Wind turbine generator health assessment method and system based on AI multi-source data fusion

Abstract

The invention relates to the technical field of data analysis, and provides a wind turbine generator health assessment method and system based on AI multi-source data fusion, which are used for realizing accurate and dynamic wind turbine generator health assessment, and the method comprises the following steps: collecting a first operation data set of a wind turbine generator through a set sensor network, carrying out multi-dimensional state feature mining to obtain a multi-dimensional state feature set; combining the feature set, utilizing a pre-constructed state deduction model to carry out state deduction iteration, introducing different initial conditions and parameter perturbation to carry out multi-path deduction, and generating a plurality of state deduction iteration results; carrying out fusion processing based on conflict resolution on a state deduction iteration result, and obtaining a state deduction fusion result in combination with the feature relevance weight matrix and a credibility evaluation mechanism; and finally, performing state evaluation based on synchronous time sequence analysis on a state deduction fusion result and a second operation data set of which the acquisition moment is later than that of the first operation data set to obtain a health state evaluation label of the wind turbine generator.

Description

TECHNICAL FIELD

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

[0002] In the field of wind turbine health assessment, early existing technologies only rely on a single type of sensor data, such as collecting only vibration data of a wind turbine to assess its health status. This way, the data obtained is limited and it is difficult to fully reflect the actual operation of the wind turbine. Later, multi-sensor data collection was adopted, but only simple data summary and analysis were performed without in-depth mining of the dynamics characteristics and potential abnormal patterns behind the data. In terms of state deduction, traditional methods often perform single-path deduction based on fixed initial conditions and parameters, and cannot consider the uncertainty and diversity in the operation of the wind turbine, resulting in a large deviation between the deduction result and the actual situation. Moreover, the existing technologies lack effective conflict resolution mechanisms, and often adopt simple discard or average processing methods for data conflicts, which cannot achieve effective complementation of information and affect the accuracy and reliability of the fusion result. As can be seen, the existing technologies lack flexibility and have poor assessment accuracy when assessing the health of a wind turbine. SUMMARY

[0003] The application provides a wind turbine health assessment method and system based on AI multi-source data fusion to achieve accurate and dynamic wind turbine health assessment.

[0004] In a first aspect, the application provides a wind turbine health assessment method based on AI multi-source data fusion, applied to a wind turbine health assessment system, and the method comprises the following steps: A first set of operation data of a wind turbine is collected through a set of sensor networks; A multi-dimensional state feature set representing the dynamics characteristics and potential abnormal patterns of the wind turbine is obtained by performing multi-dimensional state feature mining on the first set of operation data; A state deduction iteration based on dynamics characteristics and operation rules of the wind turbine is performed by using a pre-constructed state deduction model in combination with the multi-dimensional state feature set, and different initial conditions and parameter disturbances are introduced in the state deduction iteration process to perform multi-path deduction, thereby generating a plurality of state deduction iteration results reflecting the evolution trend of different operation states; The multiple state deduction iteration results are fused based on conflict resolution, and combined with the established feature correlation weight matrix and the credibility evaluation mechanism to perform conflict coordination and information complementation, to generate a state deduction fusion result, and the state deduction fusion result and a second running data set are subjected to state evaluation based on synchronous time sequence analysis to obtain the health state evaluation label of the wind turbine.

[0005] In a second aspect, an embodiment of the present application provides a wind turbine health evaluation system, comprising a processor and a memory, wherein the memory stores a computer program, and when the computer program is executed by the processor, the computer program causes the processor to execute the steps of the above method.

[0006] In a third aspect, an embodiment of the present application provides a computer readable storage medium comprising a computer program, and when the computer program is executed on a wind turbine health evaluation system, the computer program is configured to cause the wind turbine health evaluation system to execute the steps of the above method.

[0007] The embodiment of the present application can deeply analyze the dynamic characteristics and potential abnormal patterns of the wind turbine by setting up a sensor network to collect a first running data set and performing multi-dimensional state feature mining on the first running data set, and can mine key information that is difficult to find by traditional methods, thereby improving the depth of understanding of the running state of the wind turbine.

[0008] The state deduction model is used to perform state deduction iteration, and different initial conditions and parameter perturbations are introduced to perform multi-path deduction, fully considering the uncertainty and diversity in the running process of the wind turbine, to generate multiple results reflecting different running state evolution trends, which can more comprehensively predict the future state of the wind turbine compared with single-path deduction.

[0009] The multiple state deduction iteration results are fused based on conflict resolution, combined with the feature correlation weight matrix and the credibility evaluation mechanism, to effectively solve the conflict problem in the multi-source data fusion process, realize conflict coordination and information complementation, and make the state deduction fusion result more accurate and reliable.

[0010] Finally, the state deduction fusion result and a second running data set are subjected to state evaluation based on synchronous time sequence analysis to obtain the health state evaluation label of the wind turbine, and this evaluation method combining data of different time periods can dynamically and accurately reflect the actual health status of the wind turbine, improve the accuracy and timeliness of health evaluation, and can timely find potential faults of the wind turbine, reduce maintenance costs, and improve power generation efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1A flowchart of a wind turbine health assessment method based on AI multi-source data fusion provided by an embodiment of the present application.

[0012] Figure 2 A structural diagram of a wind turbine health assessment system provided by an embodiment of the present application. DETAILED DESCRIPTION

[0013] To make the purposes, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments described in the present application document, all other embodiments obtained by those of ordinary skill in the art without creative labor fall within the scope of protection of the present application technical solutions.

[0014] Reference is made to Figure 1 which is a wind turbine health assessment method based on AI multi-source data fusion provided by an embodiment of the present application. The method can be applied to a wind turbine health assessment system, and the specific process is as steps 110-140.

[0015] Step 110: Collecting a first operating data set of the wind turbine through a set of sensor networks.

[0016] In the wind turbine health assessment scenario, in order to comprehensively obtain the operating information of the wind turbine, a sensor network is set at multiple key parts of the wind turbine. These sensors are of various types and functions, and are respectively responsible for collecting different types of data. For example, a rotational speed sensor and a torque sensor are installed on the wind wheel. The rotational speed sensor converts the rotational speed of the wind wheel into an electrical signal for recording through electromagnetic induction principle, and the torque sensor measures the torque received by the wind wheel using a strain gauge and other devices. A voltage sensor and a current sensor are set at the generator. The voltage sensor measures the output voltage of the generator based on electromagnetic induction or resistance voltage division principle, and the current sensor obtains current data through Hall effect and other methods. A temperature sensor and a vibration sensor are arranged at the gear box part. The temperature sensor uses a thermistor or a thermocouple and other elements to sense the temperature change of the gear box, and the vibration sensor detects the vibration of the gear box through piezoelectric effect and other methods.

[0017] The sensor network continuously collects data at a certain sampling frequency. The setting of the sampling frequency needs to consider factors such as the timeliness of the data and the storage cost. The collected data is transmitted to the data storage device in the form of digital signals to form a first operating data set. This data set contains a variety of operating parameter information of the wind turbine within a period of time.

[0018] Step 120: performing multi-dimensional state feature mining on the first running data set to obtain a multi-dimensional state feature set representing the dynamic characteristics and potential abnormal patterns of the wind turbine.

[0019] The first running data set contains a large amount of raw data. In order to extract information reflecting the dynamic characteristics and potential abnormal patterns of the wind turbine from the raw data, multi-dimensional state feature mining needs to be performed.

[0020] Step 121: performing time series decomposition processing on the first running data set to separate a multi-component time series set containing periodic fluctuation characteristics and non-periodic fluctuation characteristics.

[0021] When performing time series decomposition processing on the first running data set, a suitable time series decomposition algorithm is used. Taking the power data of the wind turbine as an example, the data changes over time in a certain regularity, but also contains some irregular fluctuations. Common time series decomposition methods include additive models and multiplicative models. The embodiments of the present application can select a suitable model according to the characteristics of the data.

[0022] In the additive model, the time series is decomposed into a trend component, a seasonal component, and a residual component. The trend component reflects the long-term trend of the data, the seasonal component reflects the periodic fluctuation of the data, and the residual component contains non-periodic fluctuation characteristics. By decomposing the power data, periodic fluctuation characteristics such as daily or weekly periodic changes and non-periodic fluctuation characteristics such as sudden changes in weather conditions or temporary equipment failures are obtained, from which a multi-component time series set containing periodic fluctuation characteristics and non-periodic fluctuation characteristics is obtained.

[0023] Step 122: performing dynamic characteristic extraction processing on each component time series in the multi-component time series set to generate vibration mode features reflecting mechanical vibration characteristics and energy transfer features reflecting energy conversion efficiency.

[0024] Dynamic characteristic extraction is performed on each component time series in the multi-component time series set. For the vibration data component of the wind turbine, signal processing methods are used to extract vibration mode features. First, the vibration data is filtered to remove noise interference, and then the time domain signal is converted to the frequency domain signal through Fourier transform and other methods to extract the frequency, amplitude, and other features of the vibration. These features can reflect the mechanical vibration characteristics of the wind turbine, for example, different frequency vibrations may correspond to different component failures.

[0025] For energy-related data components such as power and wind speed data, by analyzing the relationship between them, the energy transfer characteristics reflecting the energy conversion efficiency are extracted. For example, the power output at different wind speeds is calculated to obtain the efficiency coefficient of energy conversion. Through statistical analysis of the efficiency coefficient at multiple time points, the distribution and variation law of the energy transfer characteristics are obtained.

[0026] Step 123: Correlation analysis and processing of the vibration mode characteristics and the energy transfer characteristics, identifying the coordinated variation law between the vibration mode characteristics and the energy transfer characteristics, and generating a feature correlation graph.

[0027] When performing correlation analysis on the extracted vibration mode characteristics and energy transfer characteristics, statistical analysis and machine learning algorithms are used. First, the correlation coefficient between the vibration mode characteristics and the energy transfer characteristics is calculated to evaluate their linear relationship. Then, machine learning algorithms such as decision trees, neural networks, etc. are used to mine their non-linear relationships.

[0028] Through analysis, it is found that in some cases, the increase of vibration frequency is accompanied by the decrease of energy conversion efficiency, which may indicate that the wind turbine has mechanical failure, affecting the normal conversion of energy. According to these coordinated variation laws, a feature correlation graph is generated. In the graph, nodes represent features, edges represent the correlation between features, and the weight of edges represents the strength of correlation. Through the feature correlation graph, the relationship between the vibration mode characteristics and the energy transfer characteristics can be intuitively displayed.

[0029] Step 124: Abnormal sensitive feature screening processing of the vibration mode characteristics and the energy transfer characteristics based on the feature correlation graph, extracting a key feature subset sensitive to potential abnormal patterns.

[0030] Based on the generated feature correlation graph, the vibration mode characteristics and the energy transfer characteristics are subjected to abnormal sensitive feature screening. In the feature correlation graph, the correlation strength and variation trend between each feature are analyzed. Those features that are closely related to potential abnormal patterns and are relatively sensitive to changes are given special attention.

[0031] For example, through analysis of historical data, it is found that when an abnormal change occurs in a specific frequency vibration mode, the energy conversion efficiency will also decrease significantly, and this situation is often accompanied by failure of the wind turbine. Therefore, the vibration mode characteristics of the specific frequency and the corresponding energy conversion efficiency characteristics are selected as key features sensitive to potential abnormal patterns. Through screening algorithms such as threshold-based screening methods, these key features are extracted to form a key feature subset.

[0032] Step 125: Dimensional reorganization processing is performed on the key feature subset to generate a multi-dimensional state feature set containing time dimension features, frequency dimension features, and amplitude dimension features.

[0033] The key feature subset is subjected to dimensional reorganization processing to generate a multi-dimensional state feature set containing time dimension features, frequency dimension features, and amplitude dimension features. First, each feature in the key feature subset is subjected to time dimension analysis to extract the variation law of the feature over time, such as the rate of change, periodicity, etc., to form time dimension features.

[0034] Then, the features are subjected to frequency dimension analysis, and time domain features are converted into frequency domain features through Fourier transform or other methods to extract the frequency components of the features and form frequency dimension features. Finally, the amplitude information of the features, such as the amplitude of vibration and the numerical value of energy conversion efficiency, is extracted to form amplitude dimension features.

[0035] The time dimension features, frequency dimension features, and amplitude dimension features are combined to obtain a multi-dimensional state feature set, which can more comprehensively and accurately represent the dynamic characteristics and potential abnormal patterns of the wind turbine.

[0036] In the exemplary application process, in step 121, an additive model is used when performing time series decomposition processing. For the power data of the wind turbine, the parameter configuration of the decomposition is as follows: the trend component is calculated using the moving average method, and the moving window size is set to 7 days, i.e., a trend value is calculated every 7 days; the seasonal component is determined by calculating the average value of the same time period each year, for example, the average power of the same month each year is taken as the seasonal component of that month; and the residual component is obtained by subtracting the trend component and the seasonal component from the original data. The above parameter configuration can better separate the periodic and aperiodic fluctuation characteristics in the power data.

[0037] In step 122, the Fourier transform algorithm is used when extracting vibration pattern features. For the vibration data of the wind turbine, the sampling frequency is set to 100 Hz, i.e., 100 data points are collected per second. When performing Fourier transform, the data length is selected to be 1024 points, so that a relatively accurate frequency resolution can be obtained in the frequency domain. In this way, the frequency and amplitude of the vibration can be accurately extracted.

[0038] In step 123, the decision tree algorithm is used when performing correlation analysis. The maximum depth of the decision tree is set to 5, i.e., the number of layers of the tree is at most 5; the minimum sample splitting number is set to 10, meaning that each internal node needs at least 10 samples to continue splitting; and the minimum sample leaf node number is set to 5, i.e., each leaf node needs at least 5 samples. This can ensure that the model complexity is moderate while improving the accuracy of the correlation analysis.

[0039] Step 130: Based on the multi-dimensional state feature set, use the pre-constructed state inference model to perform state inference iteration on the wind turbine based on its dynamic characteristics and operating rules, and introduce different initial conditions and parameter perturbations during the state inference iteration process to perform multi-path inference, generating multiple state inference iteration results reflecting the evolution trend of different operating states.

[0040] Based on the obtained multi-dimensional state feature set, use the pre-constructed state inference model to perform state inference iteration on the wind turbine, which is based on the dynamic characteristics and operating rules of the wind turbine and can simulate the evolution of the wind turbine's operating state under different conditions.

[0041] Step 131: Based on the historical operating state records of the wind turbine, standardize the multi-dimensional state feature set to generate a standardized multi-dimensional state feature set; input the standardized multi-dimensional state feature set into the initial state configuration layer of the state inference model, and generate multiple differentiated initial state vectors according to the historical operating state records of the wind turbine.

[0042] According to the historical operating state records of the wind turbine, the multi-dimensional state feature set is standardized. Since the value range and dimension of different features may be different, standardization can make these features comparable. Exemplary standardization methods include Z-score standardization and Min-Max standardization.

[0043] Taking Z-score standardization as an example, the mean and standard deviation of each feature are calculated, and the feature value is subtracted from the mean and divided by the standard deviation to obtain the standardized feature value. In this way, the multi-dimensional state feature set is converted into a standardized multi-dimensional state feature set.

[0044] The standardized multi-dimensional state feature set is input into the initial state configuration layer of the state inference model. In this layer, multiple differentiated initial state vectors are generated according to the historical operating state records of the wind turbine. Through analysis of historical data, different initial conditions such as different wind speeds and different rotor speeds are determined, and combined with the standardized multi-dimensional state feature set, multiple vectors representing different initial states are generated. These initial state vectors provide different starting points for multi-path inference.

[0045] Step 132: Perform parameter perturbation processing on each initial state vector, randomly adjust the dynamic characteristic parameters within a pre-set perturbation range, and generate a set of perturbed state vectors containing different parameter combinations.

[0046] Parameter perturbation processing is performed on each initial state vector to simulate various uncertain factors that the wind turbine may encounter during operation. Within a pre-set perturbation range, dynamic characteristic parameters such as wind speed, wind turbine blade angle, and gearbox transmission efficiency are randomly adjusted.

[0047] For example, within a preset disturbance range of wind speed, a certain wind speed value is randomly increased or decreased, and at the same time, the angle of the wind wheel blade is slightly adjusted. By randomly combining and adjusting multiple dynamic characteristic parameters, a set of disturbance state vectors containing different parameter combinations is generated, which provides more possibilities for multi-path deduction and makes the deduction results more reflect the various situations of the wind turbine in actual operation.

[0048] Step 133: input the set of disturbance state vectors into the deduction iteration layer of the state deduction model, and perform multi-step state deduction calculation based on the wind turbine dynamics equation to generate a state evolution path sequence corresponding to each disturbance state vector.

[0049] The set of disturbance state vectors is input into the deduction iteration layer of the state deduction model. In this layer, multi-step state deduction calculation is performed based on the wind turbine dynamics equation.

[0050] Step 1331: convert each disturbance state vector in the set of disturbance state vectors into a standard state vector that matches the input format of the state deduction model.

[0051] Different models may have specific requirements for the format of input data, so each disturbance state vector needs to be converted into a standard state vector that matches the input format of the state deduction model. First, perform dimension checking on the disturbance state vector to ensure that its dimensions are consistent with the model input requirements. If the dimensions are not consistent, adjust them through interpolation or dimension reduction, etc.

[0052] Then, adjust the order of the elements in the disturbance state vector to meet the order requirements of the model input. For example, the model requires the elements of the input vector to be arranged in the order of wind speed, wind wheel speed, and power, while the order of the elements in the disturbance state vector may be different and needs to be adjusted accordingly. Through these processes, the disturbance state vector is converted into a standard state vector.

[0053] Step 1332: call the dynamics calculation core of the state deduction model, and perform single-step state transition calculation on the standard state vector based on the mechanical transmission system characteristics and aerodynamic characteristics equations of the wind turbine to obtain the state prediction vector at the next time.

[0054] The dynamics calculation core of the state deduction model is called, which contains the mechanical transmission system characteristics and aerodynamic characteristics equations of the wind turbine. The standard state vector is input, and single-step state transition calculation is performed according to these equations.

[0055] For example, according to the current speed of the wind wheel, the blade angle and other parameters, combined with the aerodynamic principle and the mechanical transmission relationship, the speed, torque and other state parameters of the wind wheel at the next moment are calculated. In the calculation process, the inertia, friction and other factors of the wind turbine are considered to ensure the accuracy of the calculation results. Through single-step state transition calculation, the state prediction vector at the next moment is obtained.

[0056] Step 1333: Extract the key state parameters in the state prediction vector, normalize and compare the similarity of the same parameters in the historical running state database, and generate a parameter similarity score.

[0057] The key state parameters such as power and speed are extracted from the state prediction vector. These key state parameters are normalized with the same parameters in the historical running state database to make them comparable. The normalization process can use the same standardization method as in step 131.

[0058] Then, through the similarity comparison algorithm, the similarity between the predicted parameters and the historical parameters is calculated, and the parameter similarity score is generated. Common similarity comparison algorithms include Euclidean distance, cosine similarity, etc. Taking Euclidean distance as an example, the distance between the predicted parameters and the historical parameters is calculated. The smaller the distance, the higher the similarity. Convert the distance value to a similarity score. The higher the score, the more similar the predicted state is to the historical state.

[0059] Step 1334: According to the parameter similarity score, the state prediction vector is weighted and processed to generate a weighted state prediction vector.

[0060] According to the generated parameter similarity score, the state prediction vector is weighted and processed. If the parameter similarity score is high, it means that the predicted state is similar to the historical state, and the credibility of the prediction vector is high, giving it a high weight; on the contrary, if the similarity score is low, it gives a lower weight.

[0061] For example, the parameter similarity score is used as a weight coefficient, multiplied by each element in the state prediction vector to obtain a weighted state prediction vector. In this way, the state prediction vector is adjusted to better reflect the actual situation.

[0062] Step 1335: The weighted state prediction vector is used as the new input state vector, and the single-step state transition calculation, similarity comparison and credibility weighting process are repeated until the preset number of steps is completed. The state evolution path sequence containing the time stamp is generated.

[0063] The weighted state prediction vector is taken as the new input state vector, and the above-mentioned single-step state transition calculation, similarity comparison and credibility weighting process are repeated. The calculation result of each step is marked with a time stamp, recording the occurrence time of each state.

[0064] For example, after the state prediction vector is calculated in the first step, similarity comparison and credibility weighting process are performed to obtain the weighted state prediction vector. This vector is taken as the new input to perform the second step of single-step state transition calculation, and so on. This process is repeated until the preset number of steps of deduction is completed, and finally a state evolution path sequence containing time stamp markers is generated, which records in detail the evolution of the wind turbine's running state at different times.

[0065] Step 134: In the generation process of the state evolution path sequence, the deviation degree of the deduced state and the historical running state is monitored in real time, and when the deviation degree exceeds the preset deviation degree, the path correction mechanism is triggered to dynamically adjust the state transition matrix of the subsequent deduction step.

[0066] In the generation process of the state evolution path sequence, the deviation degree of the deduced state and the historical running state needs to be monitored in real time to ensure the accuracy of the deduction result.

[0067] Step 1341: Extract the deduced state vector of the current time from the state evolution path sequence, and extract the actual running state vector of the corresponding time from the historical running state database.

[0068] The deduced state vector of the current time is extracted from the state evolution path sequence. Since the state evolution path sequence contains state information at each time, the deduced state vector of the current time can be accurately extracted through the time stamp marker.

[0069] At the same time, the actual running state vector of the corresponding time is extracted from the historical running state database. The historical running state database records the actual running state of the wind turbine at different times in the past, and according to the time stamp, the actual running state vector corresponding to the current time can be found for deviation degree calculation.

[0070] Step 1342: Normalize the deduced state vector and the actual running state vector, calculate the cosine similarity or correlation coefficient between the normalized vectors, and convert the cosine similarity or correlation coefficient into a difference degree index as a deviation quantification index.

[0071] The deduced state vector and the actual running state vector are normalized to make them comparable. The normalization process can use the same standardization method as in step 131.

[0072] Then, the cosine similarity or correlation coefficient between the normalized vectors is calculated. The cosine similarity measures the cosine value of the included angle between two vectors, and the closer the value is to 1, the more similar the two vectors are; the correlation coefficient reflects the strength of the linear relationship between the two vectors. The calculated cosine similarity or correlation coefficient is converted into a deviation index, for example, 1 is subtracted from the cosine similarity or correlation coefficient, and the result is taken as the deviation quantification index, which is used to measure the deviation degree between the inferred state and the actual state.

[0073] Step 1343: Compare the deviation quantification index with a preset deviation threshold value, and when the deviation quantification index is greater than the deviation threshold value, start the path correction mechanism.

[0074] The calculated deviation quantification index is compared with a preset deviation threshold value. The preset deviation threshold value is determined according to the historical data and experience of the wind turbine, and is used to judge whether the deviation between the inferred state and the actual state is within an acceptable range.

[0075] If the deviation quantification index is greater than the deviation threshold value, it means that the deviation between the inferred state and the actual state is large, and the inferred path needs to be corrected. At this time, the path correction mechanism is started to ensure that the inferred result is more accurate.

[0076] Step 1344: Call the state transition matrix adjustment module in the path correction mechanism, calculate the matrix correction coefficient based on the size of the deviation quantification index, and the matrix correction coefficient is positively correlated with the deviation quantification index.

[0077] Call the state transition matrix adjustment module in the path correction mechanism, which calculates the matrix correction coefficient according to the size of the deviation quantification index. Since the matrix correction coefficient is positively correlated with the deviation quantification index, the larger the deviation, the larger the matrix correction coefficient.

[0078] For example, through a mapping function, the deviation quantification index is mapped to the value range of the matrix correction coefficient. When the deviation quantification index increases, the matrix correction coefficient also increases accordingly, so that the state transition matrix can be dynamically adjusted according to the size of the deviation.

[0079] Step 1345: Use the matrix correction coefficient to perform element-level weighted adjustment on the state transition matrix in the state inference model to generate a corrected state transition matrix.

[0080] The calculated matrix correction coefficient is used to perform element-level weighted adjustment on the state transition matrix in the state inference model. Specifically, the matrix correction coefficient is multiplied by each element of the state transition matrix to obtain a corrected state transition matrix.

[0081] For example, each element in the state transition matrix represents the probability of transition from one state to another, and the probabilities are adjusted by the matrix correction coefficient. If the bias degree is large, the matrix correction coefficient is large, which will make a large adjustment to the elements of the state transition matrix, change the probability of state transition, and thus change the deduction path.

[0082] Step 1346: Continue to perform subsequent state deduction calculation steps using the corrected state transition matrix until the bias quantification index is less than or equal to the bias threshold or the entire deduction step is completed.

[0083] Continue to perform subsequent state deduction calculation steps using the corrected state transition matrix. During the deduction process, the bias quantification index is constantly monitored.

[0084] When the bias quantification index is less than or equal to the bias threshold, it means that the bias between the deduced state and the actual state is within an acceptable range, and the path correction is stopped; or when the entire deduction step is completed, the entire deduction process is ended. In this way, the accuracy and reliability of the deduction result are guaranteed.

[0085] Step 135: Repeat the state deduction calculation and path correction mechanism until the deduction process of the preset time window is completed, and output multiple state deduction iteration results containing multiple independent evolution paths.

[0086] Repeat the above state deduction calculation and path correction mechanism to constantly update the state prediction vector and the state transition matrix. In the preset time window, the deduction is continuously performed.

[0087] Due to the introduction of different initial conditions and parameter perturbations, multiple state deduction iteration results containing multiple independent evolution paths are finally obtained, which reflect the running state evolution trend of the wind turbine under different conditions and provide rich information for state assessment.

[0088] In the above embodiment, for some non-limiting application examples, in step 131, the Z-score standardization method is used for standardization of the multi-dimensional state feature set. For the power and speed of the wind turbine, the mean and standard deviation are calculated. Taking the power feature as an example, the mean of the power data in a period of time is 500 kW, and the standard deviation is 50 kW. In standardization, each power data point is subtracted from the mean 500 kW and divided by the standard deviation 50 kW to obtain the standardized power value.

[0089] In step 1332, the dynamics calculation core of the state deduction model uses a calculation method based on physical equations. For the mechanical transmission system of the wind turbine, the moment of inertia of the wind wheel is considered to be 10000 kg·m 2The transmission efficiency of the gearbox is 0.95. When performing single-step state transition calculation, the rotational speed and torque of the wind wheel at the next moment and other state parameters are accurately calculated according to the above physical parameters and aerodynamic characteristic equations.

[0090] In step 1333, the cosine similarity algorithm is used when performing similarity comparison. For the state prediction vector and the vector in the historical operation state database, the similarity score is obtained by calculating the cosine value of the two vectors. The cosine similarity value ranges from -1 to 1, and the closer the value is to 1, the higher the similarity. In step 1342, the cosine similarity is converted into a difference index when calculating the deviation metric.

[0091] Step 140: Perform conflict resolution-based fusion processing on the plurality of state deduction iteration results, and combine the established feature correlation weight matrix and the credibility evaluation mechanism to perform conflict coordination and information complementation, generate a state deduction fusion result, and perform state evaluation based on synchronous time sequence analysis on the state deduction fusion result and a second operation data set to obtain a health state evaluation label of the wind turbine; the collection time of the second operation data set is later than the collection time of the first operation data set.

[0092] The plurality of state deduction iteration results are fused to obtain more accurate state evaluation results. At the same time, the second operation data set is combined to perform synchronous time sequence analysis to determine the health state evaluation label of the wind turbine.

[0093] Step 141: Perform time axis alignment processing on the plurality of state deduction iteration results to keep all state deduction iteration results synchronized in the time dimension; extract the key state feature sequence in each state deduction iteration result, and calculate the feature similarity matrix between different state deduction iteration results.

[0094] The plurality of state deduction iteration results are time axis aligned. Since different deduction results may have different time starting points and sampling frequencies, time axis alignment is needed to keep all results synchronized in the time dimension.

[0095] First, determine the maximum time range of all deduction results, and then perform interpolation processing on each deduction result based on the smallest sampling interval to have corresponding state values at the same time points.

[0096] Next, extract the key state feature sequence in each state deduction iteration result, such as power, rotational speed, etc. Calculate the feature similarity matrix between different state deduction iteration results. Through similarity calculation methods such as Euclidean distance and cosine similarity, the similarity between each two key state feature sequences is calculated, and the similarity values are combined into a matrix, which reflects the similarity between different deduction results.

[0097] Step 142: identifying feature sequence pairs with conflicts based on the feature similarity matrix, classifying the conflict types of the feature sequence pairs with conflicts, and determining the feature dimensions and time nodes where conflicts occur.

[0098] Step 1421: traversing the feature similarity matrix according to a preset feature similarity threshold, and marking feature sequence pairs with similarity values lower than the feature similarity threshold as potential conflict feature sequence pairs.

[0099] Traverse the feature similarity matrix according to the preset feature similarity threshold. The preset feature similarity threshold is determined based on experience and historical data to determine whether two feature sequences have conflicts.

[0100] When the similarity value is lower than the threshold, mark the corresponding feature sequence pair as a potential conflict feature sequence pair. For example, for the power feature sequence, if the power feature sequence similarity in the two deduction results is lower than the threshold, mark the two sequences as potential conflict feature sequence pairs.

[0101] Step 1422: performing time series segmentation processing on the potential conflict feature sequence pairs, and dividing each feature sequence into multiple equal-length time window segments.

[0102] Perform time series segmentation processing on the potential conflict feature sequence pairs. Divide each feature sequence into multiple equal-length time window segments according to a certain time length.

[0103] For example, divide the power feature sequence into multiple time window segments according to an hour as a time window. In this way, the distribution of conflicts at different time points can be analyzed in more detail.

[0104] Step 1423: performing normalization processing on the feature sequence pairs in each time window segment, calculating the dynamic time warping distance of the normalized sequence pairs, and marking the time window segments with dynamic time warping distances greater than a preset distance threshold as conflict time windows.

[0105] Perform normalization processing on the feature sequence pairs in each time window segment to make them comparable. The normalization processing can use the same standardization method as in step 131.

[0106] Then, calculate the dynamic time warping distance of the normalized sequence pairs. Dynamic time warping distance is a method for measuring the similarity of two time series, which can handle the problem of different sequence lengths and time shifts. Mark the time window segments with dynamic time warping distances greater than a preset distance threshold as conflict time windows. The preset distance threshold is determined based on experience and historical data to determine whether two time series have conflicts in the time window.

[0107] Step 1424: Extract the start timestamp and end timestamp of the conflict time window in the feature sequence, determine the target time node where the conflict occurs.

[0108] Extract the start timestamp and end timestamp from the segment marked as the conflict time window, which determines the target time node where the conflict occurs.

[0109] For example, if the start time of a conflict time window is 9 am and the end time is 10 am, these two time points are the target time nodes where the conflict occurs. By determining the target time node, we can further analyze the causes and effects of the conflict.

[0110] Step 1425: Perform feature dimension decomposition on the feature sequence pair in the conflict time window, identify the target feature dimension with a difference jump, and mark the target feature dimension as a conflict feature dimension.

[0111] Perform feature dimension decomposition on the feature sequence pair in the conflict time window. Each feature sequence may contain multiple feature dimensions, such as power, speed, temperature, etc.

[0112] By analyzing the changes of the feature sequence pair in different feature dimensions, identify the target feature dimension with a difference jump. For example, in one of the conflict time windows, the power feature sequence has a significant difference jump, while the speed and temperature feature sequences change less, so the power feature dimension is the conflict feature dimension. Mark these target feature dimensions as conflict feature dimensions.

[0113] Step 1426: According to the number and type of conflict feature dimensions and the distribution characteristics of conflict time windows, classify the conflict feature sequence pair into conflict types, and generate a conflict analysis report containing conflict type identifiers, conflict feature dimension sets, and conflict time node lists.

[0114] According to the number and type of conflict feature dimensions and the distribution characteristics of conflict time windows, classify the conflict feature sequence pair into conflict types.

[0115] For example, if there is only one conflict feature dimension and the conflict time windows are concentrated in one time period, it can be classified as a single-dimensional local conflict; if there are multiple conflict feature dimensions and the conflict time windows are distributed relatively dispersed, it can be classified as a multi-dimensional global conflict.

[0116] Generate a conflict analysis report containing conflict type identifiers, conflict feature dimension sets, and conflict time node lists, which records the type of conflict, the involved feature dimensions, and the time nodes where the conflict occurs.

[0117] Step 143: Call the established feature correlation weight matrix, extract the correlation weight value of the conflict feature dimension in the historical data, and calculate the credibility score of each conflict feature sequence combined with the credibility evaluation mechanism.

[0118] Step 1431: Hierarchical deconstruction of the feature correlation weight matrix, locate the coordinate index of the conflict feature dimension in the matrix, and extract the corresponding row vector as the initial correlation weight vector.

[0119] The hierarchical deconstruction of the feature correlation weight matrix is performed. The feature correlation weight matrix is a two-dimensional matrix, where each element represents the correlation weight between two feature dimensions. By locating the coordinate index of the conflict feature dimension in the matrix, the corresponding row vector is extracted as the initial correlation weight vector. For example, if the conflict feature dimension is power, find the row where power is located in the matrix, and extract the vector of this row as the initial correlation weight vector, which reflects the correlation degree between the power feature dimension and other feature dimensions.

[0120] Step 1432: Correlation mapping between the initial correlation weight vector and the historical performance characteristics of the conflict feature dimension in the historical operation database, generating a weight dynamic adjustment coefficient, the historical performance characteristics including historical conflict occurrence rate and historical fusion contribution degree.

[0121] The initial correlation weight vector is associated with the historical performance characteristics of the conflict feature dimension in the historical operation database. The historical performance characteristics include historical conflict occurrence rate and historical fusion contribution degree. Historical conflict occurrence rate reflects the frequency of conflict of this conflict feature dimension in the past, and historical fusion contribution degree reflects the importance of this feature dimension in the fusion process. Through correlation mapping processing, the initial correlation weight vector is comprehensively considered with these historical performance characteristics, and a weight dynamic adjustment coefficient is generated. For example, if the historical conflict occurrence rate of one of the conflict feature dimensions is high, and the historical fusion contribution degree is low, the weight dynamic adjustment coefficient will correspondingly reduce the weight of this feature dimension.

[0122] Step 1433: Multi-dimensional evidence collection processing of conflict feature sequence by multi-source evidence fusion module of credibility evaluation mechanism, the multi-dimensional evidence including internal consistency evidence of feature sequence, external correlation evidence with other non-conflict feature sequences and reference evidence under similar historical scenarios.

[0123] Multi-dimensional evidence collection processing of conflict feature sequence by multi-source evidence fusion module of credibility evaluation mechanism. Multi-dimensional evidence includes internal consistency evidence of feature sequence, external correlation evidence with other non-conflict feature sequences and reference evidence under similar historical scenarios.

[0124] The internal consistency evidence of the feature sequence can be obtained by analyzing the fluctuation, trend change, etc. of the sequence, reflecting the stability and reliability of the feature sequence itself. The external correlation evidence of other non-conflict feature sequences can be obtained by calculating the correlation between them, reflecting the cooperative relationship of the feature sequence with other feature sequences. The reference evidence in the historical similar scene can be obtained from the historical operation database, by searching for a historical scene similar to the current conflict feature sequence, analyzing its processing result and influence, and providing a reference for the credibility evaluation of the current conflict feature sequence.

[0125] Step 1434: Perform evidence strength quantification processing on the collected multi-dimensional evidence to generate an evidence strength value and an evidence reliability factor corresponding to each evidence.

[0126] The collected multi-dimensional evidence is subjected to evidence strength quantification processing. By setting different quantification standards, the strength of each evidence is quantified to generate an evidence strength value. At the same time, considering the source and reliability of the evidence, an evidence reliability factor is assigned to each evidence. For example, for evidence from a high-precision sensor, the reliability factor is higher; for evidence from experience judgment, the reliability factor is lower.

[0127] Step 1435: Perform weighted correction processing on the initial correlation weight vector based on the weight dynamic adjustment coefficient to generate a corrected correlation weight vector.

[0128] The initial correlation weight vector is subjected to weighted correction processing based on the weight dynamic adjustment coefficient. The weight dynamic adjustment coefficient is multiplied by each element of the initial correlation weight vector to obtain the corrected correlation weight vector. For example, if the weight dynamic adjustment coefficient is 0.8 and the initial correlation weight vector is [0.2, 0.3, 0.5], then the corrected correlation weight vector is [0.16, 0.24, 0.4]. In this way, the correlation weight of the conflict feature dimension is adjusted according to its historical performance.

[0129] Step 1436: Normalize the corrected correlation weight vector and the evidence strength value of the multi-dimensional evidence, and then perform weighted fusion operation to obtain a preliminary credibility score matrix.

[0130] The corrected correlation weight vector and the evidence strength value of the multi-dimensional evidence are subjected to normalization processing to make them comparable. The normalization processing can adopt the same standardization method as in step 131. Then, weighted fusion operation is performed. The corrected correlation weight vector and the normalized evidence strength value are weighted and multiplied, and then summed to obtain a preliminary credibility score matrix, which reflects the preliminary credibility score of each conflict feature sequence under different evidence.

[0131] Step 1437: Element-wise weighted average processing of the preliminary credibility score matrix according to the evidence reliability factor to generate the credibility sub-score of each conflict feature sequence in different conflict time windows.

[0132] Element-wise weighted average processing of the preliminary credibility score matrix according to the evidence reliability factor. Multiply each element in the preliminary credibility score matrix by the corresponding evidence reliability factor, then sum and divide by the total number of evidences to obtain the credibility sub-score of each conflict feature sequence in different conflict time windows.

[0133] For example, for one conflict feature sequence in one conflict time window, there are three evidences with preliminary credibility scores of 0.6, 0.7, and 0.8, and corresponding evidence reliability factors of 0.8, 0.9, and 0.7. The credibility sub-score of the conflict feature sequence in the conflict time window is (0.6x0.8+0.7x0.9+0.8x0.7) / 3.

[0134] Step 1438: Time series accumulation processing of the credibility sub-scores of all conflict time windows to generate the credibility score covering the entire conflict feature sequence.

[0135] Time series accumulation processing of the credibility sub-scores of all conflict time windows. Accumulate the credibility sub-scores of each conflict time window in time order to obtain the credibility score covering the entire conflict feature sequence.

[0136] For example, for a conflict feature sequence, there are three conflict time windows with credibility sub-scores of 0.7, 0.8, and 0.9. The credibility score of the conflict feature sequence is 0.7+0.8+0.9, which reflects the credibility level of the entire conflict feature sequence.

[0137] Step 144: Weighted fusion processing of the conflict feature sequence according to the credibility score, prioritizing the feature sequence with a weight value higher than the preset weight, and performing feature correction processing on the feature sequence with a weight value lower than the preset weight.

[0138] Weighted fusion processing of the conflict feature sequence according to the credibility score. The credibility score is used as the weight to perform weighted average processing on the conflict feature sequence. For the feature sequence with a weight value higher than the preset weight, it is prioritized for retention. The preset weight is determined based on experience and historical data to judge whether a feature sequence is reliable. For the feature sequence with a weight value lower than the preset weight, feature correction processing is performed. For example, interpolation or fitting with other reliable feature sequences can be used to correct the low-weight feature sequence to make it more consistent with the actual situation.

[0139] Step 145: splice the fused feature sequence with the non-conflict feature sequence to generate a state deduction fusion result containing complete state feature information; standardize the second running data set based on the historical running state record of the wind turbine to generate a standardized second running data set; synchronize the time stamps of the sampling points of the state deduction fusion result and the standardized second running data set, extract the state feature values of the corresponding time points of the synchronized state deduction fusion result and the standardized second running data set, and calculate a feature deviation degree sequence.

[0140] The fused feature sequence is spliced with the non-conflict feature sequence. The fused feature sequence after conflict resolution processing and the feature sequence without conflict are spliced in time sequence to generate a state deduction fusion result containing complete state feature information. The second running data set is standardized based on the historical running state record of the wind turbine. The standardization processing method is the same as the standardization processing of the multi-dimensional state feature set in step 131, so that the second running data set has comparability. The state deduction fusion result and the standardized second running data set are synchronized in time stamp of the sampling points. Since the sampling times of the two data sets may be different, time stamp synchronization is needed to make them have corresponding state values at the same time point. The state feature values of the corresponding time points of the synchronized state deduction fusion result and the standardized second running data set are extracted, and a feature deviation degree sequence is calculated. By calculating the difference between the state feature values at each corresponding time point, the feature deviation degree sequence is obtained, which reflects the difference between the state deduction fusion result and the actual running state.

[0141] Step 146: trend analysis processing is performed on the feature deviation degree sequence to identify the change trend type and change rate of the deviation degree.

[0142] Step 1461: input the feature deviation degree sequence into the sequence preprocessing module of the trend analysis model for sequence smoothing processing and outlier identification processing to generate a purified deviation degree sequence.

[0143] The feature deviation degree sequence is input into the sequence preprocessing module of the trend analysis model. The module first performs sequence smoothing processing to eliminate trends and seasonal components in the sequence through difference, logarithmic transformation, etc., so that the sequence becomes stationary. Then, outlier identification processing is performed by setting a threshold or using statistical methods to identify outliers in the sequence. For outliers, interpolation or deletion methods can be used for processing. Through these processes, a purified deviation degree sequence is generated, providing more accurate data for trend analysis.

[0144] Step 1462: perform multi-scale decomposition processing on the purified deviation degree sequence to decompose it into a multi-component sequence set containing fluctuation components and trend components.

[0145] The purified deviation degree sequence is subjected to multi-scale decomposition processing. Common multi-scale decomposition methods include wavelet decomposition. Through wavelet decomposition, the deviation degree sequence is decomposed into components of different scales, including fluctuation components and trend components. Fluctuation components reflect the short-term fluctuation of the sequence, and trend components reflect the long-term trend of the sequence. Through this decomposition, the change characteristics of the deviation degree sequence can be more clearly analyzed.

[0146] Step 1463: Extract the trend component from the multi-component sequence set, input it into the trend type identifier for pattern matching processing, which includes morphological comparison of the trend component with a preset standard trend pattern library containing continuous rising patterns, continuous falling patterns, fluctuating rising patterns, fluctuating falling patterns, and stable patterns.

[0147] Extract the trend component from the multi-component sequence set, input it into the trend type identifier for pattern matching processing. The trend type identifier contains a preset standard trend pattern library, which contains continuous rising patterns, continuous falling patterns, fluctuating rising patterns, fluctuating falling patterns, and stable patterns. Morphological comparison is made between the trend component and the patterns in the standard trend pattern library, and the most matching pattern is found by calculating the similarity or matching degree. For example, use methods such as dynamic time warping distance to calculate the similarity of the trend component with each standard pattern, and select the pattern with the highest similarity as the matching result.

[0148] Step 1464: Determine the main trend type corresponding to the trend component according to the principle of highest similarity of pattern matching, and extract the fluctuation frequency and fluctuation amplitude characteristics of the fluctuation component.

[0149] Determine the main trend type corresponding to the trend component according to the principle of highest similarity of pattern matching. If the trend component has the highest similarity with the continuous rising pattern, the main trend type is continuous rising. At the same time, extract the fluctuation frequency and fluctuation amplitude characteristics of the fluctuation component. Fluctuation frequency can be obtained by calculating the period of the fluctuation component, and fluctuation amplitude can be obtained by calculating the difference between the maximum and minimum values of the fluctuation component. These characteristics can further describe the change of the deviation degree sequence.

[0150] Step 1465: Based on the main trend type, fluctuation frequency and fluctuation amplitude characteristics, call the trend descriptor generator to generate a trend feature descriptor containing trend direction identification, fluctuation characteristic parameters and trend stability index.

[0151] Based on the main trend type, the fluctuation frequency and the fluctuation amplitude characteristics, the trend descriptor generator is called to generate the trend characteristic descriptor. The trend characteristic descriptor contains the trend direction identifier, the fluctuation characteristic parameters and the trend stability index. The trend direction identifier is determined according to the main trend type, such as "up", "down" or "flat". The fluctuation characteristic parameters include the fluctuation frequency and the fluctuation amplitude, and the trend stability index can be obtained by calculating the variance or standard deviation of the fluctuation component, reflecting the stability degree of the trend. Through the trend characteristic descriptor, the change trend of the deviation degree sequence can be comprehensively described.

[0152] Step 1466: The purified deviation degree sequence is processed by a sliding window, and observation windows of different lengths are set to calculate the sequence morphological change rate in each observation window.

[0153] The purified deviation degree sequence is processed by a sliding window. Observation windows of different lengths are set, such as 10 time points, 20 time points, etc.

[0154] In each observation window, the morphological change rate of the sequence is calculated, which can be obtained by calculating the slope or curvature of the sequence in the window. For example, for an observation window of 10 time points, the difference between the first time point and the last time point in the window is calculated, and the ratio of the window length is obtained, which is the morphological change rate of the sequence in the window.

[0155] Step 1467: The morphological change rates of different observation windows are statistically fused and processed, and the fluctuation characteristic parameters in the trend characteristic descriptor are combined to generate a comprehensive change trend evaluation index.

[0156] The morphological change rates of different observation windows are statistically fused and processed, which can be weighted average or other statistical methods to integrate the morphological change rates of different windows.

[0157] The fluctuation characteristic parameters in the trend characteristic descriptor are combined to generate a comprehensive change trend evaluation index. For example, the morphological change rate is combined with the fluctuation frequency and the fluctuation amplitude to obtain a comprehensive evaluation index, which can more comprehensively reflect the change trend of the deviation degree sequence.

[0158] Step 1468: The comprehensive change trend evaluation index is processed by semantic conversion through the change rate inference model, and the comprehensive change trend evaluation index is mapped to the corresponding level in the preset change rate level system to generate a trend analysis result report containing the trend type identifier and the change rate level.

[0159] The comprehensive change trend evaluation index is semantically converted by a change rate inference model. The change rate inference model includes a preset change rate level system, such as “rapid rise”, “slow rise”, “rapid fall”, “slow fall”, and the like.

[0160] The comprehensive change trend evaluation index is mapped to a corresponding level in the level system. For example, by setting different threshold ranges, the evaluation index is divided into different levels. A trend analysis result report including a trend type identifier and a change rate level is generated, which describes the change trend and change rate of the bias degree sequence in detail.

[0161] Step 147: Based on the change trend type and the change rate, a preset health state evaluation rule library is queried to determine the health state evaluation label corresponding to the wind turbine.

[0162] The change trend type and the change rate are used to query the preset health state evaluation rule library. The health state evaluation rule library includes health state evaluation labels corresponding to different change trend types and change rates, such as “healthy”, “sub-healthy”, “fault warning”, “fault”, and the like.

[0163] According to the trend type identifier and the change rate level in the trend analysis result report, the corresponding health state evaluation label is found in the rule library. For example, if the trend type is continuous rise and the change rate is rapid rise, the corresponding label in the rule library may be “fault warning”. In this way, the health state evaluation label corresponding to the wind turbine is determined.

[0164] Exemplarily, in step 141, the Euclidean distance algorithm is used when calculating the feature similarity matrix. For two key state feature sequences, such as sequence A = [1, 2, 3, 4, 5] and sequence B = [2, 3, 4, 5, 6], the Euclidean distance is obtained by calculating the square root of the sum of the squares of the differences between the corresponding elements. The calculated Euclidean distance is 2.24. The distance value is converted to a similarity value, for example, a mapping function can be used to convert the distance value to the range of 0-1 to obtain a similarity score.

[0165] In step 1423, when calculating the dynamic time warping distance of the normalized sequence pair, the length of the time window segment is set to 10 time points. For the feature sequences within the two time window segments, their dynamic time warping distances are calculated by the dynamic programming algorithm. For example, the two sequences are [0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0] and [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0, 1.1] respectively, and the calculated dynamic time warping distance is 0.5. The distance is compared with a preset distance threshold (such as 0.6) to determine whether it is a conflict time window.

[0166] In step 1431, the feature correlation weight matrix is a 10x10 matrix representing the correlation weights between the 10 feature dimensions. For example, the conflict feature dimension is power, and the row vector in the matrix where the power is located is [0.2, 0.3, 0.1, 0.15, 0.05, 0.1, 0.05, 0.03, 0.02, 0.0], which is taken as the initial correlation weight vector.

[0167] In step 1432, the calculation of the weight dynamic adjustment coefficient considers the historical conflict occurrence rate and the historical fusion contribution degree. For example, the historical conflict occurrence rate of the power feature dimension is 0.2, and the historical fusion contribution degree is 0.3. The weight dynamic adjustment coefficient is calculated by a linear combination, such as coefficient = 0.8x(1-0.2) + 0.2x0.3.

[0168] In step 1461, the sequence preprocessing module of the trend analysis model uses the first-order difference method for sequence stationarization processing. For the feature deviation degree sequence, each data point is subtracted by the previous data point to obtain the difference sequence. For example, the original sequence is [1, 2, 3, 4, 5], and the first-order difference is [1, 1, 1, 1]. In the abnormal value identification process, the threshold is set to 3 times the standard deviation, that is, if the difference between a data point and the mean value exceeds 3 times the standard deviation, it is considered as an abnormal value.

[0169] In step 1463, the trend type identifier uses dynamic time warping distance for pattern matching. The continuous rising pattern sequence in the standard trend pattern library is [1, 2, 3, 4, 5], and the fluctuating rising pattern sequence is [1, 1.5, 2, 2.5, 3], etc. For example, the trend component sequence is [1.2, 2.1, 3.2, 4.1, 5.2], and the dynamic time warping distance between it and each standard pattern is calculated to find the most matching pattern.

[0170] In an alternative embodiment, the method further comprises: Step 210: Historical data tracing analysis is performed on the generated wind turbine health status evaluation label to extract historical evaluation records that are the same or similar to the current health status evaluation label.

[0171] Historical data tracing analysis is performed on the generated wind turbine health status evaluation label. In the historical data, historical evaluation records that are the same or similar to the current health status evaluation label are searched. Through analysis of the historical evaluation records, the subsequent operation of the wind turbine and the treatment measures under the same or similar health status can be understood. For example, if the current health status evaluation label is "fault warning", all records marked as "fault warning" are searched in the historical data, and the operation parameters, fault occurrence time and treatment methods of the wind turbine corresponding to these records are analyzed.

[0172] Step 220: Collect the subsequent operation state data of the wind turbine corresponding to the historical evaluation records, and establish a health status evolution case library.

[0173] The subsequent operation state data of the wind turbine corresponding to the historical evaluation records is collected, which includes the change of operation parameters after the fault occurs, maintenance records, downtime, etc. These data are sorted and classified to establish a health status evolution case library. Each case in the case library contains health status evaluation label, corresponding operation state data and treatment result, etc. Through the case library, reference can be provided for the health status evaluation and treatment of the current wind turbine.

[0174] Step 230: Feature extraction processing is performed on the case data in the health status evolution case library to identify the state evolution path features corresponding to different health status evaluation labels.

[0175] Feature extraction processing is performed on the case data in the health status evolution case library. The change rule of the operation state data in each case is analyzed, and the features reflecting the state evolution are extracted. For example, for the case of "fault warning" label, the change trend of power, speed and other parameters after the warning is analyzed, and the state evolution path feature is extracted. By analyzing the cases of different health status evaluation labels, the state evolution path features corresponding to each label are identified, which can help to predict the development trend of the wind turbine under different health status.

[0176] Step 240: Based on the state evolution path features, a health status prediction model is constructed, and the current multi-dimensional state feature set is input into the health status prediction model to generate a health status prediction sequence within a preset time window.

[0177] The health state prediction model can be constructed based on the state evolution path characteristics, and machine learning algorithms such as neural networks, decision trees, etc. can be used to construct the model. The current multi-dimensional state feature set is input into the health state prediction model, and the model learns and predicts based on the state evolution path characteristics. Through the calculation of the model, a health state prediction sequence within a preset time window is generated. For example, the health state change of the wind turbine in the next week is predicted, and the sequence includes health state evaluation labels at different time points.

[0178] Step 250: Risk level assessment of the health state prediction sequence is performed to determine the risk level corresponding to different time nodes.

[0179] The risk level of the health state prediction sequence is assessed. According to the severity and probability of occurrence of the health state evaluation label, the risk level of the health state of each time node is assigned. For example, the risk level corresponding to the "healthy" state is low, and the risk level corresponding to the "failure" state is high. By evaluating the health state of each time node in the prediction sequence, the risk level corresponding to different time nodes is determined, which can help to develop appropriate maintenance strategies.

[0180] Step 260: According to the risk level, an appropriate maintenance recommendation scheme is developed, which includes a recommended maintenance time window and a recommended maintenance measure type.

[0181] According to the risk level, an appropriate maintenance recommendation scheme is developed. For time nodes with high risk levels, more timely and strict maintenance measures are developed. The maintenance recommendation scheme includes a recommended maintenance time window and a recommended maintenance measure type. For example, if the risk level of one of the time nodes is "high", the recommended maintenance time window can be set to the near future, and the recommended maintenance measure type can include comprehensive inspection, replacement of vulnerable components, etc. By developing a maintenance recommendation scheme, the maintenance of the wind turbine can be prepared in advance, and the probability of failure can be reduced.

[0182] Step 270: The maintenance recommendation scheme is analyzed for coordination with the current operation plan of the wind turbine, the time window of the maintenance recommendation scheme is adjusted, and the adjusted maintenance recommendation scheme is output.

[0183] The maintenance recommendation scheme is analyzed for coordination with the current operation plan of the wind turbine. Considering factors such as power generation tasks, maintenance resources, etc. of the wind turbine, it is determined whether the time window of the maintenance recommendation scheme conflicts with the current operation plan. If there is a conflict, the time window of the maintenance recommendation scheme is adjusted. For example, if the maintenance recommendation time window conflicts with the power generation peak period, the maintenance time can be adjusted to the power generation valley period. Through coordination analysis and adjustment, the adjusted maintenance recommendation scheme is output, ensuring that the maintenance work can guarantee the safe operation of the wind turbine and minimize the impact on power generation tasks.

[0184] In step 240, the health state prediction model adopts a neural network model. The number of input layer nodes of the neural network is 10, corresponding to 10 features of the multi-dimensional state feature set; the hidden layer is set to 2 layers, the number of nodes of the first hidden layer is 20, and the number of nodes of the second hidden layer is 15; the number of output layer nodes is 3, corresponding to 3 types of health state evaluation labels (such as healthy, sub-healthy, and failure). The learning rate is set to 0.01, and the number of training rounds is set to 100 rounds.

[0185] In an alternative embodiment, the method further comprises: Step 310: Collecting historical operation data sets and corresponding health state evaluation results of multiple wind turbines, and constructing a multi-turbine health state analysis database.

[0186] Collecting historical operation data sets and corresponding health state evaluation results of multiple wind turbines, which come from the operation records of different wind turbines at different time periods. These data are sorted and integrated to construct a multi-turbine health state analysis database. The database contains the operation parameters, health state evaluation labels, and fault records of each wind turbine. Through the database, the health states of multiple wind turbines can be comprehensively analyzed and compared.

[0187] Step 320: Standardizing the data in the multi-turbine health state analysis database to eliminate the effects of equipment differences and environmental differences between different turbines.

[0188] Standardizing the data in the multi-turbine health state analysis database. Due to the differences in equipment specifications, installation environments, etc. of different wind turbines, these differences will affect the comparability of the data. Standardization methods such as Z-score standardization or Min-Max standardization are used to process the operation parameters of each wind turbine. Through standardization, the effects of equipment differences and environmental differences between different turbines are eliminated, and the data are made comparable.

[0189] Step 330: Extracting the multi-dimensional state feature set of the standardized historical operation data set, and performing correlation analysis with the corresponding health state evaluation results to construct a feature-health state correlation model.

[0190] Extracting the multi-dimensional state feature set of the standardized historical operation data set. The extraction method of the multi-dimensional state feature set is the same as the processing method of the first operation data set in steps 120-125. The extracted multi-dimensional state feature set is correlated with the corresponding health state evaluation results, which can be analyzed by statistical analysis methods or machine learning algorithms to analyze the relationship between features and health states.

[0191] For example, find out the features that have a greater impact on the health status through the decision tree algorithm. Based on the correlation analysis results, build a feature-health status correlation model, which can be used to predict the health status of wind turbines, and output the corresponding health status evaluation label according to the input multi-dimensional state feature set.

[0192] Step 340: Use the feature-health status correlation model to perform similarity analysis on the multi-dimensional state feature set of different wind turbines, and identify a cluster of wind turbines with similar health status evolution trends.

[0193] Use the feature-health status correlation model to perform similarity analysis on the multi-dimensional state feature set of different wind turbines. Calculate the similarity between the multi-dimensional state feature sets of each two wind turbines. Methods such as Euclidean distance and cosine similarity can be used for calculation. According to the similarity results, wind turbines with similar health status evolution trends are divided into a cluster of wind turbines. For example, if the multi-dimensional state feature sets of two wind turbines have high similarity, it means that their health status evolution trends are similar, and they are classified into the same cluster of wind turbines. By identifying the cluster of wind turbines, the wind turbines can be classified and managed, improving the maintenance efficiency.

[0194] Step 350: Perform common feature extraction processing on the multi-dimensional state feature set of each cluster of wind turbines to determine the key common features that affect the health status of the wind turbines in the cluster.

[0195] Perform common feature extraction processing on the multi-dimensional state feature set of each cluster of wind turbines. Analyze the multi-dimensional state feature set of each wind turbine in the cluster to find out the common features among them. Methods such as principal component analysis can be used to extract key common features that represent the characteristics of the cluster. For example, for a cluster of wind turbines, it is found that features such as power and speed have similar trends in most wind turbines, and these features are the key common features. By determining the key common features, the factors that affect the health status of the wind turbines in the cluster can be understood in depth.

[0196] Step 360: Compare the differences in key common features of different clusters of wind turbines, identify the main influencing factors that cause the differences in health status, build a health status optimization model based on the main influencing factors, generate personalized operation parameter adjustment suggestions for different clusters of wind turbines, and input the personalized operation parameter adjustment suggestions into the control system of the wind turbines to adjust the operation parameters of the wind turbines in real time.

[0197] Differences in key common characteristics of different unit clusters are compared. Differences in the value range, trend, etc. of key common characteristics between different clusters are analyzed. Through comparison, the main influencing factors leading to differences in health status are identified. For example, if the power of one unit cluster is generally lower, while the power of another cluster is higher, it may be due to factors such as wind speed, equipment efficiency, etc. Based on the main influencing factors, a health status optimization model is constructed, which can use optimization algorithms such as genetic algorithm, particle swarm algorithm, etc. to construct the model. The model takes the main influencing factors as input and optimizes the health status as the goal, and calculates the optimal operating parameters. Individualized operating parameter adjustment suggestions are generated for different unit clusters. The individualized operating parameter adjustment suggestions are input into the control system of the wind turbine, and the control system adjusts the operating parameters of the wind turbine in real time according to the suggestions. For example, adjusting the blade angle of the wind wheel, the output power of the generator, etc. to improve the health status and power generation efficiency of the wind turbine.

[0198] Step 370: Continuously monitor the changes in the adjusted health status of the wind turbine, and iteratively update the health status optimization model based on the monitoring results.

[0199] The adjusted health status of the wind turbine is continuously monitored. Real-time collection of operating data of the wind turbine is performed through a sensor network, and the health status of the wind turbine is evaluated according to the health status evaluation method. The health status optimization model is iteratively updated based on the monitoring results. If it is found that the adjusted operating parameters do not achieve the expected health status optimization effect, the reasons are analyzed and the model is adjusted. For example, adjusting the parameters or optimization algorithm in the model so that the model can more accurately predict and adjust the operating parameters of the wind turbine. Through iterative updating, the accuracy and effectiveness of the health status optimization model are continuously improved.

[0200] Exemplarily, in step 330, the feature-health status correlation model is constructed using a logistic regression algorithm. The regularization parameter is set to 0.1 to control the complexity of the model and prevent overfitting. The model is trained through the training data to adjust the weight parameters of the model so that the model can accurately predict the health status of the wind turbine.

[0201] The embodiments of the present application can deeply analyze the dynamic characteristics and potential abnormal patterns of the wind turbine by setting up a sensor network to collect the first operating data set and performing multi-dimensional state feature mining on the first operating data set, and can mine key information that is difficult to find by traditional methods, thereby improving the depth of understanding of the operating state of the wind turbine.

[0202] The state deduction model is iterated by using a pre-constructed state deduction model, and different initial conditions and parameter disturbances are introduced for multi-path deduction, fully considering the uncertainty and diversity in the operation process of the wind turbine, and multiple results reflecting different evolution trends of the operation state are generated. Compared with single-path deduction, the future state of the wind turbine can be more comprehensively predicted.

[0203] The fusion processing based on conflict resolution is performed on the multiple state deduction iteration results, and the feature correlation weight matrix and the credibility evaluation mechanism are combined to effectively solve the conflict problem in the multi-source data fusion process, realize conflict coordination and information complementation, and make the state deduction fusion result more accurate and reliable.

[0204] Finally, the state deduction fusion result is subjected to state evaluation based on synchronous time sequence analysis with the second operation data set, and the health state evaluation label of the wind turbine is obtained. This evaluation method combining data of different time periods can dynamically and accurately reflect the actual health status of the wind turbine, improve the accuracy and timeliness of health evaluation, and timely discover potential faults of the wind turbine, reduce maintenance cost, and improve power generation efficiency.

[0205] Based on the same inventive concept, the embodiments of the present application also provide a wind turbine health evaluation system. Referring to Figure 2 , which is a possible structure schematic diagram of a wind turbine health evaluation system provided in the embodiments of the present application, Figure 2 , the wind turbine health evaluation system 200 includes a processor 210 and a memory 220. The memory 220 stores a computer program executable by the processor 210, and the processor 210 can execute the steps of the above-mentioned wind turbine health evaluation method based on AI multi-source data fusion by executing the instructions stored in the memory 220.

[0206] Based on the same inventive concept, the embodiments of the present application provide a computer readable storage medium including a computer program, when the computer program runs on the wind turbine health evaluation system, the computer program is used to make the wind turbine health evaluation system execute the steps of the above-mentioned wind turbine health evaluation method based on AI multi-source data fusion. In some possible implementation manners, each aspect of the wind turbine health evaluation method based on AI multi-source data fusion provided by the present application can also be implemented in the form of a program product, which includes a computer program, when the program product runs on the wind turbine health evaluation system, the computer program is used to make the wind turbine health evaluation system execute the steps in the above-mentioned wind turbine health evaluation method based on AI multi-source data fusion, for example, the wind turbine health evaluation system can execute the steps as shown in Figure 1 .

[0207] The above merely provides preferred exemplary embodiments of the present application, and is not intended to limit the implementation of the present application. Based on the main concept and spirit of the present application, the person skilled in the art can easily make corresponding changes or modifications.

Claims

1. A method for health assessment of wind turbine units based on AI multi-source data fusion, characterized in that, include: The initial operational dataset of the wind turbine was collected through the established sensor network. Multidimensional state feature mining is performed on the first running dataset to obtain a multidimensional state feature set characterizing the dynamic characteristics and potential anomaly modes of the wind turbine. Combining the multidimensional state feature set, the wind turbine is subjected to state deduction iteration based on dynamic characteristics and operating laws using a pre-built state deduction model. Different initial conditions and parameter disturbances are introduced during the state deduction iteration process to carry out multi-path deduction, generating multiple state deduction iteration results that reflect the evolution trend of different operating states. The multiple state inference iteration results are fused based on conflict resolution, and conflict coordination and information complementarity are carried out by combining the established feature correlation weight matrix and credibility assessment mechanism to generate state inference fusion results. The state inference fusion results and the second operating dataset are subjected to state assessment based on synchronous time series analysis to obtain the health status assessment label of the wind turbine. The second running dataset was collected later than the first running dataset.

2. The method as described in claim 1, characterized in that, The step of performing multidimensional state feature mining on the first operational dataset to obtain a multidimensional state feature set characterizing the dynamic characteristics and potential anomaly modes of the wind turbine includes: The first running dataset is subjected to time series decomposition processing to separate a multi-component time series set containing periodic fluctuation features and non-periodic fluctuation features; The dynamic characteristics of each component time series in the multi-component time series set are extracted to generate vibration mode features reflecting mechanical vibration characteristics and energy transfer features reflecting energy conversion efficiency. Correlation analysis is performed on the vibration mode features and the energy transfer features to identify the synergistic change patterns between them and generate a feature correlation map. Based on the feature association map, the vibration mode features and energy transfer features are subjected to anomaly-sensitive feature screening to extract a subset of key features that are sensitive to potential anomalies. The key feature subset is reorganized to generate a multidimensional state feature set containing time dimension features, frequency dimension features, and amplitude dimension features.

3. The method as described in claim 1, characterized in that, The process involves combining the multidimensional state feature set and using a pre-built state deduction model to perform state deduction iterations on the wind turbine based on its dynamic characteristics and operating laws. Different initial conditions and parameter disturbances are introduced during the state deduction iteration process to perform multi-path deductions, generating multiple state deduction iteration results reflecting the evolution trends of different operating states, including: Based on the historical operating status records of the wind turbine, the multidimensional state feature set is standardized to generate a standardized multidimensional state feature set; the standardized multidimensional state feature set is input into the initial state configuration layer of the state inference model to generate multiple differentiated initial state vectors according to the historical operating status records of the wind turbine. For each initial state vector, parameter perturbation processing is performed, and the dynamic characteristic parameters are randomly adjusted within a preset perturbation range to generate a set of perturbed state vectors containing different parameter combinations; The set of disturbance state vectors is input into the deduction iteration layer of the state deduction model, and multi-step state deduction calculation is performed based on the wind turbine dynamic equation to generate a state evolution path sequence corresponding to each disturbance state vector. During the generation of the state evolution path sequence, the deviation between the deduced state and the historical operating state is monitored in real time. When the deviation exceeds the preset deviation, the path correction mechanism is triggered to dynamically adjust the state transition matrix of the subsequent deduced steps. Repeatedly execute state deduction calculations and path correction mechanisms until the deduction process within the preset time window is completed, and output multiple state deduction iteration results containing multiple independent evolution paths.

4. The method as described in claim 3, characterized in that, The step involves inputting the set of disturbance state vectors into the iteration layer of the state deduction model, performing multi-step state deduction calculations based on the wind turbine dynamics equations, and generating a state evolution path sequence corresponding to each disturbance state vector, including: Each perturbation state vector in the set of perturbation state vectors is converted into a standard state vector that matches the input format of the state deduction model; The dynamic calculation kernel of the state deduction model is invoked, and a single-step state transition calculation is performed on the standard state vector based on the mechanical transmission system characteristics and aerodynamic characteristic equations of the wind turbine to obtain the state prediction vector for the next moment. Key state parameters are extracted from the state prediction vector and normalized and compared with similar parameters in the historical operating state database to generate parameter similarity scores. The state prediction vector is weighted based on the parameter similarity score to generate a weighted state prediction vector. The weighted state prediction vector is used as the new input state vector, and the single-step state transition calculation, similarity comparison and confidence weighting are repeatedly performed until the preset number of deduction steps are completed, generating a state evolution path sequence containing timestamps.

5. The method as described in claim 3, characterized in that, During the generation of the state evolution path sequence, the deviation between the deduced state and the historical operating state is monitored in real time. When the deviation exceeds a preset deviation, a path correction mechanism is triggered to dynamically adjust the state transition matrix of subsequent deduced steps, including: Extract the inferred state vector at the current moment from the state evolution path sequence, and extract the actual operating state vector at the corresponding moment from the historical operating state database; The inferred state vector and the actual operating state vector are normalized, and the cosine similarity or correlation coefficient between the normalized vectors is calculated. The cosine similarity or correlation coefficient is then converted into a difference index, which is used as a deviation quantification index. The deviation metric is compared with a preset deviation threshold. When the deviation metric is greater than the deviation threshold, the path correction mechanism is activated. The state transition matrix adjustment module in the path correction mechanism is invoked to calculate the matrix correction coefficient based on the magnitude of the deviation metric index. The matrix correction coefficient is positively correlated with the deviation metric index. The state transition matrix in the state deduction model is adjusted element-wise using the matrix correction coefficients to generate the corrected state transition matrix. The modified state transition matrix is ​​used to continue the subsequent state deduction calculation steps until the deviation quantification index is less than or equal to the deviation threshold or all deduction steps are completed.

6. The method as described in claim 1, characterized in that, The process involves fusing the multiple state deduction iteration results based on conflict resolution, and combining the established feature correlation weight matrix and credibility assessment mechanism for conflict coordination and information complementarity to generate a state deduction fusion result. The state deduction fusion result is then compared with the second operational dataset to perform a state assessment based on synchronous time-series analysis to obtain the health status assessment label of the wind turbine, including: The multiple state deduction iteration results are aligned on the time axis to keep all state deduction iteration results synchronized in the time dimension; the key state feature sequence in each state deduction iteration result is extracted, and the feature similarity matrix between different state deduction iteration results is calculated; Based on the feature similarity matrix, conflicting feature sequence pairs are identified, and the conflict type of the conflicting feature sequence pairs is classified to determine the feature dimension and time node where the conflict occurred. The established feature correlation weight matrix is ​​invoked to extract the correlation weight values ​​of conflict feature dimensions in historical data, and the credibility score of each conflict feature sequence is calculated in combination with the credibility assessment mechanism. Based on the credibility score, the conflict feature sequences are weighted and fused. Feature sequences with weight values ​​higher than the preset weight are retained first, and feature sequences with weight values ​​lower than the preset weight are modified. The fused feature sequence is concatenated with the non-conflicting feature sequence to generate a state inference fusion result containing complete state feature information; the second operation dataset is standardized based on the historical operation state record of the wind turbine to generate a standardized second operation dataset; the timestamp of the sampling point is synchronized between the state inference fusion result and the standardized second operation dataset, and the state feature values ​​at the corresponding time of the synchronized state inference fusion result and the standardized second operation dataset are extracted and the feature deviation sequence is calculated. The characteristic deviation sequence is subjected to trend analysis to identify the type and rate of change of the deviation. Based on the change trend type and change rate, a preset health status assessment rule base is queried to determine the health status assessment label corresponding to the wind turbine.

7. The method as described in claim 6, characterized in that, The process of identifying conflicting feature sequence pairs based on the feature similarity matrix, classifying the conflict type of the conflicting feature sequence pairs, and determining the feature dimension and time point at which the conflict occurred includes: The feature similarity matrix is ​​traversed according to a preset feature similarity threshold, and feature sequence pairs with similarity values ​​lower than the feature similarity threshold are marked as potentially conflicting feature sequence pairs. The potential conflict feature sequence pairs are subjected to time series segmentation processing, dividing each feature sequence into multiple time window segments of equal length; Normalize the feature sequence pairs within each time window segment, calculate the dynamic time bending distance of the normalized sequence pairs, and mark the time window segments with a dynamic time bending distance greater than a preset distance threshold as conflict time windows. Extract the start and end timestamps of the conflict time window from the feature sequence to determine the target time node where the conflict occurred; The feature sequence pairs within the conflict time window are decomposed into feature dimensions to identify the target feature dimension where there is a difference jump, and the target feature dimension is marked as the conflict feature dimension. Based on the number and type of the conflict feature dimensions and the distribution characteristics of the conflict time windows, the conflict feature sequence pairs are classified into conflict types, and a conflict analysis report containing a conflict type identifier, a set of conflict feature dimensions, and a list of conflict time nodes is generated.

8. The method as described in claim 6, characterized in that, The aforementioned call establishes a feature correlation weight matrix, extracts the correlation weight values ​​of conflict feature dimensions in historical data, and calculates the credibility score for each conflict feature sequence using a credibility assessment mechanism, including: The feature correlation weight matrix is ​​subjected to hierarchical deconstruction to locate the coordinate index of the conflicting feature dimension in the matrix, and the row vector corresponding to the coordinate index is extracted as the initial correlation weight vector. The initial correlation weight vector is correlated with the historical performance characteristics of the conflict feature dimension in the historical operation database to generate dynamic weight adjustment coefficients. The historical performance characteristics include historical conflict occurrence rate and historical fusion contribution. The multi-source evidence fusion module of the credibility assessment mechanism performs multi-dimensional evidence collection and processing on conflict feature sequences. The multi-dimensional evidence includes internal consistency evidence of the feature sequences, external correlation evidence with other non-conflicting feature sequences, and reference evidence in similar historical scenarios. The collected multi-dimensional evidence is subjected to evidence strength quantification processing to generate an evidence strength value and evidence reliability factor for each piece of evidence. The initial associated weight vector is weighted and corrected based on the aforementioned weight dynamic adjustment coefficient to generate the corrected associated weight vector. The modified correlation weight vector and the evidence strength values ​​of the multi-dimensional evidence are normalized, and then a weighted fusion operation is performed to obtain a preliminary credibility score matrix. The initial credibility score matrix is ​​weighted element-wise based on the evidence reliability factor to generate a credibility sub-score for each conflict feature sequence in different conflict time windows. The credibility sub-scores for all conflict time windows are accumulated over time to generate credibility scores covering the entire conflict feature sequence.

9. The method as described in claim 6, characterized in that, The step of performing trend analysis on the characteristic deviation sequence to identify the type and rate of change of the deviation trend includes: The characteristic deviation sequence is input into the sequence preprocessing module of the trend analysis model for sequence stabilization and outlier identification to generate a purified deviation sequence. The purified deviation sequence is decomposed into a multi-component sequence set containing fluctuation and trend components by multi-scale decomposition. Extract the trend component from the multi-component sequence set and input it into the trend type recognizer for pattern matching processing. The pattern matching processing includes comparing the trend component with a preset standard trend pattern library. The standard trend pattern library includes continuous upward pattern, continuous downward pattern, fluctuating upward pattern, fluctuating downward pattern, and stable pattern. The main trend type corresponding to the trend component is determined based on the principle of highest similarity in pattern matching, and the fluctuation frequency and fluctuation amplitude features of the fluctuation component are extracted. Based on the main trend type, fluctuation frequency, and fluctuation amplitude characteristics, the trend descriptor generator is invoked to generate a trend feature descriptor that includes trend direction identifier, fluctuation characteristic parameters, and trend stability indicators. The purified deviation sequence was processed by sliding window, and observation windows of different lengths were set to calculate the rate of change of sequence morphology within each observation window. The morphological change rates of different observation windows are statistically fused and combined with the fluctuation characteristic parameters in the trend feature descriptor to generate a comprehensive change trend assessment index. The comprehensive change situation assessment index is semantically transformed by the change rate inference model, and the comprehensive change situation assessment index is mapped to the corresponding level in the preset change rate level system, generating a trend analysis result report containing trend type identifier and change rate level.

10. A wind turbine health assessment system, characterized in that, It includes a processor and a memory, wherein the memory stores a computer program that, when executed by the processor, causes the processor to perform the steps of any one of the methods described in claims 1 to 9.

Images