A training method and device of a wind turbine full-load segment anomaly detection model, equipment and medium

Abstract

The present application relates to the field of wind turbine fault diagnosis, and discloses a training method and device of a wind turbine full-load segment anomaly detection model, equipment and medium, which can determine the full-load state of the wind turbine at each historical time step according to the guaranteed power features of the wind turbine at multiple historical time steps respectively. If the full-load state of the wind turbine at the historical time step is that it should be full-load but is not, the actual power generation power is subtracted from the guaranteed power in the guaranteed power feature to obtain a power difference value, which is associated with the wind speed to obtain associated data. Each associated data and each guaranteed power feature are used for model training to obtain a trained full-load segment anomaly detection model. The trained full-load segment anomaly detection model can be used to intelligently detect the full-load anomaly of the wind turbine, reduce the human resource consumption required for technicians to manually confirm the full-load anomaly of the wind turbine, and improve the full-load anomaly diagnosis efficiency of the wind turbine.

Description

Technical Field

[0001] This invention relates to the field of wind turbine fault diagnosis, and in particular to a training method, apparatus, equipment and medium for an anomaly detection model of a wind turbine at full power output. Background Technology

[0002] With the development of science and technology, the research and application technologies of renewable energy are constantly improving.

[0003] Wind power is a renewable energy source that plays a crucial role in energy structure transformation. With the widespread application and technological advancements in wind power generation, the single-unit capacity of wind turbines continues to increase. Ensuring the safe, stable, and efficient operation of wind turbines, reducing maintenance costs, and improving power generation efficiency have become core challenges in wind farm operation and management. One issue that can arise during wind turbine operation is the possibility of lower-than-expected power generation, leading to power loss and severely impacting the safety and economic viability of the wind turbine.

[0004] Currently, the relevant technologies are generally used by technicians to diagnose whether wind turbines have abnormally low actual power generation, which requires a lot of human resources and has low efficiency in anomaly diagnosis. Summary of the Invention

[0005] This invention provides a training method, apparatus, equipment, and medium for a wind turbine full-load anomaly detection model. It addresses the shortcomings of related technologies where technical personnel are generally required to diagnose whether a wind turbine has an abnormal situation with lower actual power generation, which consumes a lot of human resources and has low anomaly diagnosis efficiency. This invention enables intelligent detection of wind turbine full-load anomaly diagnosis and improves the efficiency of full-load anomaly diagnosis.

[0006] In a first aspect, the present invention provides a training method for a wind turbine full-load section anomaly detection model, comprising:

[0007] Based on the guaranteed power characteristics of the wind turbine at multiple historical time steps, the full-load state of the wind turbine at each historical time step is determined. The full-load state is full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed. For any of the historical time steps, if the wind turbine is not at full capacity when it should be at full capacity, then the actual power generation of the wind turbine at the historical time step is subtracted from the guaranteed power in the guaranteed power feature to obtain a power difference. The power difference is then correlated with the wind speed in the guaranteed power feature to obtain correlation data. The model to be trained is obtained by using each of the associated data and the guaranteed power features of the wind turbine at each of the historical time steps to obtain a trained full-load anomaly detection model.

[0008] Optionally, the guaranteed power feature also includes rated power; The step of determining the full-load status of the wind turbine at each historical time step based on the guaranteed power characteristics of the wind turbine at multiple historical time steps includes: For any given historical time step, if the guaranteed power of the wind turbine at that historical time step is not greater than the rated power, then the wind turbine at that historical time step is determined to be not fully operational. If the guaranteed power of the wind turbine at that historical time step is greater than the rated power, and the actual power generated by the wind turbine at that historical time step is not less than the guaranteed power, then the wind turbine at that historical time step is determined to be fully operational. If the guaranteed power of the wind turbine at that historical time step is greater than the rated power, and the actual power generated by the wind turbine at that historical time step is less than the guaranteed power, then the wind turbine at that historical time step is determined to be not fully operational even though it should be.

[0009] Optionally, the step of training the model to be trained using each of the associated data and the guaranteed power features of the wind turbine at each of the historical time steps to obtain a trained full-load anomaly detection model includes: Based on the order of wind speed in each of the associated data, each of the associated data is arranged to obtain a corresponding two-dimensional data sample; Perform continuous wavelet transform on the two-dimensional data samples to obtain multiple wavelet sub-band features; For any of the historical time steps, the operation-related features, guaranteed power features and each wavelet sub-band feature of the wind turbine at the historical time step are spliced ​​together to obtain spliced ​​data and used as training samples. According to the full-load status of the wind turbine at the historical time step, the full-load status label corresponding to the training sample is set. Using each training sample and its corresponding full-slot status label, the model to be trained is trained to obtain a trained full-slot anomaly detection model.

[0010] Optionally, performing continuous wavelet transform on the two-dimensional data samples to obtain multiple wavelet sub-band features includes: Obtain the data sample length of the two-dimensional data sample; The data sample length of the two-dimensional data sample is used as the sampling frequency; and the data sample length of the two-dimensional data sample is divided by 2 to obtain the corresponding result, which is used as the scale length. Based on the selected wavelet function, the sampling frequency, and the scale length, continuous wavelet transform is performed on the two-dimensional data samples to obtain the multiple wavelet sub-band features.

[0011] Optionally, the full-release status label is used to identify whether the full-release status corresponding to the training sample is that it should be full-release but is not. The step of setting the full-load status label corresponding to the training sample based on the full-load status of the wind turbine at the historical time step includes: If the wind turbine is in a state of full power generation or not at full power generation at the historical time step, then the full power generation status label corresponding to the training sample is set as the first label for identifying normal operation. If the wind power is not at full capacity at the historical time step, then the full capacity status label corresponding to the training sample is set as a second label to identify the anomaly.

[0012] Optionally, the operation-related features include at least one of time features, impeller features, and main drive features; The time feature is a feature vector extracted from quarters and / or months; The impeller features are feature vectors extracted from the blade pitch motor temperature, blade pitch battery voltage, blade pitch battery temperature, pitch control cabinet temperature, blade pitch driver temperature, blade pitch charger temperature, blade pitch charger voltage and / or blade pitch charger current. The main drive characteristics are feature vectors extracted from generator winding temperature, generator cooling system outlet temperature, generator drive end bearing temperature, generator non-drive end bearing temperature and / or generator speed measured by the frequency converter.

[0013] Optionally, after obtaining the trained full-segment anomaly detection model, the method further includes: Obtain the relevant operating characteristics and guaranteed power characteristics of the wind turbine at the target time step; The target spliced ​​data is obtained by splicing the operation-related features, guaranteed power features, and wavelet sub-band features of the wind turbine at the target time step; The target spliced ​​data is input into the trained full-segment anomaly detection model for detection, and the detection result output by the full-segment anomaly detection model is obtained.

[0014] Secondly, the present invention provides a training device for a wind turbine full-load section anomaly detection model, comprising: The determining unit is used to determine the full-load status of the wind turbine in each historical time step based on the guaranteed power characteristics of the wind turbine in multiple historical time steps. The full-load status is full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed. The difference unit is used for the guaranteed power feature of any historical time step. If the wind turbine is in a full-load state at the historical time step but is not at full load, the actual power generated by the wind turbine at the historical time step is subtracted from the guaranteed power in the guaranteed power feature to obtain the power difference. The difference unit is used to correlate the power difference with the wind speed in the guaranteed power feature to obtain correlated data; The training unit is used to train the model to be trained using each of the associated data and the guaranteed power features of the wind turbine at each of the historical time steps, so as to obtain a trained full-load anomaly detection model.

[0015] Thirdly, the present invention provides a computer device, including: 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 computer instructions to perform the training method for the wind turbine full-load section anomaly detection model described in the first aspect or any corresponding embodiment.

[0016] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the training method for a wind turbine full-load section anomaly detection model according to the first aspect or any corresponding embodiment described above.

[0017] The training method, apparatus, equipment, and medium for a wind turbine full-load anomaly detection model provided by this invention can determine the full-load state of the wind turbine at each historical time step based on the guaranteed power characteristics of the wind turbine at multiple historical time steps. The full-load state can be full-load, not full-load, or should be full-load but not full-load. The guaranteed power characteristics include guaranteed power and wind speed. For any historical time step's guaranteed power characteristics, if the wind turbine's full-load state at that historical time step is should be full-load but not full-load, the actual power generated by the wind turbine at that historical time step is subtracted from the guaranteed power in the guaranteed power characteristics to obtain a power difference. This power difference is then correlated with the wind speed in the guaranteed power characteristics to obtain correlation data. Each correlation data and the guaranteed power characteristics of the wind turbine at each historical time step are used to train the model to be trained, resulting in a trained full-load anomaly detection model. This invention can use a trained full-load anomaly detection model to detect whether a wind turbine is in a state of not being at full load at a certain time step, thereby determining whether there is a full-load anomaly in the wind turbine, realizing intelligent detection of full-load anomalies in wind turbines, reducing the human resource consumption required for technicians to manually confirm full-load anomalies in wind turbines, and improving the efficiency of diagnosing full-load anomalies in wind turbines. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in this invention or related technologies, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 A flowchart illustrating a training method for a wind turbine full-load section anomaly detection model provided in an embodiment of the present invention; Figure 2 This is a schematic diagram illustrating the wavelet frequency and amplitude changes when the full-power output of a wind turbine is low, as provided in an embodiment of the present invention. Figure 3 This is a schematic diagram illustrating the wavelet frequency and amplitude changes when the full power output of a wind turbine is low, as provided in another embodiment of the present invention. Figure 4 This is a schematic diagram of the wavelet frequency and amplitude changes when the wind turbine is operating at full power, as provided in an embodiment of the present invention. Figure 5 A flowchart illustrating another method for training a wind turbine full-load section anomaly detection model provided in an embodiment of the present invention; Figure 6 A schematic diagram of the structure of a training device for an anomaly detection model of a wind turbine at full power output provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of the structure of a computer device provided in an embodiment of the present invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0021] The following is combined Figures 1-5 The training method of the wind turbine full-load section anomaly detection model of the present invention is described.

[0022] like Figure 1 As shown in the figure, this embodiment proposes a training method for a first-type anomaly detection model for wind turbines operating at full capacity. This method may include the following steps: S101. Based on the guaranteed power characteristics of the wind turbine at multiple historical time steps, determine the full-load status of the wind turbine at each historical time step. The full-load status is full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed.

[0023] Among them, the historical time step is a certain time step in the historical period, which can be a certain moment or a certain duration in the historical period.

[0024] Specifically, the aforementioned multiple historical time steps can be consecutive time steps within a historical period, such as a moment every 5 minutes within a historical period.

[0025] Specifically, this embodiment can obtain the guaranteed power characteristics of the wind turbine at each of the aforementioned historical time steps. The guaranteed power characteristics of the wind turbine at a certain historical time step include: the guaranteed power of the wind turbine at that historical time step and the wind speed.

[0026] Optionally, the power characteristics may also include rated power. In this case, step S101 includes: For any historical time step, if the guaranteed power of the wind turbine at the historical time step is not greater than the rated power, then the wind turbine at the historical time step is determined to be not fully powered. If the guaranteed power of the wind turbine at the historical time step is greater than the rated power, and the actual power generated by the wind turbine at the historical time step is not less than the guaranteed power, then the wind turbine at the historical time step is determined to be fully powered. If the guaranteed power of the wind turbine at the historical time step is greater than the rated power, and the actual power generated by the wind turbine at the historical time step is less than the guaranteed power, then the wind turbine at the historical time step is determined to be not fully powered if it should have been fully powered.

[0027] Specifically, this embodiment can obtain the guaranteed power, rated power, and wind speed of the wind turbine at each of the above historical time steps.

[0028] Specifically, in this embodiment, if the guaranteed power of the wind turbine at a certain historical time step is not greater than the rated power of the wind turbine at that historical time step, it can be determined that the wind turbine is not operating at full capacity at that historical time step. If the guaranteed power of the wind turbine at a certain historical time step is greater than the rated power, it can be determined that the wind turbine should be operating at full capacity at that historical time step. If the actual power generated by the wind turbine at that historical time step is not less than the guaranteed power, it can be further determined that the wind turbine is operating at full capacity at that historical time step. If the actual power generated by the wind turbine at that historical time step is less than the guaranteed power, it can be further determined that the wind turbine is operating at full capacity but not at full capacity at that historical time step.

[0029] S102. For any historical time step's guaranteed power characteristic, if the wind turbine is in a state of full power generation at the historical time step but is not at full power generation, then the actual power generation of the wind turbine at the historical time step is subtracted from the guaranteed power in the guaranteed power characteristic to obtain the power difference.

[0030] Specifically, when a wind turbine is not at full capacity at a certain historical time step, this embodiment can subtract the guaranteed power from the actual power generated by the wind turbine at that historical time step to obtain the power difference of the wind turbine at that historical time step.

[0031] S103. Correlate the power difference with the wind speed in the guaranteed power characteristics to obtain the correlation data.

[0032] Specifically, when a wind turbine is not operating at full capacity at a certain historical time step, this embodiment can calculate the power difference of the wind turbine at that historical time step and correlate the wind speed of the wind turbine at that historical time step with the power difference to obtain the corresponding correlation data.

[0033] It is understandable that for any historical time step in which a wind turbine is operating at full capacity but not at full capacity, this embodiment can calculate the corresponding power difference and obtain the corresponding associated data.

[0034] S104. Use each associated data and the guaranteed power features of the wind turbine at each historical time step to train the model to be trained, so as to obtain a trained full-load section anomaly detection model.

[0035] Specifically, in this embodiment, the model to be trained can be trained based on each associated data and the guaranteed power characteristics of the wind turbine at each of the above historical time steps to obtain a trained full-load anomaly detection model.

[0036] The model to be trained can be selected by technicians according to actual needs, such as choosing a decision tree as the model to be trained. This embodiment does not impose any restrictions.

[0037] Optionally, in the training method of other wind turbine full-load section anomaly detection models proposed in this embodiment, step S104 includes: Based on the order of wind speed in each associated data, each associated data is arranged to obtain the corresponding two-dimensional data sample; Continuous wavelet transform is performed on two-dimensional data samples to obtain multiple wavelet sub-band features; For any historical time step, the operation-related features, guaranteed power features and wavelet subband features of the wind turbine at the historical time step are spliced ​​together to obtain spliced ​​data and used as training samples. According to the full-load status of the wind turbine at the historical time step, the full-load status label corresponding to the training sample is set. The model to be trained is trained using each training sample and its corresponding full-slot state label to obtain a trained full-slot anomaly detection model.

[0038] Specifically, in this embodiment, all associated data can be sorted according to the order of wind speed in the associated data to obtain the corresponding arrangement result, i.e., a two-dimensional data sample.

[0039] Optionally, the above continuous wavelet transform of the two-dimensional data samples yields multiple wavelet sub-band features, including: Obtain the data sample length of the two-dimensional data sample; The sampling frequency is taken as the data sample length of the two-dimensional data sample; and the sampling frequency is taken as the scale length by dividing the data sample length of the two-dimensional data sample by 2. Based on the selected wavelet function, sampling frequency, and scale length, continuous wavelet transform is performed on the two-dimensional data samples to obtain multiple wavelet sub-band features.

[0040] Specifically, in this embodiment, the wavelet function cmor1.5-2 can be selected to perform continuous wavelet transform on the two-dimensional data samples, decomposing the two-dimensional data samples into multiple wavelet sub-band features. Each wavelet sub-band feature includes a coefficient of positive correlation with wind speed, a frequency coefficient, and an amplitude coefficient.

[0041] like Figure 2 and Figure 3 The diagram shows the wavelet frequency and amplitude variations when the wind turbine's full-power output is low. Figure 4 The figure shows a schematic diagram of the wavelet frequency and amplitude changes when the wind turbine is operating at full power.

[0042] Optionally, the operation-related features include at least one of time features, impeller features, and main drive features; The time features are feature vectors extracted from quarters and / or months; The impeller features are feature vectors extracted from the blade pitch motor temperature, blade pitch battery voltage, blade pitch battery temperature, pitch control cabinet temperature, blade pitch driver temperature, blade pitch charger temperature, blade pitch charger voltage and / or blade pitch charger current. The main drive characteristics are feature vectors extracted from generator winding temperature, generator cooling system outlet temperature, generator drive end bearing temperature, generator non-drive end bearing temperature and / or generator speed measured by frequency converter.

[0043] Specifically, in this embodiment, for any of the aforementioned historical time steps, the operation-related features, guaranteed power features, and wavelet sub-band features of the wind turbine at that historical time step are spliced ​​together to obtain spliced ​​data, which is then used as training samples. In this way, this embodiment can create training samples corresponding to each historical time step.

[0044] like Figure 5As shown, this embodiment can collect wind speed characteristics, time characteristics, rotor characteristics, and main drive data from the Supervisory Control and Data Acquisition (SCADA) system of the wind turbine, and then extract the corresponding feature vectors. This embodiment performs continuous wavelet transform on the difference data existing in the full-load section to obtain multiple corresponding wavelet sub-band features, i.e., wavelet coefficient features. Then, the feature vectors extracted from the SCADA data and the wavelet coefficient features are concatenated, and the concatenated result is input into a trained decision tree model, i.e., the full-load section anomaly detection model, for detection.

[0045] Optionally, the full-load status label is used to identify whether the full-load status corresponding to the training sample is one that should be at full load but is not. In this case, the above-mentioned setting of the full-load status label corresponding to the training sample based on the full-load status of the wind turbine at a historical time step includes: If the wind turbine is at full power or not at full power in the historical time step, then set the full power status label corresponding to the training sample as the first label to identify normal operation. If the wind power is not at full capacity when it should be at full capacity in a historical time step, then the full capacity status label corresponding to the training sample is set as the second label to identify the anomaly.

[0046] The first and second labels can be identifiers composed of numbers, letters, or Chinese characters. For example, the first label can be set to 1 and the second label can be set to 0.

[0047] Specifically, in this embodiment, a full-spinning status label can be set for each training sample. The training model to be trained is then trained using each training sample with the set full-spinning status label to obtain a trained full-spinning segment anomaly detection model.

[0048] It is understood that this embodiment can use a trained full-load anomaly detection model to detect whether the wind turbine is in a state of not being at full load at a certain time step, thereby determining whether the wind turbine has a full-load anomaly and realizing intelligent detection of the wind turbine's full-load anomaly.

[0049] Optionally, in the training method of other wind turbine full-load section anomaly detection models proposed in this embodiment, after obtaining the trained full-load section anomaly detection model as described above, the method further includes: Obtain the relevant operating characteristics and guaranteed power characteristics of the wind turbine at the target time step; The target spliced ​​data is obtained by splicing the relevant operating characteristics, guaranteed power characteristics, and wavelet subband characteristics of the wind turbine at the target time step. The target spliced ​​data is input into the trained full-segment anomaly detection model for detection, and the detection results output by the full-segment anomaly detection model are obtained.

[0050] The target time step can be a historical time step, the current time step, or a future time step to be predicted; this embodiment does not limit this.

[0051] It is understandable that the detection results output by the full-range anomaly detection model are either normal or abnormal.

[0052] The training method for the wind turbine full-load anomaly detection model proposed in this embodiment can determine the full-load status of the wind turbine at each historical time step based on the guaranteed power characteristics of the wind turbine at multiple historical time steps. The full-load status is categorized as full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed. For any historical time step's guaranteed power characteristics, if the wind turbine's full-load status at that historical time step is should be full load but not full load, the actual power generated by the wind turbine at that historical time step is subtracted from the guaranteed power in the guaranteed power characteristics to obtain a power difference. This power difference is then correlated with the wind speed in the guaranteed power characteristics to obtain correlation data. Each correlation data and the guaranteed power characteristics of the wind turbine at each historical time step are used to train the model to obtain a trained full-load anomaly detection model. This embodiment can use a trained full-load anomaly detection model to detect whether the wind turbine is in a state of not being at full load at a certain time step, thereby determining whether the wind turbine has a full-load anomaly, realizing intelligent detection of wind turbine full-load anomalies, reducing the human resource consumption required for technicians to manually confirm wind turbine full-load anomalies, and improving the efficiency of diagnosing wind turbine full-load anomalies.

[0053] like Figure 6 As shown in the figure, this embodiment proposes a training device for an anomaly detection model of a wind turbine during full-load operation. The device may include: The determining unit 101 is used to determine the full-load status of the wind turbine in each historical time step based on the guaranteed power characteristics of the wind turbine in multiple historical time steps. The full-load status is full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed. The difference unit 102 is used for the guaranteed power characteristic of any historical time step. If the wind turbine is in the full-power state of the historical time step but is not in the full-power state, the actual power generation of the wind turbine in the historical time step is subtracted from the guaranteed power in the guaranteed power characteristic to obtain the power difference. The association unit 103 is used to associate the power difference with the wind speed in the guaranteed power characteristics to obtain association data; Training unit 104 is used to train the model to be trained using each associated data and the guaranteed power features of the wind turbine at each historical time step to obtain a trained full-load anomaly detection model.

[0054] It should be noted that the processing procedures of the determination unit 101, the difference unit 102, the association unit 103, and the training unit 104, and their beneficial effects, can be found in the following references: Figure 1 Steps S101 to S104 in the process will not be described again.

[0055] Optionally, the power characteristics may also include rated power; The determining unit 101 is also used for: For any historical time step, if the guaranteed power of the wind turbine at the historical time step is not greater than the rated power, then the wind turbine at the historical time step is determined to be not fully powered. If the guaranteed power of the wind turbine at the historical time step is greater than the rated power, and the actual power generated by the wind turbine at the historical time step is not less than the guaranteed power, then the wind turbine at the historical time step is determined to be fully powered. If the guaranteed power of the wind turbine at the historical time step is greater than the rated power, and the actual power generated by the wind turbine at the historical time step is less than the guaranteed power, then the wind turbine at the historical time step is determined to be not fully powered if it should have been fully powered.

[0056] Optionally, training unit 104 is also used for: Based on the order of wind speed in each associated data, each associated data is arranged to obtain the corresponding two-dimensional data sample; Continuous wavelet transform is performed on two-dimensional data samples to obtain multiple wavelet sub-band features; For any historical time step, the operation-related features, guaranteed power features and wavelet subband features of the wind turbine at the historical time step are spliced ​​together to obtain spliced ​​data and used as training samples. According to the full-load status of the wind turbine at the historical time step, the full-load status label corresponding to the training sample is set. The model to be trained is trained using each training sample and its corresponding full-slot state label to obtain a trained full-slot anomaly detection model.

[0057] Optionally, training unit 104 is also used for: Obtain the data sample length of the two-dimensional data sample; The sampling frequency is taken as the data sample length of the two-dimensional data sample; and the sampling frequency is taken as the scale length by dividing the data sample length of the two-dimensional data sample by 2. Based on the selected wavelet function, sampling frequency, and scale length, continuous wavelet transform is performed on the two-dimensional data samples to obtain multiple wavelet sub-band features.

[0058] Optionally, the full-release status label is used to identify whether the full-release status corresponding to the training sample is that it should be full-released but is not. Training unit 104 is also used for: If the wind turbine is at full power or not at full power in the historical time step, then set the full power status label corresponding to the training sample as the first label to identify normal operation. If the wind power is not at full capacity when it should be at full capacity in a historical time step, then the full capacity status label corresponding to the training sample is set as the second label to identify the anomaly.

[0059] Optionally, the operation-related features include at least one of time features, impeller features, and main drive features; The time features are feature vectors extracted from quarters and / or months; The impeller features are feature vectors extracted from the blade pitch motor temperature, blade pitch battery voltage, blade pitch battery temperature, pitch control cabinet temperature, blade pitch driver temperature, blade pitch charger temperature, blade pitch charger voltage and / or blade pitch charger current. The main drive characteristics are feature vectors extracted from generator winding temperature, generator cooling system outlet temperature, generator drive end bearing temperature, generator non-drive end bearing temperature and / or generator speed measured by frequency converter.

[0060] Optionally, the above-mentioned device further includes: The detection unit is used to acquire the operation-related features and guaranteed power features of the wind turbine at the target time step after obtaining the trained full-load anomaly detection model; to splice the operation-related features, guaranteed power features and each wavelet sub-band features of the wind turbine at the target time step to obtain the target spliced ​​data; and to input the target spliced ​​data into the trained full-load anomaly detection model for detection to obtain the detection results output by the full-load anomaly detection model.

[0061] The training device for the wind turbine full-load anomaly detection model proposed in this embodiment can determine the full-load status of the wind turbine at each historical time step based on the guaranteed power characteristics of the wind turbine at multiple historical time steps. The full-load status is categorized as full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed. For any historical time step's guaranteed power characteristics, if the wind turbine's full-load status at that historical time step is should be full load but not full load, the actual power generated by the wind turbine at that historical time step is subtracted from the guaranteed power in the guaranteed power characteristics to obtain a power difference. This power difference is then correlated with the wind speed in the guaranteed power characteristics to obtain correlation data. Each correlation data and the guaranteed power characteristics of the wind turbine at each historical time step are used to train the model to obtain a trained full-load anomaly detection model. This invention can use a trained full-load anomaly detection model to detect whether a wind turbine is in a state of not being at full load at a certain time step, thereby determining whether there is a full-load anomaly in the wind turbine, realizing intelligent detection of full-load anomalies in wind turbines, reducing the human resource consumption required for technicians to manually confirm full-load anomalies in wind turbines, and improving the efficiency of diagnosing full-load anomalies in wind turbines.

[0062] In this embodiment, the training device for the wind turbine full-load section anomaly detection model is presented in the form of functional units. Here, a unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that execute one or more software or fixed programs, and / or other devices that can provide the above functions.

[0063] This invention also provides a computer device having the above-described features. Figure 6 The training device shown is for the anomaly detection model of the wind turbine at full power output.

[0064] Please see Figure 7 The present invention provides a schematic diagram of the structure of a computer device according to an optional embodiment. The computer device includes one or more processors 10, a memory 20, and interfaces for connecting the various components, including high-speed interfaces and low-speed interfaces. The various components are interconnected via different buses and can be mounted on a common motherboard or otherwise installed as needed. The processors can process instructions executed within the computer device, including instructions stored in or on memory to display graphical information of a GUI on an external input / output device (such as a display device coupled to the interface). In some optional embodiments, multiple processors and / or multiple buses can be used with multiple memories, if desired. Similarly, multiple computer devices can be connected, each providing some of the necessary operations (e.g., as a server array, a group of blade servers, or a multiprocessor system). Figure 7 Take a processor 10 as an example.

[0065] Processor 10 may be a central processing unit, a network processor, or a combination thereof. Processor 10 may further include a hardware chip. The hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD), or a combination thereof. The programmable logic device may be a complex programmable logic device (CAMP), a field-programmable gate array (FPGA), a general-purpose array logic (GDA), or any combination thereof.

[0066] The memory 20 stores instructions executable by at least one processor 10 to cause at least one processor 10 to perform the method shown in the above embodiments.

[0067] The memory 20 may include a program storage area and a data storage area. The program storage area may store the operating system and applications required for at least one function. The data storage area may store data created based on the use of the computer device. Furthermore, the memory 20 may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some alternative embodiments, the memory 20 may optionally include memory remotely located relative to the processor 10, which can be connected to the computer device via a network. Examples of such networks include, but are not limited to, the Internet, intranets, local area networks, mobile communication networks, and combinations thereof.

[0068] Memory 20 may include volatile memory, such as random access memory. Memory may also include non-volatile memory, such as flash memory, hard disk, or solid-state drive. Memory 20 may also include combinations of the above types of memory.

[0069] The computer device also includes a communication interface 30 for communicating with other devices or communication networks.

[0070] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code, which, when accessed and executed by the computer, processor, or hardware, implements the methods shown in the above embodiments.

[0071] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for training a wind turbine full-power segment anomaly detection model, characterized in that, include: Based on the guaranteed power characteristics of the wind turbine at multiple historical time steps, the full-load state of the wind turbine at each historical time step is determined. The full-load state is full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed. For any of the historical time steps, if the wind turbine is not at full capacity when it should be at full capacity, then the actual power generation of the wind turbine at the historical time step is subtracted from the guaranteed power in the guaranteed power feature to obtain a power difference. The power difference is then correlated with the wind speed in the guaranteed power feature to obtain correlation data. The model to be trained is obtained by using each of the associated data and the guaranteed power features of the wind turbine at each of the historical time steps to obtain a trained full-load anomaly detection model.

2. The method of claim 1, wherein, The guaranteed power feature also includes rated power; The step of determining the full-load status of the wind turbine at each historical time step based on the guaranteed power characteristics of the wind turbine at multiple historical time steps includes: For any given historical time step, if the guaranteed power of the wind turbine at that historical time step is not greater than the rated power, then the wind turbine at that historical time step is determined to be not fully operational. If the guaranteed power of the wind turbine at that historical time step is greater than the rated power, and the actual power generated by the wind turbine at that historical time step is not less than the guaranteed power, then the wind turbine at that historical time step is determined to be fully operational. If the guaranteed power of the wind turbine at that historical time step is greater than the rated power, and the actual power generated by the wind turbine at that historical time step is less than the guaranteed power, then the wind turbine at that historical time step is determined to be not fully operational even though it should be.

3. The method of claim 2, wherein, The step of training the model to be trained using each of the associated data and the guaranteed power features of the wind turbine at each of the historical time steps to obtain a trained full-load anomaly detection model includes: Based on the order of wind speed in each of the associated data, each of the associated data is arranged to obtain a corresponding two-dimensional data sample; Perform continuous wavelet transform on the two-dimensional data samples to obtain multiple wavelet sub-band features; For any of the historical time steps, the operation-related features, guaranteed power features and each wavelet sub-band feature of the wind turbine at the historical time step are spliced ​​together to obtain spliced ​​data and used as training samples. According to the full-load status of the wind turbine at the historical time step, the full-load status label corresponding to the training sample is set. Using each training sample and its corresponding full-slot status label, the model to be trained is trained to obtain a trained full-slot anomaly detection model.

4. The method of claim 3, wherein, The continuous wavelet transform of the two-dimensional data samples yields multiple wavelet sub-band features, including: Obtain the data sample length of the two-dimensional data sample; The data sample length of the two-dimensional data sample is used as the sampling frequency; and the data sample length of the two-dimensional data sample is divided by 2 to obtain the corresponding result, which is used as the scale length. Based on the selected wavelet function, the sampling frequency, and the scale length, continuous wavelet transform is performed on the two-dimensional data samples to obtain the multiple wavelet sub-band features.

5. The method of claim 3, wherein, The full-release status label is used to identify whether the full-release status corresponding to the training sample is that it should be full-released but was not. The step of setting the full-load status label corresponding to the training sample based on the full-load status of the wind turbine at the historical time step includes: If the wind turbine is in a state of full power generation or not at full power generation at the historical time step, then the full power generation status label corresponding to the training sample is set as the first label for identifying normal operation. If the wind power is not at full capacity at the historical time step, then the full capacity status label corresponding to the training sample is set as a second label to identify the anomaly.

6. The method of claim 3, wherein, The operation-related features include at least one of time features, impeller features, and main drive features; The time feature is a feature vector extracted from quarters and / or months; The impeller features are feature vectors extracted from the blade pitch motor temperature, blade pitch battery voltage, blade pitch battery temperature, pitch control cabinet temperature, blade pitch driver temperature, blade pitch charger temperature, blade pitch charger voltage and / or blade pitch charger current. The main drive characteristics are feature vectors extracted from generator winding temperature, generator cooling system outlet temperature, generator drive end bearing temperature, generator non-drive end bearing temperature and / or generator speed measured by the frequency converter.

7. The method of claim 3, wherein, After obtaining the trained full-segment anomaly detection model, the method further includes: Obtain the relevant operating characteristics and guaranteed power characteristics of the wind turbine at the target time step; The target spliced ​​data is obtained by splicing the operation-related features, guaranteed power features, and wavelet sub-band features of the wind turbine at the target time step; The target spliced ​​data is input into the trained full-segment anomaly detection model for detection, and the detection result output by the full-segment anomaly detection model is obtained.

8. A training device of a wind turbine generator full power segment anomaly detection model, characterized in that, include: The determining unit is used to determine the full-load status of the wind turbine in each historical time step based on the guaranteed power characteristics of the wind turbine in multiple historical time steps. The full-load status is full load, not full load, or should be full load but not full load. The guaranteed power characteristics include guaranteed power and wind speed. The difference unit is used for the guaranteed power feature of any historical time step. If the wind turbine is in a full-load state at the historical time step but is not at full load, the actual power generated by the wind turbine at the historical time step is subtracted from the guaranteed power in the guaranteed power feature to obtain the power difference. The difference unit is used to correlate the power difference with the wind speed in the guaranteed power feature to obtain correlated data; The training unit is used to train the model to be trained using each of the associated data and the guaranteed power features of the wind turbine at each of the historical time steps, so as to obtain a trained full-load anomaly detection model.

9. A computer device, comprising: include: The system includes a memory and a processor, which are interconnected. The memory stores computer instructions, and the processor executes the computer instructions to perform the training method for the wind turbine full-load section anomaly detection model according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer instructions for causing the computer to execute the training method of the wind turbine full-load section anomaly detection model according to any one of claims 1 to 7.

Images