Wind turbine operation status monitoring method based on SCADA and FISSA-DBN model

Abstract

The present invention discloses a method for monitoring the operating status of a wind turbine based on SCADA and FISSA-DBN models, comprising the following steps: step (1) selecting data of wind turbine operation detected by SCADA; step (2) preprocessing the wind turbine operation data; step (3) determining the parameters of the prediction model; step (4) standardizing the selected input and output parameter data; step (5) FISSA-DBN model topology structure; step (6) DBN parameter optimization; step (7) model training; step (8) prediction model application. The present invention determines whether a wind turbine and its components are abnormal based on reconstruction error and EWMA threshold. The prediction result of the FISSA-DBN model is consistent with the actual operation record of the wind turbine. The prediction model finds that the abnormal state is earlier than the actual operation record time, which is conducive to the wind farm staff to make judgments as early as possible and reduce unnecessary losses.

Description

Technical Field

[0001] The invention relates to the technical field of wind turbine operating status monitoring, in particular to a wind turbine operating status monitoring method based on SCADA and FISSA-DBN models. Background Art

[0002] As a green and clean energy, wind energy is gaining more and more attention. Wind turbines are the core components for realizing wind energy conversion. Monitoring their operating status is particularly important. Considering that wind turbines are usually located in harsh conditions and high altitude environments, abnormal conditions of generators, inverters and various sensors often occur. However, there are many inconveniences in daily maintenance. Therefore, it is necessary to build a wind turbine operating status monitoring model.

[0003] The defects of the wind turbine operating status monitoring method in the prior art are:

[0004] 1. Patent document CN110005580B discloses a method for monitoring the operating status of a wind turbine set. "The method comprises: obtaining sensor data of the historical operation of the wind turbine set, wherein the sensor data comprises wind speed signal, rotation speed signal, pitch angle signal, power signal, strain signal and vibration signal; constructing a wind speed, rotation speed, pitch angle and power matching monitoring model and a rotation speed, power, strain and vibration spectrum monitoring model according to the wind speed signal, rotation speed signal, pitch angle signal, power signal, strain signal and vibration signal; collecting real-time operating data of the wind turbine set, and comparing the real-time operating data of the wind turbine set with the real-time operating data of the wind turbine set. The wind speed, rotation speed, pitch angle, power matching monitoring model, rotation speed, power, strain, and vibration spectrum monitoring model are compared to determine the current operating status of the wind turbine. The wind turbine operating status monitoring method provided by the present invention can not only reduce the maintenance cost of the wind farm, but also detect faults in time or predict faults in advance. However, it does not fully utilize the correlation characteristics between data parameters, and it is difficult to determine whether the abnormal operating status is caused by the abnormal state of the measured object or the abnormal state of the sensor itself. It can only determine the current operating status, and cannot predict faults in advance, which is not conducive to the staff to make judgments as soon as possible, resulting in unnecessary losses. Summary of the invention

[0005] The purpose of the present invention is to provide a wind turbine operating status monitoring method based on SCADA and FISSA-DBN models to solve the problems raised in the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for monitoring the operating status of a wind turbine generator set based on SCADA and FISSA-DBN model, comprising the following steps:

[0007] The step (1) selects data from SCADA detecting the operation of the wind turbine generator set;

[0008] The step (2) pre-processes the wind turbine operating data;

[0009] The step (3) predicts model parameters are determined;

[0010] The step (4) performs standardization processing on the selected input and output parameter data;

[0011] The step (5) FISSA-DBN model topology structure;

[0012] The step (6) DBN parameter optimization;

[0013] The step (7) model training;

[0014] The step (8) predicts the application of the model.

[0015] Preferably, in step (1): original operation data of the wind turbine set is extracted from SCADA data according to the operation record of the wind turbine set.

[0016] Preferably, in step (2): based on expert domain knowledge, the overall prediction model condition parameters are selected from the original SCADA data, the data information is further processed according to the accident probability of the wind farm, and the erroneous data and unnecessary data in the original SCADA data are deleted.

[0017] Preferably, in step (3): according to the data sample, the input condition parameters and output parameters of the wind turbine condition monitoring model are selected, the preprocessed data group obtained in step (2) is used to calculate the correlation between each parameter and the output parameter using the Pearson correlation coefficient, and the parameters with large correlation are selected as the input parameters of the model.

[0018] Preferably, in step (4): the selected input and output parameter data are standardized, the values ​​are normalized to the interval [0, 1], and the prediction model is trained or tested using the data sample. 20,000 sets of historical normal data in each state are selected as samples, of which 80% are randomly selected as training data and 20% are selected as test data.

[0019] Preferably, in step (5): the population size of sparrows, the number of iterations, the producer ratio and the relevant parameters of the DBN network structure are set.

[0020] Preferably, in step (6): FISSA is used to optimize the learning rate of the RBM layer and the number of neurons in the hidden layer, improve the local optimization caused by randomly setting the DBN initial value, and obtain the optimal initial vector of the DBN model.

[0021] Preferably, in step (7): according to the research object of wind turbine operating status monitoring, the model is trained to obtain the upper and lower limit thresholds of the EWMA control chart.

[0022] Preferably, in step (8): the SCADA data of the actual wind farm is input into the prediction model, and the operating status of the wind turbine is monitored using an exponentially weighted average threshold method.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] 1. The present invention uses a deep belief network model to learn and extract the internal rules between the SCADA parameters of the wind turbine set layer by layer, and establishes a wind turbine operation status monitoring model based on FISSA-DBN. When the operation status and components of the wind turbine set are abnormal, the internal relationship of the SCADA data is destroyed. The wind turbine set and its components are judged whether they are abnormal based on the reconstruction error and the EWMA threshold. The prediction results of the FISSA-DBN model are consistent with the actual operation records of the wind turbine set. The prediction model finds that the abnormal state is earlier than the actual operation record time, which is conducive to the wind farm staff to make judgments as soon as possible and reduce unnecessary losses. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 A schematic diagram of the structure of the FISSA-DBN model of the present invention is established;

[0026] Figure 2 A parameter table of a wind turbine state assessment model of the present invention;

[0027] Figure 3 A schematic diagram of correlation between parameters of a wind turbine state monitoring model according to the present invention;

[0028] Figure 4 It is a δE curve diagram of the wind turbine state monitoring of the present invention;

[0029] Figure 5 It is a δE curve diagram of stator abnormal state monitoring of the present invention;

[0030] Figure 6 This is a test data performance comparison result diagram of the present invention. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0032] In the description of the present invention, it should be noted that the terms "upper", "lower", "inner", "outer", "front end", "rear end", "two ends", "one end", "the other end" and the like indicate positions or positional relationships based on the positions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as limiting the present invention. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0033] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the terms "installed", "provided with", "connected", etc. should be understood in a broad sense. For example, "connected" can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0034] See also Figure 1 , Figure 2 and Figure 3 , Wind turbine operation status monitoring method based on SCADA and FISSA-DBN model;

[0035] Example 1: Data source: SCADA system data of the same wind farm provided by a wind power company. The capacity of each unit is 2MW. 10-minute data of 5 years are selected for analysis, from January 1, 2013 to December 30, 2017. The data is sorted according to the wind farm operation data and accident records. The wind speed of the SCADA data used for modeling training and testing is between 3.5m / s and 22m / s. Fault data is eliminated to ensure healthy data for normal operation. The wind turbine transmits the field data to the background monitoring system through the SCADA system. By analyzing various parameters in the monitoring system, the operating status of the wind turbine is evaluated and each component is monitored. When various alarm information appears, the operating conditions of the wind turbine can be analyzed to make targeted decisions on whether to repair or shut down.

[0036] In order to verify the effectiveness of the FISSA-DBN model for wind turbine fault monitoring, a comparative analysis was conducted on three other models, namely the SSA-DBN model without firefly disturbance, the DBN model optimized by the whale optimization algorithm (WOA) (WOA-DBN) and the unoptimized DBN model. The analysis objects are the monitoring of two different conditions: the operating status of the wind turbine and the abnormal state of the generator stator. The monitoring of the operating status of the wind turbine is based on the network-measured active power, and the monitoring of the abnormal state of the generator stator is based on the stator winding temperature 2 of the generator. After expert system judgment and Pearson method calculation, the Pearson correlation criterion value is 0.8, and the model input state parameters in two different cases are obtained. The model input state parameters for the operating status of the wind turbine are 8, and the model input state parameters for the abnormal state of the generator stator are 9. The specific parameters are as follows: Figure 2 As shown, Figure 3 This is a correlation coefficient diagram between the model parameters of the wind turbine operating state. The correlation coefficient diagrams between the model parameters of the other two states are similar.

[0037] See also Figure 1 and Figure 4 , Wind turbine operation status monitoring method based on SCADA and FISSA-DBN model;

[0038] Example 2: Use the historical normal state model input and output SCADA data to establish a sample training set and a test sample set and perform normalization processing, then set the model initialization parameters, use the FISSA algorithm to find the best fitness value, and when the iteration is completed, the best fitness value is obtained according to the formula

[0039]

[0040] The values ​​of α, β, and λ are 82, 90, and 0.0668 respectively. Use the optimal values ​​to set the model parameters, train the DBN model, test the test sample set with the trained model, and use the formula

[0041] Z t =λδ Et +(1-λ)Z t-1 ;

[0042]

[0043] Get the upper and lower limits of the EWMA control chart, and take the alarm threshold as the upper limit.

[0044] The No. 5 wind turbine in 2013 provided by the wind power company was used for example verification. The data selected were from 18:05 on June 25, 2013 to 10:45 on June 26, 2013, that is, data points 25310 to 25410. After prediction by the FISSA-DBN model, the results are as follows Figure 4 As shown in (a), Figure 4 (b) is the SSA-DBN model detection result. Figure 4 (c) is the detection result of WOA-DBN model. Figure 4 (d) is the unoptimized DBN result. Figure 4 In (a), before 25333 points, i.e., 21:55 on June 25, the δE value was below the upper limit of the alarm threshold. After that time, the δE value continued to exceed the upper limit of the alarm threshold, indicating that the wind turbine was abnormal. The curve fell back later, but it still exceeded the upper limit of the alarm threshold by a lot. By comparing with the actual situation, it was found that the anemometer of the wind turbine had an abnormal condition starting from 23:25. The prediction result of the model was 90 minutes ahead of the actual situation. Figure 4 (b) Although the δE value of the SSA-DBN model exceeds the upper limit of the alarm threshold at point 25333, the subsequent curve does not fall back beyond the upper limit of the alarm threshold by much, which can easily be ignored. Figure 4 (c) The WOA-DBN model can also exceed the upper limit of the alarm threshold at 25333 points, but the subsequent fallback curve is below the upper limit of the alarm threshold, so the actual abnormal situation cannot be detected. Figure 4 (d) is the unoptimized wind turbine state detection model, which exceeds the upper limit of the alarm threshold at 25334 points, 10 minutes later than the FISSA-DBN model, and the δE value is relatively large, indicating that the unoptimized DBN model has a large error. The curve finally falls below the threshold because the anemometer was checked during the shutdown, and it was found that the anemometer itself was faulty and replaced.

[0045] See also Figure 1 , Figure 2 and Figure 5 , Wind turbine operation status monitoring method based on SCADA and FISSA-DBN model;

[0046] Example 3: Its input and output parameters are as follows Figure 2As shown in the figure, the correlation between stator temperature sensor 2 and the other five stator temperature sensors is very high, and the highest stator temperature 6 is selected as the relevant input parameter. The input and output parameter training and test sample sets are established using historical normal data, and then the model initialization parameters are set. The FISSA algorithm is used to find the best fitness value, and the values ​​of α, β, and λ in the formula are 4, 82, and 0.0215 respectively. After obtaining the optimal value, the DBN model is trained, and the test sample set is tested with the trained model, and the upper and lower limits of the EWMA control chart are obtained according to the formula.

[0047] The SCADA data of wind turbine No. 19 in 2013 provided by a wind power company was verified. The data selected was from 14:55 on October 2, 2013 to 11:35 on October 7, 2013, that is, data points 39400 to 40100. The results are as follows Figure 5 As shown, Figure 5 In (a), at point 39409, the δE value first appeared above the upper limit of the alarm threshold, indicating that the stator winding temperature was abnormal. The subsequent curve fluctuated above and below the alarm threshold until point 39704, when δE began to rise linearly, indicating that the abnormal state of stator winding 2 was obvious. At point 40002, that is, at 19:15 on October 6, the δE value dropped below the alarm threshold. By checking the records, the wind turbine was shut down at this time. It can be seen that the alarm signal appeared more than 4 days earlier than the actual shutdown time. Figure 5 (b) The first alarm point is 39412. Figure 5 (a) Model is 30 minutes late, while Figure 5 (c) The first alarm point is 39418. Figure 5 (a) Model is 90 minutes late, Figure 5 (d) is the prediction of DBN without optimization. It can be seen that the DBN model without optimization detected the abnormality at 39705 points when the stator winding temperature was obviously abnormal, which was nearly 50 hours later than the prediction using the FISSA optimization model. Therefore, the FISSA optimization model can detect abnormal conditions earlier and more accurately, and handle them in time. After shutdown detection, it was found that the temperature sensor itself had a problem. The sensor was replaced and installed correctly, and the fault disappeared after restarting.

[0048] See also Figure 6 , Wind turbine operation status monitoring method based on SCADA and FISSA-DBN model;

[0049] Example 4: The performance of four prediction models was compared and analyzed using root mean square error, mean absolute percentage error and mean absolute error. The evaluation data was 4000 sets of test data under three conditions. The performance comparison results are as follows: Figure 6As shown, it can be seen that the error of the optimized DBN model is smaller than that of the unoptimized DBN, which is beneficial to improving the accuracy of diagnosis.

[0050] It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above and that the invention can be implemented in other specific forms without departing from the spirit or essential features of the invention. Therefore, the embodiments should be considered exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description, and it is intended that all variations falling within the meaning and scope of the equivalent elements of the claims be included in the invention. Any reference numeral in a claim should not be considered as limiting the claim to which it relates.

Claims

1. A wind turbine operating status monitoring method based on SCADA and FISSA-DBN model, characterized in that: The following steps are involved: Step (1): selecting the data of wind turbine operation detected by SCADA, and extracting the original operation data of the wind turbine from the SCADA data according to the operation record of the wind turbine; Step (2): Preprocess the wind turbine operation data, select the total FISSA-DBN model condition parameters from the original SCADA data, further process the data information according to the accident probability of the wind farm, and delete the erroneous data and unnecessary data in the original SCADA data; Step (3): Determine the parameters of the FISSA-DBN model. According to the data sample, select the input condition parameters and output parameters of the FISSA-DBN model. Use the Pearson correlation coefficient to calculate the correlation between each parameter and the output parameter of the preprocessed data set obtained in step (2). Select the parameters with the largest correlation as the input parameters of the FISSA-DBN model. Step (4): Standardize the selected input and output parameter data, normalize the values ​​to the interval [0,1], use the obtained data samples to train or test the FISSA-DBN model, and select 20,000 sets of historical normal data in each state as samples, of which 80% are randomly used as training data and 20% as test data; Step (5): FISSA-DBN model topology structure setting, setting the sparrow population size, number of iterations, producer ratio and related parameters of DBN network structure; Step (6): DBN parameter optimization, using FISSA to optimize the learning rate of the RBM layer and the number of neurons in the hidden layer, improve the local optimization caused by randomly setting the DBN initial value, and obtain the optimal initial vector of the DBN model; Step (7): FISSA-DBN model training; Step (8): FISSA-DBN model application.

2. The wind turbine operating status monitoring method based on SCADA and FISSA-DBN model according to claim 1 is characterized in that: In the step (7): according to the research object of wind turbine operating status monitoring, the FISSA-DBN model is trained to obtain the upper and lower limit thresholds of the EWMA control chart.

3. The method for monitoring the operation status of a wind turbine generator set based on SCADA and FISSA-DBN model according to claim 1 is characterized in that: In the step (8), the SCADA data of the actual wind farm is input into the FISSA-DBN model, and the operating status of the wind turbine is monitored using an exponentially weighted average threshold method.

Images