Hydraulic turbine speed regulation optimization control method and system based on neural network

Abstract

The application relates to the technical field of water turbine control, in particular to a water turbine speed regulation optimization control method and system based on a neural network, which comprises the following steps: collecting each kind of monitoring data of each group of data under each kind of working condition of a water turbine, obtaining interference significant coefficients of each modal component of each kind of monitoring data in each group of data under the same working condition, performing frequency domain conversion on each modal component of each kind of monitoring data, obtaining interference influence coefficients of each modal component of each kind of monitoring data in each group of data under the same working condition, then obtaining adjustment coefficients corresponding to each modal component, performing noise reduction processing on each modal component to obtain adjustment thresholds of each modal component of each kind of monitoring data in each group of data under the same working condition, performing regional division after linearization processing of each kind of monitoring data, constructing a multi-objective optimization function of a linear model of each region, and then performing speed regulation optimization control on the water turbine through the neural network. The application improves the precision of water turbine speed regulation optimization control.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of water turbine control, in particular to a water turbine speed regulation optimization control method and system based on a neural network. BACKGROUND

[0002] A water turbine regulation system is a core link for stable operation of a power system, and can control voltage and power grid frequency within a certain range to ensure safe and stable power consumption. However, due to the strong nonlinearity, time-varying and coupling characteristics of the water turbine, a traditional PID controller is difficult to adapt to full-condition operation. For example, under high penetration of new energy, load fluctuation is intensified, and the traditional control method is prone to response lag and large overshoot. Therefore, the self-adaptability of the nonlinear model constructed based on hardware, combined with intelligent algorithms and neural network technology, to optimize the PID parameters, is a key direction to improve the performance of the speed regulation system, and can meet the efficient and intelligent operation requirements.

[0003] In the process of water turbine speed regulation optimization, the data collected under different conditions are usually processed by modal decomposition and wavelet denoising. In the actual processing process, due to sudden jumps, power grid short-circuit faults, line switching and other situations, the interference of water turbine monitoring data has the characteristics of multi-band, multi-source and time-varying, resulting in large differences in the interference of collected data under different conditions, which further affects the data processing under different conditions and the accuracy of the neural network model in judging the characteristics of the conditions, and thus affects the optimization effect of the PID parameters, and the accuracy of the water turbine speed regulation optimization based on the neural network is generally low. SUMMARY

[0004] To solve the above technical problems, the purpose of the present application is to provide a water turbine speed regulation optimization control method and system based on a neural network, and the technical solution adopted is as follows:

[0005] The application embodiment provides a water turbine speed regulation optimization control method based on a neural network, which comprises the following steps:

[0006] Collecting each kind of monitoring data of each group of data under each condition of the water turbine;

[0007] By analyzing the differences in the same frequency band modal components of the same kind of monitoring data of different groups of data under the same condition after modal decomposition, and combining the feature similarity of different kinds of monitoring data in each modal component of each group of data under the same condition, the interference significant coefficient of each modal component of each kind of monitoring data in each group of data under the same condition is obtained.

[0008] The frequency domain conversion is performed on each modal component of each monitoring data in each group of data under the same working condition, the interference influence coefficient of each modal component of each monitoring data in each group of data under the same working condition is obtained according to the fluctuation of all peak values in the frequency domain, and the adjustment coefficient of each modal component of each monitoring data in each group of data under the same working condition is obtained in combination with the interference significant coefficient of each corresponding modal component, so that the threshold noise reduction processing is performed on each modal component of each monitoring data in each group of data under the same working condition, and then all modal components of each monitoring data in each group of data under the same working condition are reconstructed.

[0009] After linear processing of each monitoring data of each group of data under different working conditions is performed, the linear model of each region is obtained through regional division, a multi-objective optimization function of the linear model of each region is constructed, the optimal PID parameter corresponding to the obtained optimal result is input into the neural network, and then the speed regulation optimization of the water turbine is performed through the neural network.

[0010] Preferably, the calculation method of the interference significant coefficient of each modal component of each monitoring data in each group of data under the same working condition is as follows:

[0011] ;

[0012] In the formula, is the interference significant coefficient of the i th modal component of each monitoring data in the j th group of data under the same working condition; is the eigenvalue of the i th modal component of each monitoring data in the j th group of data under the same working condition; is the eigenvalue of the i th modal component of each monitoring data in the j th group of data under the same working condition; is the eigenvalue of the i th modal component of each monitoring data in the j th group of data under the same working condition; is the eigenvalue of the i th modal component of each monitoring data in the j th group of data under the same working condition; is the eigenvalue of the i th modal component of each monitoring data in the j th group of data under the same working condition; is the eigenvalue of the i th modal component of each monitoring data in the j th group of data under the same working condition; indicates an exponential function with a natural constant as a base number; is the number of groups collected under each working condition.

[0013] Preferably, the eigenvalue further includes:

[0014] The mean value of all DTW distances of the same monitoring data in the same frequency band modal component between each group of data and other groups of data under the same working condition is taken as the eigenvalue of each modal component of each monitoring data of each group of data under the same working condition.

[0015] Preferably, the acquisition method of the eigenvalue is as follows:

[0016] The eigenvalues of each modal component of each monitoring data in each group of data under the same working condition are counted to form an eigenvector of each monitoring data in each group of data under the same working condition; and the average of all cosine similarities between each monitoring data and the eigenvectors of other monitoring data in each group of data under the same working condition is taken as a characteristic coefficient of each monitoring data in each group of data under the same working condition.

[0017] Preferably, the interference influence coefficient of each modal component of each monitoring data in each group of data under the same working condition comprises:

[0018] The frequency spectrum diagram of each modal component of each monitoring data in each group of data under the same working condition is converted to obtain all peak values in each modal component spectrum diagram, the average of the kurtosis values of all peak values in each modal component spectrum diagram is counted to obtain a first eigenvalue of each modal component, and the average of the Euclidean distances between all adjacent peak values in each modal component spectrum diagram is counted as a second eigenvalue of each modal component, and then the product of the first eigenvalue of each modal component and the second eigenvalue of each modal component is taken as the interference influence coefficient of each modal component of each monitoring data in each group of data under the same working condition.

[0019] Preferably, the calculation method of the adjustment coefficient of each modal component of each monitoring data in each group of data under the same working condition is:

[0020] ;

[0021] In the formula, is the adjustment coefficient of the i-th modal component of each monitoring data in the j-th group of data under the same working condition; is the adjustment coefficient of the i-th modal component of each monitoring data in the j-th group of data under the same working condition; is the interference significant coefficient of the i-th modal component of each monitoring data in each group of data under the same working condition; is the interference significant coefficient of the i-th modal component of each monitoring data in each group of data under the same working condition; is the interference influence coefficient of the i-th modal component of each monitoring data in the j-th group of data under the same working condition; is the interference influence coefficient of the i-th modal component of each monitoring data in the j-th group of data under the same working condition. Preferably, when the threshold noise reduction processing is performed, the adjustment coefficient is used to set the adjustment threshold to realize the threshold noise reduction processing, and the method of setting the adjustment threshold further comprises:

[0022]

[0023] ;

[0024] In the formula, is the adjustment threshold of each modal component of each monitoring data in each group of data under the same working condition; is the initial threshold of each modal component of each monitoring data in each group of data under the same working condition; ​​a normalization result of an adjustment coefficient of each modal component of each monitoring data in each group of data under the same working condition; a preset adjustment judgment threshold value of each modal component of each monitoring data in each group of data under the same working condition.

[0025] Preferably, the method for obtaining the initial threshold value is:

[0026] each modal component of each monitoring data in each group of data under the same working condition is subjected to soft threshold processing, to obtain an initial threshold value of each modal component of each monitoring data in each group of data under the same working condition.

[0027] Preferably, the method for calculating the multi-objective optimization function of the linear model of each region is:

[0028] ;

[0029] In the formula, is an overshoot of each regional linear model; is a stable time of each regional linear model; is a steady-state error of each regional linear model; is a first weight coefficient; is a second weight coefficient; is a third weight coefficient; is an optimal result of the target optimization function; wherein, ; the overshoot, the stable time and the steady-state error of the linear model are dynamic performance indexes of the linear model constructed through system identification technology, and are used to obtain optimal PID parameters.

[0030] The embodiment of the application further provides a neural network-based water turbine speed regulation optimization control system, including a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements the steps of the neural network-based water turbine speed regulation optimization control method.

[0031] As can be seen from the above, the neural network-based water turbine speed regulation optimization control method and system provided by the application has at least the following beneficial effects:

[0032] The application fully considers that there are many interference sources in the operation environment of the water turbine, and the corresponding interference characteristics are complex, which leads to large differences in the interference of the collected data under different working conditions, affects the accuracy of the working condition characteristic judgment of the neural network model, and makes the accuracy of the water turbine speed regulation optimization low; therefore, the data under the full working condition of the water turbine is collected, and the influence of the complex environmental interference source on the actual control process of the water turbine is considered, which may cause large judgment errors, the interference characteristics of the monitoring data in different frequency bands under different working conditions are analyzed, according to the strong coupling and the concentration characteristics of high-frequency interference in the operation process, the significant characteristics of the interference of each monitoring data in different frequency bands under different working conditions are analyzed, to determine the adjustment coefficient of the optimization adjustment of each monitoring data under different working conditions. By fully considering the influence of the operation environment of the water turbine on the speed regulation optimization control judgment, the collected data under different working conditions is accurately processed, and then a precise water turbine speed regulation optimization sample library is constructed, the accuracy of the working condition characteristic judgment of the constructed neural network model is improved, and the accuracy of the water turbine speed regulation optimization control based on the neural network is improved. BRIEF DESCRIPTION OF DRAWINGS

[0033] In order to more clearly illustrate the technical solutions and advantages of the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0034] Figure 1 The step flow chart of the water turbine speed regulation optimization control method based on the neural network provided by the present application is provided.

[0035] Figure 2 The step flow chart of the adjustment coefficient acquisition method provided by the present application is provided. DETAILED DESCRIPTION

[0036] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined invention purposes, the following describes the water turbine speed regulation optimization control method and system based on the neural network according to the present application, its specific implementation, structure, features and effects in detail, combined with the drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0037] Unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be apparent to those skilled in the art that various modifications and variations can be made to the present application without departing from the spirit or scope of the application. Thus, it is intended that the present application cover modifications and variations of this application provided they come within the scope of the appended claims and their equivalents.

[0038] The specific scheme of the neural network-based water turbine speed regulation optimization control method and system provided by the present application will be specifically described below in combination with the drawings.

[0039] Please refer to Figure 1 which shows the step flow chart of the neural network-based water turbine speed regulation optimization control method provided by an embodiment of the present application, including the following steps:

[0040] Step one: collect each kind of monitoring data of each group of data under each working condition of the water turbine.

[0041] The monitoring data under different working conditions of the water turbine is collected. Through the operation of the water turbine under each working condition, for each kind of monitoring data collected in the operation process, at least 30 groups of data are collected under each working condition, each group of data corresponds to a time period, all kinds of monitoring data are collected in each time period, and then the collected data is analyzed and processed. Preferably, in the present embodiment, the monitoring data includes speed deviation, real-time water head, power grid load and guide vane opening; the working conditions involved in the actual operation of the water turbine include but are not limited to starting, grid connection, load increase and decrease and load shedding.

[0042] Step two: through the difference of the same kind of monitoring data of different groups of data under the same working condition after modal decomposition, and combining the characteristic similarity of different kinds of monitoring data in each group of data under the same working condition in each modal component, the interference significant coefficient of each modal component of each kind of monitoring data in each group of data under the same working condition is obtained.

[0043] Due to the energy output fluctuation and industrial load change in the operation process of the water turbine, the power grid load suddenly jumps, showing the characteristics of high frequency, short time and high amplitude, and further causing the speed and guide vane opening to fluctuate sharply. In addition, the short-circuit fault and line switching of the power grid also produce high-frequency transient characteristics. At the same time, the hydraulic pulsation and mechanical vibration in the water turbine will cause periodic non-stationary characteristics. Therefore, due to the complexity of the interference source in the actual operation environment, the interference characteristics of the water turbine under different working conditions may be more different, and the change characteristics of each kind of monitoring data under different frequency bands may also be more significant.

[0044] Based on the above analysis, in the process of building a neural network model to analyze and extract the characteristics of the monitoring data collected under each working condition, if the differences in the interference of the monitoring data under each working condition are large, it may cause the judgment error of the neural network model for the running state of each working condition to increase. Therefore, in order to accurately realize the speed optimization control of the water turbine based on the neural network model, avoid the deviation of each kind of monitoring data under each working condition caused by the differences in the interference source and the interference characteristics, and further affect the recognition and adjustment effect of the neural network model, it is necessary to analyze the characteristics of each kind of monitoring data collected under each working condition, so as to improve the accuracy of the neural network in the speed optimization control of the water turbine.

[0045] Considering that in the operation process of the water turbine under each working condition, due to the influence of complex interference sources, each kind of monitoring data collected will show sudden and periodic high-frequency interference characteristics in different frequency bands. For each group of data collected when the water turbine operates under each working condition, align each kind of monitoring data in each group of data according to the collection time sequence. In addition, the interference will cause the differences in the interference changes in different frequency bands, so arrange each kind of monitoring data in each group of data according to the collection time sequence, and the sequence after the arrangement is used as the monitoring time sequence of each kind of monitoring data in each group of data. The ensemble empirical mode decomposition (EEMD) algorithm is used to decompose the monitoring time sequence of each kind of monitoring data, so as to obtain 5 modal components after the decomposition of each kind of monitoring data, and the 5 modal components correspond to different frequency bands. Preferably, in the present embodiment, the empirical mode decomposition algorithm is a known technology, and the specific process will not be described here.

[0046] For the same monitoring data in each group under the same operating conditions, this study analyzes the characteristic differences between the same monitoring data in different frequency bands due to interference factors, thereby accurately analyzing the interference impact characteristics of turbine operation under the same operating conditions. Specifically, for the same monitoring data in all groups under the same operating conditions, the DTW distance of each group of data with the same monitoring data in the same frequency band is calculated. The larger the DTW distance, the more significant the data variation difference may occur due to interference under the same operating conditions. The mean of all DTW distances of each group of data with the same monitoring data in the same frequency band is used as the characteristic value of each modal component of each monitoring data in each group of data under the same operating conditions.

[0047] Furthermore, considering the strong coupling characteristics among different monitoring data in the turbine during actual monitoring, i.e., the changes in different monitoring data in the turbine are correlated, a comprehensive analysis is conducted on the interference differences between different monitoring data in each data set under the same operating conditions. Specifically, for each data set under the same operating conditions, the eigenvalues ​​of all modal components of each monitoring data in each data set are statistically analyzed to form the eigenvector of each monitoring data; the cosine similarity between the eigenvectors of each monitoring data and other monitoring data in each data set is calculated; and the mean of all cosine similarities in each data set is used as the characteristic coefficient of each monitoring data in each data set under the same operating conditions.

[0048] Based on the above analysis, according to the characteristic values ​​of each modal component of each monitoring data in each set of data under the same working conditions, and combined with the characteristic similarity of each monitoring data in each set of data with other monitoring data in each modal component, the interference significance coefficient of each modal component of each monitoring data in each set of data under the same working conditions is calculated. In this embodiment, the specific calculation formula is as follows:

[0049] ;

[0050] In the formula, For each monitoring data in each set of data under the same working conditions, the first... The significance coefficient of interference for each modal component; For the first under the same working conditions The first of each monitoring data in the group data Eigenvalues ​​of each modal component; For the first under the same working conditions Characteristic coefficients for each monitoring data point in the dataset; Represents an exponential function with the natural constant as the base; This represents the number of data sets collected under each operating condition.

[0051] in, The greater the value, the more significant the difference in interference influence of each monitoring data in each group of data on different frequency bands due to the complex interference source in the water turbine operating environment under the same working condition. The greater the value, the more significant the difference in data change of the current monitoring data in the corresponding frequency band due to the interference under the same working condition. The smaller the value, the greater the difference in data change between different monitoring data due to the high-frequency interference characteristics of the complex interference source during the same group of data acquisition under the same working condition, and the more significant the difference in data change of the current monitoring data due to the interference of the water turbine operating environment.

[0052] Step three: frequency domain conversion is performed on each modal component of each monitoring data in each group of data under the same working condition, the interference influence coefficient of each modal component of each monitoring data in each group of data under the same working condition is obtained according to the fluctuation of all peak values in the frequency domain, and the adjustment coefficient of each modal component of each monitoring data in each group of data under the same working condition is obtained by combining the interference significant coefficient of each modal component, so as to perform threshold noise reduction processing on each modal component of each monitoring data in each group of data under the same working condition, and obtain the adjustment threshold of each modal component of each monitoring data in each group of data under the same working condition, and then reconstruct all modal components of each monitoring data in each group of data under the same working condition.

[0053] Therefore, combined with the characteristics of the complex interference source in the actual monitoring process of the water turbine and the coupling characteristics of the multi-source monitoring data of the water turbine, the interference influence difference of different monitoring data in different frequency bands in each group of data under the same working condition can be obtained through comparative analysis of each group of data under the same working condition. Specifically, considering the interference influence peak characteristics of different monitoring data in different frequency bands during the operation of the water turbine, for each group of data under the same working condition, each modal component of each monitoring data in each group of data is taken as input, Fourier transform is used to obtain the frequency spectrum diagram of each modal component, and the findpeaks function of MATLAB is used to extract all peak values in the frequency spectrum diagram of each modal component. Then, the kurtosis value corresponding to each peak value in the frequency spectrum diagram is counted, and the Euclidean distance between adjacent peak values in the frequency spectrum diagram is counted. The mean value of all kurtosis values of each modal component is taken as the first characteristic value of each modal component, the mean value of all Euclidean distances between adjacent peak values in each modal frequency spectrum is taken as the second characteristic value of each modal component, and the product value of the first characteristic value of each modal component and the second characteristic value of each modal component is counted, and the product value is taken as the interference influence coefficient of each modal component of each monitoring data in each group of data under the same working condition. Preferably, in this embodiment, Fourier transform and the findpeaks function of MATLAB are both known technologies, and the specific process will not be described again.

[0054] According to the interference significant coefficient of each modal component of each monitoring data in each group of data under the same working condition, and in combination with the fluctuation of the peak value of each modal component of each monitoring data in each group of data under the same working condition, the adjustment coefficient of each modal component of each monitoring data in each group of data under the same working condition is calculated, and the specific calculation formula in the embodiment is as follows:

[0055] ;

[0056] In the formula, is the adjustment coefficient of the i th modal component of each monitoring data in the j th group of data under the same working condition; is the interference significant coefficient of the i th modal component of each monitoring data in each group of data under the same working condition; is the interference influence coefficient of the i th modal component of each monitoring data in the j th group of data under the same working condition.

[0057] Wherein, The greater the value is, the higher the degree of influence of the current monitoring data on the corresponding frequency band modal component under the high-frequency interference of the water turbine under the monitoring contrast result of each monitoring data in each group of data under the same working condition and the monitoring data of each monitoring data in each group of data under the same working condition under different frequency band modal components. The greater the value is, the greater the degree of influence of the water turbine operation under the high-frequency interference in the current data acquisition process. In addition, the step flow chart of the adjustment coefficient acquisition method provided in the embodiment is shown in Figure 2 .

[0058] Considering that the noise caused by the complex interference source in the water turbine operation environment is usually high-frequency interference, and the interference influence characteristics tend to concentrate in the high-frequency band under the influence of environmental interference. Therefore, in order to reduce the influence of high-frequency interference, the threshold noise reduction processing is performed on each modal component of each monitoring data in each group of data under the same working condition according to the interference situation of different frequency band modal components to optimize and adjust. For each monitoring data of each group of data under the same working condition, the adjustment coefficient of all modal components of each monitoring data is normalized by using the Softmax function. In addition, in order to accurately reduce the interference influence under different working conditions and avoid large deviation of the water turbine monitoring control due to excessive processing, the preset adjustment judgment threshold of each modal component in each monitoring data in the embodiment is set to 0.3. Preferably, in the embodiment, the Softmax function is a known technology, and the specific process is not described again.

[0059] ​​​​​​Further, the soft threshold processing method is adopted, and each modal component of each monitoring data is processed by a Sure Shrink threshold to obtain an initial threshold of each modal component of each monitoring data , and then each modal component of each monitoring data is subjected to noise reduction processing by wavelet transform to realize wavelet threshold denoising processing. Preferably, in the embodiment, the wavelet transform and the Sure Shrink threshold are both known technologies, and the specific process is not described again.

[0060] Through the above analysis, according to the initial threshold of each modal component of each monitoring data in each group of data under the same working condition, and in combination with the normalization result of the adjustment coefficient of all modal components of each monitoring data in each group of data under the same working condition, the adjustment threshold of each modal component of each monitoring data in each group of data under the same working condition is calculated, and in the embodiment, the specific calculation formula is:

[0061] ;

[0062] In the formula, is the adjustment threshold of each modal component of each monitoring data in each group of data under the same working condition; is the initial threshold of each modal component of each monitoring data in each group of data under the same working condition; is the normalization result of the adjustment coefficient of each modal component of each monitoring data in each group of data under the same working condition; is the preset adjustment judgment threshold of each modal component of each monitoring data in each group of data under the same working condition.

[0063] If the adjustment coefficient normalization result of each modal component of each monitoring data is less than the preset adjustment judgment threshold, the initial threshold of each modal component of each monitoring data is taken as the adjustment threshold for the noise reduction processing of the corresponding modal component.

[0064] Therefore, each monitoring data of each group of data under different working conditions after noise reduction processing avoids the problem of large monitoring control deviation caused by the influence of complex interference sources in the water turbine running environment, and further prevents large deviation of the speed optimization control of the water turbine based on the neural network model. Further, all modal components of each monitoring data in each group of data under the same working condition are reconstructed, and preferably, in the embodiment, the signal reconstruction is a known technology, and the specific process is not described again.

[0065] Step four: linearizing each monitoring data of each group of data under different working conditions after reconstruction, and dividing the region to obtain a linear model of each region, to construct a multi-objective optimization function of the linear model of each region, and then to optimize the speed of the water turbine by the neural network.

[0066] According to each kind of monitoring data of each group of data under different working conditions after reconstruction, each kind of monitoring data is subjected to local linearization processing, and a segmented linear model is constructed. Specifically, in each kind of monitoring data of each group of data under different working conditions, an approximate linear equation of the non-linear water turbine characteristics is constructed through system identification technology, including but not limited to the approximate linear equation of torque speed, torque head, torque guide vane opening, and a transfer function corresponding to each kind of monitoring data is calculated to convert the non-linear model into a linear model.

[0067] Further, the working condition range involved in each kind of monitoring data of each group of data under each working condition is divided, the horizontal interval is divided at intervals of 10% of the design head, for example, the horizontal interval of 70% design head-80% design head; the guide vane opening is divided into longitudinal intervals at intervals of 5%, for example, the longitudinal interval of 5% guide vane opening-10% guide vane opening; the horizontal interval and the longitudinal interval form a grid continuous region, the grid continuous region represents all working condition characteristic regions running within the interval range divided by the real-time head and the guide vane opening, each region is the real-time acquisition data of all kinds of monitoring data, and the corresponding transfer function, i.e. the corresponding linear model, is obtained in turn, and the linear model of each kind of monitoring data corresponding to the center point of each region is the linear model of the region. Preferably, in the present embodiment, the system identification technology is a known technology, and the specific process will not be described again.

[0068] Through the above analysis, the overshoot, the stable time and the steady-state error of each region linear model in the grid continuous region are constructed as the core target according to the fuzzy control theory, a multi-objective optimization function of each region linear model is calculated, and in the present embodiment, the specific calculation formula is:

[0069] ;

[0070] In the formula, C is the overshoot of each region linear model; is the stable time of each region linear model; is the steady-state error of each region linear model; is the first weight coefficient; is the second weight coefficient; is the third weight coefficient; is the optimization result of the target optimization function; wherein, ; the overshoot, the stable time and the steady-state error of the linear model are dynamic performance indexes presented by the linear model constructed through the system identification technology, and are used to obtain the optimal PID parameters. Preferably, in the present embodiment , , ; the constraint condition is , , .

[0071] Further, for each monitoring data of each group of data under each working condition, a linear model of each region is obtained, and the overshoot, the settling time and the steady-state error of the corresponding region after optimization are taken as the evaluation indexes of the PID control effect, so as to construct a PID control model, and the IPSO algorithm is adopted to obtain the PID parameters when the overshoot, the settling time and the steady-state error are optimal, that is, the optimal PID parameters. Specifically, 30 particles are used to initialize the population, the iteration number is 50 times, and the inertia weight is decreased from 0.9 to 0.4. Preferably, in the present embodiment, the IPSO algorithm is a known technology, and the specific process will not be repeated.

[0072] Therefore, the interference characteristics under different working conditions are optimized, and then the accurate optimal PID parameters under different working conditions are extracted, so as to construct a precise PID sample library of water turbine speed regulation optimization. Specifically, at least 200 groups of "working condition characteristic parameters-optimal PID parameters" mapping sample library are obtained through the above processing. The working condition characteristic parameters are water head, real-time load, speed deviation and guide vane opening, and the optimal PID parameters are proportional coefficient Kp, integral coefficient Ki and differential coefficient Kd.

[0073] Further, the "working condition characteristic parameters" in the sample library are taken as the input, and the "optimal PID parameters" are taken as the output, to construct a BP neural network model for training. Specifically, the data in the sample library are divided into a training set and a validation set in a ratio of 7:3, the loss function adopts a mean square error function, the network weight is iteratively adjusted by using a back propagation algorithm, and the hyperparameters are optimized by using an IWOA optimization algorithm, and the convergence condition is MSE≤0.001 and the prediction error stability≤1%, wherein the iteration number of the hyperparameters is 500-1500, and the learning rate is 0.01. After the above processing, the trained model is integrated into a real-time control module to realize dynamic speed regulation optimization of the water turbine. Preferably, in the present embodiment, the BP neural network model training process, the mean square error function and the IWOA optimization algorithm are known technologies, and the specific process will not be repeated; the real-time control module adopts a NARUI NGS6100 controller.

[0074] Each kind of monitoring data of each group of data collected by the water turbine under the current working condition is input into the neural network model, and each kind of monitoring data of each group of data under different working conditions obtained by the above reconstruction is input into the neural network model, to obtain the preliminary PID parameters of the water turbine speed control, and the water turbine is controlled under different working conditions in combination with the preliminary PID parameters. When the load fluctuation is less than or equal to 5%, the PID parameters output by the model are used to adjust the guide vane opening; when the load fluctuation is greater than 10% or there is load rejection, the PID parameters are dynamically corrected by the piecewise linear model to avoid overshoot or oscillation. Further, through the real-time optimization of the neural network and the dynamic correction of the piecewise linear model, and in combination with the influence of the interference characteristics of the water turbine under different working conditions on the speed optimization analysis, the processing of the monitoring data of the water turbine under different working conditions is optimized and adjusted, and then the sample library of the accurate water turbine speed optimization model is obtained, the influence of the interference of different working conditions on the water turbine speed optimization is avoided, and high-precision adaptive speed regulation under all working conditions is realized.

[0075] Based on the same inventive concept as the above method, the embodiment of the present application also provides a neural network-based water turbine speed optimization control system, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, and the processor implements the steps of the neural network-based water turbine speed optimization control method according to any one of the above methods when executing the computer program.

[0076] It can be understood that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. And the above describes the specific embodiments of the present application. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are possible or may be advantageous.

[0077] Each embodiment in the present specification is described in a progressive manner, and the same or similar parts between each embodiment can be referred to each other, and each embodiment mainly describes the difference from other embodiments.

[0078] The above is only the embodiment of the present application, and is not used to limit the scope of the present application. Any equivalent structure or equivalent process conversion using the content of the present application specification and drawings, or direct or indirect application in other related technical fields, are also included in the protection scope of the present application.

Claims

1. A method for optimizing speed control of a hydro turbine based on neural networks, characterized in that, Includes the following steps: Collect each type of monitoring data for each group of data under each operating condition in the water turbine; By analyzing the differences in modal components in the same frequency band after modal decomposition of the same monitoring data from different groups of data under the same working conditions, and combining the characteristic similarity of different types of monitoring data in each modal component under the same working conditions, the interference significance coefficient of each modal component of each type of monitoring data in each group of data under the same working conditions is obtained. For each modal component of each monitoring data in each set of data under the same working condition, frequency domain transformation is performed. Based on the fluctuation of all peaks in the frequency domain, the interference influence coefficient of each modal component of each monitoring data in each set of data under the same working condition is obtained. Combined with the interference significance coefficient of each corresponding modal component, the adjustment coefficient of each modal component of each monitoring data in each set of data under the same working condition is obtained. Threshold noise reduction processing is then performed on each modal component of each monitoring data in each set of data under the same working condition, and then all modal components of each monitoring data in each set of data under the same working condition are reconstructed. After linearizing each monitoring data of each group of data under different operating conditions after reconstruction, the regions are divided to obtain the linear models of each region. This allows for the construction of multi-objective optimization functions for the linear models of each region. The optimal PID parameters corresponding to the obtained optimization results are then input into a neural network, which is used to optimize the speed regulation of the water turbine.

2. The speed regulation optimization control method for water turbines based on neural networks as described in claim 1, characterized in that, The method for calculating the interference significance coefficient of each modal component of each monitoring data in each set of data under the same operating conditions is as follows: ； In the formula, For each monitoring data in each set of data under the same working conditions, the first... The significance coefficient of interference for each modal component; For the first under the same working conditions The first monitoring data in each group of data Eigenvalues ​​of each modal component; For the first under the same working conditions Characteristic coefficients for each monitoring data point in the dataset; Represents an exponential function with the natural constant as its base; This represents the number of data sets collected under each operating condition.

3. The speed regulation optimization control method for water turbines based on neural networks as described in claim 2, characterized in that, The feature value further includes: The mean of all DTW distances of the same monitoring data in the same frequency band modal component between each group of data and other groups of data under the same operating conditions is used as the feature value of each modal component of each monitoring data in each group of data under the same operating conditions.

4. The speed regulation optimization control method for water turbines based on neural networks as described in claim 2, characterized in that, The method for obtaining the characteristic coefficients is as follows: The eigenvalues ​​of all modal components of each monitoring data in each data set under the same working conditions are statistically analyzed to form the eigenvector of each monitoring data in each data set under the same working conditions; and the mean of all cosine similarities between the eigenvectors of each monitoring data in each data set and the eigenvectors of other monitoring data are used as the eigencoefficient of each monitoring data in each data set under the same working conditions.

5. The speed regulation optimization control method for hydro-turbines based on neural networks as described in claim 1, characterized in that, The interference influence coefficient of each modal component of each monitoring data in each set of data under the same operating conditions includes: For each modal component of each monitoring data in each set of data under the same operating conditions, after frequency domain transformation, obtain all peak values ​​in the spectrum of each modal component. Calculate the mean kurtosis value of all peak values ​​in the spectrum of each modal component to obtain the first eigenvalue of each modal component. Calculate the mean Euclidean distance between all adjacent peak values ​​in the spectrum of each modal component as the second eigenvalue of each modal component. Then, use the product of the first eigenvalue and the second eigenvalue of each modal component as the interference influence coefficient of each modal component of each monitoring data in each set of data under the same operating conditions.

6. The speed regulation optimization control method for hydro-turbines based on neural networks as described in claim 1, characterized in that, The method for calculating the adjustment coefficient of each modal component of each type of monitoring data in each set of data under the same operating conditions is as follows: ； In the formula, For the first under the same working conditions The first monitoring data in each group of data The adjustment coefficients for each modal component; For each monitoring data in each set of data under the same working conditions, the first... The significance coefficient of interference for each modal component; For the first under the same working conditions The first monitoring data in each group of data Interference influence coefficient of each modal component.

7. The speed regulation optimization control method for hydro-turbines based on neural networks as described in claim 1, characterized in that, The threshold noise reduction process utilizes the adjustment coefficient to set the adjustment threshold, thereby achieving threshold noise reduction. The method for setting the adjustment threshold further includes: ； In the formula, The adjustment threshold for each modal component of each type of monitoring data in each set of data under the same working conditions; The initial threshold for each modal component of each type of monitoring data in each set of data under the same working conditions; This represents the normalized result of the adjustment coefficient for each modal component of each monitoring data in each set of data under the same working conditions; This refers to the preset adjustment and judgment threshold for each modal component in each type of monitoring data under the same working conditions.

8. The speed regulation optimization control method for hydro-turbines based on neural networks as described in claim 7, characterized in that, The method for obtaining the initial threshold is as follows: For each modal component of each monitoring data in each set of data under the same working conditions, soft thresholding is performed to obtain the initial threshold for each modal component of each monitoring data in each set of data under the same working conditions.

9. The speed regulation optimization control method for hydro-turbines based on neural networks as described in claim 1, characterized in that, The method for calculating the multi-objective optimization function for constructing the linear model of each region is as follows: ； In the formula, The overshoot of the linear model for each region; The settling time of the linear model for each region; The steady-state error of the linear model for each region; The first weighting coefficient; This is the second weighting coefficient; This is the third weighting coefficient; The result of optimizing the objective function; where, The overshoot, settling time, and steady-state error of the linear model are dynamic performance indicators presented by the linear model constructed through system identification technology, which are used to obtain the optimal PID parameters.

10. A neural network-based speed regulation optimization control system for a hydro-turbine, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the neural network-based turbine speed regulation optimization control method as described in any one of claims 1-9.

Images