Health state assessment-based active power optimization control method and system for wind farm

Abstract

The application discloses a kind of wind farm active power optimization control method and system based on health state evaluation, the method of the present application includes training component normal behavior model based on the historical operation data of wind farm SCADA system, the mapping relationship between key index, component predicted temperature is established;Based on the real-time operation data of wind farm SCADA system, the input of component normal behavior model is used to obtain component temperature predicted value;The actual measurement of component temperature and its deviation from component temperature predicted value are constructed multidimensional state feature vector, the Mahalanobis distance of multidimensional state feature vector and normal data sample space is calculated;The degradation state type of wind turbine and health degree evaluation result are determined, and the output active power of each wind turbine is adjusted to realize dynamic optimization of wind farm active power scheduling.The present application aims at optimizing the active power distribution of wind farm, reducing the operating temperature and fatigue load of fault wind turbine, prolonging the service life of wind turbine and improving economic benefits.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of wind farm active power optimization, and particularly relates to a wind farm active power optimization control method and system based on health state evaluation. BACKGROUND

[0002] With the rapid development of wind power technology, the scale of wind farms is expanding, and it has become a great challenge to maintain the safety, continuity and profitability of wind turbines. Modern wind turbines can quickly control their output active power, which has stimulated research interest in active power distribution optimization (APC) in wind farms. There are many methods proposed for optimization scheduling in wind farms, which are committed to reducing wind turbine fatigue load and system power loss as much as possible under the conditions of wind power prediction and power system load constraints, avoiding frequent start-stop of units, and also reducing operating costs and improving the power quality of wind farm output power. However, the existing power distribution methods usually do not consider the actual health state of individual wind turbines, and different wind turbines are often considered to be in the same health state, which leads to the fact that the specific conditions of each wind turbine cannot be fully reflected when the power distribution is carried out. SUMMARY

[0003] The technical problem solved by the present application: In view of the above problems of the prior art, the present application provides a wind farm active power optimization control method and system based on health state evaluation, which aims to optimize the active power distribution of the wind farm, reduce the operating temperature and fatigue load of the faulty wind turbine, prolong the service life of the wind turbine and improve the economic benefits.

[0004] In order to solve the above technical problems, the technical scheme adopted by the present application is:

[0005] A wind farm active power optimization control method based on health state evaluation, comprising the following steps:

[0006] 1) Based on the historical operation data of the wind farm SCADA system, normal data samples without fault alarm records are screened, and for the specified multiple components of the wind turbine, multiple key indicators related to component temperature are selected as inputs, and the component temperature is taken as a label to train the component normal behavior model offline, so as to establish the mapping relationship between the key indicators and the component predicted temperature;

[0007] 2) Based on the real-time operation data of the wind farm SCADA system, multiple key indicators of each component are taken as inputs of the component normal behavior model to obtain the component temperature prediction value; a multi-dimensional state feature vector is constructed based on the actual measurement value of the component temperature and the deviation of the component temperature prediction value; and the Mahalanobis distance of the multi-dimensional state feature vector and the normal data sample space is calculated.

[0008] 3) The type of deterioration state of the wind turbine is determined by the difference between the Mahalanobis distance and the preset threshold, and the Mahalanobis distance at each time point in the monitoring period is mapped to the health assessment result of the wind turbine.

[0009] 4) Adjust the output active power of each wind turbine according to the degradation status type and health assessment results of each wind turbine in the wind farm to achieve dynamic optimization of active power scheduling of the wind farm.

[0010] Optionally, the multiple components of the wind turbine specified in step 1) include the main bearing of the wind turbine, and part or all of the components in the nacelle.

[0011] Optionally, the functional expression of the multidimensional state feature vector in step 2) is:

[0012] ;

[0013] in, For the first time during the monitoring period Multidimensional state feature vectors at each time point For the first time during the monitoring period The error between the predicted component temperature and the actual measured component temperature at n time points. The monitoring period from the real-time operation data of the wind farm SCADA system Actual measured temperatures of n components at n time points. ~ These are the 1st to mth state features, where each state feature represents the error between the predicted and actual measured temperatures of a component, or the actual measured temperature of the component, where m = 2n, and n is the number of components specified for the fan. The functional expression for calculating the Mahalanobis distance between the multidimensional state feature vector and the normal data sample space in step 2) is:

[0014] ;

[0015] in, For the first time during the monitoring period Mahalanobis distance between the multidimensional state feature vectors at each time point and the normal data sample space For the first time during the monitoring period Multidimensional state feature vectors at each time point and Let be the mean vector and covariance matrix of the normal data sample space, respectively. for The inverse matrix, in the superscript This is a transpose operation.

[0016] Optionally, in step 3), when determining the deterioration state type of the wind turbine by the difference between the Mahalanobis distance and the preset threshold, the calculation function expression for the difference between the Mahalanobis distance and the preset threshold is:

[0017] ;

[0018] in, This is the difference between the Mahalanobis distance and a preset threshold. For the first time during the monitoring period Mahalanobis distance between the multidimensional state feature vectors at each time point and the normal data sample space The threshold is set; if the difference between the Mahalanobis distance of a certain wind turbine and the preset threshold is... If the value is greater than 0, the fan is determined to be in a deteriorated state; otherwise, the fan is determined to be in a healthy state.

[0019] Optionally, a preset threshold is used. The determination includes: defining events with a probability less than a preset probability as impossible events; subtracting the preset probability from 1 to obtain the value of the two-parameter Weibull probability distribution; and determining the preset threshold according to the following formula. :

[0020] ;

[0021] in, It is a two-parameter Weibull probability distribution. For shape parameters, This is the scale parameter.

[0022] Optionally, in step 3), when mapping the Mahalanobis distance at each time point within the monitoring period to the health assessment result of the wind turbine, the following formula is used to dynamically calculate the degree of wind turbine degradation using a sliding time window:

[0023] ;

[0024] in, Indicates the first Typhoon machine in Degradation rate within each monitoring period The window length represents the monitoring period. Indicates the first The difference between the Mahalanobis distance of the typhoon turbine and a preset threshold; using a fuzzy evaluation algorithm, the degradation degree is mapped to a membership degree by combining a preset membership function, and the membership degree of each typhoon turbine is calculated at the [missing value]. Comprehensive health assessment indicators within each monitoring period:

[0025] ;

[0026] in, For the first Typhoon machine in Comprehensive health assessment indicators within each monitoring period, The membership matrix is ​​obtained by combining the degradation degree with the membership function. Evaluation standard vector The transpose of the evaluation criterion vector It includes a set of boundary values ​​to divide the comprehensive health assessment index into multiple level ranges; the corresponding health assessment result is determined based on the level range into which each wind turbine's comprehensive health assessment index falls.

[0027] Optionally, in step 4), when adjusting the output active power of each wind turbine in the wind farm based on its degradation status type and health assessment results to achieve dynamic optimization of the wind farm's active power dispatch, this includes switching healthy wind turbines to maximum power mode for active power output, and using the comprehensive health assessment index as a load reduction factor for degraded wind turbines, multiplying their active power in the default dispatch mode by the load reduction factor to obtain their adaptive derating active power, thereby allocating wind farm power output commands according to the following formula:

[0028] ;

[0029] in, Output command for the power to be allocated to the wind farm within the i-th power allocation window. For a healthy number of wind turbines, For the k-th healthy wind turbine, the active power output based on the maximum power mode within the i-th power allocation window. The number of deteriorated wind turbines, For the first The comprehensive health assessment index of the typhoon turbine within the i-th power allocation window. The active power output of the kth degraded wind turbine within the i-th power allocation window based on the default scheduling mode.

[0030] Furthermore, the present invention also provides a wind farm active power optimization control system based on health status assessment, including a microprocessor and a memory interconnected thereto, wherein the microprocessor is programmed or configured to execute the wind farm active power optimization control method based on health status assessment.

[0031] Furthermore, the present invention also provides a computer-readable storage medium storing a computer program or instructions that are programmed or configured to execute the wind farm active power optimization control method based on health status assessment by a processor.

[0032] In addition, the application also provides a computer program product comprising a computer program or instructions programmed or configured to execute the health state assessment-based wind farm active power optimization control method by a processor.

[0033] Compared with the prior art, the application has the following beneficial effects: the health state assessment-based wind farm active power optimization control method comprises training a component normal behavior model based on offline modeling of historical operation data of a wind farm SCADA system, establishing a mapping relationship between key indicators and component predicted temperature; based on real-time operation data of the wind farm SCADA system, a plurality of key indicators of each component are taken as inputs of the component normal behavior model to obtain a component temperature prediction value; a multi-dimensional state feature vector is constructed based on a component temperature actual measurement value and a deviation of the component temperature actual measurement value from the component temperature prediction value, a Mahalanobis distance between the multi-dimensional state feature vector and a normal data sample space is calculated; a degradation state type of a wind turbine is determined by a difference between the Mahalanobis distance and a preset threshold value, and the Mahalanobis distance at each time point in a monitoring period is mapped to a health degree assessment result of the wind turbine; and the output active power of each wind turbine is adjusted according to the degradation state type and the health degree assessment result of each wind turbine in the wind farm to realize dynamic optimization of active power scheduling of the wind farm. The application can not only more sensitively detect wind turbine failure, but also significantly reduce wind turbine operating temperature and fatigue load, which helps to prolong the service life of the wind turbine and improve economic benefits. BRIEF DESCRIPTION OF DRAWINGS

[0034] Figure 1 It is a basic flowchart of the embodiment method of the application.

[0035] Figure 2 It is a principle diagram of the embodiment method of the application.

[0036] Figure 3 It is a membership function used in the embodiment of the application.

[0037] Figure 4 It is a framework diagram of the active power optimization distribution of the embodiment of the application.

[0038] Figure 5 It is a simulation diagram of the health degree assessment method of the embodiment of the application on a healthy wind turbine.

[0039] Figure 6 It is a simulation diagram of the health degree assessment method of the embodiment of the application on a faulty wind turbine.

[0040] Figure 7 It is a simulation diagram of the converter operating temperature of a wind turbine under different power control methods in the embodiment of the application. DETAILED DESCRIPTION

[0041] In order for those skilled in the art to better understand the technical solutions of the present application, the technical solutions of the present application will be further described in detail below with reference to the drawings in the embodiments of the present application.

[0042] As shown in Figure 1 and Figure 2 The wind farm active power optimization control method based on health state evaluation in the present embodiment includes the following steps:

[0043] 1) Offline modeling stage: Based on the historical operation data of the wind farm SCADA system, normal data samples with no fault alarm records are screened, and for a plurality of components specified for the wind turbine, a plurality of key indicators related to the component temperature are selected as inputs, and the component temperature is taken as a label to train the component normal behavior model offline, so as to establish a mapping relationship between the key indicators and the component predicted temperature;

[0044] 2) Online monitoring stage: Based on the real-time operation data of the wind farm SCADA system, a plurality of key indicators of each component are taken as inputs of the component normal behavior model to obtain a component temperature prediction value; a multi-dimensional state feature vector is constructed by the actual measurement value of the component temperature and the deviation of the component temperature prediction value; and the Mahalanobis distance of the multi-dimensional state feature vector and the normal data sample space is calculated.

[0045] 3) Health degree evaluation stage: The difference between the Mahalanobis distance and the preset threshold value is used to determine the degradation state type of the wind turbine, and the Mahalanobis distance at each time point in the monitoring period is mapped to the health degree evaluation result of the wind turbine;

[0046] 4) Power optimization stage: The output active power of each wind turbine is adjusted according to the degradation state type of each wind turbine in the wind farm and the health degree evaluation result, so as to realize dynamic optimization of the active power dispatch of the wind farm.

[0047] The plurality of components specified for the wind turbine in step 1) can be specified according to actual needs. For example, as an optional embodiment, the plurality of components specified for the wind turbine in step 1) of the present embodiment includes the main shaft and the nacelle of the wind turbine. For the main shaft, a plurality of key indicators related to the main shaft temperature are selected as inputs from the historical operation data of the wind farm SCADA system, and the main shaft temperature is taken as a label to train the component normal behavior model of the main shaft offline, so as to establish a mapping relationship between the key indicators of the main shaft and the component predicted temperature of the main shaft; and for the nacelle, a plurality of key indicators related to the nacelle temperature are selected as inputs from the historical operation data of the wind farm SCADA system, and the nacelle temperature is taken as a label to train the component normal behavior model of the nacelle offline, so as to establish a mapping relationship between the key indicators of the nacelle and the component predicted temperature of the nacelle.

[0048] In step 1), when multiple key indicators related to the component temperature are selected as inputs, the component temperature, the correlation between the indicators can be calculated according to the required method, the correlation is sorted, and the multiple indicators with the strongest correlation are selected as the multiple key indicators related to the component temperature. The number of key indicators can also be adjusted according to actual needs. For example, as an optional embodiment, the ReliefF method is used to calculate the correlation between the component temperature and the indicators, and the correlation is sorted, and the 15 indicators with the strongest correlation are selected as the 15 key indicators related to the component temperature. The ReliefF method is a known correlation calculation method, and the core principle is that key features can more clearly distinguish different categories of samples, that is, to make the same samples closer and non-same samples farther apart.

[0049] In step 1), the component normal behavior model can use the required machine learning model according to the needs. For example, as an optional embodiment, the gated recurrent unit neural network (GRU) is used as the component normal behavior model in this embodiment. The gated recurrent unit neural network (GRU) is a multi-input single-output model, the input features are the multiple key indicators in the SCADA with the highest relevance to the prediction target, and the output is the corresponding component prediction temperature, such as the main bearing temperature and the engine room temperature, so as to respectively establish the behavior benchmarks of the main bearing temperature and the engine room temperature under the normal working condition of the wind turbine.

[0050] Through the offline modeling stage of step 1), the component normal behavior model of each component is established, which establishes the mapping relationship between the key indicators and the component prediction temperature, so that it can be used for the prediction of the component prediction temperature. In step 2), by using the real-time running data based on the SCADA system of the wind farm, the multiple key indicators of each component are used as the input of the component normal behavior model, and the component temperature prediction value can be obtained.

[0051] The function expression of the multi-dimensional state feature vector in step 2) of the embodiment is:

[0052] ;

[0053] Wherein, is the multi-dimensional state feature vector at the i th time point in the monitoring period, is the error of the component temperature prediction value and the actual measured value of the component temperature of the n components at the i th time point in the monitoring period, is the actual measured value of the component temperature of the n components at the i th time point in the monitoring period from the real-time running data of the SCADA system of the wind farm, ​​​​respectively, are the first to mth state features, the state features are errors of component temperature predicted values and component temperature actual measured values of the components or the component temperature actual measured values, m = 2n, wherein n is the number of components designated for the fan; the function expression for calculating the Mahalanobis distance of the multidimensional state feature vector and the normal data sample space in step 2) is:

[0054] ;

[0055] wherein, is the Mahalanobis distance of the multidimensional state feature vector at the i th time point in the monitoring period and the normal data sample space, is the multidimensional state feature vector at the i th time point in the monitoring period, and are a mean vector and a covariance matrix of the normal data sample space respectively, is an inverse matrix of , and the superscript is a transposition operation. The function expression for calculating the difference between the Mahalanobis distance and the preset threshold value in step 3) of the embodiment for determining the degradation state type of the fan is:

[0056]

[0057] ;

[0058] wherein, is the difference between the Mahalanobis distance and the preset threshold value, is the Mahalanobis distance of the multidimensional state feature vector at the i th time point in the monitoring period and the normal data sample space, is the preset threshold value; if the difference between the Mahalanobis distance and the preset threshold value of a certain fan is greater than 0, it is determined that the degradation state type of the fan is degradation, otherwise it is determined that the degradation state type of the fan is healthy. The preset threshold value corresponding to the Mahalanobis distance can be set according to actual needs. For example, as an optional embodiment, the preset threshold value corresponding to the Mahalanobis distance in the embodiment is fitted by using a two-parameter Weibull probability distribution. The determination of the preset threshold value includes: defining an event with a probability less than a preset probability as an impossible event, taking 1 minus the preset probability as the value of the two-parameter Weibull probability distribution, and determining the preset threshold value according to the following formula:

[0059] ;

[0060] wherein, ​​​​It is a two-parameter Weibull probability distribution. For shape parameters, This is the scaling parameter. For example, events with a probability less than 0.05 are generally considered impossible. Therefore, the value of the two-parameter Weibull probability distribution can be defined as 0.95. Substituting this value into the function expression of the two-parameter Weibull probability distribution yields the preset threshold. The value of overcomes the shortcoming of the traditional threshold method where the error value characteristics do not change significantly.

[0061] In step 3) of this embodiment, when mapping the Mahalanobis distance at each time point within the monitoring period to the health assessment result of the wind turbine, the following formula is used to dynamically calculate the degree of deterioration of the wind turbine using a sliding time window:

[0062] ;

[0063] in, Indicates the first Typhoon machine in Degradation rate within each monitoring period The window length represents the monitoring period. Indicates the first The difference between the Mahalanobis distance of the typhoon generator and the preset threshold; using a fuzzy evaluation algorithm, the degradation degree is mapped to the [0,1] interval by combining the preset membership function, and the membership degree of each typhoon generator is calculated in the [0,1] interval. Comprehensive health assessment indicators within each monitoring period:

[0064] ;

[0065] in, For the first Typhoon machine in Comprehensive health assessment indicators within each monitoring period, The membership matrix is ​​obtained by combining the degradation degree with the membership function. Evaluation standard vector The transpose of the evaluation criterion vector It includes a set of boundary values ​​to divide the comprehensive health assessment index into multiple level ranges; the corresponding health assessment result is determined based on the level range into which each wind turbine's comprehensive health assessment index falls.

[0066] Figure 3 The membership functions used in this embodiment are g1 to g4, which represent the degree of degradation. They include the membership functions of the four health assessment results: "Good", "Attention", "Warning" and "Danger". The membership functions used for "Attention" and "Warning" are trigonometric functions, while the membership functions used for "Good" and "Danger" are semi-trapezoidal functions.

[0067] The evaluation standard vector is introduced The purpose is to re-fuzzify the health degree and quantify the health degree as an index, maximize the use of the information of the four membership degrees, and facilitate the optimization of active power distribution by using the comprehensive evaluation index. The four membership degrees are integrated into a comprehensive health evaluation index by the evaluation standard vector , which simplifies the judgment of the health state, quantifies the fuzzy evaluation result, realizes the intuitive classification of the health state, and realizes the rapid decision. As an optional implementation, the evaluation standard vector in this embodiment takes the value as:

[0068] ;

[0069] When the degradation degree corresponding to the membership degree matrix at a certain moment is , the membership degree corresponding to the evaluation of "attention" is the largest, that is, the preliminary evaluation result of the health degree of the fan at the moment is "attention" by the fuzzy evaluation algorithm, and the corresponding comprehensive health evaluation index is 0.5*0.8+0.6*0.6+0*0.4+0*0.2=0.76.

[0070] The evaluation standard vector includes a group of boundary values for dividing the comprehensive health evaluation index into multiple level ranges, that is, it can be divided into four level ranges greater than or equal to 0.8, [0.6, 0.8), [0.4, 0.6), and less than 0.4. Finally, if the comprehensive health evaluation index is greater than or equal to 0.8, the health degree evaluation result is "good"; if the comprehensive health evaluation index is between [0.6, 0.8), the health degree evaluation result is "attention"; if the comprehensive health evaluation index is between [0.4, 0.6), the health degree evaluation result is "warning"; if the comprehensive health evaluation index is less than 0.4, the health degree evaluation result is "dangerous". Since 0.76 is between [0.6, 0.8), the final corresponding health degree evaluation result of the fan is also "attention".

[0071] In step 4) of this embodiment, the output active power of each fan is adjusted according to the degradation state type and the health degree evaluation result of each fan in the wind farm to realize dynamic optimization of the active power scheduling of the wind farm. The dynamic optimization strategy for the active power scheduling of the wind farm is to implement adaptive capacity reduction operation on the degraded fan, and at the same time, to improve the power quota of the healthy fan, including switching the healthy fan to the maximum power mode for active power output, and taking the comprehensive health evaluation index of the degraded fan as a load shedding coefficient, and multiplying the active power under the default scheduling mode by the load shedding coefficient as the active power after adaptive capacity reduction, so as to distribute the wind farm power output instruction according to the following formula:

[0072] ;

[0073] in, Output command for the power to be allocated to the wind farm within the i-th power allocation window. For a healthy number of wind turbines, For the k-th healthy wind turbine, the active power output based on the maximum power mode within the i-th power allocation window. The number of deteriorated wind turbines, For the first The comprehensive health assessment index of the typhoon turbine within the i-th power allocation window. Let the active power output of the k-th degraded wind turbine be based on the default scheduling mode within the i-th power allocation window. Let the time period of the power allocation window be denoted as . The time period of the monitoring cycle is as follows: This allows for the integration of the power allocation window's time period. Time period of monitoring cycle Optimize the rolling time of the wind turbine. Referring to the formula for allocating power output commands to the wind farm mentioned above, it can be seen that for... For a healthy wind turbine, the active power output is based on the maximum power mode. For example, MPPT technology is used to extract maximum power from the wind until the rated power is reached, thereby improving the power quota of the healthy wind turbine; for For a deteriorated wind turbine, its comprehensive health assessment index is used as a load reduction factor, thereby achieving the goal of adaptive capacity reduction operation for the deteriorated wind turbine.

[0074] Figure 4 This is a framework diagram of the active power optimization allocation in this embodiment. After receiving the wind farm power output command to be allocated, the power control center uses the comprehensive health assessment index of the deteriorated wind turbine as the load reduction factor, and multiplies the active power of the turbine in the default scheduling mode by the load reduction factor to obtain its adaptive derating active power. Based on the above formula, the power output command of the wind farm is allocated to N wind turbines WT1 to WT2. N , where P1~P N Then, the power allocated to N wind turbines (i.e., or Meanwhile, the wind farm's SCADA system monitors N wind turbines WT1 to WT2. N Status monitoring is performed to obtain real-time operational data. The health assessment method described in this embodiment is used to obtain the WT1~WT values ​​for N wind turbines. N The comprehensive health assessment indicators are sent to the power control center for power allocation in the next cycle.

[0075] The simulation verification is performed on the wind farm active power optimization control method based on the health state evaluation of the embodiment, and the obtained results are as shown in Figures 5 to 7 Figure 5 The simulation graph of the healthy wind turbine using the method of the embodiment is shown in the figure. Although the Mahalanobis distance of the healthy wind turbine exceeds the threshold value instantaneously, this is caused by the abnormality of the instantaneous SCADA data acquisition or the transient operation process of the wind turbine. Since the sliding time window comprehensively considers the Mahalanobis distance error within three days of the monitoring period, the unit degradation degree remains at a very low level. It can be seen that the method can correctly evaluate the state of the healthy wind turbine and has a certain anti-interference ability. Figure 6 The simulation graph of the faulty wind turbine using the method of the embodiment is shown in the figure. Before the fault is cleared, the unit degradation degree of the faulty wind turbine will obviously rise. Compared with the abnormal monitoring method using the error value and the simple threshold value, the method of the embodiment detects the state abnormality through the Mahalanobis distance statistical outlier detection technology, and overcomes the shortcoming that the error value feature change is not obvious using the traditional threshold method. Figure 7 The converter operating temperature simulation graph of the wind turbine under the power control method of the embodiment is shown in the figure. The faulty wind turbine is automatically controlled by the power optimization method to run in the reduced capacity mode during the fault time. Compared with the traditional proportional power distribution method, the method of the embodiment can keep the converter temperature at a lower level, and avoid further damage to the device due to the excessively high temperature during the fault. At the same time, the lower power output can significantly reduce the torque of the wind turbine, thereby reducing the fatigue load of the wind turbine, and being more conducive to the protection of the faulty wind turbine.

[0076] In summary, in view of the technical problems existing in the existing wind farm active power distribution technology, the method of the embodiment establishes the normal behavior model of each key component of the wind turbine in the offline stage through the neural network, constructs a multi-dimensional state feature vector through the model output prediction value and the actual measured value, monitors the abnormality by using the Mahalanobis distance, and quantifies the overall health state of the wind turbine by using the fuzzy evaluation algorithm. The active power distribution of the wind farm is optimized, the operating temperature and fatigue load of the faulty wind turbine can be reduced, which is helpful to prolong the service life of the wind turbine and improve the economic benefit.

[0077] ​In addition, the embodiment also provides a wind farm active power optimization control system based on health state evaluation, comprising a microprocessor and a memory connected with each other, the microprocessor is programmed or configured to execute the wind farm active power optimization control method based on health state evaluation. The embodiment also provides a computer readable storage medium, the computer readable storage medium stores a computer program or instructions, the computer program or instructions are programmed or configured to execute the wind farm active power optimization control method based on health state evaluation by the processor. The embodiment also provides a computer program product, comprising a computer program or instructions, the computer program or instructions are programmed or configured to execute the wind farm active power optimization control method based on health state evaluation by the processor.

[0078] The present application can realize all or part of the processes in the above-mentioned embodiment methods, and can also be completed by computer program instruction related hardware. The computer program can be stored in a computer readable storage medium or provided in the form of a software mall product for online installation. When the computer program is executed by a processor, the steps of the above-mentioned method embodiment can be realized. The computer program includes computer program code, which can be in the form of source code, object code, executable files or some intermediate forms. The computer readable storage medium includes any entity or device capable of carrying computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. The memory is used to store computer programs and / or modules, and the processor realizes various functions by running or executing the computer programs and / or modules stored in the memory and calling the data stored in the memory. The memory can include high-speed random access memory and can also include non-volatile memory such as hard disk, memory, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one magnetic disk storage device, flash memory device, or other volatile solid-state memory device, etc.

[0079] The above is only the preferred embodiment of the present application, and the protection scope of the present application is not limited to the above-mentioned embodiment. Any technical solution falling within the concept of the present application shall fall within the protection scope of the present application. It should be noted that, for ordinary skilled persons in the technical field, some improvements and refinements without departing from the principle of the present application shall be regarded as the protection scope of the present application.

Claims

1. A health state evaluation-based wind farm active power optimization control method, characterized in that, The method comprises the following steps: 1) Based on the historical operation data of the wind farm SCADA system, normal data samples of non-fault alarm records are screened out, and for the specified multiple components of the wind turbine, multiple key indicators related to the component temperature are selected as inputs, and the component temperature is taken as a label to train the component normal behavior model offline, so as to establish the mapping relationship between the key indicators and the predicted component temperature; 2) Based on the real-time operation data of the wind farm SCADA system, the multiple key indicators of each component are taken as the input of the component normal behavior model to obtain the predicted component temperature; the actual measured value of the component temperature and the deviation of the predicted component temperature are used to construct a multi-dimensional state feature vector, and the Mahalanobis distance between the multi-dimensional state feature vector and the normal data sample space is calculated; 3) The difference between the Mahalanobis distance and the preset threshold value is used to determine the degradation state type of the wind turbine, and the Mahalanobis distance at each time point in the monitoring period is mapped to the health degree evaluation result of the wind turbine; 4) The output active power of each wind turbine is adjusted according to the degradation state type and the health degree evaluation result of each wind turbine in the wind farm to realize dynamic optimization of the active power dispatch of the wind farm.

2. The health status assessment based wind farm active power optimization control method according to claim 1, characterized in that, The multiple components specified by the wind turbine in step 1) include the main bearing and the nacelle of the wind turbine.

3. The health status assessment based wind farm active power optimization control method according to claim 1, characterized in that, In step 2), the function expression of the multi-dimensional state feature vector is: ； wherein, is a multi-dimensional state feature vector at the i-th time point in the monitoring period, is an error of the component temperature predicted value and the component temperature actual measurement value of the n components at the i-th time point in the monitoring period, is a component temperature actual measurement value of the n components at the i-th time point in the monitoring period from the real-time operation data of the wind farm SCADA system, are the 1st to mth state features, respectively, which are the error of the component temperature predicted value and the component temperature actual measurement value or the component temperature actual measurement value of the components, and m = 2n, wherein n is the number of components designated for the wind turbine; and the function expression for calculating the Mahalanobis distance of the multi-dimensional state feature vector and the normal data sample space in step 2) is:​​​​ ； wherein is the Mahalanobis distance of the multi-dimensional state feature vector at the time point within the monitoring period to the normal data sample space, is the multi-dimensional state feature vector at the time point within the monitoring period, and are the mean vector and the covariance matrix of the normal data sample space, respectively, is the inverse matrix of , and is the transpose operation.

4. The health status assessment based wind farm active power optimization control method according to claim 1, characterized in that, In step 3), when the difference between the Mahalanobis distance and the preset threshold value is used to determine the degradation state type of the wind turbine, the calculation function expression of the difference between the Mahalanobis distance and the preset threshold value is: ； in, This is the difference between the Mahalanobis distance and a preset threshold. For the first time during the monitoring period Mahalanobis distance between the multidimensional state feature vectors at each time point and the normal data sample space The threshold is set; if the difference between the Mahalanobis distance of a certain wind turbine and the preset threshold is... If the value is greater than 0, the fan is determined to be in a deteriorated state; otherwise, the fan is determined to be in a healthy state.

5. The health status assessment based wind farm active power optimization control method according to claim 4, characterized in that, Pre-set threshold The determination includes: defining an event with a probability of occurrence less than a pre-set probability as an impossible event, subtracting the pre-set probability from 1 as a value of the two-parameter Weibull probability distribution, and determining the pre-set threshold according to the following formula : ； wherein, is a two-parameter Weibull probability distribution, is a shape parameter, is a scale parameter.

6. The health status assessment based wind farm active power optimization control method according to claim 1, wherein, In step 3), when the Mahalanobis distance at each time point in the monitoring period is mapped to the health degree evaluation result of the wind turbine, the degradation degree of the wind turbine is dynamically calculated by using a sliding time window according to the following formula: ； wherein, represents the degradation degree of the typhoon machine in the first monitoring period, represents the degradation degree of the typhoon machine in the first monitoring period, represents the window length of the monitoring period, represents the Mahalanobis distance of the typhoon machine at the first time point in the monitoring period and the preset threshold value; the degradation degree is mapped to the membership degree by using a fuzzy evaluation algorithm combined with a preset membership function, and a comprehensive health evaluation index of each typhoon machine in the first monitoring period is calculated. ； wherein, is the comprehensive health evaluation index of the typhoon machine in the first monitoring period, is the comprehensive health evaluation index of the typhoon machine in the second monitoring period, is the comprehensive health evaluation index of the typhoon machine in the monitoring period, is a membership matrix, which is obtained by combining the degradation degree with a membership function, is a transpose of an evaluation standard vector , and the evaluation standard vector includes a set of boundary values for dividing the comprehensive health evaluation index into a plurality of grade ranges; and the health degree evaluation result is determined according to the grade range into which the comprehensive health evaluation index of each typhoon machine falls.

7. The health status assessment based wind farm active power optimization control method according to claim 6, characterized in that, In step 4), when the output active power of each wind turbine is adjusted according to the degradation state type and the health degree evaluation result of each wind turbine in the wind farm to realize dynamic optimization of the active power dispatch of the wind farm, the wind turbine is switched to the maximum power mode for active power output for the healthy wind turbine, and the comprehensive health evaluation index of the degraded wind turbine is taken as a load shedding coefficient, and the active power of the degraded wind turbine under the default dispatch mode is multiplied by the load shedding coefficient to obtain the active power after adaptive capacity reduction, so that the wind farm power output instruction is distributed according to the following formula: ； in, Output command for the power to be allocated to the wind farm within the i-th power allocation window. For a healthy number of wind turbines, For the k-th healthy wind turbine, the active power output based on the maximum power mode within the i-th power allocation window. The number of deteriorated wind turbines, For the first The comprehensive health assessment index of the typhoon turbine within the i-th power allocation window. For the first The degraded wind turbine outputs active power based on the default scheduling mode within the i-th power allocation window.

8. A health status assessment based wind farm active power optimization control system comprising a microprocessor and a memory connected to each other, characterized in that, The microprocessor is programmed or configured to execute the wind farm active power optimization control method based on health state evaluation according to any one of claims 1-7.

9. A computer-readable storage medium having stored therein a computer program or instructions, characterized in that, The computer program or instructions are programmed or configured to execute the wind farm active power optimization control method based on health state evaluation according to any one of claims 1-7 by the processor.

10. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are programmed or configured to execute the wind farm active power optimization control method based on health state evaluation according to any one of claims 1-7 by the processor.

Images