotemporal correlation feature matrix based on the time dimension and the space dimension for the characteristic parameters; the health assessment module constructs an adaptive health assessment model, trains the coupling correlation relationship between the collected characteristic parameters and the spatiotemporal correlation feature matrix as input, maps the characteristic parameters to a health index and outputs it; the fault prediction module constructs a fault prediction model, takes the health index and characteristic parameters as input, outputs the prediction results, and generates a graded warning based on the prediction results.

[0034] A computer device includes a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned method for intelligent fault early warning of power equipment based on a health assessment model are implemented.

[0035] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the above-mentioned method for intelligent fault early warning of power equipment based on a health assessment model.

[0036] The present invention has the following beneficial effects: By constructing a spatiotemporal correlation feature matrix, it comprehensively considers changes in equipment operating status and environmental conditions from both temporal and spatial dimensions, enhancing the ability to comprehensively assess equipment health in complex environments. By quantifying the correlation strength between equipment parameters and environmental conditions, the accuracy and adaptability of early warnings are further improved.

[0037] The present invention designs a fault evolution prediction module based on deep belief network (DBN), which can provide fault warning 4-6 operation and maintenance cycles in advance, significantly shortening the response time, so that preventive measures can be taken before the fault occurs and avoid losses caused by sudden equipment failure. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0039] Figure 1 This is a flow chart of an intelligent fault warning method for power equipment based on a health assessment model in Example 1. DETAILED DESCRIPTION

[0040] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0041] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0042] Example 1, reference Figure 1 , which is the first embodiment of the present invention, provides an intelligent fault warning method for power equipment based on a health assessment model, including: Figure 1 As shown:

[0043] Step 1: Collect the characteristic parameters of the power equipment in real time through the distributed sensor network and preprocess the characteristic parameters.

[0044] The characteristic parameters of power equipment include electrical parameters, mechanical parameters, chemical parameters and environmental parameters, specifically:

[0045] Electrical parameters: partial discharge, dielectric loss angle, leakage current.

[0046] Mechanical parameters: vibration spectrum, stress-strain data.

[0047] Chemical parameters: concentration of dissolved gas (DGA) components in oil and trace water content.

[0048] Environmental parameters: ambient temperature and humidity, equivalent salt density, wind speed and direction.

[0049] The collected signals are preprocessed by using the wavelet packet transform and adaptive Kalman filter combined algorithm to preprocess the characteristic parameters. Specifically:

[0050] The high-frequency pulse interference is decomposed and reconstructed by 6-layer db4 wavelet packet, a dynamic updating mechanism of non-stationary noise covariance matrix is ​​established, and data stream standardization is achieved through sliding time window (window length 30min, overlap rate 50%).

[0051] Step 2: Based on the feature parameters from the time dimension and space dimension, construct a spatiotemporal correlation feature matrix.

[0052] Extract the trend item and cycle item of each characteristic parameter from the time dimension; the extraction of trend item includes obtaining the time series of characteristic parameters from the time dimension as {x t |t=1,2,...,T z}, x t Represented as the data point of the characteristic parameter at time t, T z Expressed as the total time series, the triple exponential smoothing equation is constructed as:

[0053]

[0054] Among them, L t Expressed as the horizontal component at time t, T t Expressed as the trend component at time t, S t It is represented as the seasonal component at time t, α, β, and γ are smoothing coefficients obtained through grid search optimization, for example, α = 0.2, β = 0.1, and γ = 0.3, and m is the length of the seasonal cycle, such as the daily cycle m = 24.

[0055] The extraction of periodic terms involves performing discrete Fourier transform on the time series to express it as,

[0056]

[0057] Among them, X k It represents the kth frequency component, N represents the number of sampling points, and i represents the imaginary unit.

[0058] To X k To extract the main frequency, the formula is:

[0059] and|X k |≥0.7max|X|)

[0060] Among them, f dominant Indicates the main frequency component, |X k | represents the amplitude of the kth frequency component, and (|X|) represents the set of all frequency components.

[0061] The characteristics of trend items and period items include the characteristics F corresponding to the trend items trend =[L t ,T t ], the characteristic F corresponding to the periodic term cycle =[f dominant ,A peak ], A peak Indicates the amplitude corresponding to the main frequency.

[0062] Quantify the correlation strength between the equipment parameters and the environmental conditions from the spatial dimension, and define the equipment parameter set D = {d1, d2, ..., d g}(such as leakage current, vibration amplitude, etc.).

[0063] Define the environmental condition parameter set E={e1,e2,...,e j}(such as ambient temperature, equivalent salt density, etc.).

[0064] For each pair of parameters (d g ;e j ) performs the following operations:

[0065] Divide the data into a G×G grid (G=10 is recommended) and calculate the mutual information I(d g ;e j ), the formula is expressed as,

[0066]

[0067] Among them, d g Expressed as device parameters, e j is represented as the environment parameter, x∈X is represented as the ergodic device parameter d g All possible values ​​of y∈Y are represented by the traversal environment parameter e j All possible values ​​of p(x,y) are expressed as parameter pairs (d g ;e j ), p(x) is expressed as parameter d g The marginal probability of , p(y) is expressed as parameter e j The marginal probability of .

[0068] The standardized processing results in the large information coefficient MIC, which is expressed as follows:

[0069]

[0070] Among them, G represents the grid division base, G x Represented as parameter d g The number of grid divisions, G y Expressed as parameter e j The number of mesh divisions.

[0071] Construct the mutual information matrix between the device and the environment, element M g,j =MIC(d g ,e j ), for example, constructing a 3×2 matrix, expressed as,

[0072]

[0073] Among them, M 1,2 =0.92 indicates a strong correlation between d1 and e2.

[0074] The trend characteristics, period characteristics and the mutual information matrix between equipment and environment are integrated to construct the spatiotemporal correlation matrix Contains information on time and space dimensions. The time dimension information is indexed by the time axis, such as the number of sliding time windows (e.g., window length 30 days, step length 7 days). The space dimension information is the element of the mutual information matrix between the device and the environment. Expressed as,

[0075] [Trend Level L t , trend slope T t , main frequency intensity A peak , spatial correlation M g,j ]

[0076] For each space-time unit Assign weight w t,g,j Expressed as,

[0077]

[0078] Among them, λ1 represents the time series weight coefficient, λ2 represents the space weight coefficient, and the denominator ∑t ′,i′,j′ exp(λ1L t′ +λ2M g′,j′ ) is expressed as performing exponential operations on the linear combinations of all spatiotemporal units (t′, i′, j′) and summing them up to achieve global normalization.

[0079] This embodiment introduces an aging correction factor to quantify the impact of equipment aging on spatiotemporal characteristics to avoid baseline drift. The aging compensation term is introduced into the trend component and expressed as:

[0080]

[0081] Among them, L t is represented as the trend component after compensation, ∈, κ is represented as the aging sensitivity parameter, and is fitted by the continuous probability distribution Weibull distribution, t age Expressed as the cumulative operating time of the equipment, t design Expressed as the design life of the equipment.

[0082] This embodiment takes a specific example as follows: the scenario is vibration monitoring of a 220kV GIS equipment.

[0083] Input data:

[0084] Time series: 30 days of vibration data (sampling interval 10 minutes).

[0085] Environmental parameters: temperature, humidity, and salt density.

[0086] The construction process includes time dimension processing:

[0087] Trend extraction: L t =52.3μm, T t =+0.15μm / day.

[0088] Period analysis: main frequency f = 0.0417 Hz (corresponding to a 24-hour period), main frequency intensity A peak =28.7μm.

[0089] Spatial dimension processing: vibration-temperature mutual information M 1,1 =0.88, vibration-salt density mutual information M 1,3 =0.93.

[0090] Assume that when t=15 days, the space-time matrix slice is expressed as,

[0091]

[0092] Assuming the equipment has been in operation for 10 years, the aging correction is expressed as:

[0093] L′ t =52.3×[1-0.15×(10 / 40) 1.2 ]=52.3×0.941=49.2μm.

[0094] Step 3: Build an adaptive health assessment model, use the coupled correlation relationship between the collected feature parameters and the spatiotemporal correlation feature matrix as input for training, map the feature parameters into a health index HI and output it.

[0095] The input layer includes 24-dimensional original features and 8-dimensional derived features. The 24-dimensional original features are the time domain and frequency domain features of the preprocessed feature parameters, and the 8-dimensional derived features are the environmental coupling features of the spatiotemporal correlation matrix. Suppose the input vector is X = [x1, x2, ..., x 32 [.

[0096] The bidirectional LSTM network structure specifically includes:

[0097] Forward propagation:

[0098]

[0099] Backward Propagation:

[0100]

[0101] Output splicing:

[0102]

[0103] in, They are respectively represented as the forward hidden states at time t and t-1, Xt is represented as the input feature vector at time t, Represented as the reverse hidden state at time t and t-1, θ f ,θ b Represented as the forward and reverse LSTM parameter sets, h t It is represented as the bidirectional hidden state at time t, and || is represented as the vector concatenation operation.

[0104] The particle swarm algorithm is used to dynamically update the weights, and the health index HI is output after the training is completed.

[0105] Step 4: Build a fault prediction model, take the health index HI and characteristic parameters as input, output the prediction results, and generate graded warnings based on the prediction results.

[0106] The value of the health index HI is embedded in the physical degradation equation, and the deviation of the pure data-driven model is corrected to obtain the remaining life RUL expressed as:

[0107]

[0108] Among them, Ea is the activation energy, which is used to reflect the material characteristics of the equipment, kB is the Boltzmann constant, T is the operating temperature of the equipment, which is obtained from the collected characteristic parameters, and HI c It is represented as the critical threshold of health index, HI(t) is represented as the current health index, and α and β are represented as material aging coefficients, which are calibrated by historical data.

[0109] The probability of failure P is calculated by the time series change rate of the health index HI value. f (t) is expressed as,

[0110]

[0111] Where λ is the baseline failure rate, which is related to the device type, and δ is the nonlinear correction factor.

[0112] Based on the health index, remaining life and probability of failure, a graded warning is carried out. Specifically, the judgment conditions for the primary warning are:

[0113]

[0114] The judgment conditions for the intermediate warning are:

[0115] P f (t)≥0.8OR RUL≤t design ×15%

[0116] The judgment conditions for advanced warning are:

[0117] P f (t)≥0.95AND RUL≤72hours

[0118] in, It is expressed as the rate of change of health index over time. -0.05 / day means a decrease of 0.05 per day, that is, the HI value decreases by 0.05 every day.

[0119] It is further explained that when a primary warning is triggered, the operation and maintenance plan needs to be changed, when an intermediate warning is triggered, an on-site investigation is required, and when a high-level warning is triggered, emergency intervention is required.

[0120] Example 2 is the second embodiment of the present invention, which is different from the first embodiment in that: an intelligent fault warning system for power equipment based on a health assessment model, including a data acquisition module, a spatiotemporal correlation module, a health assessment module and a fault prediction module; the data acquisition module collects characteristic parameters of power equipment in real time through a distributed sensor network, and preprocesses the characteristic parameters; the spatiotemporal correlation module constructs a spatiotemporal correlation feature matrix based on the time dimension and space dimension of the characteristic parameters; the health assessment module constructs an adaptive health assessment model, takes the coupled correlation relationship between the collected characteristic parameters and the spatiotemporal correlation feature matrix as input for training, maps the characteristic parameters to a health index and outputs it; the fault prediction module constructs a fault prediction model, takes the health index and characteristic parameters as input, outputs the prediction results, and generates a graded warning based on the prediction results.

[0121] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0122] The logic and / or steps represented in the flowcharts or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing the logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (e.g., a computer-based system, a system including a processor, or other system that can fetch and execute instructions from an instruction execution system, apparatus, or device). For purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0123] More specific examples (a non-exhaustive list) of computer-readable media include the following: an electrical connection with one or more wires (electronic devices), a portable computer disk cartridge (magnetic devices), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disc read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering, or processing in another suitable manner as necessary, and then stored in a computer memory.

[0124] It should be understood that various parts of the present invention can be implemented using hardware, software, firmware, or a combination thereof. In the above-described embodiments, multiple steps or methods can be implemented using software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented using hardware, as in another embodiment, any one of the following technologies known in the art or a combination thereof can be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application-specific integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0125] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An intelligent fault early warning method for power equipment based on a health assessment model, characterized by: include, The characteristic parameters of power equipment are collected in real time through a distributed sensor network, and the characteristic parameters are preprocessed. Based on the characteristic parameters from the time dimension and the space dimension, a spatiotemporal correlation feature matrix is ​​constructed; Construct an adaptive health assessment model, use the coupled correlation relationship between the collected characteristic parameters and the spatiotemporal correlation feature matrix as input for training, map the characteristic parameters into health index and output; Construct a fault prediction model, take the health index and characteristic parameters as input, output the prediction results, and generate graded warnings based on the prediction results.

2. The method for intelligent fault early warning of power equipment based on a health assessment model according to claim 1, characterized in that: The characteristic parameters of the power equipment include electrical parameters, mechanical parameters, chemical parameters and environmental parameters; The preprocessing of the characteristic parameters includes preprocessing the collected signals using a wavelet packet transform and an adaptive Kalman filter combined algorithm.

3. The method for intelligent fault early warning of power equipment based on a health assessment model according to claim 2, characterized in that: The said constructing of the spatiotemporal correlation feature matrix includes extracting the trend item and the period item of each feature parameter from the time dimension; The extraction of the trend item includes obtaining the time series of the characteristic parameters from the time dimension as {x t |t=1,2,...,T z }, x t Represented as the data point of the characteristic parameter at time t, T z Expressed as the total time series, the triple exponential smoothing equation is constructed as: Among them, L t Expressed as the horizontal component at time t, T t Expressed as the trend component at time t, S t It is represented as the seasonal component at time t, α, β, and γ are smoothing coefficients, and m is the length of the seasonal cycle; The extraction of the periodic term includes performing discrete Fourier transform on the time series to express it as: Among them, X k It is represented by the kth frequency component, N is the number of sampling points, and i is the imaginary unit; To X k To extract the main frequency, the formula is: and|X k |≥0.7max(|X|) Among them, f dominant Indicates the main frequency component, |X k | represents the amplitude of the kth frequency component, and (|X|) represents the set of all frequency components; The characteristics of the trend item and the period item include the characteristic F corresponding to the trend item. trend =[L t ,T t ], the characteristic F corresponding to the periodic term cycle =[f dominant ,A peak ], A peak Indicates the amplitude corresponding to the main frequency.

4. The method for intelligent fault early warning of power equipment based on a health assessment model according to claim 3, characterized in that: The construction of the spatiotemporal correlation feature matrix also includes quantifying the correlation strength between the device body parameters and the environmental working conditions from the spatial dimension, and constructing the mutual information matrix M between the device and the environment. g,j Expressed as, Among them, MIC(.) represents the maximum information coefficient, d g Expressed as device parameters, e j It is represented as the environmental parameter, and I(.) is represented as the mutual information, which is used to measure d g With e j Correlation, G represents the grid division base, G x Represented as parameter d g The number of grid divisions, G y Expressed as parameter e j The number of grid divisions; The trend characteristics, periodic characteristics and the mutual information matrix between equipment and environment are integrated to construct the spatiotemporal correlation matrix.

5. The method for intelligent fault early warning of power equipment based on a health assessment model according to claim 4, characterized in that: The constructing of the adaptive health assessment model includes constructing an input layer including 24-dimensional original features and 8-dimensional derived features; The 24-dimensional original features are the time domain and frequency domain features of the pre-processed feature parameters, and the 8-dimensional derived features are the environmental coupling features of the spatiotemporal correlation matrix; Constructing the hidden layer includes adopting a bidirectional LSTM network structure and dynamically updating the weights using the particle swarm algorithm update rule; The output layer outputs the health index.

6. The method for intelligent fault early warning of power equipment based on a health assessment model according to claim 5, characterized in that: The fault prediction model includes physical degradation equation correction and fault probability calculation; The method of taking the health index and characteristic parameters as input and outputting the prediction results includes embedding the value of the health index into the physical degradation equation, correcting the deviation of the pure data-driven model, and obtaining the remaining life RUL expressed as: Among them, E a Expressed as activation energy, used to reflect the material properties of the device, k B It is expressed as Boltzmann constant, T is the operating temperature of the equipment, which is obtained from the collected characteristic parameters, HI c It is expressed as the critical threshold of health index, HI(t) is expressed as the current health index, α and β are expressed as material aging coefficients, which are calibrated by historical data; The probability of failure P is calculated by the time series change rate of the health index. f (t) is expressed as, Where λ is the baseline failure rate, which is related to the device type, and δ is the nonlinear correction factor.

7. The method for intelligent fault early warning of power equipment based on a health assessment model according to claim 6, characterized in that: Generating a graded warning based on the prediction results includes generating a graded warning based on the health index, remaining life and probability of failure, specifically including: The judgment conditions for primary warning are: The judgment conditions for the intermediate warning are: P f (t)≥0.8OR RUL≤t design ×15% The judgment conditions for advanced warning are: P f (t)≥0.95AND RUL≤72hours in, It is expressed as the rate of change of health index over time, -0.05 / day means a decrease of 0.05 per day, t design Expressed as the design life of the equipment.

8. An intelligent fault early warning system for power equipment based on a health assessment model, applying the intelligent fault early warning method for power equipment based on a health assessment model according to any one of claims 1 to 7, characterized in that: It includes data acquisition module, spatiotemporal correlation module, health assessment module and fault prediction module; The data acquisition module collects characteristic parameters of power equipment in real time through a distributed sensor network and pre-processes the characteristic parameters; The spatiotemporal correlation module constructs a spatiotemporal correlation feature matrix based on the time dimension and the space dimension of the feature parameters; The health assessment module constructs an adaptive health assessment model, takes the coupled correlation relationship between the collected characteristic parameters and the spatiotemporal correlation characteristic matrix as input for training, maps the characteristic parameters into a health index and outputs it; The fault prediction module constructs a fault prediction model, takes the health index and characteristic parameters as input, outputs prediction results, and generates graded warnings based on the prediction results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for intelligent fault early warning of electric power equipment based on a health assessment model according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for intelligent fault early warning of electric power equipment based on a health assessment model according to any one of claims 1 to 7 are implemented.

Images