Photovoltaic cleaning unmanned aerial vehicle and cleaning equipment fault diagnosis system
By constructing a fault diagnosis system with a three-level parameterized model and dual sample set training, the problem of low fault detection efficiency of photovoltaic cleaning drones and cleaning equipment is solved, and accurate fault diagnosis and reliable fault prediction are achieved.
Patent Information
- Application Number
- CN202511281922.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-09
- Publication Date
- 2025-12-19
AI Technical Summary
In the current technology, fault detection of photovoltaic cleaning drones and cleaning equipment relies on manual inspection, which is inefficient and greatly affected by subjective factors, making it difficult to detect early potential faults.
The fault diagnosis system employs a parametric model and dual-sample set training. By constructing a three-level parametric model consisting of a system sub-model, a parameter coupling layer, and a risk output layer, it selects key parameters and, combined with supervised and semi-supervised learning, automatically labels fault dimensions that conform to preset conditions, reducing reliance on manual intervention and improving diagnostic efficiency.
It enables accurate fault diagnosis of photovoltaic cleaning drones and cleaning equipment, reduces diagnostic redundancy costs, and improves fault prediction capabilities and diagnostic reliability.
Smart Images
Figure CN121167413A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of fault diagnosis, and more particularly to a photovoltaic cleaning unmanned aerial vehicle and cleaning equipment fault diagnosis system. BACKGROUND
[0002] At present, with the rapid growth of global demand for renewable energy, photovoltaic power generation as a clean and sustainable energy acquisition method, its industry scale is expanding. A large number of photovoltaic power stations have been built in various places, from vast desert areas to complex mountainous areas, from open plains to large areas of water, photovoltaic panels are widely laid out.
[0003] However, photovoltaic panels are exposed to the outdoor environment for a long time, and are easily attacked by dust, bird droppings, leaves and various pollutants. In areas with less rainfall and more wind and sand, dust accumulation on the surface of photovoltaic panels is particularly serious. Related research shows that moderate and severe dust cover can cause photovoltaic components to generate 10%-15% less power, and even if it is lightly polluted, the power loss can reach 3%-10%. The traditional manual cleaning method not only has low efficiency and consumes a lot of labor cost, but also in complex terrain, manual cleaning is difficult to reach some areas, and cleaning work cannot be carried out comprehensively and timely. Delegating cleaning to external units not only incurs high costs, but also cannot guarantee the cleaning period and effect; photovoltaic cleaning unmanned aerial vehicles and cleaning equipment may fail due to various factors during operation. Electrical faults, mechanical faults and sensor faults occur from time to time, and the causes include environmental factors, aging and wear caused by long-term use of equipment, and human errors during operation. At present, fault detection mainly relies on manual inspection and simple sensor data monitoring. Manual inspection not only has a huge workload and extremely low efficiency, but is also greatly affected by subjective factors and difficult to detect early fault hazards.
[0004] Therefore, how to realize accurate fault diagnosis of photovoltaic cleaning unmanned aerial vehicles and cleaning equipment is a problem that those skilled in the art need to solve. SUMMARY
[0005] Therefore, the present application provides a photovoltaic cleaning unmanned aerial vehicle and cleaning equipment fault diagnosis system, which realizes accurate fault diagnosis of photovoltaic cleaning unmanned aerial vehicles and cleaning equipment.
[0006] In order to achieve the above purpose, the present application adopts the following technical solutions:
[0007] A photovoltaic cleaning unmanned aerial vehicle and cleaning equipment fault diagnosis system is arranged on the cleaning unmanned aerial vehicle and cleaning equipment, and comprises:
[0008] a parameter selection module, obtaining original parameters related to the fault, building a current parameterized model according to the original parameters and a current fault feature of the equipment, obtaining related data of the parameterized model under the fault scene and processing the related data to construct a quantization database, and selecting key parameters of all parameterized models in the quantization database based on the quantization database and the current parameterized model;
[0009] a model training module, establishing a label sample set and a parameter sample set, training a fault prediction model and a fault mapping model based on the label sample set, the fault prediction model being used for predicting labels of the key parameters in each fault dimension, inputting the parameter sample set into the fault prediction model to obtain predicted labels of the key parameters in each fault dimension, the label sample set including the key parameters and labels of the key parameters in each different fault dimension, and the parameter sample set including the key parameters without labels, and the fault mapping model being used for predicting a label of a current fault dimension according to labels of the same key parameter in other fault dimensions;
[0010] a label mapping module, inputting predicted fault labels of the same key parameter in other fault dimensions into the fault mapping model to obtain a mapping label of the corresponding key parameter in the current fault dimension;
[0011] a model obtaining module, analyzing consistency degrees of predicted labels and mapping labels of the same key parameter in the same fault dimension, selecting fault dimensions with consistent degrees meeting preset conditions for automatic labeling, and the automatic labeling value of each fault dimension being a weighted average value of the corresponding dimension predicted label and the mapping label, and continuously training the fault prediction model according to newly labeled data to obtain a fault diagnosis model.
[0012] Preferably, the parameter selection module comprises:
[0013] a model building module, constructing a system submodel, a parameter coupling layer and a risk output layer, and a three-level current parameterized model M;
[0014]
[0015] wherein R is a current risk level, S s (X s ) is a submodel of the s th system, X s is an original parameter vector of the system, ω s is a system weight, σ(·) is a Sigmoid mapping function, and ε is an error correction term;
[0016] a database construction module, obtaining related data of the parameterized model under the fault scene, standardizing fault feature parameters in the related data, and constructing a quantization database;
[0017] The key parameter screening module selects common parameters of the parameterized model under all fault scenarios in the quantification database, calculates the cumulative contribution rate of each common parameter, sorts based on the cumulative contribution rate, and analyzes the correlation degree of each common parameter and the current parameterized model in turn. When the correlation degree is greater than a preset degree, the corresponding common parameter is retained. All retained common parameters are used as key parameters.
[0018] Preferably, the key parameter screening module calculates the cumulative contribution rate of each common parameter, including:
[0019]
[0020] Wherein Y is the cumulative contribution rate of a certain common parameter, M is the number of device types, H m is the fault contribution rate of the common parameter in the mth device, S k is the fault contribution rate of the kth associated parameter, K is the number of other parameters related to the current common parameter, is the average value of the contribution rate proportion of the associated parameter, is the minimum value of the contribution rate proportion of the associated parameter, ∑ l≠k S l is the sum of the fault contribution rates of all associated parameters except the kth associated parameter.
[0021] Preferably, the model training module includes:
[0022] The first model training module trains a fault prediction model for the tth fault dimension based on the key parameters and labels of the tth fault dimension in the label sample set, and obtains T fault prediction models, 1≤t≤T.
[0023] The second model training module inputs the parameter sample set into the T fault prediction models, wherein the tth fault prediction model generates a predicted label of the key parameter in the tth fault dimension.
[0024] Preferably, the model acquisition module includes: calculating the absolute difference between the predicted label and the mapping label of the same key parameter in the same fault dimension; traversing the T absolute differences of the same key parameter in the T fault dimensions, and determining that the consistency of the predicted label and the mapping label of the fault dimension with an absolute difference less than a given threshold meets a preset condition.
[0025] Preferably, the model acquisition module further includes: selecting a part of key parameters with the greatest deviation from the preset condition for manual annotation, and continuing to train the fault prediction model according to the manually annotated data.
[0026] Preferably, each fault dimension automatic labeling value is a weighted average value of the corresponding dimension prediction label and the mapping label, wherein the weighted coefficient of the prediction label is 0.5, and the weighted coefficient of the mapping label is 0.5.
[0027] Compared with the prior art, the photovoltaic cleaning unmanned aerial vehicle and the cleaning equipment fault diagnosis system provided by the present application can realize accurate parameter screening to reduce the diagnosis redundancy cost by constructing a three-level parameterization model combining a system sub-model, a parameter coupling layer and a risk output layer and calculating the cumulative contribution rate to screen the key parameters. BRIEF DESCRIPTION OF DRAWINGS
[0028] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.
[0029] Figure 1 The structural schematic diagram provided by the present application. DETAILED DESCRIPTION
[0030] The technical solutions in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.
[0031] The embodiments of the present application disclose a photovoltaic cleaning unmanned aerial vehicle and cleaning equipment fault diagnosis system, which is arranged on a cleaning unmanned aerial vehicle and a cleaning equipment, as shown in Figure 1 The system comprises:
[0032] The parameter selection module acquires original parameters related to faults, builds a current parameterization model according to the original parameters and the current fault characteristics of the equipment, acquires related data of the parameterization model under the fault scene and processes the data to build a quantization database, selects key parameters of all parameterization models in the quantization database based on the quantization database and the current parameterization model,
[0033] The model training module establishes a label sample set and a parameter sample set, trains a fault prediction model and a fault mapping model based on the label sample set, the fault prediction model is used to predict the label of the key parameter in each fault dimension, the parameter sample set is input into the fault prediction model, and the predicted label of the key parameter in each fault dimension is obtained; the label sample set includes the key parameter and the label of the key parameter in each different fault dimension, and the parameter sample set includes the key parameter without a label; the fault mapping model is used to predict the label of the current fault dimension according to the label of the same key parameter in other fault dimensions;
[0034] The label mapping module inputs the predicted fault label of the same key parameter in other fault dimensions into the fault mapping model, and obtains the mapping label of the corresponding key parameter in the current fault dimension;
[0035] The model obtaining module analyzes the consistency degree of the predicted label and the mapping label of the same key parameter in the same fault dimension, selects the fault dimension with consistent degree meeting a preset condition for automatic labeling, and the automatic labeling value of each fault dimension is a weighted average value of the corresponding dimension prediction label and mapping label; the fault prediction model is continuously trained according to the newly labeled data to obtain a fault diagnosis model. Wherein, the newly labeled data includes not only the automatic labeling value of each fault dimension, but also the unlabeled key parameter in the parameter sample set, which is bound to the automatic labeling value after screening to form a sample pair.
[0036] In one specific embodiment, the parameter selection module includes:
[0037] The model building module constructs a system submodel, a parameter coupling layer and a risk output layer, a three-level current parameterized model M;
[0038]
[0039] Wherein, R∈[0,3] is the current risk level (0=no fault, 1=mild abnormality, 2=severe fault, 3=emergency shutdown), S s (X s ) is the submodel of the s-th system (s=1: power system, s=2: execution system, s=3: control system, s=4: environment system), X s is the original parameter vector of the system, ω s is the system weight, σ(·) is the Sigmoid mapping function, and ε is the error correction term, |ε|≤0.1;
[0040] Wherein, the cleaning unmanned aerial vehicle includes:
[0041] The power system, the original parameters correspond to the flight motor temperature, the cleaning motor current and the power battery voltage;
[0042] The execution system, the original parameter corresponds to the cleaning water pump pressure, the brush head rotation speed;
[0043] The control system, the original parameter corresponds to the flight inclination deviation, the GPS positioning deviation;
[0044] The environment adaptation system, the original parameter corresponds to the operation wind speed;
[0045] The ground cleaning equipment comprises:
[0046] The power system, the original parameter corresponds to the walking motor vibration;
[0047] The execution system, the original parameter corresponds to the cleaning agent flow, the brush head wear amount;
[0048] The control system, the original parameter corresponds to the sensor response delay;
[0049] The database construction module obtains the related data of the parameterized model under the fault scene, standardizes the fault characteristic parameters in the related data, and constructs a quantitative database;
[0050] The standardization processing formula is specifically as follows:
[0051]
[0052] Wherein, A i is the measured value of the i th fault characteristic parameter, r i is the acquisition error of the i th fault characteristic parameter, n is the total number of fault characteristic parameters, N i is the parameter value after standardization processing.
[0053] The key parameter screening module selects the common parameters of the parameterized model under all fault scenes in the quantitative database, calculates the cumulative contribution rate of each common parameter, sorts based on the cumulative contribution rate, analyzes the correlation degree of each common parameter and the current parameterized model in turn, and when the correlation degree is greater than a preset degree, the corresponding common parameter is retained; all the retained common parameters are taken as key parameters.
[0054] In one specific embodiment, the key parameter screening module calculates the cumulative contribution rate of each common parameter, comprising:
[0055]
[0056] Wherein, Y is the cumulative contribution rate of a certain common parameter, M is the number of device types, H m is the fault contribution rate of the common parameter in the m th device, S k is the fault contribution rate of the k th associated parameter, K is the number of other parameters related to the current common parameter, is the average value of the associated parameter contribution rate proportion, The minimum value of the contribution rate of the correlation parameter is ∑ l≠k S l The sum of the failure contribution rates of all correlation parameters except the kth correlation parameter.
[0057] In one specific embodiment, the model training module includes:
[0058] The first model training module trains a failure prediction model corresponding to the tth failure dimension based on the key parameters and labels of the tth failure dimension in the labeled sample set, obtaining T failure prediction models, 1≤t≤T;
[0059] The second model training module inputs the parameter sample set into the T failure prediction models, wherein the tth failure prediction model generates a predicted label of the key parameter on the tth failure dimension.
[0060] In one specific embodiment, the model acquisition module includes: calculating the absolute difference between the predicted label and the mapped label of the same key parameter on the same failure dimension; traversing the T absolute differences of the same key parameter on the T failure dimensions, and determining that the consistency of the predicted label and the mapped label of the failure dimension whose absolute difference is less than a given threshold meets a preset condition.
[0061] In one specific embodiment, the model training module establishes a labeled sample set X L , Wherein, is the standardized value of the key parameter, is the failure label corresponding to each key parameter, N L is the number of labeled samples; the parameter sample set X U , N U is the number of unlabeled samples;
[0062] The failure prediction model f t is trained, and for each key parameter dimension t, an independent prediction model f t is trained to predict the failure label of the dimension:
[0063] Model input:
[0064] Model output: (the tth dimension failure label);
[0065] Model type: gradient boosting tree, training loss function is mean square error:
[0066]
[0067] The failure label mapping model g tTraining, for each dimension t, predict the fault label of the current dimension using the predicted labels of other dimensions, train the mapping model g t :
[0068] Model input: (fault labels of other 2 dimensions except the t-th dimension);
[0069] Model output: (fault label of the t-th dimension);
[0070] Model type: use multi-layer perceptron, training loss function is mean square error:
[0071]
[0072] For the sample U in the parameter sample set X Input the prediction model f t to obtain the predicted label:
[0073]
[0074] Input the predicted labels of other dimensions into the mapping model g t to obtain the mapping label:
[0075]
[0076] In one specific embodiment, the model obtaining module further comprises: selecting the part of key parameters with the maximum deviation of consistency degree from the preset condition for manual labeling, and continuing to train the fault prediction model according to the manually labeled data.
[0077] In one specific embodiment, the automatically labeled value of each fault dimension is the weighted average of the corresponding dimension predicted label and mapping label, wherein the weighted coefficient of the predicted label is 0.5 and the weighted coefficient of the mapping label is 0.5.
[0078] In the specification, each embodiment is described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between each embodiment can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the related parts can be referred to the method part.
[0079] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A photovoltaic cleaning drone and cleaning equipment failure diagnosis system provided on a cleaning drone and cleaning equipment, characterized in that, The application relates to a fault diagnosis method and device. The parameter selection module obtains original parameters related to faults, and builds a current parameterized model according to the original parameters and current fault characteristics of equipment; Obtain relevant data of the parameterized model under the fault scene and process it to build a quantitative database; Based on the quantitative database and the current parameterized model, key parameters of all parameterized models in the quantitative database are selected; The model training module establishes a label sample set and a parameter sample set, trains a fault prediction model and a fault mapping model based on the label sample set, the fault prediction model is used for predicting labels of key parameters in various fault dimensions, the parameter sample set is input into the fault prediction model, and predicted labels of the key parameters in various fault dimensions are obtained; the label sample set comprises key parameters and labels of the key parameters in various different fault dimensions, and the parameter sample set comprises key parameters without labels; the fault mapping model is used for predicting labels of a current fault dimension according to labels of the same key parameter in other fault dimensions; The label mapping module inputs predicted fault labels of the same key parameter in other fault dimensions into the fault mapping model, and obtains mapping labels of the corresponding key parameter in the current fault dimension; The model obtaining module analyzes consistency degrees of predicted labels and mapping labels of the same key parameter in the same fault dimension, selects fault dimensions with consistent degrees meeting preset conditions for automatic labeling, and the automatic labeling values of various fault dimensions are weighted average values of corresponding dimension predicted labels and mapping labels; the fault prediction model is continuously trained according to newly labeled data, and a fault diagnosis model is obtained.
2. The photovoltaic cleaning drone and cleaning equipment fault diagnosis system according to claim 1, characterized in that, The parameter selection module comprises: The model building module builds a system submodel, a parameter coupling layer and a risk output layer, and a three-level current parameterized model M; where R ∈ [0, 3] is the current risk level, S s (X s ) is the sub-model of the s-th system, X s is the original parameter vector of the system, ω s is the system weight, σ(·) is the Sigmoid mapping function, and ε is the error correction term. The database building module obtains relevant data of the parameterized model under the fault scene, standardizes fault characteristic parameters in the relevant data, and builds a quantitative database; The key parameter screening module selects common parameters of the parameterized model under the fault scene in the quantitative database, calculates cumulative contribution rates of each common parameter, sorts the cumulative contribution rates, analyzes correlation degrees of each common parameter and the current parameterized model in sequence, retains corresponding common parameters when the correlation degrees are greater than a preset degree, and takes all retained common parameters as key parameters.
3. The photovoltaic cleaning drone and cleaning equipment fault diagnosis system according to claim 2, characterized in that, The model training module comprises: Wherein Y is the cumulative contribution rate of a certain common parameter, M is the number of equipment types, H m is the failure contribution rate of the common parameter in the mth type of equipment, S k is the failure contribution rate of the kth associated parameter, K is the number of other parameters related to the current common parameter, is the average of the proportion of the contribution rate of the associated parameter, is the minimum value of the proportion of the contribution rate of the associated parameter, ∑ l≠k S l is the sum of the failure contribution rates of all associated parameters except the kth associated parameter.
4. The photovoltaic cleaning drone and cleaning equipment fault diagnosis system of claim 1, wherein, The first model training module trains a fault prediction model of a tth fault dimension based on a key parameter and a label of the tth fault dimension in the label sample set, obtains T fault prediction models, and 1<=t<=T; The second model training module inputs the parameter sample set into the T fault prediction models, wherein the tth fault prediction model generates a predicted label of the key parameter in the tth fault dimension. 5. The photovoltaic cleaning drone and cleaning equipment fault diagnosis system of claim 1, wherein, The model acquisition module comprises: calculating an absolute difference value of a predicted label and a mapping label of a same key parameter in a same fault dimension; traversing T absolute difference values of the same key parameter in T fault dimensions, and determining consistency of the predicted label and the mapping label of the fault dimension with the absolute difference value less than a given threshold value to meet a preset condition.
6. The photovoltaic cleaning drone and cleaning equipment fault diagnosis system of claim 1, wherein, The model acquisition module further comprises: selecting a part of key parameters with the greatest deviation from the preset condition in terms of consistency to perform manual labeling, and continuing to train the fault prediction model according to the manually labeled data.
7. The photovoltaic cleaning drone and cleaning equipment fault diagnosis system of claim 1, wherein, The automatic labeling value of each fault dimension is a weighted average value of the corresponding dimension predicted label and the mapping label, wherein the weighted coefficient of the predicted label is 0.5, and the weighted coefficient of the mapping label is 0.5.