Blade icing diagnosis method and device, intelligent deicing system and storage medium

Abstract

The invention relates to the technical field of wind power generation, in particular to a blade icing diagnosis method and device, an intelligent deicing system and a storage medium. According to the invention, blade icing diagnosis is carried out by adopting multi-modal data, the early detection accuracy of blade icing is improved, the false alarm rate of single sensor detection is reduced, and accurate diagnosis of blade icing is realized. When the multi-modal data is processed, a dynamic alignment algorithm is adopted, the limitation that traditional data alignment only depends on timestamps is broken through, meteorological data is introduced to serve as an alignment constraint condition, and the problem that physical relevance of the multi-modal data is missing is solved. And meanwhile, a hybrid expert architecture is adopted to process the multi-modal data, so that effective fusion and high-robustness diagnosis of the multi-modal data are realized.

Description

Technical Field

[0001] The present invention relates to the technical field of wind power generation, and in particular to a method and device for diagnosing blade icing, an intelligent deicing system, and a storage medium. Background Art

[0002] Blade icing is a common fault phenomenon in industrial equipment, energy equipment and transportation facilities, especially in low temperature environments, which can easily cause equipment operation to be blocked, efficiency to be reduced or even damaged. Traditional blade icing detection methods mainly rely on a single sensor, but because the detection results of a single sensor are greatly affected by environmental conditions and equipment status, it is easy to produce false positives or missed positives, resulting in inaccurate diagnosis of blade icing problems.

[0003] At present, there have been many studies on the detection and diagnosis technology of blade icing. For example, the blade temperature anomaly detection method based on infrared thermal imager analyzes the temperature distribution of the blade, identifies the temperature anomaly area, and then determines the blade icing situation. In addition, the method based on vibration sensor detects the early signals of blade icing by analyzing the changes in the blade vibration spectrum. These methods have good detection effects in certain specific scenarios, but there are still many problems. For example, it is difficult to fully capture the complex characteristics of blade icing using a single infrared thermal imager or vibration sensor, which is prone to false alarms or missed alarms, and it is difficult to maintain a stable detection effect in a complex environment. Summary of the invention

[0004] In view of this, the present invention provides a method and device for diagnosing blade icing, an intelligent deicing system, and a storage medium to solve one of the problems existing in the above-mentioned prior art.

[0005] In a first aspect, the present invention provides a method for diagnosing blade icing, the method comprising: acquiring multimodal data of the blade, the multimodal data including meteorological data of the blade's environment and temperature data and vibration data obtained by monitoring the blade; using a timestamp-based dynamic alignment algorithm to align the multimodal data using the meteorological data as a constraint item to obtain aligned multimodal data; using a hybrid expert architecture to extract features of the aligned multimodal data, and outputting a diagnostic result of blade icing.

[0006] In the present invention, multimodal data is used to diagnose blade icing, which improves the accuracy of early detection of blade icing, reduces the false alarm rate of single sensor detection, and realizes accurate diagnosis of blade icing. When processing multimodal data, a dynamic alignment algorithm is used to break through the limitation of traditional data alignment that only relies on timestamps, introduce meteorological data as alignment constraints, and solve the problem of missing physical correlation of multimodal data. At the same time, a hybrid expert architecture is used to process multimodal data to achieve effective fusion of multimodal data and high robustness diagnosis.

[0007] In an optional embodiment, the method further includes: determining the de-icing timing and presetting the power of the de-icing device according to the multimodal data and the diagnosis result.

[0008] In the present invention, the deicing timing and the power of the preset deicing device are further determined based on multimodal data and diagnostic results, thereby upgrading the icing diagnosis from a passive mode of "detection-execution" to an active mode of "prediction-optimization", significantly reducing maintenance costs.

[0009] In an optional embodiment, a timestamp-based dynamic alignment algorithm is used to align multimodal data with meteorological data as a constraint to obtain aligned multimodal data, including: sorting the multimodal data according to timestamps, and interpolating data with low sampling frequency in the multimodal data based on the sorting results to obtain a time series corresponding to the multimodal data; using meteorological data as a constraint, using a dynamic programming algorithm to calculate the cumulative distance matrix between time series, and aligning the multimodal data based on an alignment path with the smallest distance to obtain aligned multimodal data.

[0010] In the present invention, a dynamic alignment algorithm based on timestamp is used to align data to solve the problem of inconsistent sampling frequencies of different sensors (such as low frame rate of infrared images and high sampling rate of vibration data). At the same time, meteorological data is introduced as an alignment constraint to ensure the consistency of multimodal data in time dimension and physical mechanism. This improves the accuracy and efficiency of blade icing diagnosis.

[0011] In an optional embodiment, a hybrid expert architecture includes an expert module and a gating network, the expert module includes a convolutional network expert module, a long short-term memory network expert module and a meteorological feature expert module, and the hybrid expert architecture is used to extract features of the aligned multimodal data and output the diagnosis result of blade icing, including: using the convolutional network expert module, the long short-term memory network expert module and the meteorological feature expert module to respectively extract the first feature of the temperature data, the second feature of the vibration data and the third feature of the meteorological data; using the weights assigned by the gating network to fuse the first feature, the second feature and the third feature and input them into a random forest classifier; using the random forest classifier to process the fused features to obtain the diagnosis result of blade icing.

[0012] In the present invention, different expert modules are used to extract features from data of different modes, which can adapt to different data changes.

[0013] In an optional embodiment, a random forest classifier is used to process the fused features to obtain a diagnosis result of blade icing, including: using a random forest classifier to process the fused features to obtain a blade icing probability; obtaining a blade icing diagnosis result based on a relationship between the blade icing probability and a classification threshold, wherein the classification threshold is dynamically adjusted according to meteorological data; and performing a weighted summation of the blade icing probability and the feature weights of the multimodal data to obtain an explainable diagnosis result.

[0014] In the present invention, the classification threshold is dynamically adjusted according to meteorological data to improve the robustness of the model under different environmental conditions. At the same time, an interpretable diagnosis result is generated, which enhances the interpretability of the diagnosis result.

[0015] In an optional embodiment, the de-icing timing is determined based on the multimodal data and the diagnosis results, including: using a reinforcement learning algorithm to construct and train an icing risk energy consumption game model; inputting the multimodal data and the diagnosis results into the trained icing risk energy consumption game model to obtain the de-icing timing, which includes triggering de-icing or delaying the triggering of de-icing.

[0016] In the present invention, reinforcement learning is used to construct an icing risk energy consumption game model to optimize the deicing timing, thereby reducing ineffective energy consumption.

[0017] In an optional implementation, the diagnosis result includes the probability of blade icing, the meteorological data includes humidity and wind speed, and the power of the preset deicing device is determined using the following formula:

[0018] In the formula, P Indicates power, represents the probability of icing, H Indicates humidity, H 0 represents the critical humidity threshold, Indicates wind speed, Indicates the reference wind speed, k and a Represents the adjustment coefficient.

[0019] In the present invention, the power of the deicing device is dynamically adjusted through the correlation function between the icing probability and the meteorological data, thereby realizing the nonlinear optimization of the power output of the deicing device.

[0020] In a second aspect, the present invention provides a diagnostic device for blade icing, the device comprising: a data acquisition module, used to acquire multimodal data of the blade, the multimodal data including meteorological data of the blade's environment and temperature data and vibration data obtained by monitoring the blade; a data alignment module, used to align the multimodal data using a timestamp-based dynamic alignment algorithm with meteorological data as a constraint item to obtain aligned multimodal data; a diagnostic module, used to extract features of the aligned multimodal data using a hybrid expert architecture, and output a diagnostic result of blade icing.

[0021] In a third aspect, the present invention provides an intelligent de-icing system, which includes: a sensor for acquiring multi-modal data of blades, the multi-modal data including meteorological data of the blade environment and temperature data and vibration data obtained by monitoring the blades; an intelligent de-icing device for aligning the multi-modal data using a timestamp-based dynamic alignment algorithm with meteorological data as a constraint to obtain aligned multi-modal data; extracting features of the aligned multi-modal data using a hybrid expert architecture, and outputting a diagnosis result of blade icing; and determining the de-icing timing and de-icing power for de-icing based on the multi-modal data and the diagnosis result.

[0022] In a fourth aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the method for diagnosing blade icing according to the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0023] In a fifth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to enable a computer to execute the method for diagnosing blade icing according to the first aspect or any corresponding embodiment thereof.

[0024] In a sixth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the method for diagnosing blade icing according to the first aspect or any corresponding embodiment thereof. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0026] Figure 1 is a schematic flow chart of a method for diagnosing blade icing according to an embodiment of the present invention; Figure 2 is a structural block diagram of a device for diagnosing blade icing according to an embodiment of the present invention; Figure 3 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0028] According to an embodiment of the present invention, an embodiment of a method for diagnosing blade icing is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0029] In this embodiment, a method for diagnosing blade icing is provided, which can be used in electronic devices such as computers, mobile phones, tablet computers, etc. Figure 1 FIG. 4 is a flow chart of a method for diagnosing blade icing according to an embodiment of the present invention. Figure 1 As shown, the process includes the following steps: Step S101, obtaining multimodal data of the blade, the multimodal data including meteorological data of the environment in which the blade is located and temperature data and vibration data obtained by monitoring the blade. In this embodiment, the blade may refer to a blade on a wind turbine. In actual application, the blade may also be other types of blades, such as blades of an air conditioner outdoor unit, blades of a car radiator, blades of an aircraft engine, and the like. When monitoring blade icing, this embodiment provides various types of sensors to collect multimodal data.

[0030] In order to monitor the temperature of the blades, the present embodiment sets an infrared thermal imager to collect the temperature distribution image of the blades. Moreover, when the blades are frozen, their vibration spectrum will also change, so a vibration sensor is further set to collect the vibration spectrum data of the blades. In addition, environmental factors are also the main factors affecting the freezing of blades, so a meteorological sensor is set to collect the environmental parameters of the environment in which the blades are located, and the environmental parameters include data such as ambient temperature, ambient humidity, and wind speed.

[0031] For the acquired multimodal data, the data is first preprocessed, including cleaning and normalizing the data. In addition, for the temperature distribution image data acquired by the infrared thermal imager, edge detection, texture analysis and other processing can be performed to highlight the abnormal temperature area of ​​the blade. For the vibration data collected by the vibration sensor, the frequency and amplitude changes of the vibration signal can also be extracted.

[0032] Step S102, using a timestamp-based dynamic alignment algorithm, aligning the multimodal data with meteorological data as a constraint item, and obtaining aligned multimodal data. Specifically, the dynamic alignment algorithm used in this embodiment is a dynamic time warping algorithm (Dynamic Time Warping with Environmental Constraints, DTW-EC) that considers environmental constraints. Among them, the dynamic time warping algorithm calculates the distance matrix between the two time series, and then finds the optimal path between the two time series through a dynamic programming method based on the distance matrix. This optimal path determines the alignment relationship between the two time series. DTW-EC further considers the impact of environmental factors on the alignment of time series, converts environmental information into constraints, and considers not only the distance matrix but also the impact of environmental factors when determining the optimal path. Therefore, the dynamic alignment algorithm can not only solve the problem of inconsistent sampling frequencies of different sensors, but also ensure the consistency of multimodal data in the time dimension and physical mechanism by introducing environmental parameters as constraints.

[0033] Step S103, using a hybrid expert architecture to extract features of the aligned multimodal data, and output the diagnosis result of blade icing. Specifically, the hybrid expert architecture includes expert modules and gating networks, wherein each expert model is used as an independent model to process different data modes, and the gating network is used to determine the weight of each expert model according to the input data. In this embodiment, different expert modules in the hybrid expert architecture are used to extract features of different modal data, and then the features are fused based on the weights assigned by the gating network, and finally the diagnosis result of blade icing is determined based on the fused features. The diagnosis result can be the probability of blade icing, or it can be a diagnosis result of icing or non-icing.

[0034] The method for diagnosing blade icing provided in the embodiment of the present invention uses multimodal data to diagnose blade icing, thereby improving the accuracy of early detection of blade icing, reducing the false alarm rate of single sensor detection, and achieving accurate diagnosis of blade icing. When processing multimodal data, a dynamic alignment algorithm is used to break through the limitation that traditional data alignment only relies on timestamps, introduce meteorological data as alignment constraints, and solve the problem of missing physical correlation of multimodal data. At the same time, a hybrid expert architecture is used to process multimodal data to achieve effective fusion of multimodal data and high-robustness diagnosis.

[0035] In this embodiment, a method for diagnosing blade icing is provided, and the method comprises the following steps: Step S201, obtaining multimodal data of the blade, the multimodal data including meteorological data of the environment in which the blade is located and temperature data and vibration data obtained by monitoring the blade.

[0036] Step S202: Use a timestamp-based dynamic alignment algorithm to align the multimodal data using the meteorological data as a constraint to obtain aligned multimodal data. Figure 1 Step S104 of the illustrated embodiment will not be described in detail here.

[0037] Specifically, the above step S202 includes: Step S2021, sort the multimodal data according to the timestamp, and interpolate the data with low sampling frequency in the multimodal data based on the sorting result to obtain the time series corresponding to the multimodal data; specifically, by sorting the multimodal data according to the timestamp, the common time range of the multimodal data can be determined, and the data with low sampling frequency in the multimodal data can also be determined. For example, the temperature data is 50 data collected per unit time, and the vibration data is 100 data collected per unit time. After sorting by the timestamp, some timestamps may only correspond to vibration data but no temperature data, which indicates that the sampling frequency of the temperature data is low. At this time, interpolation methods such as linear interpolation or Lagrange interpolation can be used to interpolate the temperature data to estimate the temperature data at the high-frequency sampling point. In this way, the temperature data and the vibration data can be aligned with the timestamp. The data after timestamp sorting and interpolation can be used as a time series for subsequent processing.

[0038] Step S2022, using meteorological data as a constraint, using a dynamic programming algorithm to calculate the cumulative distance matrix between time series, and aligning the multimodal data based on the alignment path with the minimum distance to obtain aligned multimodal data. Specifically, when determining the alignment path with the minimum distance, the traditional dynamic time planning algorithm first calculates the local distance between two time series (such as the temperature data time series and the vibration data time series in this embodiment) using Euclidean distance or other methods, and then uses a recursive formula to calculate the optimal cumulative distance of each position to obtain a cumulative distance matrix, and finally determines the alignment path with the minimum distance by backtracking from the cumulative distance matrix.

[0039] In this embodiment, meteorological data is introduced as a constraint item when determining the alignment path. First, the impact of meteorological data on temperature data and vibration data is analyzed. For example, wind speed may affect the propagation speed and amplitude of vibration signals. Based on this impact analysis, when calculating the local distance, the corresponding weight is introduced to correct the local distance. For example, when the wind speed is high, the alignment weight of the vibration signal is appropriately adjusted to reflect the degree to which it is affected by the environment. In addition, when calculating the local distance, meteorological data, temperature data, and vibration data can also be used as optimization variables together, and the local distance is calculated by joint optimization. At this time, in addition to using Euclidean distance to calculate the distance between the temperature time series and the vibration time series, the Euclidean distance between the meteorological data can also be calculated, and then the two are weighted fused to obtain the local distance.

[0040] Therefore, when introducing meteorological data to calculate local distance, meteorological data, temperature data, and vibration data can be used as optimization variables, and local distance can be calculated by joint optimization. Then, corresponding weights are introduced to correct the local distance, thereby obtaining the final local distance. The way to accumulate distance matrix and align paths can refer to the traditional dynamic time planning algorithm, which will not be described here.

[0041] Step S203, using a hybrid expert architecture to extract features of the aligned multimodal data and output a diagnosis result of blade icing; the hybrid expert architecture includes an expert module and a gating network, and the expert module includes a convolutional network expert module, a long short-term memory network expert module, and a meteorological feature expert module.

[0042] Specifically, the above step S203 includes: Step S2031, using a convolutional network expert module, a long short-term memory network expert module and a meteorological feature expert module to respectively extract the first feature of the temperature data, the second feature of the vibration data and the third feature of the meteorological data; specifically, a convolutional neural network is constructed in the convolutional network expert module, which can process the image data of the infrared thermal imager, extract the high-dimensional spatial features of the temperature distribution of the blades, and generate a temperature anomaly feature map; a long short-term memory network is constructed in the long short-term memory network expert module, which can process the time series data of the vibration sensor, extract the dynamic features in the vibration spectrum, and capture the dynamic changes in the time series. A statistical analysis model or a shallow neural network model is constructed in the meteorological feature expert module to extract meteorological features, such as humidity change rate or temperature gradient.

[0043] Step S2032, the first feature, the second feature and the third feature are fused using the weights assigned by the gating network and then input into the random forest classifier; specifically, the gating network is constructed using a feedforward neural network, and its output is calculated through the softmax function to calculate the activation probability (weight) of each expert. In practical applications, it can be trained with multimodal data first, so that it can learn the characteristic distribution of the data, and gradually optimize the selection strategy of each expert module, so that the gating network assigns appropriate weights to each expert module according to the characteristics of the data. For example, in a low temperature and high humidity environment, the gating network may assign higher weights to the meteorological feature expert module. Afterwards, based on the weights assigned by the gating network, the features output by each expert module are fused, and the fused data is input into the random forest classifier.

[0044] Step S2033, using a random forest classifier to process the fused features to obtain the diagnosis results of blade icing. After the fused features are input into the random forest classifier, the random forest classifier can further evaluate the contribution of the features to the results, helping to identify and select the most important features. In addition, before using the expert module, the gated network and the random forest classifier to extract features and determine the diagnosis results, this embodiment first uses the labeled historical data to train it, and uses cross-validation and other methods to adjust the parameters to obtain the final model.

[0045] In an optional implementation, the above step S2033 includes: Step a1: Use a random forest classifier to process the fused features to obtain the probability of leaf icing.

[0046] Step a2, based on the relationship between the blade icing probability and the classification threshold, the diagnosis result of blade icing is obtained, and the classification threshold is dynamically adjusted according to the meteorological data; specifically, after the blade icing probability is obtained, it can be compared with the preset classification threshold, when it is greater than or equal to the classification threshold, the diagnosis result is icing; when it is less than the classification threshold, the diagnosis result is not icing. In addition, the classification threshold can also be dynamically adjusted according to the meteorological data, for example, the threshold can be appropriately lowered in a low temperature and high humidity environment to increase the sensitivity to icing conditions.

[0047] Step a3, weighted sum the blade icing probability and the feature weights of the multimodal data to obtain an interpretable diagnosis result. Specifically, the feature weights of the multimodal data can be determined by mutual information calculation, for example, calculating the mutual information between different modal features and the blade icing label. The larger the mutual information, the stronger the correlation between the feature and the label, and the higher the contribution. However, weights are assigned to each modal feature according to the size of the mutual information. The interpretable diagnosis result can also be understood as a credibility score, which is specifically determined by weighted summation of the icing probability and the feature weights of each modal data, that is, credibility score = Σ (feature weight × icing probability). In addition, the contribution (i.e., weight) of each modal feature to the diagnosis result, such as the degree of influence of temperature anomaly features, vibration features, and meteorological features on the icing probability, can be visualized in the form of a table to enhance the interpretability of the results.

[0048] In this embodiment, a method for diagnosing blade icing is provided, and the method comprises the following steps: Step S301, obtaining multimodal data of the blade, the multimodal data including meteorological data of the environment in which the blade is located and temperature data and vibration data obtained by monitoring the blade; see Figure 1 Step S101 of the illustrated embodiment will not be described in detail here.

[0049] Step S302: Use a timestamp-based dynamic alignment algorithm to align the multimodal data using the meteorological data as a constraint to obtain aligned multimodal data. Figure 1 Step S102 of the illustrated embodiment will not be described in detail here.

[0050] Step S303: extract the features of the aligned multimodal data using a hybrid expert architecture and output the diagnosis result of blade icing; see Figure 1 Step S103 of the illustrated embodiment will not be described in detail here.

[0051] Step S304, determining the deicing timing and the power of the preset deicing device according to the multimodal data and the diagnosis results. Specifically, the deicing timing is the timing of starting the deicing device. In this embodiment, the timing is limited to two modes: immediate start and delayed start. By accurately determining the deicing timing, ineffective energy consumption can be reduced. The preset deicing device can be an electric heating deicing device used in the relevant technology. By determining the specific power according to the multimodal data and the diagnosis results, a more efficient deicing effect can be achieved, while maximizing energy saving. Specifically, the above step S304 includes: Step S3041, using a reinforcement learning algorithm to construct and train an icing risk energy consumption game model; specifically, this embodiment uses the Q-Learning algorithm in the reinforcement learning algorithm to construct the model, which is a reinforcement learning algorithm based on a value function, and guides the behavior decision of the intelligent agent by learning the state-action value function. The Q-Learning algorithm has the advantages of simplicity, high efficiency, and easy implementation, and is suitable for constructing an icing risk-energy consumption game model.

[0052] The constructed icing risk-energy consumption game model includes state space, action space and reward function. Among them, the state space includes multiple dimensions such as ambient humidity, wind speed, and icing probability. For example, the humidity is divided into three intervals of low humidity, medium humidity, and high humidity, the wind speed is divided into three intervals of low wind speed, medium wind speed, and high wind speed, and the icing probability is divided into three intervals of low risk, medium risk, and high risk. Each interval corresponds to a state value, and by combining these state values, a three-dimensional state space can be obtained.

[0053] The action space includes two actions: triggering deicing and delaying deicing. Triggering deicing means starting the electric heating deicing device immediately, and delaying deicing means not starting the deicing device temporarily and continuing to monitor data and icing probability. The reward function is used to measure the pros and cons of each action. When icing is successfully avoided and energy consumption is low, a higher positive reward is given; when icing occurs or energy consumption is too high, a negative reward is given. For example, successfully avoiding icing with low energy consumption can get a reward of +10, icing can get a penalty of -20, and excessive energy consumption can get a penalty of -5.

[0054] After the model is built, it is trained. Specifically, a large amount of training is performed in a virtual environment. By continuously trying different actions, the Q value is updated according to the feedback of the reward function, and finally converges to the optimal strategy. During the training process, the ε-greedy strategy is used to balance exploration and utilization, that is, the current optimal action is selected with a certain probability, and the action is randomly selected with a certain probability to explore more state-action combinations. The trained model can be directly deployed to the intelligent deicing system to make deicing trigger decisions based on the actions output by the model.

[0055] Step S3042, input the multimodal data and diagnostic results into the trained icing risk energy consumption game model to obtain the deicing timing, which includes triggering deicing or delaying triggering deicing. Specifically, the currently acquired diagnostic results and corresponding data can be input into the trained icing risk energy consumption game model to obtain the deicing timing. In addition, in practical applications, the system can also be verified through actual operating data to evaluate its performance under different working conditions, such as deicing effect, energy consumption level, etc. The model is further optimized according to the verification results to improve the stability and reliability of the system.

[0056] In the present invention, the de-icing timing is determined by constructing an icing risk energy consumption game model, and icing diagnosis is upgraded from a passive mode of "detection-execution" to an active mode of "prediction-optimization", which significantly reduces maintenance costs.

[0057] For the power of the preset de-icing device, the correlation function of the icing probability and meteorological data (humidity, wind speed) can be used. With the help of power dynamic adjustment technology, the nonlinear optimization of the power output of the de-icing device can be achieved by combining environmental parameters and diagnostic results. The correlation function is expressed by the following formula:

[0058] In the formula, P Indicates power, represents the probability of icing, H Indicates humidity, H 0 represents the critical humidity threshold, Indicates wind speed, Indicates the reference wind speed, k and a Represents the adjustment coefficient.

[0059] The probability of icing It reflects the current icing risk of the blade. The higher the icing probability, the greater the power output. When ≥0.8, the system determines that there is a high risk of icing and triggers high-power deicing mode 。 Represents a dynamic adjustment factor based on ambient humidity, using a Sigmoid function as a humidity adjustment item. The critical humidity threshold can be determined based on actual conditions, for example, it can be set to 80%. In this humidity adjustment item, if the current humidity exceeds the critical humidity threshold, the humidity factor rapidly approaches 1, significantly increasing power output; when it is lower than the critical humidity threshold, power is suppressed to avoid ineffective energy consumption. Ratio of wind speed to reference wind speed , as a wind speed compensation item, is used to correct the power output. For example, high wind speed can accelerate natural deicing and reduce the demand for electric deicing; low wind speed requires increased power output to compensate for the insufficient environmental deicing efficiency.

[0060] The reference wind speed is a pre-set reference value used to compare the current wind speed to determine whether the power output of the deicing device needs to be adjusted. It is usually set according to the operating parameters of the equipment and the actual application scenario, and can be a fixed value or a dynamically adjusted value. For example, suppose the reference wind speed is set to 10m / s. When the actual wind speed is higher than 10m / s, it means that the natural wind is strong and can assist in deicing. At this time, the power output of the electric deicing device can be appropriately reduced to reduce energy consumption; when the actual wind speed is lower than 10m / s, the natural wind is weak and the deicing efficiency is low. The power output needs to be increased to compensate for the lack of environmental deicing efficiency. The setting of the reference wind speed needs to comprehensively consider factors such as the performance of the equipment, the climate characteristics of the installation site, and historical operating data to ensure that efficient deicing effects can be achieved under different wind speed conditions while maximizing energy savings.

[0061] The adjustment coefficient k represents the system calibration coefficient, which is determined according to the equipment power upper limit and energy consumption limit; the adjustment coefficient The sensitivity used to control humidity regulation needs to be optimized experimentally to avoid over-response or delay.

[0062] In the present invention, the formula for determining the power of the preset deicing device realizes the construction of a dynamic game model through multi-parameter coupling of humidity, wind speed and icing probability, that is, the power determined by this formula can balance energy consumption and deicing efficiency. For example, under low humidity and high wind speed conditions, the system actively reduces power and uses natural wind to assist deicing, which reduces energy consumption by 40%-50% compared with the traditional fixed power mode. Critical humidity threshold H 0 and reference wind speed It can be adjusted dynamically according to season and geographical location. For example, in winter in cold regions, H 0 It can be adjusted down to 70% to trigger the high power mode in advance to cope with the risk of low temperature icing. The generation of relies on cross-modal feature alignment of infrared thermal imagers, vibration sensors and meteorological data, for example, improving diagnostic accuracy through timestamp synchronization and physical constraints (such as the impact of wind speed on vibration signals).

[0063] In addition, it should be noted that in practical applications, the de-icing timing can be first determined through the icing risk energy consumption game model. When the de-icing timing is to trigger de-icing, the above formula is used to determine the power of the preset de-icing device for de-icing.

[0064] The blade icing diagnosis method provided by the embodiment of the present invention combines the data of infrared thermal imagers, vibration sensors and meteorological sensors, and uses deep learning algorithms and machine learning algorithms to achieve early detection and accurate diagnosis of blade icing. In addition, the method significantly improves the accuracy and efficiency of blade icing diagnosis through multimodal data fusion analysis, cross-modal feature alignment and intelligent deicing control, while reducing energy consumption and maintenance costs.

[0065] As a specific application example of the embodiment of the present invention, the blade icing diagnosis method is implemented by the following process: 1. Multimodal data collection and cross-modal data alignment.

[0066] 1.1. Obtain data collected by infrared thermal imagers, vibration sensors, and meteorological sensors. Preprocess the blade temperature distribution image in the data collected by the infrared thermal imager, including edge detection, texture analysis, and other operations to highlight the abnormal blade temperature area. Perform time series analysis on the vibration spectrum data in the data collected by the vibration sensor to extract the characteristic frequency and amplitude changes of the vibration signal. Normalize the environmental data such as humidity, temperature, and wind speed in the data collected by the meteorological sensor so that it can be fused with other modal data.

[0067] 1.2, cross-modal data alignment. The acquired multimodal data is aligned using a dynamic alignment algorithm based on timestamps (Dynamic Time Warping with Environmental Constraints). This solves the problem of inconsistent sampling frequencies of different sensors (such as low frame rate of infrared images and high sampling rate of vibration data). At the same time, environmental parameters (such as wind speed) are introduced as alignment constraints to ensure the consistency of multimodal data in time dimension and physical mechanism.

[0068] 2. Multimodal data fusion and intelligent diagnosis of blade icing based on hybrid expert architecture (MoE). The feature data obtained by infrared thermal imager, vibration sensor and meteorological sensor are fused to extract comprehensive features, and the comprehensive features are classified using random forest classifier to output the probability value of blade icing.

[0069] 2.1,Mixed Expert Architecture Design.,The core of the Mixed Expert Architecture (MoE) is to design the feature extraction modules of different,modalities as “experts” and dynamically select the expert module that best suits the current input,through a gating network.

[0070] The expert module specifically includes the following experts: CNN expert: responsible for processing infrared thermal imager data and extracting high-dimensional spatial features of blade temperature distribution.

[0071] LSTM Expert: Processing time series data from vibration sensors and extracting dynamic features from vibration spectra.

[0072] Meteorological feature experts: process standardized environmental parameters such as temperature and humidity, and extract shallow statistical features.

[0073] Each expert module focuses on feature extraction of specific modality data, similar to different experts handling different tasks in the MoE architecture.

[0074] The gating network is a simple feed-forward neural network (FFN), whose output is calculated through the softmax function to calculate the activation probability of each expert. The most suitable expert module is dynamically selected according to the characteristics of the input data. For example, in a low temperature and high humidity environment, the gating network may be more inclined to activate meteorological feature experts. The gating network gradually optimizes the expert selection strategy by learning the feature distribution of the input data to ensure the robustness of the model in different scenarios.

[0075] The expert output features selected by the gating network are fused and input into the random forest classifier to obtain the probability of icing. Random forest not only serves as the final classifier, but also plays the role of integrating the outputs of different experts. The obtained probability of icing is compared with the classification threshold dynamically optimized according to the ambient temperature and humidity to determine the final diagnosis result.

[0076] 2.2, Credibility Scoring Mechanism. Mutual information calculation is used to assign weights according to the contribution of multimodal features, such as temperature anomaly feature weight > vibration feature weight > meteorological feature weight. The icing probability output by the random forest is combined with the feature weight to generate an explainable diagnostic result.

[0077] 3. Intelligent de-icing control strategy based on dynamic game.

[0078] 3.1, the reinforcement learning (Q-Learning) algorithm is used to build an icing risk-energy consumption game model to optimize the de-icing triggering timing. When the ambient humidity is lower than the critical value and the wind speed is high (accelerating natural de-icing), the triggering is delayed to reduce ineffective energy consumption.

[0079] The icing risk-energy consumption game model includes state space, action space and reward function. For the constructed model, a large number of trainings are carried out in a virtual environment. By constantly trying different actions, the Q value is updated according to the feedback of the reward function, and finally converges to the optimal strategy. During the training process, the ε-greedy strategy is used to balance exploration and utilization, that is, the current optimal action is selected with a certain probability, and the action is randomly selected with a certain probability to explore more state-action combinations. The trained Q-Learning model is deployed to the actual intelligent deicing system. The environmental parameters and icing probability are collected in real time through sensors, and the deicing trigger decision is made according to the optimal action output by the model. The system is verified through actual operation data to evaluate its performance under different working conditions, such as deicing effect, energy consumption level, etc. The model is further optimized according to the verification results to improve the stability and reliability of the system.

[0080] 3.2, Dynamic power adjustment.

[0081] Design the correlation function between the power output of the electric heating deicing device and the diagnosis probability value and environmental parameters (humidity, wind speed). With the help of power dynamic adjustment technology, the nonlinear optimization of the power output of the deicing device is achieved by combining environmental parameters and diagnosis results:

[0082] 4. Use the following Table 1 to evaluate the effect.

[0083] Table 1

[0084] In this embodiment, a blade icing diagnosis device is also provided, which is used to implement the above-mentioned embodiments and preferred implementations, and the descriptions that have been made will not be repeated. As used below, the term "module" can be a combination of software and / or hardware that implements a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and conceivable.

[0085] This embodiment provides a blade icing diagnosis device, such as Figure 2 As shown, including: A data acquisition module 21 is used to acquire multimodal data of the blade, where the multimodal data includes meteorological data of the environment in which the blade is located and temperature data and vibration data obtained by monitoring the blade; A data alignment module 22 is used to align the multimodal data using a dynamic alignment algorithm based on a timestamp and taking the meteorological data as a constraint item to obtain aligned multimodal data; The diagnosis module 23 is used to extract features of the aligned multimodal data using a hybrid expert architecture and output a diagnosis result of blade icing.

[0086] Specifically, the blade icing diagnostic device can be deployed in an intelligent deicing device to achieve intelligent deicing, or the diagnostic device can be deployed separately, and the diagnostic result of the diagnostic device controls the intelligent deicing device to perform deicing.

[0087] The further functional description of each of the above modules is the same as that of the above corresponding embodiments and will not be repeated here.

[0088] The present embodiment provides an intelligent deicing system, which includes: a sensor for acquiring multimodal data of blades, the multimodal data including meteorological data of the environment in which the blades are located and temperature data and vibration data obtained by monitoring the blades; an intelligent deicing device for aligning the multimodal data using a dynamic alignment algorithm based on timestamps and using meteorological data as a constraint item to obtain aligned multimodal data; extracting features of the aligned multimodal data using a hybrid expert architecture and outputting a diagnosis result of blade icing; and determining a deicing timing and a deicing power to perform deicing according to the multimodal data and the diagnosis result.

[0089] The embodiment of the present invention also provides a computer device having the above Figure 2 The diagnostic device for blade icing is shown.

[0090] See also Figure 3 , Figure 3 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Figure 3 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Figure 3 A processor 10 is taken as an example.

[0091] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0092] The memory 20 stores instructions executable by at least one processor 10, so that at least one processor 10 executes the method shown in the above embodiment.

[0093] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created by the use of a computer device based on the presentation of a small program landing page, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0094] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0095] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0096] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0097] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0098] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.

Claims

1. A method for diagnosing blade icing, characterized in that: The method comprises: Acquire multimodal data of the blade, wherein the multimodal data includes meteorological data of the environment in which the blade is located and temperature data and vibration data obtained by monitoring the blade; Using a timestamp-based dynamic alignment algorithm, the multimodal data is aligned using the meteorological data as a constraint item to obtain aligned multimodal data; A hybrid expert architecture is used to extract the features of the aligned multimodal data and output the diagnosis results of blade icing.

2. The method according to claim 1, characterized in that The method further comprises: The de-icing timing and the power of the de-icing device are preset according to the multimodal data and the diagnosis result.

3. The method according to claim 1, characterized in that: A dynamic alignment algorithm based on timestamp is used to align the multimodal data with the meteorological data as a constraint condition to obtain aligned multimodal data, including: The multimodal data are sorted according to timestamps, and based on the sorting result, data with low sampling frequency in the multimodal data are interpolated to obtain a time series corresponding to the multimodal data; Taking meteorological data as constraints, a dynamic programming algorithm is used to calculate the cumulative distance matrix between time series, and the multimodal data are aligned based on the alignment path with the minimum distance to obtain the aligned multimodal data.

4. The method according to claim 1, characterized in that: The hybrid expert architecture includes an expert module and a gating network. The expert module includes a convolutional network expert module, a long short-term memory network expert module, and a meteorological feature expert module. The hybrid expert architecture is used to extract features of the aligned multimodal data and output a diagnosis result of blade icing, including: A convolutional network expert module, a long short-term memory network expert module, and a meteorological feature expert module are used to extract the first feature of the temperature data, the second feature of the vibration data, and the third feature of the meteorological data respectively; The first feature, the second feature, and the third feature are fused using the weights assigned by the gating network and then input into the random forest classifier; The random forest classifier is used to process the fused features to obtain the diagnosis results of leaf icing.

5. The method according to claim 1, characterized in that The random forest classifier is used to process the fused features to obtain the diagnosis results of blade icing, including: The random forest classifier is used to process the fused features to obtain the probability of leaf icing; Obtaining a diagnosis result of blade icing based on a relationship between the blade icing probability and a classification threshold, wherein the classification threshold is dynamically adjusted according to the meteorological data; The blade icing probability and the feature weights of the multimodal data are weighted and summed to obtain an explainable diagnosis result.

6. The method according to claim 2, characterized in that Determining deicing timing according to the multimodal data and the diagnosis result includes: A reinforcement learning algorithm is used to build and train an icing risk energy consumption game model; The multimodal data and the diagnosis result are input into a trained icing risk energy consumption game model to obtain a deicing timing, wherein the deicing timing includes triggering deicing or delaying triggering deicing.

7. The method according to claim 2, characterized in that The diagnostic results include the probability of blade icing, the meteorological data include humidity and wind speed, and the power of the preset deicing device is determined by the following formula: In the formula, P Indicates power, represents the probability of icing, H Indicates humidity, H 0 represents the critical humidity threshold, Indicates wind speed, Indicates the reference wind speed, k and a Represents the adjustment coefficient.

8. A blade icing diagnosis device, characterized in that: The device comprises: A data acquisition module, used to acquire multimodal data of the blade, wherein the multimodal data includes meteorological data of the environment in which the blade is located and temperature data and vibration data obtained by monitoring the blade; A data alignment module, used to align the multimodal data using a timestamp-based dynamic alignment algorithm and taking the meteorological data as a constraint item to obtain aligned multimodal data; The diagnosis module is used to extract features of the aligned multimodal data using a hybrid expert architecture and output a diagnosis result of blade icing.

9. An intelligent deicing system, characterized in that: The system comprises: A sensor for acquiring multimodal data of the blade, wherein the multimodal data includes meteorological data of the environment in which the blade is located and temperature data and vibration data obtained by monitoring the blade; An intelligent deicing device is used for aligning multimodal data using a timestamp-based dynamic alignment algorithm with the meteorological data as a constraint item to obtain aligned multimodal data; extracting features of the aligned multimodal data using a hybrid expert architecture and outputting a diagnosis result of blade icing; and determining a deicing timing and a deicing power for deicing according to the multimodal data and the diagnosis result.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable a computer to execute the blade icing diagnosis method according to any one of claims 1 to 7.

Images