An extreme weather power generation equipment fault prediction method based on a time sequence large model

Abstract

The present application relates to the technical field of generator fault prediction, and particularly relates to an extreme weather power generation equipment fault prediction method based on a time sequence large model, which comprises the following steps: multi-parameter synchronous acquisition; abnormal event determination; trend type identification; risk level assessment; dynamic threshold adjustment; risk sequence prediction and early warning. By introducing the three-axis vibration peak value of the middle part of the blade, the oil viscosity of the gearbox, the current harmonic of the stator, the starting torque increment of the yaw system, the wind speed change rate of the top of the cabin and the temperature rise rate of the main shaft bearing, multiple parameters closely related to structural load, electrical disturbance and thermal-lubrication state are realized, full-chain monitoring of the operation state of the wind turbine under strong wind extreme weather is realized, and the problem that the fault early warning is not timely and the unit shutdown loss is intensified due to the lag of the lubrication abnormal trend identification under extreme weather caused by the incomplete monitoring parameter coverage and the fixed threshold setting is effectively solved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of generator fault prediction, in particular to an extreme weather power generation equipment fault prediction method based on a time sequence large model. BACKGROUND

[0002] With the continuous growth of wind power installed capacity, the operation stability of wind power generation equipment under extreme weather conditions has become increasingly prominent. In particular, during strong winds, typhoons and other extreme weather processes, wind turbines are affected by aerodynamic loads, structural responses and electrical disturbances, and are prone to induce lubrication degradation, component fatigue and system instability.

[0003] Traditional wind power operation and maintenance methods are mostly based on single-point monitoring or fixed threshold judgment, which cannot dynamically perceive the coordinated change trend of multiple parameters, and cannot meet the demand for early fault trend identification and early warning under extreme environments. In recent years, the development of time sequence prediction models has provided a new means for time domain evolution analysis of multi-source monitoring data, but it is still necessary to build a monitoring parameter system that matches the actual unit structure response and introduce a dynamic adjustment mechanism to adapt to the strong nonlinearity and rapid evolution of extreme wind operation scenarios. Therefore, it is urgent to propose a power generation equipment fault prediction method that can integrate key structural feature parameters, dynamic judgment mechanism and high-precision time sequence prediction capability to realize early identification and risk warning of fault trends under extreme weather conditions. SUMMARY

[0004] Therefore, the application provides an extreme weather power generation equipment fault prediction method based on a time sequence large model to overcome the problem that the existing technology causes the fault warning to be not timely and the unit shutdown loss to be aggravated due to the incomplete monitoring parameter coverage and the fixed threshold setting, which leads to the lag of lubrication abnormal trend identification under extreme weather.

[0005] To achieve the above purpose, the application provides an extreme weather power generation equipment fault prediction method based on a time sequence large model, comprising:

[0006] Real-time synchronous acquisition of three-axis vibration peak values of a blade middle portion of each wind turbine, oil viscosity of a gear box input shaft, current harmonic content of a generator stator, yaw system startup torque increment, wind speed variation rate of a cabin top portion and temperature rise rate of a main shaft bearing during strong wind passage;

[0007] According to the three-axis vibration peak values in a preset abnormal judgment period and a preset gust impact threshold, it is determined that an abnormal event exists, and an abnormal judgment result is obtained;

[0008] Based on the abnormal judgment result, according to the oil viscosity change rate and the temperature rise rate, it is determined that the type of the abnormal event is a lubrication abnormal trend, and a type judgment result is obtained;

[0009] Based on the type determination result, according to each of the current harmonic content, the three-axis vibration peak value and the preset risk evolution threshold, the risk level of the fault trend is determined as a serious warning, and a risk determination result is obtained;

[0010] Based on the risk determination result, the preset gust impact threshold is adjusted according to the starting torque increment and the wind speed change rate in the next preset first adjustment period, and the risk evolution threshold is adjusted according to the number of times of adjusting the gust impact threshold and the oil viscosity in the next preset second adjustment period;

[0011] According to all the risk determination results obtained after adjusting the risk evolution threshold in a preset prediction period, the three-axis vibration peak value and a preset time sequence model, a risk prediction sequence is generated, and a warning alarm is sent according to the time stamp of the risk prediction sequence.

[0012] Further, according to the oil viscosity change rate and the temperature rise rate, the type of the abnormal event is determined as a lubrication abnormal trend, and the process of obtaining a type determination result includes:

[0013] According to the oil viscosity change rate and the temperature rise rate, the change correlation amplitude is determined in a preset abnormal trend determination period;

[0014] According to the comparison result of the change correlation amplitude and the preset correlation amplitude threshold, the type of the abnormal event is determined as a lubrication abnormal trend, and the type determination result is obtained.

[0015] Further, according to each of the current harmonic content, the three-axis vibration peak value and the preset risk evolution threshold of each wind turbine, the risk level of the fault trend is determined as a serious warning, and the process includes:

[0016] According to the current harmonic content from the initial time to each time in a preset level determination period, a harmonic content fluctuation set is determined;

[0017] According to the harmonic content fluctuation set of each of any two adjacent wind turbines, a number of concerned wind turbines are determined from the wind turbine generator unit;

[0018] According to the three-axis vibration peak value and the preset risk evolution threshold of each of any two adjacent concerned wind turbines, the risk level of the fault trend is determined as a serious warning.

[0019] Further, according to the harmonic content fluctuation set of each of any two adjacent wind turbines, a number of concerned wind turbines are determined from the wind turbine generator unit, and the process includes:

[0020] According to two harmonic content fluctuation sets, a fluctuation consistency degree is determined;

[0021] determining several concerned generators from the wind turbine based on a comparison result of the fluctuation consistency degree and a preset consistency degree threshold.

[0022] Further, the process of determining the risk level of the failure trend as a serious warning according to the three-axis vibration peak values of each arbitrary adjacent two concerned generators and a preset risk evolution threshold includes:

[0023] determining several peak value change values according to the three-axis vibration peak values of each arbitrary adjacent two concerned generators;

[0024] determining a peak value change fluctuation value according to all the peak value change values;

[0025] determining the risk level of the failure trend as a serious warning according to a comparison result of the peak value change fluctuation value and the preset risk evolution threshold.

[0026] Further, the process of adjusting the preset gust impact threshold according to the start torque increment and the wind speed change rate in a next preset first adjustment period includes:

[0027] determining an adjustment synchronization degree according to all the start torque increments and all the wind speed change rates;

[0028] adjusting the preset gust impact threshold according to a comparison result of the adjustment synchronization degree and a preset adjustment synchronization threshold.

[0029] Further, the process of adjusting the risk evolution threshold according to the number of times of adjusting the gust impact threshold and the oil viscosity in a next preset second adjustment period includes:

[0030] determining an adjustment times fluctuation value according to all the number of times of adjusting the gust impact threshold, and determining an oil viscosity fluctuation value according to all the oil viscosities;

[0031] adjusting the risk evolution threshold according to the adjustment times fluctuation value and the oil viscosity fluctuation value.

[0032] Further, the process of adjusting the risk evolution threshold according to the adjustment times fluctuation value and the oil viscosity fluctuation value includes:

[0033] determining a times fluctuation deviation according to the adjustment times fluctuation value and a preset times fluctuation threshold, and determining a viscosity fluctuation deviation according to the oil viscosity fluctuation value and a preset viscosity fluctuation threshold;

[0034]

[0035] adjusting the risk evolution threshold according to a comparison result of the times fluctuation deviation and a preset times fluctuation deviation threshold, and a comparison result of the viscosity fluctuation deviation and a preset viscosity fluctuation deviation threshold.​

[0036] Further, the process of issuing a pre-warning alarm according to the time stamp of the risk prediction sequence comprises:

[0037] determining a time distribution degree according to all time stamps;

[0038] determining that the wind turbine is in failure according to a comparison result of the time distribution degree and a preset distribution degree threshold, and issuing the pre-warning alarm.

[0039] Further, the process of determining an abnormal event according to the three-axis vibration peak value in a preset abnormality determination period and a preset gust impact threshold to obtain an abnormality determination result comprises:

[0040] determining an abnormal distribution degree according to a comparison result of the three-axis vibration peak value and a preset vibration peak value threshold;

[0041] determining that the abnormal event exists according to a comparison result of the abnormal distribution degree and the preset gust impact threshold to obtain the abnormality determination result.

[0042] Compared with the prior art, the beneficial effects of the present application are that by introducing the three-axis vibration peak value in the middle of the blade, the oil viscosity of the gearbox, the current harmonic of the stator, the starting torque increment of the yaw system, the wind speed change rate at the top of the cabin and the temperature rise rate of the main shaft bearing, multiple parameters closely related to structural load, electrical disturbance and thermal-lubrication state are realized. The full-chain monitoring of the running state of the wind turbine under strong wind extreme weather is realized. The gust impact directly causes the blade vibration to intensify, which is conducted to the gearbox to cause the oil viscosity fluctuation, and at the same time, the temperature rise rate changes. Lubrication degradation further causes the component friction to intensify, which is reflected in the changes of the yaw torque and the stator current harmonic. The risk level evaluation comprehensively considers the coupling trend of electrical and mechanical signals, and further drives the dynamic threshold adjustment to realize the adaptive optimization of the model. Finally, a continuous failure risk prediction sequence is generated through the time sequence large model, which can accurately judge the trend and give early warning under the condition of multi-source coupled disturbance, improve the failure perception sensitivity and operation safety of the wind power equipment, and effectively solve the problem that the lubrication abnormal trend under extreme weather is not recognized in time due to incomplete monitoring parameter coverage and fixed threshold setting, thereby causing the failure warning to be not timely and the unit shutdown loss to be intensified.

[0043] Further, by normalizing the oil viscosity change rate and the main shaft bearing temperature rise rate (prior art, not described again), the two types of data are analyzed on a unified scale, thereby improving the measurement accuracy of the consistency of their trend changes. The Pearson correlation coefficient is used as a measure of the change correlation strength, which helps to accurately determine whether the lubrication performance degradation is synchronized with the temperature rise. When the two are highly positively correlated, it means that the decline in lubrication performance has had a significant impact on heat conduction and mechanical friction, showing the typical characteristics of lubrication degradation leading to increased temperature rise. This method effectively reveals the linkage evolution law between the lubrication state and thermal effect in the abnormal trend formation process of lubrication, thereby enhancing the accuracy and explanatory power of type determination.

[0044] Further, first, the standard deviation of the harmonic content of the current of each fan in the determination period is calculated to quantify the degree of dynamic load instability caused by wind shear or control fluctuations on the electrical side; then, the electrical fluctuations of adjacent fans are compared to select "concerned units" with synchronous harmonic abnormalities, so as to lock the devices that may be affected by the same local extreme wind field; finally, combined with the three-axis vibration peak values of these units and the preset risk evolution threshold, the units most prone to failure under electrical-mechanical dual stress can be accurately identified, avoiding missed judgments caused by relying solely on overall average indicators, and issuing the most targeted severe warnings in advance.

[0045] Further, by calculating the correlation coefficient of the current harmonic fluctuations of adjacent fans, the units that show synchronous electrical response under the same wind field disturbance can be identified. Because if the harmonic amplitude of adjacent units fluctuates at the same time, it means that they may be affected by similar gusts or load switching, and at this time these units are more likely to produce a chain effect on the structure and lubrication system. After selecting the units with a high fluctuation consistency as "concerned generators", the subsequent vibration peak analysis and risk assessment can focus on the key areas that are truly impacted by extreme weather, reducing the overall calculation amount and avoiding misjudgment of local disturbances as field-wide risks, thereby significantly improving the accuracy and response efficiency of fault warnings.

[0046] Further, by focusing on the difference changes of the three-axis vibration peak values between the generators, the fluctuation characteristics representing the severity of structural response are extracted, and the overall instability is quantified in the form of standard deviation; then, compared with the preset risk evolution threshold, the severe warning level is determined when the fluctuation exceeds the threshold, reflecting the dynamic coordination relationship between the mechanical states of the units. When the disturbance degree of adjacent units under extreme weather appears a significant deviation, it can be considered as a direct signal of increased possibility of fault evolution, thereby realizing the pre-identification of fault trend and the risk escalation judgment.

[0047] Further, by introducing the adjustment synchronization degree and cosine similarity as indicators to measure the dynamic coupling relationship between the wind speed change rate and the fluctuation of the starting torque increment, the embodiment can effectively identify the response consistency of the yaw system to wind speed disturbance. Furthermore, according to the relative deviation degree of the two, the preset gust impact threshold is quantitatively gain-processed by using the set threshold adjustment coefficient, which not only avoids misjudgment of short-time synchronization fluctuation as abnormal, but also realizes dynamic adjustment of the threshold, so that the system has self-adaptive adjustment capability. This method combines time series volatility and trend consistency at the numerical level, improves the rational judgment ability of the wind turbine operating boundary under complex weather conditions, and enhances the robustness of abnormal identification and the accuracy of the warning threshold.

[0048] Further, by introducing the strategy of adjusting the risk evolution threshold value by the fluctuation value of the adjustment frequency and the fluctuation value of the oil viscosity, the running stability and the change characteristics of the lubrication state of the wind turbine during the strong wind passage can be dynamically reflected. When the gust impact threshold is frequently adjusted and its volatility increases, it often means that the external wind field disturbance is severe or the control system response is abnormal. At the same time, the fluctuation of oil viscosity reflects the stress and thermal state change trend of the gear box lubrication system. Taking the standard deviation of the two as the quantitative basis to adjust the risk evolution threshold value, it not only can enhance the sensitivity of the system to potential fault risk, but also can effectively filter incidental disturbance, improve the stability and robustness of the early warning mechanism, and realize the adaptive optimization of the fault risk trend judgment threshold by establishing the linkage mechanism between the control system adjustment frequency and the key mechanical performance indicators.

[0049] Further, by comprehensively analyzing the relative changes of the adjustment frequency fluctuation deviation and the oil viscosity fluctuation deviation, the risk evolution threshold value is reasonably increased to dynamically adapt to the running state of the wind turbine under extreme weather conditions. The adjustment frequency reflects the frequency of adjustment of the gust impact threshold, and the viscosity fluctuation reflects the stability of the lubrication system, both of which affect the risk level of equipment failure development. By introducing the preset risk adjustment coefficient and the corresponding weight, the two deviations are weighted calculated to realize the precise adjustment of the risk threshold, ensuring that the fault warning system can not only sensitively capture potential risks, but also avoid misjudgment caused by short-term abnormal fluctuations, thereby improving the accuracy and stability of the prediction and ensuring the safe operation of the wind turbine under complex weather conditions.

[0050] Further, by statistically analyzing the time difference between each early warning time point in the risk prediction sequence and the initial moment, and measuring the standard deviation (time distribution degree) thereof, the distribution characteristics of the risk event within the prediction window are measured: when the time distribution degree is greater than a preset distribution degree threshold, it indicates that the prediction model continuously gives a high risk signal throughout the prediction period, rather than sporadic or occasional isolated moment fluctuations; this distribution width across time periods reflects the persistence and universality of fault evolution, thereby ensuring that the system triggers an early warning only when it is truly facing a continuous risk, avoiding false positives due to single burst predictions, and timely warning when the overall risk spreads, significantly improving the reliability and timeliness of the early warning.

[0051] Further, by quantitatively analyzing the duration of consecutive over-limit vibration periods, the standard deviation (abnormal distribution degree) of the "distribution duration" is extracted, which can accurately capture the repeated-recovery-repeated characteristics of blade vibration under gust impact. When the abnormal distribution degree exceeds the preset threshold, it indicates that the vibration anomaly is not just a single short-time fluctuation, but a cluster response that occurs multiple times and for a long period of time, which corresponds to the cumulative structural impact of strong wind gusts on the unit, and can filter normal small amplitude flutter and timely issue an early warning when multiple repeated impacts truly threaten the stability of the unit, thereby significantly improving the accuracy and advance of fault identification. BRIEF DESCRIPTION OF DRAWINGS

[0052] Figure 1 Flowchart of the extreme weather power generation equipment fault prediction method based on the time sequence large model of the present embodiment;

[0053] Figure 2 Determination logic diagram for determining the lubrication abnormality trend of the present embodiment;

[0054] Figure 3 Determination logic diagram for determining the generator of interest of the present embodiment;

[0055] Figure 4 Determination logic diagram for adjusting the preset gust impact threshold of the present embodiment. DETAILED DESCRIPTION

[0056] In order to make the objects and advantages of the present application clearer, the present application will be further described below with reference to the embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0057] The preferred embodiments of the present application will be described below with reference to the accompanying drawings. Those skilled in the art should understand that these embodiments are only used to explain the technical principles of the present application and do not limit the protection scope of the present application.

[0058] Please refer to Figure 1As shown, it is the flow chart of the extreme weather power generation equipment fault prediction method based on the time sequence large model of the embodiment, the embodiment provides an extreme weather power generation equipment fault prediction method based on a time sequence large model, including:

[0059] Real-time synchronous acquisition of three-axis vibration peak values of the middle part of the blade in each wind turbine during strong wind passage, oil viscosity of the gear box input shaft, current harmonic content of the generator stator, yaw system starting torque increment, wind speed variation rate at the top of the cabin and temperature rise rate of the main shaft bearing;

[0060] According to the three-axis vibration peak values in the preset abnormality determination period and the preset gust impact threshold, it is determined that there is an abnormal event, and an abnormality determination result is obtained;

[0061] Based on the abnormality determination result, according to the oil viscosity change rate and the temperature rise rate, it is determined that the type of the abnormal event is lubrication abnormality trend, and a type determination result is obtained;

[0062] Based on the type determination result, according to each current harmonic content, the three-axis vibration peak value and the preset risk evolution threshold, the risk level of the fault trend is determined to be serious warning, and a risk determination result is obtained;

[0063] Based on the risk determination result, according to the starting torque increment and the wind speed variation rate in the next preset first adjustment period, the preset gust impact threshold is adjusted, and according to the number of times of adjusting the gust impact threshold in the next preset second adjustment period and the oil viscosity, the risk evolution threshold is adjusted;

[0064] According to the risk determination result obtained again after adjusting the risk evolution threshold in the preset prediction period, the three-axis vibration peak value and the preset time sequence large model, a risk prediction sequence is generated, and a warning alarm is sent according to the time stamp of the risk prediction sequence.

[0065] The embodiment adopts an industrial Ethernet data acquisition system based on IEEE 1588 precision clock synchronization (PTP), and connects each sensor to an edge computing gateway in the cabin through high-bandwidth optical fiber or shielded twisted pair. Specifically, a MEMS three-axis accelerometer is installed in the middle of the blade, with a sampling frequency of 1 kHz; a microfluidic online viscosity meter is built in the input shaft oil circuit of the gearbox, which outputs the viscosity value of the oil in real time; an electric energy quality analyzer is connected to the stator side of the generator to monitor the current harmonic content; a torque sensor is installed on the bearing of the yaw system to record the starting torque increment; an ultrasonic weather station or a laser Doppler anemometer is arranged on the top of the cabin to measure the wind speed variation rate; a high-precision K-type thermocouple is arranged in the front bearing box of the main shaft to detect the temperature rise rate. All sensor data are aligned at the nanosecond level through PTP calibrated timestamps, and after preprocessing, they are uploaded to the central timing large model server through the optical fiber link by the edge gateway, realizing real-time synchronous acquisition and high-precision correlation analysis of parameters.

[0066] The preset abnormality determination period is a sampling time window for determining whether the vibration peak value is out of limit, which depends on the unit vibration response characteristics and the data sampling frequency, and is usually set between 10 seconds and 60 seconds, and is set to 30 seconds in the embodiment, which can timely capture the vibration mutation caused by gust impact.

[0067] The preset risk evolution threshold is a comprehensive index threshold for risk level division, which depends on the statistical distribution of multi-source features (harmonic content, vibration amplitude, etc.), and is usually set between 0.7 and 0.9, and is set to 0.85 in the embodiment, which can effectively identify high-risk trends.

[0068] The preset first adjustment period is a time interval for adaptively adjusting the gust threshold, which depends on the dynamic rate of yaw torque and wind speed change, and is usually set between 1 minute and 5 minutes, and is set to 2 minutes in the embodiment, which can respond to wind condition mutation in real time.

[0069] The preset second adjustment period is a trigger period for correcting the risk evolution threshold, which depends on the threshold adjustment frequency and the viscosity fluctuation characteristics of the oil, and is usually set between 10 minutes and 30 minutes, and is set to 15 minutes in the embodiment, which can balance the stability and sensitivity of the threshold.

[0070] The preset prediction period is a time range for generating a risk prediction sequence, which depends on the model prediction accuracy and operation and maintenance decision requirements, and is usually set between 24 hours and 72 hours, and is set to 48 hours in the embodiment, which can provide sufficient advance for scheduling deployment.

[0071] The preset timing large model is a deep timing network specially designed for multi-step risk prediction on the basis of multi-source monitoring data of a wind turbine generator. In this embodiment, a multi-channel self-attention prediction model based on the Informer architecture is adopted, and the structure and process thereof include: all risk judgment results and corresponding moment blade middle three-axis vibration peak values in a preset prediction period after adjustment by a risk evolution threshold are staggered and spliced into a feature matrix of Mx2 according to minute granularity, wherein M is the total number of time points in the prediction period; channel identifier embedding is added to the two columns of the feature matrix, and sine-cosine position coding is further added to obtain an input tensor of MxD, wherein D is the hidden layer dimension (the vector length mapped to each time step); the input tensor is input into a multi-layer self-attention encoder, and long and short term dependence features across time steps and across channels are efficiently extracted through a sparse global attention mechanism; the decoder receives the feature representation output by the encoder and the "historical target" sequence (in this embodiment, the risk judgment results of the last 60 time steps), and adopts a sliding window and a multi-step recursion strategy to accurately capture the risk evolution inertia and trend; the decoder generates a risk probability value sequence of the future T time steps (in this embodiment, the preset prediction period T = 48x60 minutes) in turn to form the final risk prediction sequence.

[0072] By introducing the three-axis vibration peak value of the blade middle part, the oil viscosity of the gearbox, the current harmonic of the stator, the starting torque increment of the yaw system, the wind speed change rate of the cabin top and the temperature rise rate of the main shaft bearing, multiple parameters closely related to structural load, electrical disturbance and thermal-lubrication state are realized to monitor the running state of the wind turbine generator under strong wind extreme weather. The gust impact directly causes the blade vibration to intensify, which is transmitted to the gearbox to cause the oil viscosity fluctuation, and at the same time, the temperature rise rate changes; the lubrication degradation further causes the component friction to intensify, which is reflected in the changes of the yaw torque and the stator current harmonic; the risk level evaluation comprehensively considers the coupling trend of electrical and mechanical signals, thereby driving the dynamic threshold adjustment to realize the adaptive optimization of the model, and finally generating a continuous fault risk prediction sequence through the timing large model, which can accurately judge the trend and give an early warning under the condition of multi-source coupled disturbance, improve the fault perception sensitivity and operation safety of the wind power equipment, and effectively solve the problem that the lubrication abnormal trend recognition lags under extreme weather due to incomplete monitoring parameter coverage and fixed threshold setting, thereby causing the fault warning to be not timely and the unit shutdown loss to intensify.

[0073] Referring to FIG. 1, Figure 2 FIG. 1 is a determination logic diagram for determining the lubrication abnormal trend in this embodiment, and the process of determining the type of the abnormal event as the lubrication abnormal trend according to the oil viscosity change rate and the temperature rise rate in this embodiment includes:

[0074] The maximum-minimum value normalization processing is performed on all viscosity change rates to obtain a viscosity change normalized value set, and the maximum-minimum value normalization processing is performed on all temperature rise rates to obtain a temperature rise rate normalized value set;

[0075] The Pearson correlation coefficient of the viscosity change normalized value set and the temperature rise rate normalized value set is calculated to obtain a change correlation amplitude;

[0076] When the change correlation amplitude is greater than a preset correlation amplitude threshold, it is determined that the type of the abnormal event is a lubrication abnormal trend, and the type determination result is obtained.

[0077] The preset correlation amplitude threshold refers to a Pearson correlation coefficient threshold for determining the synchronization degree of oil viscosity change and temperature rise rate, and is determined by the correlation distribution of the two parameters under historical extreme weather conditions. It is usually set between 0.6 and 0.8, and is set to 0.7 in this embodiment, which can effectively distinguish between the synchronous temperature rise trend caused by real lubrication degradation and the unrelated fluctuations.

[0078] By normalizing the oil viscosity change rate and the main shaft bearing temperature rise rate (prior art, not described again), the two types of data are analyzed on a unified scale, thereby improving the measurement accuracy of the consistency of their trend changes. Using the Pearson correlation coefficient as a measure of change correlation strength helps to accurately determine whether the lubrication performance degradation is synchronized with the temperature rise. When the two are highly positively correlated, it indicates that the decline in lubrication performance has had a significant impact on heat conduction and mechanical friction, showing the typical characteristics of lubrication degradation leading to increased temperature rise. This method effectively reveals the linkage evolution law between the lubrication state and the thermal effect in the formation process of the lubrication abnormal trend, thereby enhancing the accuracy and explanatory power of the type determination.

[0079] Specifically, according to the current harmonic content, the three-axis vibration peak value of each wind turbine, and a preset risk evolution threshold, the process of determining the risk level of the fault trend as a serious warning includes:

[0080] Calculate the standard deviation of all current harmonic contents from the initial time to each time within the preset level determination period to obtain a plurality of harmonic content fluctuation values, forming a harmonic content fluctuation set;

[0081] According to the harmonic content fluctuation set of each arbitrary adjacent two wind turbines, a plurality of concerned generators are determined from the wind turbine generator unit;

[0082] According to the three-axis vibration peak value of each arbitrary adjacent two concerned generators and a preset risk evolution threshold, the risk level of the fault trend is determined as a serious warning.

[0083] The preset level determination period is a time window for collecting current harmonic and vibration fluctuation data to evaluate the risk level, which depends on the response rate of wind turbine electrical and mechanical signals, and is usually set between 1 minute and 10 minutes, and is set to 5 minutes in the embodiment, which can balance the timeliness and data stability of early warning.

[0084] First, the standard deviation of the current harmonic content of each fan in the determination period is calculated to quantify the degree of dynamic load instability caused by wind shear or control fluctuation on the electrical side; then, the electrical fluctuations of adjacent fans are compared to screen out "concerned units" with synchronous harmonic abnormalities, so as to lock the devices that may be affected by the same local extreme wind field; finally, combined with the three-axis vibration peak values of these units and the preset risk evolution threshold, the units most prone to failure under electrical-mechanical dual stress can be accurately identified, avoiding missed judgments caused by relying solely on overall average indicators, and issuing the most targeted and severe early warning in advance.

[0085] Please refer to Figure 3 The determination logic diagram of the concerned generator in the embodiment is shown in the figure, and the process of determining a plurality of concerned generators from the wind turbine generator unit according to the harmonic content fluctuation set of any two adjacent wind generators in the embodiment includes:

[0086] The Pearson correlation coefficient of the harmonic content fluctuation set of any two adjacent wind generators is calculated to obtain the fluctuation consistency degree;

[0087] When the fluctuation consistency degree is greater than a preset consistency threshold, it is determined that the corresponding wind generators are concerned generators, so as to determine a plurality of concerned generators from the wind turbine generator unit.

[0088] The preset consistency threshold refers to the Pearson correlation coefficient threshold of the harmonic content fluctuation set of adjacent wind generators, which depends on the statistical distribution of the synchronous electrical response of adjacent units under historical extreme weather conditions, and is usually set between 0.7 and 0.9, and is set to 0.8 in the embodiment, which can effectively distinguish between synchronous fluctuations caused by gusts and occasional fluctuations caused by random noise.

[0089] By calculating the correlation coefficient of the current harmonic fluctuation of adjacent wind generators, the units showing synchronous electrical response under the same wind field disturbance can be identified. Because if the harmonic amplitude of adjacent units fluctuates at the same time, it means that they may be affected by similar gusts or load switching, and at this time these units are more likely to produce a chain effect on the structure and lubrication system. After the units with fluctuation consistency degree higher than the threshold are determined as "concerned generators", the subsequent vibration peak analysis and risk evaluation can be focused on the key areas that are really affected by extreme weather, which not only reduces the overall calculation amount, but also avoids misjudging local disturbance as overall risk, thereby significantly improving the accuracy and response efficiency of fault warning.

[0090] Specifically, the process of determining the risk level of the fault trend as a serious warning according to the three-axis vibration peak values of each arbitrary adjacent two of the concerned generators and a preset risk evolution threshold value comprises:

[0091] calculating the difference of the three-axis vibration peak values of each arbitrary adjacent two of the concerned generators to obtain a plurality of peak value change values;

[0092] calculating the standard deviation of all the peak value change values to obtain a peak value change fluctuation value;

[0093] determining the risk level of the fault trend as a serious warning when the peak value change fluctuation value is greater than the preset risk evolution threshold value.

[0094] By observing the difference change of the three-axis vibration peak values between the concerned generators, the fluctuation characteristics representing the degree of structural response is extracted, and the overall instability is quantified in the form of standard deviation. Then, the preset risk evolution threshold value is compared, and the serious warning level is triggered when the fluctuation exceeds the threshold value, which reflects the dynamic coordination relationship between the mechanical states of the units. When the disturbance degree of the adjacent units under extreme weather appears obvious deviation, it can be regarded as a direct signal of the increased possibility of fault evolution, thereby realizing the pre-identification of the fault trend and the risk escalation judgment.

[0095] Referring to FIG. 8, Figure 4 the determination logic diagram for adjusting the preset gust impact threshold value in the embodiment, the process of adjusting the preset gust impact threshold value according to the starting torque increment and the wind speed change rate in the next preset first adjustment period in the embodiment comprises:

[0096] calculating the standard deviation of all the starting torque increments from the initial time to each time in the next preset first adjustment period to obtain a plurality of torque increment fluctuation values, and calculating the wind speed change rate from the initial time to each time in the next preset first adjustment period to obtain a plurality of wind speed change fluctuation values;

[0097] drawing a curve of the torque increment fluctuation value changing with time to obtain a first curve, and drawing a curve of the wind speed change rate changing with time to obtain a second curve;

[0098] vectorizing the first curve to obtain a first vector, and vectorizing the second curve to obtain a second vector;

[0099] calculating the cosine similarity of the first vector and the second vector to obtain an adjustment synchronization degree;

[0100] When the adjustment synchronization degree is greater than the preset adjustment synchronization threshold, the preset gust impact threshold is increased according to a relative deviation between the adjustment synchronization degree and the preset adjustment synchronization threshold and a preset threshold adjustment coefficient, F' = F × [1 + k × (S - S0) / S0], wherein F' is the preset gust impact threshold after the increase, F is the preset gust impact threshold before the increase, S is the adjustment synchronization degree, S0 is the preset adjustment synchronization threshold, and k is the preset threshold adjustment coefficient.

[0101] The preset adjustment synchronization threshold is a reference value for measuring whether the fluctuation trends of the wind speed change rate and the starting torque increment in the time dimension are consistent, depends on the cosine similarity level of the two in the normal response state in the historical operation data, and is usually set to be between 0.6 and 0.85, and is set to 0.75 in the embodiment, which can effectively identify the consistency of the system response and determine whether dynamic adjustment is needed.

[0102] The preset threshold adjustment coefficient is a proportional coefficient for controlling the increase range of the preset gust impact threshold when it is determined that the wind speed change rate and the starting torque increment have high synchronization, depends on the structural tolerance and response strategy demand of the wind turbine in different meteorological change scenarios, and is usually set to be between 0.1 and 0.3, and is set to 0.2 in the embodiment, which can achieve fine control of the threshold value and avoid false judgment or missed warning caused by excessive adjustment.

[0103] By introducing the adjustment synchronization degree and the cosine similarity as indexes for measuring the dynamic coupling relationship between the wind speed change rate and the starting torque increment fluctuation, the embodiment can effectively identify the response consistency of the yaw system to the wind speed disturbance. Furthermore, according to the relative deviation degree of the two, the preset gust impact threshold is quantitatively gained by using the set threshold adjustment coefficient, which not only avoids that short-time synchronization fluctuation is misjudged as an exception, but also realizes dynamic adjustment of the threshold value, so that the system has self-adaptive adjustment capability. The method combines the time sequence fluctuation and the trend consistency at the numerical level, improves the reasonable judgment ability of the operation boundary of the wind turbine under complex meteorological conditions, and thus enhances the robustness of the abnormal identification and the accuracy of the early warning threshold.

[0104] Specifically, the process of adjusting the risk evolution threshold according to the number of times of adjusting the gust impact threshold in the next preset second adjustment period and the oil viscosity includes:

[0105] The standard deviation of all the times of adjusting the gust impact threshold is calculated to obtain an adjustment frequency fluctuation value, and the standard deviation of all the oil viscosities is calculated to obtain an oil viscosity fluctuation value.

[0106] The risk evolution threshold is adjusted according to the adjustment frequency fluctuation value and the oil viscosity fluctuation value.

[0107] By introducing a strategy of co-correcting the risk evolution threshold using the fluctuation values ​​of adjustment frequency and oil viscosity, the operational stability and lubrication status changes of wind turbine units during strong winds can be dynamically reflected. Frequent adjustments to the gust impact threshold, coupled with increased volatility, often indicate severe external wind field disturbances or abnormal control system response. Simultaneously, oil viscosity fluctuations reflect the stress and thermal state changes in the gearbox lubrication system. Using the standard deviation of both as a quantitative basis to adjust the risk evolution threshold not only enhances the system's sensitivity to potential fault risks but also effectively filters out occasional disturbances, improving the stability and robustness of the early warning mechanism. Furthermore, by establishing a linkage mechanism between the control system's adjustment frequency and key mechanical performance indicators, adaptive optimization of the fault risk trend judgment threshold can be achieved.

[0108] Specifically, the process of adjusting the risk evolution threshold based on the fluctuation value of the number of adjustments and the fluctuation value of the oil viscosity includes:

[0109] Calculate the absolute difference between the fluctuation value of the number of adjustments and the preset fluctuation threshold to obtain the fluctuation deviation of the number of adjustments; and calculate the absolute difference between the fluctuation value of the oil viscosity and the preset viscosity fluctuation threshold to obtain the viscosity fluctuation deviation.

[0110] When the frequency fluctuation deviation is greater than the preset frequency fluctuation deviation threshold and the viscosity fluctuation deviation is greater than the preset viscosity fluctuation deviation threshold, the risk evolution threshold is increased according to the relative deviation between the frequency fluctuation deviation and the preset frequency fluctuation deviation threshold, and the relative deviation between the viscosity fluctuation deviation and the preset viscosity fluctuation deviation threshold. Y'=Y×{1+u×[i×(H-H0) / H0+j×(P-P0) / P0]}, where Y' is the increased risk evolution threshold, Y is the original risk evolution threshold, u is the preset risk adjustment coefficient, i is the preset frequency deviation weight, H is the frequency fluctuation deviation, H0 is the preset frequency fluctuation deviation threshold, j is the preset viscosity deviation weight, P is the viscosity fluctuation deviation, and P0 is the preset viscosity fluctuation deviation threshold.

[0111] The preset fluctuation threshold is a benchmark value used to determine whether the fluctuation of the number of times the wind turbine adjusts the impact threshold is abnormal. It depends on the historical statistical characteristics of the wind turbine's adjustment frequency under extreme weather conditions and is usually set between 1 and 5 times. In this embodiment, it is set to 3 times, which can effectively distinguish between normal fluctuations and abnormal adjustment behavior.

[0112] The preset viscosity fluctuation threshold is a threshold used to determine whether the viscosity fluctuation of the oil exceeds the normal range. It depends on the performance changes of the lubricating oil under different operating conditions and is usually set between 0.01 and 0.05. In this embodiment, it is set to 0.03, which can accurately reflect the abnormal trend of the lubrication system.

[0113] The preset number fluctuation deviation threshold is the maximum deviation allowed between the number fluctuation value and the preset number fluctuation threshold, which depends on the fluctuation amplitude of the actual operation of the device, and is usually set to be between 0.5 and 1.5, and is set to 1.0 in the embodiment, which can reasonably define the difference between normal and abnormal fluctuations.

[0114] The preset viscosity fluctuation deviation threshold is the maximum tolerance deviation between the oil viscosity fluctuation value and the preset viscosity fluctuation threshold, which depends on the stability of the lubrication system fluctuation, and is usually set to be between 0.005 and 0.02, and is set to 0.01 in the embodiment, which ensures the scientificity and accuracy of the threshold adjustment.

[0115] The preset risk adjustment coefficient is a proportional factor for adjusting the risk evolution threshold increment, which depends on the demand of the wind power device for risk sensitivity, and is usually set to be between 0.1 and 0.5, and is set to 0.3 in the embodiment, which balances the timeliness and stability of risk response.

[0116] The preset number deviation weight is a weighting factor for weighting the influence of the number fluctuation deviation on the risk threshold adjustment, which depends on the contribution of the number fluctuation to the fault risk, and is usually set to be between 0.4 and 0.7, and is set to 0.6 in the embodiment, which strengthens the role of the adjustment number in risk judgment.

[0117] The preset viscosity deviation weight is a weighting factor for measuring the proportion of the oil viscosity fluctuation deviation in the overall risk evolution threshold adjustment, which depends on the influence of the lubrication system on the operation stability of the power generation device. It is usually set to be between 0.3 and 0.6, and is set to 0.4 in the embodiment, which can effectively reflect the important role of lubrication performance anomaly in overall risk assessment, thereby improving the response accuracy and regulation rationality of fault prediction.

[0118] By comprehensively analyzing the relative changes of the adjustment number fluctuation deviation and the oil viscosity fluctuation deviation, the risk evolution threshold is reasonably increased, so as to dynamically adapt to the operating state of the wind turbine under extreme weather conditions. The adjustment number reflects the frequency of the gust impact threshold adjustment, and the viscosity fluctuation reflects the stability of the lubrication system, both of which affect the risk level of the development of device faults. By introducing the preset risk adjustment coefficient and the corresponding weight, the two deviations are weighted and calculated, the risk threshold is accurately adjusted, the fault warning system can sensitively capture potential risks, and false judgments caused by short-term abnormal fluctuations are avoided, thereby improving the accuracy and stability of prediction, and ensuring the safe operation of the wind turbine under complex weather conditions.

[0119] Specifically, the process of issuing a warning alarm according to the time stamp of the risk prediction sequence includes:

[0120] Calculate the difference between each of the total timestamps in the risk prediction sequence and the initial time, obtain a plurality of distribution durations, and calculate the standard deviation of all the distribution durations to obtain a time distribution degree;

[0121] When the time distribution degree is greater than a preset distribution degree threshold, it is determined that the wind turbine has a fault, and the warning alarm is issued.

[0122] The preset distribution degree threshold is a standard deviation threshold for determining the dispersion degree of the risk prediction time point distribution, depends on the time distribution fluctuation characteristics of the historical fault events within the prediction window, and is usually set to be between 300 seconds and 900 seconds, and is set to 600 seconds in this embodiment, which can effectively distinguish between sporadic short-time high-risk signals and persistent fault evolution trends.

[0123] By statistically analyzing the time difference between each warning time point in the risk prediction sequence and the initial time, and using the standard deviation (time distribution degree) to measure the distribution characteristics of the risk event within the prediction window, when the time distribution degree is greater than the preset distribution degree threshold, it indicates that the prediction model continuously gives high-risk signals within the entire prediction period, rather than sporadic or occasional isolated time fluctuations. The distribution width across the time period reflects the persistence and universality of fault evolution, thereby ensuring that the system triggers a warning only when it is truly facing a persistent risk, avoiding false positives due to single burst predictions, and timely warning when the overall risk spreads, significantly improving the reliability and timeliness of the warning.

[0124] Specifically, the process of determining the existence of an abnormal event according to the three-axis vibration peak value within the preset abnormality determination period and the preset gust impact threshold includes:

[0125] When the three-axis vibration peak value within the preset abnormality determination period is greater than the preset vibration peak threshold, record the current timestamp, and stop recording when the three-axis vibration peak value is less than or equal to the preset vibration peak threshold, to obtain a plurality of continuous durations;

[0126] Obtain the time period from the initial time of the preset abnormality determination period to the timestamp at the center of each continuous duration, to obtain a plurality of distribution durations;

[0127] Calculate the standard deviation of all the distribution durations to obtain an abnormal distribution degree;

[0128] When the abnormal distribution degree is greater than the preset gust impact threshold, it is determined that the abnormal event exists, and the abnormality determination result is obtained.

[0129] The preset vibration peak threshold is a peak threshold for the three-axis acceleration sensor in the middle of the blade to determine abnormal vibration, depends on the historical gust peak statistics and the blade structure tolerance, and is usually set to be between 1.5g and 4g, and is set to 3g in this embodiment, which can effectively distinguish between normal operation flutter and abnormal large vibration caused by gust impact.

[0130] The preset gust impact threshold refers to a critical value of the standard deviation of the duration of abnormal vibration caused by continuous gusts in a judgment period, depends on the statistical distribution of the duration of blade vibration anomaly during the passage of historical typhoons, and is usually set between 5 seconds and 20 seconds, and is set to 10 seconds in the embodiment, and can distinguish between normal short-time vibration fluctuation and long-time vibration distribution deviation caused by gust impact.

[0131] By quantitatively analyzing the duration of the continuous over-limit vibration period, the standard deviation of the "distribution duration" (abnormal distribution degree) is extracted, which can accurately capture the repeated-recovery-again characteristics of blade vibration under gust impact. When the abnormal distribution degree exceeds the preset threshold, it indicates that the vibration anomaly is not only a single short-time fluctuation, but also a cluster response that occurs multiple times and for a long period of time. This corresponds to the cumulative structural impact of strong wind gusts on the unit, which can not only filter normal small amplitude flutter, but also issue an early warning when multiple repeated impacts truly threaten the stability of the unit, thereby greatly improving the accuracy and advance of fault identification.

[0132] The above only describes the preferred embodiments of the present application and is not intended to limit the present application; for those skilled in the art, the present application can have various modifications and changes. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1.A method for predicting failure of an extreme weather power generation device based on a time-series large model, characterized by, The method comprises: real-time synchronous acquisition of three-axis vibration peak values of middle parts of blades of each wind turbine, oil viscosity of a gear box input shaft, current harmonic content of a generator stator, yaw system starting torque increment, wind speed variation rate at the top of a nacelle, and temperature rise rate of a main shaft bearing during strong wind passage; determination of an abnormal event according to the three-axis vibration peak values and a preset gust impact threshold within a preset abnormality determination period to obtain an abnormality determination result; determination of a type of the abnormal event as a lubrication abnormality trend according to the oil viscosity change rate and the temperature rise rate based on the abnormality determination result to obtain a type determination result; determination of a risk level of the lubrication abnormality trend as a serious warning according to each current harmonic content, the three-axis vibration peak values, and a preset risk evolution threshold based on the type determination result to obtain a risk determination result; adjustment of the preset gust impact threshold according to the starting torque increment and the wind speed variation rate within a next preset first adjustment period and adjustment of the risk evolution threshold according to the number of times of adjustment of the gust impact threshold and the oil viscosity within a next preset second adjustment period based on the risk determination result; generation of a risk prediction sequence according to all the risk determination results obtained after adjustment of the risk evolution threshold within a preset prediction period, the three-axis vibration peak values, and a preset time sequence large model, and issuance of a warning alarm according to a time stamp of the risk prediction sequence; the process of determining an abnormal event according to the three-axis vibration peak values and a preset gust impact threshold within a preset abnormality determination period to obtain an abnormality determination result comprises: determination of an abnormal distribution degree according to a comparison result of the three-axis vibration peak values and a preset vibration peak threshold; determination of the abnormal event according to a comparison result of the abnormal distribution degree and the preset gust impact threshold to obtain the abnormality determination result; the process of determining a risk level of the lubrication abnormality trend as a serious warning according to each current harmonic content of the wind turbine, the three-axis vibration peak values, and a preset risk evolution threshold comprises: determination of a harmonic content fluctuation set according to the current harmonic content from an initial time to each time within a preset level determination period; determination of a number of concerned wind turbines from the wind turbine group according to the harmonic content fluctuation set of each arbitrary adjacent two wind turbines; determination of a risk level of the lubrication abnormality trend as a serious warning according to the three-axis vibration peak values of each arbitrary adjacent two concerned wind turbines and a preset risk evolution threshold; the process of generating a risk prediction sequence according to all the risk determination results obtained after adjustment of the risk evolution threshold within a preset prediction period, the three-axis vibration peak values, and a preset time sequence large model comprises: The all risk judgment results and the three-axis vibration peak values of the middle part of the blade at the corresponding moment in the preset prediction period are staggered and spliced into a feature matrix of M*2 according to minute granularity, wherein M is the time step number; channel identification embedding is added to two columns of the feature matrix respectively, and then sine-cosine position coding is superimposed to obtain an input tensor of M*D shape, wherein D is the hidden layer dimension; the input tensor is sent into a multi-layer self-attention encoder, and long and short term dependence features across time steps and across channels are efficiently extracted through a sparse global attention mechanism; the decoder receives the feature representation output by the encoder and the "history target" sequence, adopts a sliding window and a multi-step recursion strategy to accurately capture the risk evolution inertia and trend; the decoder generates a risk probability value sequence of future T time steps in turn to form a final risk prediction sequence. 2.The extreme weather power plant failure prediction method based on the timing large model according to claim 1, wherein, The process of determining the type of the abnormal event as a lubrication abnormal trend according to the oil viscosity change rate and the temperature rise rate includes: determining a change correlation amplitude according to the oil viscosity change rate and the temperature rise rate in a preset abnormal trend determination period; determining the type of the abnormal event as a lubrication abnormal trend according to a comparison result of the change correlation amplitude and a preset correlation amplitude threshold, to obtain the type determination result. 3.The time-series large model-based extreme weather power plant failure prediction method of claim 2, wherein, The process of determining a number of concerned generators from the wind turbine generator unit according to the harmonic content fluctuation set of each arbitrary adjacent two wind turbine generators includes: determining a fluctuation consistency degree according to the two harmonic content fluctuation sets; determining a number of concerned generators from the wind turbine generator unit based on a comparison result of the fluctuation consistency degree and a preset consistency degree threshold. 4.The time-series large model-based extreme weather power plant failure prediction method of claim 3, wherein, The process of determining the risk level of the lubrication abnormal trend as a serious warning according to the three-axis vibration peak values of each arbitrary adjacent two concerned generators and a preset risk evolution threshold includes: determining a number of peak value change values according to the three-axis vibration peak values of each arbitrary adjacent two concerned generators; determining a peak value change fluctuation value according to all the peak value change values; determining the risk level of the lubrication abnormal trend as a serious warning according to a comparison result of the peak value change fluctuation value and the preset risk evolution threshold. 5.The extreme weather power plant failure prediction method based on the timing large model according to claim 4, characterized in that, The process of adjusting the preset gust impact threshold according to the starting torque increment and the wind speed change rate in a next preset first adjustment period includes: determining an adjustment synchronization degree according to all the starting torque increments and all the wind speed change rates; adjusting the preset gust impact threshold according to a comparison result of the adjustment synchronization degree and a preset adjustment synchronization threshold. 6.The extreme weather power plant failure prediction method based on the timing large model according to claim 5, wherein, The process of adjusting the risk evolution threshold according to the number of times of adjusting the gust impact threshold and the oil viscosity in a next preset second adjustment period includes: determining an adjustment times fluctuation value according to all the number of times of adjusting the gust impact threshold, and determining an oil viscosity fluctuation value according to all the oil viscosities; adjusting the risk evolution threshold according to the adjustment times fluctuation value and the oil viscosity fluctuation value. 7.The extreme weather power plant failure prediction method based on the timing large model according to claim 6, wherein, The process of adjusting the risk evolution threshold according to the adjustment times fluctuation value and the oil viscosity fluctuation value includes: According to the adjustment number of times fluctuation value and preset number of times fluctuation threshold value, determine number of times fluctuation deviation, and according to the oil viscosity fluctuation value and preset viscosity fluctuation threshold value, determine viscosity fluctuation deviation; According to the comparison result of the number of times fluctuation deviation and preset number of times fluctuation deviation threshold value, the comparison result of the viscosity fluctuation deviation and preset viscosity fluctuation deviation threshold value, adjust the risk evolution threshold value. The process of issuing a warning alarm according to the time stamp of the risk prediction sequence includes: 8.The extreme weather power plant failure prediction method based on the timing large model according to claim 7, wherein, According to all time stamps, determine time distribution degree; According to the comparison result of the time distribution degree and preset distribution degree threshold value, determine that the wind turbine generator set has a fault, and issue the warning alarm. ​

Images