Fault diagnosis method and system for wind turbine based on multi-source monitoring

Abstract

The application provides a wind turbine fault diagnosis method and system based on multi-source monitoring, and relates to the field of wind power generation. The method comprises the following steps: determining an optimal sensor arrangement scheme through a multi-population and double-iteration butterfly optimization algorithm, and collecting multi-source fault response data corresponding to multiple fault types; constructing a fault response knowledge graph and a fault diagnosis model based on the multi-source fault response data corresponding to the multiple fault types; collecting real-time multi-source monitoring data of the wind turbine based on the optimal sensor arrangement scheme; generating a first fault diagnosis result based on the fault response knowledge graph and the real-time multi-source monitoring data of the wind turbine; generating a second fault diagnosis result based on the fault diagnosis model and the real-time multi-source monitoring data of the wind turbine; and generating a fault diagnosis result of the wind turbine based on the first fault diagnosis result and the second fault diagnosis result. The method has the advantages of improving the efficiency and accuracy of wind turbine fault diagnosis.

Description

Technical Field

[0001] This invention relates to the field of wind power generation, and in particular to a fault diagnosis method and system for wind turbines based on multi-source monitoring. Background Technology

[0002] The wind-solar hybrid water supply system has been widely used in some remote areas, effectively solving the problems of water intake and supply. Most water supply points have stable water supply capabilities and can meet daily water needs.

[0003] Fault diagnosis of wind turbines in wind-solar hybrid water supply systems primarily relies on the experience and judgment of experts or technicians. These personnel have accumulated rich professional knowledge through long-term practice, enabling them to locate and identify equipment or system faults by combining historical experience. Fault diagnosis methods based on human experience can generally be divided into two categories: sensory diagnosis and reasoning diagnosis. Sensory diagnosis relies on the visual, auditory, and tactile senses of staff to directly observe the equipment's operating status and determine the fault type. Reasoning diagnosis, on the other hand, involves consulting historical cases and combining analysis and reasoning to gradually narrow down the fault range, achieving fault location and identification. These methods require personnel to have a solid theoretical foundation, familiarity with equipment structure and operating mechanisms, and rich practical experience. However, due to differences in individual knowledge and experience, diagnostic results often exhibit significant subjectivity and inconsistency. With the development of machine learning algorithms, the traditional experience-based manual diagnosis model is gradually shifting towards a data-driven approach. Although machine learning-based fault diagnosis methods have made significant progress, in practical applications they still heavily rely on expert experience for feature selection, making full automation and intelligence difficult to achieve. For wind turbines, the operating environment is complex and the failure modes are diverse. A single sensor signal is insufficient to fully reflect the health status of the equipment. Furthermore, existing data-driven methods still suffer from problems such as strong human intervention and weak adaptability in the feature extraction stage.

[0004] Therefore, there is a need to provide a fault diagnosis method and system for wind turbines based on multi-source monitoring to improve the efficiency and accuracy of fault diagnosis for wind turbines. Summary of the Invention

[0005] This invention provides a fault diagnosis method for wind turbines based on multi-source monitoring, comprising: determining an optimal sensor deployment scheme using a multi-group and double-iteration butterfly optimization algorithm; collecting multi-source fault response data corresponding to multiple fault types based on the optimal sensor deployment scheme; constructing a fault response knowledge graph and a fault diagnosis model based on the multi-source fault response data corresponding to multiple fault types; collecting real-time multi-source monitoring data of the wind turbine based on the optimal sensor deployment scheme; generating a first fault diagnosis result based on the fault response knowledge graph and the real-time multi-source monitoring data of the wind turbine; generating a second fault diagnosis result based on the fault diagnosis model and the real-time multi-source monitoring data of the wind turbine; and generating a fault diagnosis result for the wind turbine based on the first and second fault diagnosis results.

[0006] Furthermore, the optimal sensor deployment scheme is determined through a multi-swarm and double-iteration butterfly optimization algorithm, including: setting up multiple temperature sensing locations, vibration sensing locations, and current sensing locations on the wind turbine; collecting temperature data from multiple temperature sensing locations, vibration data from multiple vibration sensing locations, and current data from multiple current sensing locations corresponding to multiple fault types; for each temperature sensing location, calculating the sensing matrix of the temperature sensing location based on the temperature data corresponding to the temperature sensing locations for multiple fault types; for each vibration sensing location, calculating the sensing matrix of the vibration sensing location based on the temperature data corresponding to the vibration sensing locations for multiple fault types; for each current sensing location, calculating the sensing matrix of the current sensing location based on the temperature data corresponding to the current sensing locations for multiple fault types; and determining the optimal sensor deployment scheme based on the sensing matrices of the temperature sensing locations, vibration sensing locations, and current sensing locations using a multi-swarm and double-iteration butterfly optimization algorithm.

[0007] Furthermore, using a multi-group and double-iteration butterfly optimization algorithm, the optimal sensor deployment scheme is determined based on the sensing matrices of temperature sensing locations, vibration sensing locations, and current sensing locations. This includes: S1, calculating the sensing difference values ​​of temperature sensing locations, vibration sensing locations, and current sensing locations based on the sensing matrices; S2, calculating the sensing coordination coefficients between any temperature sensing location and any vibration sensing location, any temperature sensing location and any current sensing location, and any current sensing location and any vibration sensing location based on the sensing matrices; S3, calculating the sensing coordination coefficients between temperature sensing locations, vibration sensing locations, and current sensing locations based on the sensing matrices; S4. Initialize multiple populations by taking the sensing difference values ​​of the temperature sensing location and the current sensing location, the sensing coordination coefficient between any temperature sensing location and any vibration sensing location, the sensing coordination coefficient between any temperature sensing location and any current sensing location, and the sensing coordination coefficient between any current sensing location and any vibration sensing location. Each butterfly in the population represents at least one combination of temperature sensing location, vibration sensing location, and current sensing location. S5. Construct a fitness function. S6. For each population, iteratively optimize the population according to the fitness function until the population iteration termination condition is met. S7. Determine whether the global iteration termination condition is met. If yes, determine the optimal sensor deployment scheme. If not, proceed to S8. S9. Perform adaptive updates on multiple populations and proceed to S5.

[0008] Further, multiple populations are initialized, including: S31, for each population of butterflies, randomly sampling the number of temperature-sensing locations, vibration-sensing locations, and current-sensing locations included in the butterflies; S32, for each population of butterflies, extracting temperature-sensing locations from multiple temperature-sensing locations based on the number of temperature-sensing locations included in the butterflies and the sensing difference value of each temperature-sensing location; S33, for each population of butterflies, extracting vibration-sensing locations from multiple vibration-sensing locations based on the number of vibration-sensing locations included in the butterflies, the sensing difference value of each vibration-sensing location, and the sensing coordination coefficient between any vibration-sensing location and any extracted temperature-sensing location; S34, for each population of butterflies, extracting current-sensing locations from multiple current-sensing locations based on the number of current-sensing locations included in the butterflies, the sensing difference value of each current-sensing location, the sensing coordination coefficient between any current-sensing location and any extracted temperature-sensing location, and the sensing coordination coefficient between any current-sensing location and any extracted vibration-sensing location.

[0009] Furthermore, the population is iteratively optimized based on the fitness function, including: calculating the fitness value of each butterfly in the population based on the fitness function, determining the current globally optimal butterfly, the butterfly performing global search, and the butterfly performing local search; for the butterfly performing global search, performing a position replacement operation based on the currently globally optimal butterfly; for the butterfly performing local search, determining the butterfly used for flight guidance, and performing a position replacement operation based on the butterfly used for flight guidance.

[0010] Furthermore, adaptive updates are performed on multiple populations, including: for each population, determining the response matrix of the current globally optimal butterfly for each fault type; for any two populations, calculating the substitution coefficients of the two populations based on the response matrices of the current globally optimal butterfly for each fault type; and performing butterfly substitution on multiple populations based on the substitution coefficients of any two populations.

[0011] Furthermore, based on multi-source fault response data corresponding to multiple fault types, a fault response knowledge graph is constructed, including: determining the response matrix corresponding to each fault type based on multi-source fault response data corresponding to multiple fault types; and constructing a fault response knowledge graph based on the response matrix corresponding to each fault type.

[0012] Furthermore, based on the fault response knowledge graph and the real-time multi-source monitoring data of the wind turbine, a first fault diagnosis result is generated, including: determining the current response matrix based on the real-time multi-source monitoring data of the wind turbine; and determining the fault matching degree for each fault type based on the fault response knowledge graph and the current response matrix, wherein the first fault diagnosis result includes the fault matching degree for each fault type.

[0013] Furthermore, the second fault diagnosis result includes the confidence level of each fault type; based on the first fault diagnosis result and the second fault diagnosis result, a fault diagnosis result for the wind turbine is generated, including: for each fault type, calculating the comprehensive probability of occurrence of the fault type based on the fault matching degree and confidence level of the fault type, wherein the fault diagnosis result for the wind turbine includes the comprehensive probability of occurrence of each fault type.

[0014] This invention provides a fault diagnosis system for wind turbines based on multi-source monitoring, comprising: a sensing deployment module for determining the optimal sensor deployment scheme using a multi-group and double-iteration butterfly optimization algorithm; a data acquisition module for acquiring multi-source fault response data corresponding to various fault types based on the optimal sensor deployment scheme; and a fault diagnosis module for constructing a fault response knowledge graph and a fault diagnosis model based on the multi-source fault response data corresponding to various fault types. The data acquisition module is further configured to acquire real-time multi-source monitoring data of the wind turbine based on the optimal sensor deployment scheme. The fault diagnosis module is further configured to generate a first fault diagnosis result based on the fault response knowledge graph and the real-time multi-source monitoring data of the wind turbine. The fault diagnosis module is further configured to generate a second fault diagnosis result based on the fault diagnosis model and the real-time multi-source monitoring data of the wind turbine. The fault diagnosis module is further configured to generate a fault diagnosis result for the wind turbine based on the first and second fault diagnosis results.

[0015] Compared with existing technologies, the fault diagnosis method and system for wind turbines based on multi-source monitoring provided by this invention has at least the following beneficial effects: The optimal sensor deployment scheme is determined by using a multi-group and dual-iteration butterfly optimization algorithm, which ensures the efficiency and representativeness of data acquisition from the source and avoids resource waste and information conflicts caused by redundant sensor deployment. Secondly, a fault response knowledge graph and fault diagnosis model are constructed based on the optimal deployment scheme, realizing bidirectional knowledge accumulation from the physical perception space to the data-driven space. Thirdly, the first fault diagnosis result provides fault location from the perspective of multimodal perception fusion based on the knowledge graph, and the second fault diagnosis result provides fault type confidence from the perspective of deep features based on convolutional neural network. The two complement each other and are integrated, which significantly improves the comprehensiveness, accuracy and robustness of wind turbine fault diagnosis. Attached Figure Description

[0016] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a flowchart illustrating a fault diagnosis method for wind turbines based on multi-source monitoring, according to some embodiments of this specification. Figure 2 The optimal sensor deployment scheme is determined based on some embodiments shown in this specification; Figure 3 This is a flowchart illustrating the multi-group and double-iteration butterfly optimization algorithm according to some embodiments of this specification; Figure 4This is a schematic diagram of the process for initializing multiple populations according to some embodiments of this specification; Figure 5 This is a block diagram of a fault diagnosis system for wind turbines based on multi-source monitoring, as shown in some embodiments of this specification. Figure 6 This is a schematic diagram of an electronic device according to some embodiments of this specification. Detailed Implementation

[0017] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0018] Figure 1 This is a flowchart illustrating a fault diagnosis method for wind turbines based on multi-source monitoring, as shown in some embodiments of this specification. Figure 1 As shown, the fault diagnosis method for wind turbines based on multi-source monitoring may include the following steps.

[0019] Step 1: Determine the optimal sensor deployment scheme using a multi-group and double-iteration butterfly optimization algorithm.

[0020] Figure 2 The optimal sensor deployment scheme is determined based on some embodiments shown in this specification, such as... Figure 2 As shown, step 1 specifically includes: Multiple temperature sensing points, vibration sensing points, and current sensing points are installed on the wind turbine. Collect temperature data from multiple temperature sensing locations, vibration data from vibration sensing locations, and current data from current sensing locations corresponding to various fault types. For each temperature sensing location, a sensing matrix is ​​calculated based on the temperature data of the temperature sensing location corresponding to various fault types. For each vibration sensing location, the sensing matrix of the vibration sensing location is calculated based on the temperature data of the vibration sensing location corresponding to various fault types. For each current sensing location, the sensing matrix of the current sensing location is calculated based on the temperature data of the current sensing location corresponding to various fault types. By employing a multi-group and double-iteration butterfly optimization algorithm, the optimal sensor deployment scheme is determined based on the sensing matrices of temperature sensing location, vibration sensing location, and current sensing location.

[0021] Specifically, multiple temperature sensing locations are deployed in key areas such as the generator stator winding, gearbox bearing housing, converter power module, and nacelle environment. Different temperature sensing locations reflect the thermal state of different components.

[0022] Multiple vibration sensing points are arranged at the input shaft and output shaft of the gearbox, the front bearing of the generator, and the bottom of the tower. Vibration at the input end of the gearbox reflects the meshing state of the blade load and the transmission chain, vibration at the output end reflects the alignment of the coupling, vibration of the generator bearing is directly related to the rotor dynamic balance and the degree of bearing wear, and vibration at the bottom of the tower can monitor problems such as foundation loosening and blade imbalance.

[0023] Multiple current sensing points are set at the generator output terminal, the DC side of the converter, and each circuit feeder. The generator output current reflects the matching status of the power generation and the load. Abnormal current on the DC side of the converter can identify rectifier or inverter module faults. Uneven current in each feeder indicates that there is a risk of poor contact or partial short circuit in the power distribution circuit.

[0024] The temperature data of the temperature sensing location corresponding to the fault type can include the temperature of the temperature sensing location at multiple time points.

[0025] For each temperature sensing location and each fault type, a temperature feature vector can be extracted from the temperature data of the temperature sensing location corresponding to the fault type. This feature vector includes, for example, maximum temperature, minimum temperature, temperature standard deviation, number of temperature peaks, and spectral energy. The sensing matrix for each temperature sensing location is constructed using the temperature feature vectors of each fault type as row vectors.

[0026] For each vibration sensing location and each fault type, vibration feature vectors can be extracted from the vibration data of the vibration sensing location corresponding to the fault type. These feature vectors include, for example, maximum vibration amplitude, minimum vibration amplitude, root mean square vibration value, peak power factor, kurtosis, skewness, and principal components of vibration frequency. The sensing matrix for each vibration sensing location is constructed using the vibration feature vectors of the vibration sensing location corresponding to each fault type as row vectors.

[0027] For each current sensing location and each fault type, current feature vectors can be extracted from the current data of the current sensing location corresponding to the fault type. These feature vectors include, for example, maximum current, minimum current, current standard deviation, number of current peaks, and spectral energy. The sensing matrix for each current sensing location is constructed using the current feature vectors of the current sensing locations corresponding to each fault type as row vectors.

[0028] Figure 3 This is a flowchart illustrating the multi-group and double-iteration butterfly optimization algorithm according to some embodiments of this specification, such as... Figure 3As shown, in some embodiments, the optimal sensor deployment scheme is determined based on the sensing matrices of temperature sensing locations, vibration sensing locations, and current sensing locations using a multi-group and double-iteration butterfly optimization algorithm, including: S1. Based on the sensing matrices of the temperature sensing location, vibration sensing location, and current sensing location, calculate the sensing difference values ​​of the temperature sensing location, vibration sensing location, and current sensing location, respectively. S2. Based on the sensing matrices of the temperature sensing location, vibration sensing location, and current sensing location, calculate the sensing coordination coefficient between any temperature sensing location and any vibration sensing location, the sensing coordination coefficient between any temperature sensing location and any current sensing location, and the sensing coordination coefficient between any current sensing location and any vibration sensing location. S3. Based on the perception difference values ​​of temperature sensing position, vibration sensing position and current sensing position, the perception coordination coefficient of any temperature sensing position and any vibration sensing position, the perception coordination coefficient of any temperature sensing position and any current sensing position and the perception coordination coefficient of any current sensing position and any vibration sensing position, initialize multiple populations, wherein each butterfly in the population represents at least one combination of temperature sensing position, vibration sensing position and current sensing position. S4. Construct the fitness function; S5. For each population, the population is iteratively optimized according to the fitness function until the population iteration termination condition is met. The population iteration termination condition can be that the number of iterations reaches the preset maximum number of iterations (e.g., 20, 30, etc.) or the change in the global optimal fitness value is less than a preset threshold (e.g., 0.5) for several consecutive generations. S6. Determine whether the global iteration termination condition is met. If yes, determine the optimal sensor deployment scheme. If no, execute S7. The global iteration termination condition can be that the substitution coefficients of any two populations are greater than a preset threshold (e.g., 0.7). The optimal sensor deployment scheme can be the combination of temperature sensing location, vibration sensing location, and current sensing location of the butterfly with the highest fitness value among multiple populations. S7. Perform adaptive updates for multiple populations, then execute S5.

[0029] Specifically, for each temperature sensing location, the Euclidean distance between any two row vectors in the sensing matrix of the temperature sensing location is calculated, and the average of the Euclidean distances between all two row vectors is taken as the sensing difference value of the temperature sensing location.

[0030] The calculation method for the sensing difference between vibration sensing location and current sensing location is similar to that for temperature sensing location, and will not be repeated here.

[0031] Under fault-free operating conditions, temperature data from multiple temperature sensing locations, vibration data from vibration sensing locations, and current data from current sensing locations are collected to construct temperature feature vectors, vibration feature vectors, and current feature vectors for each temperature sensing location, vibration sensing location, and current sensing location, respectively, for each fault-free operating condition.

[0032] For each temperature sensing location and each fault type, calculate the cosine similarity between the temperature feature vector of the temperature sensing location corresponding to the fault type and the temperature feature vector of the temperature sensing location corresponding to the fault-free operating condition. Based on the cosine similarity, determine the response identifier of the temperature sensing location corresponding to the fault type. For example, if the cosine similarity is less than the cosine similarity threshold (e.g., 40%), the response identifier of the temperature sensing location corresponding to the fault type is 1; otherwise, it is 0.

[0033] This generates a response identifier sequence corresponding to the temperature sensing location, wherein the response identifier sequence consists of response identifiers for each fault type corresponding to the temperature sensing location.

[0034] Based on the above method, the response identifier sequence corresponding to each vibration sensing location and current sensing location is determined.

[0035] For any temperature sensing location and vibration sensing location, calculate the cosine similarity between the response identifier sequence corresponding to the temperature sensing location and the response identifier sequence corresponding to the vibration sensing location, and use it as the sensing coordination coefficient between the temperature sensing location and the vibration sensing location.

[0036] Based on the above method, the sensing coordination coefficients of any temperature sensing location and any current sensing location, as well as the sensing coordination coefficients of any current sensing location and any vibration sensing location, are determined.

[0037] Figure 4 This is a schematic diagram illustrating the process of initializing multiple populations according to some embodiments of this specification, such as... Figure 4 As shown, in some embodiments, multiple populations are initialized, including: S31. For each population of butterflies, randomly sample the number of temperature-sensing locations, vibration-sensing locations, and current-sensing locations of the butterflies. For example, the range of the number of temperature-sensing locations, vibration-sensing locations, and current-sensing locations can be preset, such as the range of the number of temperature-sensing locations being (3, 10), the range of the number of vibration-sensing locations being (4, 15), and the range of the number of current-sensing locations being (2, 8), etc., and randomly sample the number of temperature-sensing locations, vibration-sensing locations, and current-sensing locations of the butterflies from their respective ranges. S32. For each population of butterflies, extract temperature sensing locations from multiple temperature sensing locations based on the number of temperature sensing locations included in the butterflies and the sensing difference value of each temperature sensing location. S33. For each population of butterflies, based on the number of vibration sensing locations included in the butterfly, the sensing difference value of each vibration sensing location, and the sensing synergy coefficient between any vibration sensing location and any extracted temperature sensing location, vibration sensing locations are extracted from multiple vibration sensing locations. S34. For each population of butterflies, based on the number of current sensing locations included in the butterfly population, the sensing difference value of each current sensing location, the sensing coordination coefficient between any current sensing location and any extracted temperature sensing location, and the sensing coordination coefficient between any current sensing location and any extracted vibration sensing location, extract current sensing locations from multiple current sensing locations.

[0038] Specifically, for each temperature sensing location, the ratio of its sensing difference value to the sum of the sensing differences values ​​of all temperature sensing locations is calculated as the sampling probability of that temperature sensing location. Based on the number of temperature sensing locations included by the butterfly and the sampling probability of each location, temperature sensing locations are drawn from multiple locations. For example, a roulette wheel selection method can be used based on this sampling probability, drawing locations one by one without replacement from all temperature sensing locations until the maximum number of temperature sensing locations included by the butterfly is reached, thus forming a subset of the butterfly's sensing locations in the temperature dimension. This sampling mechanism combines randomness and guidance: on the one hand, random selection ensures the diversity of sensing location combinations among individuals within the population, preventing the algorithm from getting trapped in local optima; on the other hand, the probability weights based on sensing difference values ​​ensure that temperature sensing locations with high information contribution are more likely to be selected, ensuring that the sensing location combinations carried by each butterfly can effectively capture key feature information in the temperature field.

[0039] For each vibration sensing location, the ratio of the sensing difference value of the vibration sensing location to the sum of the sensing difference values ​​of all vibration sensing locations is calculated as the first ratio. The mean of the sensing coordination coefficients of the vibration sensing location and all the extracted temperature sensing locations included in the butterfly is calculated as the first mean of the sensing coordination coefficients. The first ratio and the first mean of the sensing coordination coefficients are weighted and summed to obtain the weight of the vibration sensing location. Using a roulette wheel selection, based on the weight of the vibration sensing location, it is extracted one by one from all vibration sensing locations without replacement until the number of vibration sensing locations included in the butterfly is reached. This constitutes a subset of the sensing locations of the butterfly in the vibration dimension.

[0040] For each current sensing location, the ratio of the sensing difference value of the current sensing location to the sum of the sensing differences values ​​of all current sensing locations is calculated as the second ratio. The mean of the sensing coordination coefficients of the current sensing location and all the extracted temperature sensing locations included in the butterfly is calculated as the second mean of the sensing coordination coefficients. The mean of the sensing coordination coefficients of the current sensing location and all the extracted vibration sensing locations included in the butterfly is calculated as the third mean of the sensing coordination coefficients. The first ratio, the second mean of the sensing coordination coefficients, and the third mean of the sensing coordination coefficients are weighted and summed to obtain the weight of the current sensing location. Using a roulette wheel selection, based on the weight of the current sensing location, it is extracted one by one from all current sensing locations without replacement until the number of current sensing locations included in the butterfly is reached. This constitutes a subset of the butterfly's sensing locations in the current dimension.

[0041] The extraction of vibration sensing locations further incorporates the average sensing coordination coefficient with the extracted temperature sensing locations, based on the sensing difference value. This ensures that the selected vibration locations are not only rich in their own information but also form good collaborative sensing with the temperature locations. The extraction of current sensing locations considers the average coordination coefficient with both temperature and vibration locations, achieving deep coupling optimization among the three sensing dimensions. This effectively maintains the diversity of sensing configurations among individuals within the population to avoid the algorithm getting trapped in local optima, while ensuring that the sensing subset of each individual can maximize the coverage of key features of multi-source sensing information and cross-modal collaborative features.

[0042] The fitness function is related to the response of the butterfly to different fault types based on a combination of at least one temperature-sensing location, vibration-sensing location, and current-sensing location represented by the butterfly. For example, for each fault type, the total number of response dimensions for the corresponding fault type is determined. If at least one temperature-sensing location, vibration-sensing location, and current-sensing location respond to the fault type, the total number of response dimensions (temperature, vibration, current) for the corresponding fault type is 3; if only at least one temperature-sensing location responds to the fault type, the total number of response dimensions is 1; if only at least one vibration-sensing location responds to the fault type, the total number of response dimensions is 1; and if at least one temperature-sensing location and one vibration-sensing location respond to the fault type, the total number of response dimensions is 2. For each fault type, the total number of locations responding to each dimension (temperature, vibration, current) is determined, for example, 2 temperature-sensing locations, 3 vibration-sensing locations, and 4 current-sensing locations. A fitness function can be constructed based on the total number of response dimensions corresponding to each fault type and the total number of locations in each dimension that respond to each fault type. The greater the total number of response dimensions corresponding to each fault type and the smaller the total number of locations in each dimension that respond to each fault type, the larger the fitness value.

[0043] For example, the fitness function can be: , in, For the fitness function, and As weight, and Greater than 0, , This represents the total number of response dimensions corresponding to the nth fault type. The total number of fault types. The total number of temperature sensing locations in response to the nth type of fault, The total number of temperature sensing locations for the preset response (e.g., 3). The total number of vibration sensing locations in response to the nth type of fault, The total number of vibration sensing locations for a preset response (e.g., 4). The total number of current sensing locations in response to the nth fault type. The total number of current sensing locations for the preset response (e.g., 2).

[0044] This fitness function comprehensively evaluates the quality of the position combinations represented by the butterfly from two perspectives: multi-dimensional coverage and single-dimensional redundancy. Regarding multi-dimensional coverage, the first term of the function measures the breadth of the butterfly's perceptual modal response to all fault types. The total number of response dimensions for the fault type, ranging from 1 to 3, indicates that the butterfly's multimodal perception of the fault is more comprehensive and the location information is richer when all three dimensions—temperature, vibration, and current—can respond to a certain fault type, thus improving the reliability of fault diagnosis. Regarding single-dimensional redundancy, the second term of the function measures the concentration of response locations in each dimension. Taking the temperature dimension as an example... A smaller value indicates a greater number of locations that closely approximate the ideal response in that dimension; the same applies to vibration and current dimensions. This function guides the algorithm to prioritize location combinations that exhibit multi-dimensional responses to fault types and have highly concentrated positioning results across all dimensions, thereby ensuring comprehensive fault type coverage while effectively suppressing single-dimensional redundancy.

[0045] In some embodiments, iterative optimization of the population is performed according to a fitness function, including: Based on the fitness function, calculate the fitness value of each butterfly in the population, and determine the current globally optimal butterfly, the butterfly performing global search, and the butterfly performing local search. The method for determining the butterfly performing global search and the butterfly performing local search is the same as that of existing butterfly optimization algorithms, and will not be elaborated here. For butterflies performing global search, a position replacement operation is performed based on the current globally optimal butterfly. Specifically, the optimal combination of temperature sensing position, vibration sensing position, and current sensing position represented by the current globally optimal butterfly is used as a guiding benchmark to replace the position combinations carried by butterflies performing global search in the population. For example, some temperature sensing positions not included in the globally optimal butterfly are used to replace some temperature sensing positions included in the globally search butterfly; some vibration sensing positions not included in the globally optimal butterfly are used to replace some vibration sensing positions included in the globally search butterfly; and some current sensing positions not included in the globally optimal butterfly are used to replace some current sensing positions included in the globally search butterfly. This allows the replaced butterflies to retain some of their original sensing positions while integrating more advantageous sensing position information from the globally optimal solution, accelerating the convergence of the entire population towards the optimal position combination with high coverage and low redundancy. For a butterfly performing a local search, identify the butterfly used for flight guidance and perform a position replacement operation based on the butterfly used for flight guidance.

[0046] Specifically, for each butterfly, a response matrix is ​​calculated for each fault type. The response matrix consists of three row vectors, each corresponding to a dimension (temperature, vibration, current). The row vectors are encoded by the positions of the corresponding dimensions that respond to the fault type. For example, for the temperature dimension, if the butterfly performing the local search responds to the fault type at temperature sensing positions 2, 4, and 6, then the row vector corresponding to the temperature dimension is (2, 4, 6).

[0047] For each butterfly performing a global search and for each fault type, the similarity between the response matrix of that butterfly for that fault type and the response matrices of other butterflies for the same fault type is calculated. This similarity is used as the matrix similarity for that fault type. Specifically, the position overlap rate for each dimension can be calculated based on the response matrices of the butterfly performing the global search and the response matrices of other butterflies for the same fault type. The position overlap rate is the ratio of the total number of positions in the row vector corresponding to that dimension that are included in the response matrices of both butterflies to the total number of positions in that dimension included by the butterfly performing the global search. For example, if butterfly A has 4 response positions (2, 5, 7, 9) in the temperature dimension, and butterfly B has 4 response positions (5, 7, 9, 10), then both butterflies have 3 positions (5, 7, 9), and the position overlap rate for that dimension is 3 divided by 4, which equals 0.75. The average position overlap rate across all dimensions is used to obtain the matrix similarity for that fault type. The average matrix similarity for each fault type is then used to obtain the similarity between the butterfly performing the global search and other butterflies. For each other butterfly, calculate a reference value for each other based on their fitness and similarity scores. For example, the reference value for each other butterfly can be calculated using the following formula: , in, This serves as a reference value for other butterflies. To perform a global search, the similarity between the butterfly and other butterflies is calculated. This represents the fitness value for other butterflies.

[0048] Other butterflies with the highest reference values ​​were selected as the guiding butterflies for flight. The optimal combination of temperature-sensing, vibration-sensing, and current-sensing positions represented by the guiding butterflies was used as the guiding benchmark. The position combinations carried by the butterflies performing local searches in the population were then replaced, in the same way as the position replacement operation based on the current globally optimal butterfly.

[0049] When the similarity is high, it indicates that the butterfly's response pattern is similar to that of the globally searched butterfly, and the reference value is appropriately lowered to avoid excessive convergence in the population. When the similarity is low, it indicates that the butterfly carries differentiated fault perception information, and the reference value is more dominated by fitness, making high-quality but different solutions more likely to be selected. This mechanism ensures that the flight-guided butterfly is neither a purely fitness-optimal solution nor a random solution, but rather an optimal compromise between high quality and high diversity. This provides a more representative and complementary guiding benchmark for locally searched butterflies, effectively preventing the population from getting trapped in local optima and improving the global search capability and convergence accuracy for multi-fault type diagnosis.

[0050] In some embodiments, adaptive updates are performed on multiple populations, including: For each population, determine the response matrix for each fault type corresponding to the current globally optimal butterfly of the population; For any two populations, calculate the substitution coefficients for the two populations based on the response matrix of the current globally optimal butterfly for each fault type. Based on the replacement coefficients of any two populations, butterflies are replaced in multiple populations.

[0051] Specifically, the method for determining the response matrix of the current globally optimal butterfly for each fault type is the same as the method for determining the response matrix of the butterfly for each fault type, and will not be repeated here.

[0052] For any two populations, calculate the similarity of the response matrix of the current global best butterfly in each population for each fault type, and use it as the matrix similarity between the two populations for that fault type. This method is consistent with the method described above, which calculates the similarity between the response matrix of each butterfly performing a global search and the response matrices of other butterflies for each fault type, and uses it as the matrix similarity between the two populations for that fault type. This method will not be repeated here.

[0053] The mean of the matrix similarity for all fault types for the two populations is used as the substitution coefficient for the two populations.

[0054] For each population, select the population with the smallest replacement coefficient, and replace the current global best butterfly of the current population with the current global best butterfly of the population with the current global best butterfly of the population with the smallest replacement coefficient.

[0055] By calculating the similarity of the response matrices of the current globally optimal butterflies of any two populations under each fault type, and averaging the similarities across all fault types, a substitution coefficient is obtained. This coefficient comprehensively measures the consistency of responses of the two populations across all fault modes. A smaller substitution coefficient indicates a greater difference in the response patterns of the global optimal solutions of the two populations under each fault type, suggesting that the current population may have fallen into a local optimum for a particular fault type. The population with the smallest substitution coefficient represents the one with the greatest difference in response pattern and the richest complementary information from the current population. By selecting the population with the smallest substitution coefficient and replacing the current population's globally optimal butterfly with its globally optimal butterfly, information exchange and complementarity between populations are achieved, effectively overcoming the premature convergence stagnation problem caused by long-term convergence of a single population. Furthermore, since the substitution coefficient is calculated based on the matrix similarity across all fault types, this substitution strategy is not blind replacement but rather a targeted introduction of high-quality solutions with the greatest difference from the current population, ensuring population diversity while accelerating convergence towards the global optimum.

[0056] Step 2: Based on the optimal sensor deployment scheme, collect multi-source fault response data corresponding to various fault types.

[0057] Specifically, based on the optimal sensor deployment scheme, temperature sensors, vibration sensors, and current sensors can be set at corresponding locations to collect multi-source fault response data corresponding to various fault types. Among them, multi-source fault response data can include temperature data, vibration data, and current data.

[0058] Step 3: Based on multi-source fault response data corresponding to various fault types, construct a fault response knowledge graph and a fault diagnosis model.

[0059] In some embodiments, a fault response knowledge graph is constructed based on multi-source fault response data corresponding to multiple fault types, including: Based on multi-source fault response data corresponding to various fault types, determine the response matrix corresponding to each fault type; A fault response knowledge graph is constructed based on the response matrix corresponding to each fault type.

[0060] Specifically, the response matrix corresponding to the fault type consists of three row vectors, each row vector corresponding to a dimension (temperature, vibration, current). The row vector is composed of the response identifier of each location included in the dimension in the optimal sensor deployment scheme for the fault type.

[0061] The fault response knowledge graph is used to record the response matrix corresponding to each fault type.

[0062] Step 4: Based on the optimal sensor deployment scheme, collect real-time multi-source monitoring data of the wind turbine.

[0063] Specifically, real-time multi-source monitoring data can include temperature data, vibration data, and current data collected in real time by temperature sensors, vibration sensors, and current sensors.

[0064] Step 5: Generate the first fault diagnosis result based on the fault response knowledge graph and real-time multi-source monitoring data of the wind turbine.

[0065] Specifically, it includes: Based on real-time multi-source monitoring data of wind turbines, the current response matrix is ​​determined; Based on the fault response knowledge graph and the current response matrix, the fault matching degree of each fault type is determined, wherein the first fault diagnosis result includes the fault matching degree of each fault type.

[0066] Specifically, determining the current response matrix is ​​similar to determining the response matrix for each fault type, and will not be elaborated here.

[0067] For each fault type, the cosine similarity between the current response matrix and the response matrix corresponding to the fault type recorded in the fault response knowledge graph is calculated, and this is used as the fault matching degree for that fault type.

[0068] Step 6: Based on the fault diagnosis model and the real-time multi-source monitoring data of the wind turbine, generate a second fault diagnosis result.

[0069] The second fault diagnosis result includes the confidence level for each fault type.

[0070] Specifically, the fault diagnosis model comprises three parallel feature extraction branches, which receive real-time temperature, vibration, and current data collected by temperature, vibration, and current sensors, respectively, as input. Each branch sequentially includes a one-dimensional convolutional layer, a batch normalization layer, and an activation function layer: First, the one-dimensional convolutional layer uses multiple sets of convolutional kernels of different sizes (e.g., kernel sizes of 3, 5, and 7) to perform sliding convolution on the original sensor data along the time axis, extracting local feature patterns at different time scales. The number of convolutional kernels is set to 32, 64, and 128 to progressively increase the number of feature channels. Second, the batch normalization layer standardizes the convolutional output, accelerating model convergence and preventing overfitting. Then, the activation function layer introduces nonlinearity using the ReLU function, enhancing the model's ability to express complex fault patterns. After each branch undergoes 2 to 3 sets of convolution-normalization-activation stacking, the features are downsampled through a max-pooling layer, retaining the most significant feature responses and compressing the data dimensionality. Subsequently, the pooled features of the three branches are concatenated in a fully connected fusion layer to form a fused multimodal high-dimensional feature vector. After passing through two fully connected layers and a Softmax output layer, the confidence probability distribution of each fault type is finally output, which is the second fault diagnosis result.

[0071] Step 7: Based on the first fault diagnosis result and the second fault diagnosis result, generate the fault diagnosis result of the wind turbine.

[0072] Specifically, it includes: For each fault type, the comprehensive probability of occurrence of the fault type is calculated based on the fault matching degree and confidence degree of the fault type. The fault diagnosis results of wind turbines include the comprehensive probability of occurrence of each fault type.

[0073] Specifically, for each fault type, the average of the fault matching degree and confidence score is calculated to obtain the comprehensive probability of occurrence of the fault type. The first diagnostic result provides high-precision fault location with consistent multi-source data from the physical perception level, possessing strong interpretability; the second diagnostic result provides quantitative confidence scores for each fault type from the data-driven level, possessing strong classification capabilities; the fusion of the two leverages both the advantages of the butterfly algorithm in multimodal redundancy suppression and location accuracy, and the advantages of convolutional neural networks in complex fault pattern recognition, achieving a dual improvement in location accuracy and classification accuracy. This effectively avoids the risk of misjudgment when a single method is subject to noise interference or missing data, significantly improving the overall reliability and practicality of wind turbine fault diagnosis.

[0074] Figure 5 This is a block diagram of a fault diagnosis system for wind turbines based on multi-source monitoring, as shown in some embodiments of this specification. Figure 5 As shown, a fault diagnosis system for wind turbines based on multi-source monitoring can include a sensing deployment module, a data acquisition module, and a fault diagnosis module.

[0075] The sensor deployment module is used to determine the optimal sensor deployment scheme through a multi-group and double-iteration butterfly optimization algorithm. The data acquisition module is used to collect multi-source fault response data corresponding to various fault types based on the optimal sensor deployment scheme. The fault diagnosis module is used to build a fault response knowledge graph and a fault diagnosis model based on multi-source fault response data corresponding to various fault types. The data acquisition module is also used to collect real-time multi-source monitoring data of wind turbines based on the optimal sensor deployment scheme; The fault diagnosis module is also used to generate the first fault diagnosis result based on the fault response knowledge graph and real-time multi-source monitoring data of the wind turbine. The fault diagnosis module is also used to generate a second fault diagnosis result based on the fault diagnosis model and real-time multi-source monitoring data of the wind turbine. The fault diagnosis module is also used to generate fault diagnosis results for the wind turbine based on the first fault diagnosis results and the second fault diagnosis results.

[0076] The fault diagnosis system for wind turbines based on multi-source monitoring can apply the aforementioned fault diagnosis method for wind turbines based on multi-source monitoring, which will not be elaborated further here.

[0077] Figure 6 These are schematic diagrams of electronic devices illustrated according to some embodiments of this specification, such as... Figure 6 As shown, the electronic device includes a Central Processing Unit (CPU), which can perform various appropriate actions and processes based on a program stored in Read-Only Memory (ROM) or a program loaded from storage into Random Access Memory (RAM), such as executing the methods described in the above embodiments. Various programs and data required for system operation are also stored in the RAM. The CPU, ROM, and RAM are interconnected via a bus. Input / output (I / O) interfaces are also connected to the bus.

[0078] The following components are connected to the I / O interface: input sections including keyboards, mice, etc.; output sections including cathode ray tubes (CRTs), liquid crystal displays (LCDs), and speakers; storage sections including hard drives; and communication sections including network interface cards such as LAN (Local Area Network) cards and modems. The communication sections perform communication processing via networks such as the Internet. Drives are also connected to the I / O interface as needed. Removable media, such as disks, optical discs, magneto-optical discs, semiconductor memories, etc., are installed on the drive as needed so that computer programs read from them can be installed into the storage section as required.

[0079] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication component, and / or installed from a removable medium. When the computer program is executed by a central processing unit (CPU), it performs various functions defined in the system of this application.

[0080] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0081] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0082] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0083] Another aspect of this application provides a computer-readable storage medium storing a computer program thereon, which, when executed by a computer's processor, causes the computer to perform the method as described above. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0084] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the methods provided in the various embodiments described above.

[0085] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A fault diagnosis method for wind turbines based on multi-source monitoring, characterized in that, include: The optimal sensor deployment scheme is determined by using a multi-group and double-iteration butterfly optimization algorithm. Based on the optimal sensor deployment scheme, multi-source fault response data corresponding to various fault types are collected. Based on multi-source fault response data corresponding to various fault types, a fault response knowledge graph and a fault diagnosis model are constructed. Based on the optimal sensor deployment scheme, real-time multi-source monitoring data of wind turbines are collected. Based on the fault response knowledge graph and real-time multi-source monitoring data of the wind turbine, the first fault diagnosis result is generated. A second fault diagnosis result is generated based on the fault diagnosis model and real-time multi-source monitoring data of the wind turbine. Based on the first fault diagnosis result and the second fault diagnosis result, the fault diagnosis result of the wind turbine is generated. Among them, the optimal sensor deployment scheme is determined by using a multi-group and double-iteration butterfly optimization algorithm, including: Multiple temperature sensing points, vibration sensing points, and current sensing points are installed on the wind turbine. Collect temperature data from multiple temperature sensing locations, vibration data from vibration sensing locations, and current data from current sensing locations corresponding to various fault types. For each temperature sensing location, a sensing matrix is ​​calculated based on the temperature data of the temperature sensing location corresponding to various fault types. For each vibration sensing location, the sensing matrix of the vibration sensing location is calculated based on the temperature data of the vibration sensing location corresponding to various fault types. For each current sensing location, the sensing matrix of the current sensing location is calculated based on the temperature data of the current sensing location corresponding to various fault types. The optimal sensor deployment scheme is determined by using a multi-group and double-iteration butterfly optimization algorithm based on the sensing matrices of temperature sensing location, vibration sensing location, and current sensing location. Based on multi-source fault response data corresponding to various fault types, a fault response knowledge graph is constructed, including: Based on multi-source fault response data corresponding to various fault types, determine the response matrix corresponding to each fault type; A fault response knowledge graph is constructed based on the response matrix corresponding to each fault type.

2. The fault diagnosis method for wind turbines based on multi-source monitoring according to claim 1, characterized in that, Using a multi-group and double-iteration butterfly optimization algorithm, the optimal sensor deployment scheme is determined based on the sensing matrices of temperature sensing location, vibration sensing location, and current sensing location, including: S1. Based on the sensing matrices of the temperature sensing location, vibration sensing location, and current sensing location, calculate the sensing difference values ​​of the temperature sensing location, vibration sensing location, and current sensing location, respectively. S2. Based on the sensing matrices of the temperature sensing location, vibration sensing location, and current sensing location, calculate the sensing coordination coefficient between any temperature sensing location and any vibration sensing location, the sensing coordination coefficient between any temperature sensing location and any current sensing location, and the sensing coordination coefficient between any current sensing location and any vibration sensing location. S3. Based on the perception difference values ​​of temperature sensing position, vibration sensing position and current sensing position, the perception coordination coefficient of any temperature sensing position and any vibration sensing position, the perception coordination coefficient of any temperature sensing position and any current sensing position and the perception coordination coefficient of any current sensing position and any vibration sensing position, initialize multiple populations, wherein each butterfly in the population represents at least one combination of temperature sensing position, vibration sensing position and current sensing position. S4. Construct the fitness function; S5. For each population, iteratively optimize the population according to the fitness function until the population iteration termination condition is met. S6. Determine whether the global iteration termination condition is met. If yes, determine the optimal sensor deployment scheme. If not, execute S7. S7. Perform adaptive updates for multiple populations, then execute S5.

3. The fault diagnosis method for wind turbines based on multi-source monitoring according to claim 2, characterized in that, Initialize multiple populations, including: S31. For each population of butterflies, the number of temperature-sensing locations, vibration-sensing locations, and current-sensing locations included in the random sampling of butterflies. S32. For each population of butterflies, extract temperature sensing locations from multiple temperature sensing locations based on the number of temperature sensing locations included in the butterflies and the sensing difference value of each temperature sensing location. S33. For each population of butterflies, based on the number of vibration sensing locations included in the butterfly, the sensing difference value of each vibration sensing location, and the sensing synergy coefficient between any vibration sensing location and any extracted temperature sensing location, vibration sensing locations are extracted from multiple vibration sensing locations. S34. For each population of butterflies, based on the number of current sensing locations included in the butterfly population, the sensing difference value of each current sensing location, the sensing coordination coefficient between any current sensing location and any extracted temperature sensing location, and the sensing coordination coefficient between any current sensing location and any extracted vibration sensing location, extract current sensing locations from multiple current sensing locations.

4. The fault diagnosis method for wind turbines based on multi-source monitoring according to claim 2, characterized in that, Based on the fitness function, the population is iteratively optimized, including: Based on the fitness function, calculate the fitness value of each butterfly in the population, and determine the current global best butterfly, the butterfly performing the global search, and the butterfly performing the local search. For butterflies performing a global search, perform a position replacement operation based on the current globally optimal butterfly; For a butterfly performing a local search, identify the butterfly used for flight guidance and perform a position replacement operation based on the butterfly used for flight guidance.

5. The fault diagnosis method for wind turbines based on multi-source monitoring according to claim 2, characterized in that, Adaptive updates are performed on multiple populations, including: For each population, determine the response matrix for each fault type corresponding to the current globally optimal butterfly of the population; For any two populations, calculate the substitution coefficients for the two populations based on the response matrix of the current globally optimal butterfly for each fault type. Based on the replacement coefficients of any two populations, butterflies are replaced in multiple populations.

6. The fault diagnosis method for wind turbines based on multi-source monitoring according to claim 1, characterized in that, Based on the fault response knowledge graph and real-time multi-source monitoring data of the wind turbine, a first fault diagnosis result is generated, including: Based on real-time multi-source monitoring data of wind turbines, the current response matrix is ​​determined; Based on the fault response knowledge graph and the current response matrix, the fault matching degree of each fault type is determined, wherein the first fault diagnosis result includes the fault matching degree of each fault type.

7. The fault diagnosis method for wind turbines based on multi-source monitoring according to claim 6, characterized in that, The second fault diagnosis result includes the confidence level for each fault type; Based on the first fault diagnosis result and the second fault diagnosis result, the fault diagnosis result of the wind turbine is generated, including: For each fault type, the comprehensive probability of occurrence of the fault type is calculated based on the fault matching degree and confidence degree of the fault type, wherein the fault diagnosis result of the wind turbine includes the comprehensive probability of occurrence of each fault type.

8. A fault diagnosis system for wind turbines based on multi-source monitoring, characterized in that, The fault diagnosis method for wind turbines based on multi-source monitoring according to any one of claims 1-7 includes: The sensor deployment module is used to determine the optimal sensor deployment scheme through a multi-group and double-iteration butterfly optimization algorithm. The data acquisition module is used to collect multi-source fault response data corresponding to various fault types based on the optimal sensor deployment scheme. The fault diagnosis module is used to build a fault response knowledge graph and a fault diagnosis model based on multi-source fault response data corresponding to various fault types. The data acquisition module is also used to collect real-time multi-source monitoring data of the wind turbine based on the optimal sensor deployment scheme. The fault diagnosis module is also used to generate a first fault diagnosis result based on the fault response knowledge graph and real-time multi-source monitoring data of the wind turbine. The fault diagnosis module is also used to generate a second fault diagnosis result based on the fault diagnosis model and real-time multi-source monitoring data of the wind turbine. The fault diagnosis module is also used to generate fault diagnosis results for the wind turbine based on the first fault diagnosis results and the second fault diagnosis results.

Images