Composite apparatus health analysis method and system based on multi-source data fusion

Abstract

The invention discloses a combined electric appliance health analysis method and system based on multi-source data fusion, and the method comprises the steps: cleaning text data, carrying out the semantic mapping of the text data to a preset state quantity index, extracting a basic feature value from image data, and generating a standardized evaluation index set; calculating a weight vector of each evaluation index by constructing a judgment matrix, and performing consistency verification; weibull distribution fitting is carried out based on historical state quantity data, and grading threshold values of the positive degradation index and the negative degradation index are calculated; dividing a state interval based on the normal cloud model and calculating a membership degree vector of each state level; the membership degree vector is converted into a basic probability distribution function, evidence fusion is carried out in combination with the weight vector, and a comprehensive state level is output; and identifying the high-conflict evidence according to the Pignistic probability distance, performing refusion after correction, and outputting a final state level. Deep fusion and intelligent evaluation of multi-source heterogeneous data are realized, and the accuracy and robustness of health state judgment of the combined electric appliance are improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of electrical appliance health analysis, and particularly relates to a combined electrical appliance health analysis method and system based on multi-source data fusion. BACKGROUND

[0002] As the core equipment of the power system, the health state of the combined electrical appliance is directly related to the safe operation of the power grid. The evaluation method commonly used in the industry has significant limitations: first, the monitoring data sources are highly heterogeneous, including sensor real-time data (temperature, vibration, SF6 gas pressure), device operation logs, manually entered text records (operation and maintenance reports, fault descriptions), and infrared thermal imaging image data. These multi-source heterogeneous information lacks a unified processing standard, making it difficult to effectively integrate key state characteristics; second, the evaluation weight distribution relies on expert experience, which is highly subjective and lacks consistency verification, affecting the objectivity of the importance determination of the indicators; third, the state threshold setting uses a static standard, which cannot adapt to the dynamic change law of the actual degradation process of the equipment; fourth, conflicts are easily generated during multi-source data fusion, and traditional methods lack conflict resolution mechanisms, which seriously reduces the reliability of the evaluation results. The above problems restrict the accuracy and robustness of the health state determination of the combined electrical appliance, and there is an urgent need for an analysis method that can deeply fuse multi-source data, quantize weight distribution, dynamically model thresholds, and intelligently resolve conflicts. SUMMARY

[0003] The application provides a combined electrical appliance health analysis method and system based on multi-source data fusion, which solves the technical problem of restricting the accuracy and robustness of the health state determination of the combined electrical appliance.

[0004] In a first aspect, the application provides a combined electrical appliance health analysis method based on multi-source data fusion, comprising:

[0005] Obtaining text data and image data of the combined electrical appliance, cleaning and semantically mapping the text data to a preset state quantity indicator, extracting basic feature values from the image data, and generating a standardized evaluation indicator set;

[0006] Calculating the weight vector of each evaluation indicator by constructing a judgment matrix and performing consistency verification;

[0007] Based on the historical state quantity data, Weibull distribution fitting is performed to calculate the grading thresholds of the positive and negative degradation indicators;

[0008] Dividing the state interval based on the normal cloud model and calculating the membership vector of each state grade;

[0009] Converting the membership vector into a basic probability assignment function, combining the weight vector to perform evidence fusion, and outputting the comprehensive state grade;

[0010] High conflict evidence is identified according to Pignistic probability distance, and the final state level is output after re-fusion and correction.

[0011] In a second aspect, the present application provides a combined electrical appliance health analysis system based on multi-source data fusion, comprising:

[0012] A generation module is configured to acquire text data and image data of the combined electrical appliance, clean and semantically map the text data to preset state quantity indicators, and extract basic feature values from the image data to generate a standardized evaluation indicator set.

[0013] A first calculation module is configured to calculate a weight vector of each evaluation indicator by constructing a judgment matrix and perform consistency check.

[0014] A second calculation module is configured to perform Weibull distribution fitting based on historical state quantity data, and calculate grading thresholds of positive and negative degradation indicators.

[0015] A third calculation module is configured to divide state intervals based on a normal cloud model and calculate membership vectors of each state level.

[0016] An output module is configured to convert the membership vectors into basic probability assignment functions, combine the weight vectors to perform evidence fusion, and output a comprehensive state level.

[0017] A correction module is configured to identify high conflict evidence according to Pignistic probability distance, re-fuse and output the final state level after correction.

[0018] In a third aspect, an electronic device is provided, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the combined electrical appliance health analysis method based on multi-source data fusion of any embodiment of the present application.

[0019] In a fourth aspect, the present application further provides a computer readable storage medium having a computer program stored thereon, wherein the program instructions are executed by a processor to enable the processor to perform the steps of the combined electrical appliance health analysis method based on multi-source data fusion of any embodiment of the present application.

[0020] The multi-source data fusion-based combined electrical appliance health analysis method and system of the application collects sensor monitoring data, device operation logs and environmental data in real time through an Internet of Things protocol, and cleans, extracts features and performs semantic mapping on non-standardized data such as manually entered operation and maintenance records, fault report texts and infrared thermal imaging diagrams, to generate a structured state quantity index set; a judgment matrix is constructed based on the T.L.Satty 1-9 consistency scale method, the characteristic value and characteristic vector are calculated, and consistency checking is performed; the classification threshold values of positive and negative degradation indicators are calculated by Weibull distribution fitting of historical data; the state interval is divided based on the normal cloud model, and the membership vector of the measured data is calculated; the membership is converted into a basic probability assignment function, the weight-corrected evidence is combined, and the comprehensive state grade is output through evidence fusion; for high-conflict evidence, the evidence weight is corrected based on the Pignistic probability distance and re-fused to improve the evaluation reliability, realizing deep fusion and intelligent evaluation of multi-source heterogeneous data, and improving the accuracy and robustness of the combined electrical appliance health state determination. BRIEF DESCRIPTION OF DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiment description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0022] Figure 1 A flowchart of a multi-source data fusion-based combined electrical appliance health analysis method provided by an embodiment of the present application;

[0023] Figure 2 A structural block diagram of a multi-source data fusion-based combined electrical appliance health analysis system provided by an embodiment of the present application;

[0024] Figure 3 A structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0025] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0026] Please refer to Figure 1 which shows a flowchart of a multi-source data fusion-based combined electrical appliance health analysis method of the present application.

[0027] As Figure 1 shown, the combined electrical appliance health analysis method based on multi-source data fusion specifically includes the following steps:

[0028] Step S101, obtaining text data and image data of the combined electrical appliance, cleaning and semantically mapping the text data to a preset state quantity index, extracting basic feature values from the image data, and generating a standardized evaluation index set.

[0029] In this step, real-time collection of standardized data through Internet of Things protocol, batch import of non-standardized data;

[0030] Cleaning, key information extraction and semantic mapping of non-standardized text data;

[0031] Extraction of basic feature values from non-standardized image data;

[0032] Integrating the processed data into a structured set, each index containing a unique identifier, physical quantity name and value / state.

[0033] Step S102, calculating the weight vector of each evaluation index by constructing a judgment matrix, and performing consistency check.

[0034] In this step, the T.L.Satty 1-9 consistency scale method is used to construct the judgment matrix, and the expression is:

[0035] A=[a ij ] n×n i,j=1,2,…,n,

[0036] In the formula, A is the judgment matrix, a ij is the importance ratio of index I i relative to index I j ;

[0037] Calculate the maximum eigenvalue and eigenvector in the judgment matrix, determine the eigenvector ω=[ω1,ω2,...,ω max ] max corresponding to the maximum eigenvalue λ n based on Aω=λ T , and the eigenvector is the weight vector.

[0038] Step S103, Weibull distribution fitting based on historical state quantity data, calculating the classification threshold of positive and negative degradation indexes.

[0039] In this step, historical data is retrieved from the time series database, and historical monitoring data is fitted with Weibull distribution, and the probability density function is:

[0040]

[0041] The corresponding inverse cumulative distribution function is:

[0042] y=F -1 (p|η,β)=η-ln(1-p) 1 / β , p∈0,1,

[0043] In the formula, p is a cumulative probability, x is a value corresponding to the cumulative probability p, β is a shape parameter of the Weibull distribution, η is a scale parameter of the Weibull distribution, F -1 is the inverse function of the cumulative probability distribution function.

[0044] According to the device historical state distribution proportion, the grading threshold is calculated, wherein, for the positive degradation index, when p in the inverse cumulative distribution function is a%, a%+b%, a%+b%+c% respectively, the calculation results are the attention threshold y1, the abnormal threshold y2, and the serious threshold y3 respectively;

[0045] For the negative degradation index, when p in the inverse cumulative distribution function is 1-a%, 1-(a%+b%), 1-(a%+b%+c%) respectively, the calculation results are the attention threshold y1, the abnormal threshold y2, and the serious threshold y3 respectively, wherein, a%, b%, c% respectively represent the proportions of normal, attention, and abnormal state devices in the historical data.

[0046] Step S104, state intervals are divided based on the normal cloud model, and membership vectors of each state level are calculated.

[0047] In this step, the value range is divided into four state intervals according to the threshold, wherein, the state intervals include a normal interval D1=[0, y1], an attention interval D2=[y1, y2], an abnormal interval D3=[y2, y3], and a serious interval D4=[y3, +∞];

[0048] The membership of each state level k is calculated by the forward cloud generator, and the membership vector μ(x i )=[μ1(x i ), μ2(x i ), μ3(x i ), μ4(x i )] is outputted, wherein, the expression for calculating the membership of each state level k is:

[0049]

[0050] En′ k，j ~ N[En k , (He k ) 2 ],,

[0051]

[0052] In the formula, M is the number of intersecting cloud droplets, and Ex k For expectations, En k Entropy.

[0053] Step S105: Transform the membership vector into a basic probability assignment function, combine it with the weight vector to perform evidence fusion, and output the comprehensive state level.

[0054] In this step, the basic probability assignment function is constructed, with the following expression:

[0055] m i (D k )=μ i k = 1, 2, 3, 4

[0056]

[0057] In the formula, m i Indicator I k The BPA function, D k For the state level, μ i Θ represents the membership degree, Θ represents the identification frame, which is the complete set containing all possible states, and m represents the membership degree. i (Θ) represents the uncertainty, which is the probability that a specific state cannot be assigned to it;

[0058] The eigenvectors of the input judgment matrix are used as the weight vectors, and the expression is:

[0059]

[0060] α i =1-ω i ′·[1-mi(Θ)],

[0061] m i ′(D k )=α i ·m i (D k ),

[0062]

[0063] In the formula, ω i Let ω be the weight vector. i ′ represents the normalized weight, α i This is a discount factor used to weaken the credibility of low-weighted evidence; m i ' represents the modified BPA function;

[0064] Input the corrected BPA set {m1',m2',......,mn'} and perform evidence synthesis, the expression is:

[0065]

[0066] where K is the total conflict coefficient, A, B and C are non-empty subsets of the recognition frame Θ, δ(A) denotes the conflict distribution weight of subset A, m (i) (A) is the probability distribution of subset A after fusing i evidences;

[0067] Output the comprehensive state level where m final (D k ) is the final probability distribution value of state D k

[0068] Step S106, identify the high conflict evidence according to the Pignistic probability distance, revise and re-fuse and output the final state level.

[0069] In this step, the conflict degree of evidence is calculated, specifically:

[0070] Input the original BPA data, {m1, m2,..., mn}, calculate the conflict degree for each pair of evidence, the steps of conflict degree calculation are: calculate the conflict coefficient where B and C are non-empty subsets of the recognition frame Θ, m i (B), m j (C) denotes the BPA function value of the i-th evidence m i to B, the BPA function value of the j-th evidence m j to C, calculate the Pignistic probability function where A is a subset of the recognition frame, calculate the Pignistic probability distance Finally, calculate the comprehensive conflict degree The greater the comprehensive conflict degree value is, the stronger the conflict is;

[0071] Calculate the evidence weight, specifically: calculate the similarity sim(m i , m j ) = 1-conf(m i , m j ), support degree certainty degree D k is the state interval, decision degree credibility CRD(i) = SD(i) + CD(i)·DD(i), finally, calculate the normalized weight

[0072] Determine the high conflict evidence, specifically:

[0073] The determination condition is if​ then m i is high conflict evidence, wherein, is the arithmetic mean of all evidence weights, and n is the total number of evidence;

[0074] The high conflict evidence BPA is corrected, and the correction formula is:

[0075]

[0076] wherein, m i is the corrected BPA function, D k is the state level;

[0077] The evidence is re-fused and the final state level is output, specifically:

[0078] Using the corrected BPA function, the synthesis rule of evidence fusion is re-fused, and the final state level is output

[0079] In summary, the method of the application, through the Internet of Things protocol, real-time collection of sensor monitoring data, equipment operation log and environmental data, and cleaning, feature extraction and semantic mapping of non-standardized data such as artificial input operation and maintenance records, fault report texts and infrared thermal imaging images, generate a structured state quantity index set; based on the T.L.Satty 1-9 consistency scale method, a judgment matrix is constructed, the characteristic value and characteristic vector are calculated, and consistency check is performed; the Weibull distribution fitting of historical data is used to calculate the grading threshold of positive and negative degradation indicators; based on the normal cloud model, the state interval is divided and the membership vector of the measured data is calculated; the membership is converted into the basic probability assignment function, combined with the weight correction of the evidence, and the comprehensive state level is output through the evidence fusion; for high conflict evidence, the evidence weight is corrected based on the Pignistic probability distance and re-fused to improve the evaluation reliability, realizing the deep fusion and intelligent evaluation of multi-source heterogeneous data, and improving the accuracy and robustness of the health state determination of the combined electrical apparatus.

[0080] In one specific embodiment, step 1: multi-source data acquisition and standardized processing

[0081] Multi-source heterogeneous data acquisition, storage and non-standardized information conversion are realized, providing standardized information input for subsequent health evaluation. The specific process is:

[0082] 1. Data acquisition and storage

[0083] The combination electrical equipment health assessment has two types of information sources, which are standardized and non-standardized information. For standardized data collection, real-time access to sensor monitoring data (temperature, vibration, SF6 gas pressure), device operation logs (switch action times, fault records) and environmental data (temperature and humidity) is achieved through Internet of Things protocols (MQTT / CoAP). For non-standardized data collection, batch import of manually entered operation and maintenance records, fault report texts, and infrared thermal imaging images, vibration waveform images and other image data are performed.

[0084] After collecting data, for standardized data, a time series database is used for storage, and for non-standardized data, the next step of processing is performed.

[0085] 2. Standardization processing of non-standardized information

[0086] For text data, the text is cleaned to remove HTML tags, special symbols and stop words such as "of" and "and". Key information extraction is performed by word segmentation and part-of-speech tagging to identify core descriptions. Semantic mapping is used to associate text descriptions to pre-set state quantity indicators. For image data processing, basic feature values are extracted, such as the temperature rise value corresponding to the maximum temperature of the infrared image and the vibration acceleration corresponding to the vibration waveform peak value. Then a time series database is used for storage.

[0087] 3. Output of evaluation indicators

[0088] The processed standardized data and non-standardized data are integrated into a unified structured set: {I1, I2,......In}, each indicator contains a unique identifier, a physical quantity name and a numerical value or state. n

[0089] Step 2: Weight calculation

[0090] Determine the relative importance weight of each evaluation indicator I k (k = 1, 2,......n).

[0091] 1. Construct the judgment matrix A:

[0092] Construct an n-order judgment matrix A:

[0093] A = [a ij ] n×n i,j = 1, 2,..., n

[0094] Where element a ij represents the importance ratio of indicator I i relative to I j , matrix A needs to satisfy: a ij = 1 / a ji (i ≠ j), i = j, a ii = 1, a ij > 0.​

[0095] The decision matrix is ​​constructed using the TLSatty1-9 consistency scaling method. The scale values ​​and their meanings are as follows: Scale 1 represents I... i with I j Equally important. Scale 3, representing I. i Than I j Slightly more important. Scale 5, representing I. i Than I j Clearly important. Scale 7 indicates I i Than I j Strongly important. Scale 9, representing I. i Than I j Extremely important. Scales 2, 4, 6, and 8 represent the compromise values ​​for adjacent judgments.

[0096] Taking five state variables as an example, I1: relative electrical wear degree (weight 4), I2: main circuit resistance (weight 4), I3: temperature rise value (weight 3), I4: vibration acceleration (weight 3), and I5: SF6 gas pressure (weight 2).

[0097] Where I1 and I2, I3 and I4 have equal weights, then a 12 ,a 34 =1. I1 with I3, I4, I2 with

[0098] I3 and I4, with a weight of 4 to 3, are slightly more important. 13 ,a 14 ,a 23 ,a 24 Take 3. I1 and I5, I2 and I5, with a weight of 4 to 2, are clearly more important. a 15 ,a 25 The weight is 5. I3 and I5, I4 and I5, with a weight of 3 to 2, falling between slightly and obviously. 35 ,a 45 The value is 4. Therefore, the corresponding matrix is...

[0099]

[0100] Calculate the largest eigenvalue and eigenvector of a matrix, Aω = λ max ω, thus obtaining the largest eigenvalue λ max The corresponding eigenvector ω = [ω1, ω2, ..., ω n ] T The feature vector is the same as the weight vector.

[0101] The determination of weight coefficients in the construction of the judgment matrix involves subjective elements, leading to errors between the eigenvalues ​​and eigenvectors and their actual values. Therefore, their consistency needs to be verified. The formula is:

[0102]

[0103] where CI is the consistency deviation degree index, λ max is the maximum eigenvalue of the matrix, n is the order of the matrix, and RI is the average random consistency index. CR is the consistency check index, 0 < CR < 0.10, which indicates that the weight coefficient of each index is reasonable, otherwise, the weight coefficient of the state quantity index needs to be re-evaluated until it meets the consistency check.

[0104] Step 3: Dynamic threshold modeling and calculation

[0105] Statistical analysis based on historical state quantity data:

[0106] 1. Historical data is retrieved from the time series database, and the historical monitoring data is fitted with a Weibull distribution, and the probability density function is:

[0107]

[0108] The corresponding inverse cumulative distribution function is:

[0109] y = F -1 (p | η, β) = η - ln(1 - p) 1 / β , p ∈ 0, 1

[0110] where p represents the cumulative distribution probability, x represents the value corresponding to the cumulative probability p, β represents the shape parameter of the Weibull distribution, η represents the scale parameter of the Weibull distribution, and F -1 represents the inverse function of the cumulative probability distribution function.

[0111] 2. Calculate the classification threshold based on the historical state distribution proportion of the equipment:

[0112] For positive degradation indicators such as temperature rise, main loop resistance, relative electrical wear degree, and vibration acceleration amplitude, when p in the inverse cumulative distribution function is a%, a% + b%, and a% + b% + c% respectively, the calculation results are the attention threshold y1, the abnormal threshold y2, and the serious threshold y3 respectively.

[0113] For negative degradation indicators such as SF6 gas pressure, when p in the formula is 1-a%, 1-(a% + b%), and 1-(a% + b% + c%) respectively, the calculation results are the attention threshold y1, the abnormal threshold y2, and the serious threshold y3 respectively.

[0114] 1-(a% + b% + c%) respectively.

[0115] Where a%, b%, and c% represent the proportions of normal, attention, and abnormal state equipment in the historical data respectively.

[0116] Step 4: Determine the membership based on the normal cloud model

[0117] 1. Divide the state interval:

[0118] According to the threshold, the value range is divided into four state intervals: normal interval D1 = [0, y1], attention interval D2 = [y1, y2], abnormal interval D3 = [y2, y3], and serious interval D4 = [y3, +∞].

[0119] 2. Calculate the cloud model digital features:

[0120] For each interval D k (k = 1, 2, 3, 4), calculate the expectation Ex k , entropy En k , and hyper-entropy He k :

[0121] He j = 0.005

[0122] For the measured data x i of a specific evaluation index, use the forward cloud generator to generate a normal cloud model and calculate the membership of each state grade k. The specific steps are as follows:

[0123] Generate random disturbance value of entropy:

[0124] En′ k,j ~ N[En k , (He k ) 2 ]

[0125] Generate cloud titration value:

[0126] x k,j ~ N[Ex k , (En′ k,j ) 2 ]

[0127] Calculate cloud droplet membership:

[0128]

[0129] Generate cloud droplets (x i , μ k,j (x)), and filter out the intersecting cloud droplets, that is, keep the cloud droplets that satisfy |x k,j -x i | < θ, where θ is a small threshold value.

[0130] For state grade k, take the mean value of the membership of the intersecting cloud droplets to calculate the average membership:

[0131]

[0132] Where M represents the number of intersecting cloud droplets. Then, the membership vector is output:

[0133] μ(x i )=[μ1(x i ),μ2(x i ),μ3(x i ),μ4(x i )]

[0134] Step 5: Evidence Fusion

[0135] 1. Construct the Basic Probability Assignment (BPA) function:

[0136] Input membership vector μ(x) i )=[μ1(x i ),μ2(x i ),μ3(x i ),μ4(x i The conversion formula for the BPA function is:

[0137] m i (D k )=μ i k = 1, 2, 3, 4

[0138]

[0139] Where, m i Indicator I k The BPA function, D k For the state level, μ i Θ represents the membership degree, Θ represents the identification frame, which is the complete set containing all possible states, and m represents the membership degree. i (Θ) represents the uncertainty, which is the probability that a specific state cannot be assigned to it.

[0140] 2. Adjust the weight of evidence

[0141] The eigenvectors of the input judgment matrix are used as weight vectors, ω = [ω1, ω2, ..., ω]. n ] T The corrected formula is:

[0142]

[0143] α i =1-ω i ′·[1-mi(Θ)]

[0144] m i ′(D k )=α i ·m i (D k )

[0145]

[0146] Where, ω i Let ω be the weight vector. i ' represents the normalized weights, α i As a discount factor, m can weaken the credibility of low-weighted evidence. i 'This is the modified BPA function.

[0147] 3. Evidence synthesis

[0148] The input modified BPA set is {m1', m2', ..., m n The formula for evidence synthesis is:

[0149]

[0150] Where K is the total conflict coefficient, A, B, and C are non-empty subsets of the identification frame Θ, δ(A) represents the conflict assignment weight of subset A, and m (i) (A) represents the probability assignment of subset A after fusing i pieces of evidence.

[0151] 4. Status Determination

[0152] Output overall status level Where, m final (D k ) is state D k The final probability assignment value.

[0153] Step 6: Conflict Resolution

[0154] In actual evaluation, highly conflicting evidence may be generated due to inherent contradictions in the data or preprocessing errors. Based on pignistic probability distance, highly conflicting evidence is identified, corrected, and re-fused to improve reliability.

[0155] 1. Calculate the degree of conflict of evidence

[0156] Input the raw BPA data, {m1,m2,......,m n For each pair of evidence, calculate the degree of conflict. The steps for calculating the degree of conflict are: calculate the conflict coefficient. Where B and C are non-empty subsets of the identification frame Θ, m i (B),m j (C) indicates evidence m i ,m j Calculate the BPA function values ​​for B and C. Calculate the Pignistic probability function. Where A is a subset of the identification frame. Calculate the Pidnistic probabilistic distance. Finally, calculate the overall conflict degree. The higher the overall conflict level value, the stronger the conflict.

[0157] 2. Calculate the weight of evidence

[0158] The calculation steps are as follows: Calculate the similarity sim(mi,m) j ) = 1 - conf(m i ,m j ), support Certainty D k For the state interval, the decision degree The credibility CRD(i) is calculated as SD(i) + CD(i)·DD(i). Finally, the normalized weights are calculated.

[0159] 3. Identifying highly conflicting evidence

[0160] The judgment condition is like Then m i This constitutes highly conflicting evidence. Among them, Let n be the arithmetic mean of the weights of all evidence, where n is the total number of pieces of evidence.

[0161] 4. Revise the BPA (Best Practices) for highly conflicting evidence.

[0162] The corrected formula is:

[0163]

[0164] Where mi' is the modified BPA function, D k This refers to the status level.

[0165] 5. Reintegrate the evidence and output the final status level.

[0166] Using the modified BPA function {m1',m2',......,m n Following the evidence fusion rules in step 6, the evidence is fused again to output the final state level.

[0167] The final output of the comprehensive status level (normal, alert, abnormal, severe) is directly connected to the power system early warning platform. Normal level: continuous monitoring, no intervention required; alert level: equipment is marked and the data verification process is initiated.

[0168] Abnormal level: Triggers a preventative maintenance work order; Critical level: Executes an emergency shutdown command and pushes an alarm. The result is simultaneously updated to the equipment management database, supporting degradation trend analysis and operational decision optimization.

[0169] Please see Figure 2The diagram shows a structural block diagram of a combined electrical appliance health analysis system based on multi-source data fusion according to this application.

[0170] like Figure 2 As shown, the combined electrical appliance health analysis system 200 includes a generation module 210, a first calculation module 220, a second calculation module 230, a third calculation module 240, an output module 250, and a correction module 260.

[0171] The system includes the following modules: a generation module 210, configured to acquire text and image data of the combined electrical appliances; clean and semantically map the text data to preset state quantity indicators; extract basic feature values ​​from the image data; and generate a standardized set of evaluation indicators. A first calculation module 220 is configured to calculate the weight vectors of each evaluation indicator by constructing a judgment matrix and perform consistency verification. A second calculation module 230 is configured to perform Weibull distribution fitting based on historical state quantity data and calculate the grading thresholds for positive and negative degradation indicators. A third calculation module 240 is configured to divide state intervals based on a normal cloud model and calculate the membership vectors of each state level. An output module 250 is configured to convert the membership vectors into a basic probability assignment function, combine it with the weight vectors for evidence fusion, and output a comprehensive state level. A correction module 260 is configured to identify high-conflict evidence based on the Pignistic probabilistic distance, correct it, re-fuse it, and output the final state level.

[0172] It should be understood that Figure 2 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 2 The various modules in the document will not be described in detail here.

[0173] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the combined appliance health analysis method based on multi-source data fusion in any of the above method embodiments.

[0174] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0175] Acquire text data and image data of the combined electrical appliances, clean the text data and semantically map it to preset state quantity indicators, extract basic feature values ​​from the image data, and generate a standardized evaluation index set.

[0176] The weight vector of each evaluation index is calculated by constructing a judgment matrix, and consistency verification is performed.

[0177] Based on historical state data, a Weibull distribution is fitted to calculate the grading thresholds for positive and negative degradation indices.

[0178] The state intervals are divided based on the normal cloud model, and the membership vector of each state level is calculated.

[0179] The membership vector is transformed into a basic probability assignment function, and evidence is fused by combining the weight vector to output the comprehensive state level.

[0180] High-conflict evidence is identified based on the Pignistic probabilistic distance, corrected, re-fused, and the final state level is output.

[0181] Computer-readable storage media may include a stored program area and a stored data area, wherein the stored program area may store an operating system and an application program required for at least one function; the stored data area may store data created based on the use of the multi-source data fusion-based combined appliance health analysis system, etc. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, which can be connected to the multi-source data fusion-based combined appliance health analysis system via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0182] Figure 3 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 3 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 3 Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the combined appliance health analysis method based on multi-source data fusion described in the above embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the combined appliance health analysis system based on multi-source data fusion. The output device 340 may include a display screen or other display device.

[0183] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0184] In one implementation, the above-described electronic device is applied in a combined electrical appliance health analysis system based on multi-source data fusion, for a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0185] Acquire text data and image data of the combined electrical appliances, clean the text data and semantically map it to preset state quantity indicators, extract basic feature values ​​from the image data, and generate a standardized evaluation index set.

[0186] The weight vector of each evaluation index is calculated by constructing a judgment matrix, and consistency verification is performed.

[0187] Based on historical state data, a Weibull distribution is fitted to calculate the grading thresholds for positive and negative degradation indices.

[0188] The state intervals are divided based on the normal cloud model, and the membership vector of each state level is calculated.

[0189] The membership vector is transformed into a basic probability assignment function, and evidence is fused by combining the weight vector to output the comprehensive state level.

[0190] High-conflict evidence is identified based on the Pignistic probabilistic distance, corrected, re-fused, and the final state level is output.

[0191] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0192] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for health analysis of combined electrical appliances based on multi-source data fusion, characterized in that, include: Acquire text data and image data of the combined electrical appliances, clean the text data and semantically map it to preset state quantity indicators, extract basic feature values ​​from the image data, and generate a standardized evaluation index set. The weight vector of each evaluation index is calculated by constructing a judgment matrix, and consistency verification is performed. Based on historical state data, a Weibull distribution is fitted to calculate the grading thresholds for positive and negative degradation indices. The state intervals are divided based on the normal cloud model, and the membership vector of each state level is calculated. The membership vector is transformed into a basic probability assignment function, and evidence is fused by combining the weight vector to output the comprehensive state level. High-conflict evidence is identified based on the Pignistic probabilistic distance, corrected, re-fused, and the final state level is output.

2. The method for health analysis of combined electrical appliances based on multi-source data fusion according to claim 1, characterized in that, The steps of cleaning the text data and semantically mapping it to preset state quantity indicators, extracting basic feature values ​​from the image data, and generating a standardized evaluation index set include: Standardized data is collected in real time using IoT protocols, and non-standardized data is imported in batches. Cleaning, key information extraction, and semantic mapping of non-standardized text data; Extracting basic feature values ​​from non-standardized image data; The processed data is integrated into a structured set, with each indicator containing a unique identifier, a physical quantity name, and a value / status.

3. The method for health analysis of combined electrical appliances based on multi-source data fusion according to claim 1, characterized in that, The step of constructing a judgment matrix to calculate the weight vector of each evaluation index and performing consistency verification includes: The decision matrix is ​​constructed using the TLSatty1-9 consistency scaling method, and its expression is: A=[a ij ] n×n ,j=1,2,...,n In the formula, A is the judgment matrix, a ij For indicator I i Relative to index I j Importance ratio; Calculate the maximum eigenvalue and eigenvector in the judgment matrix, based on Aω = λ max ω is determined by the largest eigenvalue λ max The corresponding eigenvectors ω = [ω1, ω2, ..., ω n ] T The feature vector is the same as the weight vector.

4. The method for health analysis of combined electrical appliances based on multi-source data fusion according to claim 1, characterized in that, The step of fitting a Weibull distribution to historical state data and calculating the grading thresholds for positive and negative degradation indices includes: Historical data was retrieved from the time-series database, and a Weibull distribution was fitted to the historical monitoring data. The probability density function is: The corresponding inverse cumulative distribution function is: y=F -1 (p|η,β)=η-ln(1-p) 1 / β ,p∈0,1, In the formula, p is the cumulative probability, x represents the value corresponding to the cumulative probability p, β is the shape parameter of the Weibull distribution, η is the scale parameter of the Weibull distribution, and F -1 It is the inverse function of the cumulative probability distribution function. The grading thresholds are calculated based on the distribution ratio of the equipment's historical status. For positive deterioration indicators, when p takes a%, a%+b%, and a%+b%+c% in the inverse cumulative distribution function, the calculated results are attention threshold y1, abnormal threshold y2, and severe threshold y3, respectively. For the negative degradation index, when p takes the values ​​of 1-a%, 1-(a%+b%), and 1-(a%+b%+c%) in the inverse cumulative distribution function, the calculated results are the attention threshold y1, the abnormal threshold y2, and the severe threshold y3, respectively. Here, a%, b%, and c% represent the proportions of devices in normal, attention, and abnormal states in historical data, respectively.

5. The method for health analysis of combined electrical appliances based on multi-source data fusion according to claim 1, characterized in that, The process of dividing state intervals based on the normal cloud model and calculating the membership vector of each state level includes: The value range is divided into four state intervals based on the threshold. The state intervals include the normal interval D1 = [0, y1], the attention interval D2 = [y1, y2], the abnormal interval D3 = [y2, y3], and the severe interval D4 = [y3, +∞]. The membership degree of each state level k is calculated using a forward cloud generator, and the membership degree vector μ(x) is output. i )=[μ1(x i ),μ2(x i ),μ3(x i ),μ4(x i ]], where the expression for calculating the membership degree of each state level k is: In' k，j ~N[In k ,(He k ) 2 ],, In the formula, M is the number of intersecting cloud droplets, and Ex k For expectations, En k Entropy.

6. The method for health analysis of combined electrical appliances based on multi-source data fusion according to claim 1, characterized in that, The process of transforming the membership vector into a basic probability assignment function, combining it with the weight vector for evidence fusion, and outputting a comprehensive state level includes: Construct the basic probability assignment function, with the following expression: m i (D k )=μ i ,k=1,2,3,4 In the formula, m i Indicator I k The BPA function, D k For the state level, μ i Θ represents the membership degree, Θ represents the identification frame, which is the complete set containing all possible states, and m represents the membership degree. i (Θ) represents the uncertainty, which is the probability that a specific state cannot be assigned to it; The eigenvectors of the input judgment matrix are used as the weight vectors, and the expression is: a i =1-h i ′·[1-mi(Θ)], m i ′(D k )=α i ·m i (D k ), In the formula, ω i Let ω be the weight vector. i ' represents the normalized weights, α i This is a discount factor used to weaken the credibility of low-weighted evidence; m i ' represents the modified BPA function; Input the corrected BPA set {m1', m2', ..., mn'} and perform evidence synthesis, expressed as: In the formula, K is the total conflict coefficient, A, B, and C are non-empty subsets of the identification frame Θ, δ(A) represents the conflict assignment weight of subset A, and m (i) (A) is the probability assignment of subset A after fusing i pieces of evidence; Output overall status level Where, m final (D k ) is state D k The final probability assignment value.

7. The method for health analysis of combined electrical appliances based on multi-source data fusion according to claim 1, wherein identifying high-conflict evidence based on Pignistic probability distance, correcting and re-fusion, and outputting the final state level includes: The degree of conflict of evidence is calculated as follows: Input the original BPA data, {m1,m2,......,mn}, and calculate the conflict degree for each pair of evidence. The steps for calculating the conflict degree are: calculate the conflict coefficient. Where B and C are non-empty subsets of the identification frame Θ, m i (B), m j (C) represents the i-th piece of evidence m. i The BPA function value of B, the j-th piece of evidence m j Calculate the Pignistic probability function for the BPA function value of C. Where A is a subset of the identification frame, the Pidnistic probability distance is calculated. Finally, calculate the overall conflict degree. The higher the overall conflict level value, the stronger the conflict. The evidence weights are calculated as follows: The similarity sim(m) is calculated. i ,m j ) = 1 - conf(m i ,m j ), support Certainty D k For the state interval, the decision degree The credibility CRD(i) is calculated as SD(i) + CD(i)·DD(i). Finally, the normalized weights are calculated. The criteria for determining highly conflicting evidence are as follows: The judgment condition is like Then m i This is highly conflicting evidence, among which, The arithmetic mean of the weights of all evidence is given, where n is the total number of pieces of evidence. The revised formula for high-conflict evidence BPA is as follows: Where, m i ′ represents the modified BPA function, D k Status level; The evidence is re-integrated and a final state level is output, specifically as follows: Using the revised BPA function, the evidence fusion synthesis rules are re-fused to output the final state level.

8. A combined electrical appliance health analysis system based on multi-source data fusion, characterized in that, include: The generation module is configured to acquire text data and image data of the combined electrical appliances, clean the text data and semantically map it to preset state quantity indicators, extract basic feature values ​​from the image data, and generate a standardized evaluation index set. The first calculation module is configured to calculate the weight vector of each evaluation index by constructing a judgment matrix and perform consistency verification. The second calculation module is configured to fit a Weibull distribution based on historical state data and calculate the grading thresholds for positive and negative degradation indices. The third calculation module is configured to divide state intervals based on the normal cloud model and calculate the membership vector of each state level. The output module is configured to transform the membership vector into a basic probability assignment function, combine it with the weight vector to perform evidence fusion, and output the comprehensive state level. The correction module is configured to identify high-conflict evidence based on the pignistic probabilistic distance, correct it, re-fuse it, and output the final state level.

9. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method according to any one of claims 1 to 7.

Images