DDPG algorithm-based adaptive PID control method and device for water turbine

Abstract

The application relates to the technical field of water turbine control, in particular to a water turbine adaptive PID control method and device based on a DDPG algorithm, which comprises the following steps: acquiring various types of monitoring data of a water turbine under various working conditions; dividing the entire data acquisition time into multiple time intervals; acquiring response characteristic values of various types of monitoring data in the time intervals according to the data characteristics of various types of monitoring data in the time intervals and the importance of various types of monitoring data in the time intervals; adjusting the initial weights of various types of monitoring data in the DDPG algorithm according to the differences between the importance and the response characteristic values of various types of monitoring data in the time intervals; and then obtaining a PID parameter adjustment model to adjust preset PID parameters at the current moment. The initial weights of various types of monitoring data are optimized according to the data characteristics, and the accuracy of the water turbine adaptive PID control is improved.

Description

TECHNICAL FIELD

[0001] The application relates to the technical field of water turbine control, in particular to a water turbine adaptive PID control method and device based on a DDPG algorithm. BACKGROUND

[0002] A water turbine regulating system is a key to stable operation of a power system, and can reduce frequent load fluctuations and water head changes under high penetration of new energy. However, the regulation of the water turbine has the characteristics of nonlinearity and time variation, which leads to the fact that the traditional PID control method cannot effectively handle the stable operation of the water turbine in all operating conditions. The adaptive optimization and adjustment of the PID regulation process of the water turbine through the DDPG algorithm can adjust the accuracy and response speed, effectively solve the problems of large overshoot and response lag in the traditional control process, and further realize autonomous control by using the NGS6100 controller, thereby guaranteeing the stability of the power grid.

[0003] However, in actual application, the importance of the state characteristics of the water turbine under different operating conditions is quite different. For example, in transient operating conditions (load mutation), the speed change rate is a key factor affecting the accuracy of control; and in steady-state operating conditions, the water head change has a higher degree of influence on real-time control. In the PID control of the water turbine, the accuracy of the real-time PID parameter evaluation of the water turbine is determined by the weight assigned to each type of monitoring data by the Critic network in the DDPG algorithm. However, the weight distribution of the Critic network of the traditional DDPG algorithm does not fully consider the dynamic change characteristics of the water turbine under different operating conditions, resulting in low accuracy of real-time PID control adjustment of the water turbine under different operating conditions, and thus reducing the adaptive PID control accuracy of the water turbine. SUMMARY

[0004] To solve the above technical problems, the purpose of the present application is to provide a water turbine adaptive PID control method and device based on a DDPG algorithm, and the technical solution adopted is as follows:

[0005] In a first aspect, the embodiments of the present application provide a water turbine adaptive PID control method based on a DDPG algorithm, which comprises the following steps:

[0006] acquiring various types of monitoring data of the water turbine under various operating conditions multiple times;

[0007] The entire data collection time is divided into multiple time intervals based on the degree of anomaly in all types of monitoring data at each time point; based on the data change trends of various types of monitoring data in each time interval under each working condition and various statistical characteristics of the data, feature vectors of various types of monitoring data in each time interval are obtained; based on the importance of various types of monitoring data in each time interval compared with other types of monitoring data, and the distance between the feature vectors of various types of monitoring data in each time interval and the feature vectors of other time intervals, response feature values ​​of various types of monitoring data in each time interval are obtained.

[0008] Based on the differences between the importance of various monitoring data and other types of monitoring data in each time interval and their response characteristic values, as well as the response characteristic values ​​of various monitoring data in each time interval, the adjustment coefficients of various monitoring data are obtained. Then, the initial weights assigned to various monitoring data by the Critic network in the DDPG algorithm are adjusted, thereby obtaining the PID parameter adjustment model. Combining the various monitoring data at the current moment and the preset PID parameters, the adjustment amount of the PID parameters at the current moment is obtained, and then the preset PID parameters at the current moment are adjusted.

[0009] Preferably, the specific process of dividing the entire data acquisition time into multiple time intervals is as follows: the sequence of various monitoring data according to the order of acquisition time is used as the monitoring data sequence of various monitoring data; the LOF value of each monitoring data in each monitoring data sequence is obtained; the mean of the LOF values ​​corresponding to all types of monitoring data collected at each time is used as the abnormal significance value at each time, and the time when the abnormal significance value is greater than or equal to 1 is recorded as the characteristic time. Using all characteristic times as the dividing points, the entire data acquisition time is divided into multiple time intervals.

[0010] Preferably, the process of obtaining the feature vectors of various monitoring data in each time interval is as follows: obtaining the trend statistics and multiple statistical feature values ​​of various monitoring data in each time interval, wherein the multiple statistical feature values ​​refer to the mean, variance, and mean deviation of various monitoring data in each time interval, and the mean deviation refers to the mean of the absolute differences between all data and their rated data in each time interval; and using the vector formed by the trend statistics and all statistical feature values ​​of various monitoring data in each time interval as the feature vector of various monitoring data in each time interval.

[0011] Preferably, the formula for calculating the response characteristic values ​​of the various monitoring data in each time interval is as follows: In the formula, This indicates that the x-th type of monitoring data is in the th... Response characteristic values ​​for each time interval; This indicates that the x-th type of monitoring data is in the th... Relative weights within a time interval; Euclidean distance between the feature vector of the xth kind of monitoring data in the yth time interval and the (x+1)th kind of monitoring data in the yth time interval; Euclidean distance between the feature vector of the xth kind of monitoring data in the yth time interval and the (x+1)th kind of monitoring data in the yth time interval; total number of time intervals; wherein the relative weight of each kind of monitoring data in the yth time interval is obtained by taking all kinds of monitoring data in the yth time interval as the input of the CRITIC algorithm to obtain the relative weight of each kind of monitoring data in the yth time interval.

[0012] Preferably, the calculation formula of the adjustment coefficient of each kind of monitoring data is: ; wherein, indicates the adjustment coefficient of the xth kind of monitoring data; indicates the normalized result of the response eigenvalue of the xth kind of monitoring data in the yth time interval; indicates the feature difference value of the xth kind of monitoring data in the yth time interval; indicates the sum of the feature difference values of the xth kind of monitoring data in all time intervals; total number of time intervals.

[0013] Preferably, the feature difference value of each kind of monitoring data in each time interval refers to the absolute difference value between the relative weight and the normalized response eigenvalue of each kind of monitoring data in each time interval.

[0014] Preferably, the calculation formula for adjusting the initial weight assigned to each kind of monitoring data by the Critic network in the DDPG algorithm is: ; wherein, indicates the adjusted weight of the xth kind of monitoring data; and respectively indicate the initial weight assigned to the xth kind of monitoring data and the (x+1)th kind of monitoring data by the Critic network in the traditional DDPG algorithm, and respectively indicate the adjustment coefficient of the xth kind of monitoring data and the (x+1)th kind of monitoring data, total number of kinds of monitoring data. Preferably, the obtaining process of the PID parameter adjustment model is:

[0015]

[0016] ​​​​​​​​​The dataset composed of the monitoring data at all time points and the corresponding PID parameters is denoted as a full working condition dataset; the full working condition dataset is taken as the input of the DDPG algorithm, and the adjusted weights of various types of monitoring data under various working conditions are taken as the weights of various types of monitoring data in the Critic network in the model training process of the DDPG algorithm, so as to complete the training of the model and obtain the PID parameter adjustment model.

[0017] Preferably, the specific process of obtaining the PID parameter adjustment amount at the current time point is that various types of monitoring data at the current time point and preset PID parameters are input into the PID parameter adjustment model, and the output is the PID parameter adjustment amount at the current time point.

[0018] In a second aspect, the embodiments of the present application also provide a water turbine adaptive PID control device based on a DDPG algorithm, 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 water turbine adaptive PID control method based on the DDPG algorithm in any of the above aspects when executing the computer program.

[0019] The present application has at least the following beneficial effects:

[0020] The present application provides a water turbine adaptive PID control method and device based on a DDPG algorithm, which analyzes the importance of various types of monitoring data compared to other types of monitoring data and the fluctuation characteristics of various types of monitoring data in different time periods, constructs response characteristic values, and can represent the significant degree of response of various types of monitoring data in the PID control process of the water turbine in each time interval; by analyzing the weight difference characteristics of various types of monitoring data in each time interval, the adjustment coefficients of various types of monitoring data are constructed, which can represent the importance of various types of monitoring data in the operation control of the water turbine, and provide a basis for the subsequent weight allocation of various types of data; and then the initial weight of the Critic network of the actual DDPG algorithm training process is optimized and adjusted according to the adjustment coefficient of various types of monitoring data, which has the beneficial effect of fully considering the dynamic differences of different types of monitoring data response changes of the water turbine under different working conditions, and then realizing more accurate weight allocation, and then adaptively adjusting the PID parameters at each time point, and improving the accuracy of the water turbine adaptive PID control. BRIEF DESCRIPTION OF DRAWINGS

[0021] 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 creating any creative labor.

[0022] Figure 1 FIG. 1 shows a step flow chart of the adaptive PID control method of the hydraulic turbine based on the DDPG algorithm according to an embodiment of the present application, which comprises the following steps:

[0023] Figure 2 FIG. 4 shows a flow chart of adjusting the initial weights assigned to various types of monitoring data in the Critic network of the DDPG algorithm according to an embodiment of the present application. DETAILED DESCRIPTION

[0024] In order to further illustrate the technical means and effects adopted by the present application to achieve the predetermined purposes, the specific embodiments, structures, features and effects of the adaptive PID control method and device of the hydraulic turbine based on the DDPG algorithm according to the present application are described in detail below in combination 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.

[0025] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.

[0026] The specific scheme of the adaptive PID control method and device of the hydraulic turbine based on the DDPG algorithm according to the present application is described in detail below in combination with the drawings.

[0027] Please refer to FIG. 1, Figure 1 which shows a step flow chart of the adaptive PID control method of the hydraulic turbine based on the DDPG algorithm according to an embodiment of the present application, which comprises the following steps:

[0028] Step one: obtaining various types of monitoring data of the hydraulic turbine under various operating conditions for multiple times.

[0029] For typical operating conditions such as starting of the hydraulic turbine, grid connection, load increase and decrease, load shedding, N times of monitoring data are collected under each operating condition, and each time of collected monitoring data includes rotational speed, real-time water head, grid load and guide vane opening. At the same time, the corresponding PID parameters in the PID control algorithm are collected. Wherein, N≥30, and the value of N in the present embodiment is 60. The time interval of each collection is 1s. Since the process of data collection may contain noise interference, the EEMD algorithm and the wavelet denoising algorithm are used in the present application to pretreat the monitoring data collected under different operating conditions. The data set composed of all pretreated monitoring data and corresponding PID parameters is recorded as the full operating condition data set. The EEMD algorithm and the wavelet denoising algorithm are well known to those skilled in the art, and the specific process is not described again.

[0030] Step 2: Divide the entire data collection time into multiple time intervals based on the degree of anomaly in all types of monitoring data at each time point; obtain the feature vectors of various types of monitoring data in each time interval based on the data change trends and various statistical characteristics of the data under each working condition in each time interval; obtain the response feature values ​​of various types of monitoring data in each time interval based on the importance of various types of monitoring data in each time interval compared to other types of monitoring data, and the distance between the feature vectors of various types of monitoring data in each time interval and the feature vectors of other time intervals.

[0031] During the training of the DDPG algorithm, different monitoring data under different operating conditions exhibit varying response characteristics to the current PID control. However, the Critic network in the traditional DDPG algorithm only considers the error of the calculated Q-value when assigning weights to various types of monitoring data, without taking into account the response characteristics of different monitoring data to PID control under different operating conditions. This leads to significant errors in the weight assignment. Therefore, this application aims to analyze the importance of each type of monitoring data based on its characteristics at different times, and then adjust the weights of various types of monitoring data during the training of the DDPG algorithm. The specific analysis process is as follows.

[0032] This embodiment takes the operation of a water turbine under any operating condition as an example for the following analysis. The sequence of various monitoring data collected under any operating condition, arranged chronologically by acquisition time, is taken as the monitoring data sequence for each type of monitoring data. Furthermore, considering that the response change characteristics differ at different times under different operating conditions during actual real-time control, equally weighting different response change characteristics may cause significant adjustment evaluation deviations, thereby affecting the accuracy of the water turbine's adaptive PID control. Specifically, considering the phased and continuous changes in the characteristics of the water turbine under each operating condition, and the strong coupling between different monitoring data for abnormal responses under the same operating condition, the monitoring data sequences are used as inputs. The LOF (Low-Oriented Free) anomaly monitoring algorithm is employed to obtain the LOF value of each monitoring data point in each sequence. A larger LOF value indicates a more significant operational anomaly at the corresponding acquisition time. The average LOF value of all types of monitoring data collected at each time point is used as the anomaly significance value for that time point. Time points with an anomaly significance value greater than or equal to 1 are designated as characteristic times. Using all characteristic times as dividing points, the entire data acquisition time is divided into multiple time intervals. For example, the entire data acquisition time contains t time points, where the characteristic times are respectively... and The divided time intervals are: [ ), [ ), [ ].

[0033] Furthermore, taking the data of the x-th type of monitoring data in the y-th time interval as an example, the following analysis is performed. All monitoring data of the x-th type of monitoring data in the y-th time interval are sorted in chronological order and used as input to the trend verification algorithm to obtain the trend statistics of the x-th type of monitoring data in the y-th time interval. The larger the trend statistics, the more significant the trend change of the x-th type of monitoring data in the y-th time interval during the actual control process, and the more significant the response of this segment of monitoring data to changes in the actual control of the corresponding operating condition. Various statistical characteristic values ​​of the x-th type of monitoring data in the y-th time interval are calculated, including but not limited to the mean, variance, and mean deviation. The mean deviation is the mean of the absolute differences between all data of the x-th type of monitoring data in the y-th time interval and its rated data. The vector formed by the trend statistics and all statistical characteristic values ​​of the x-th type of monitoring data in the y-th time interval is used as the feature vector of the x-th type of monitoring data in the y-th time interval. To further determine the differences in response changes among different types of monitoring data within the y-th time interval during the actual control process under the current operating conditions, all types of monitoring data within the y-th time interval are used as input. The CRITIC algorithm is used to obtain the relative weight of each type of monitoring data within the y-th time interval. The relative weight of the x-th type of monitoring data in the y-th time interval is denoted as... The larger the value, the more significant the data response of the x-type monitoring data is compared with other types of monitoring data in the y-th time interval, and the lower the correlation between it and other types of monitoring data.

[0034] As a preferred implementation, response characteristic values ​​of various monitoring data in each time interval are obtained based on the importance of each type of monitoring data compared to other types of monitoring data in each time interval, and the distance between the characteristic vectors of each type of monitoring data in each time interval and the other time intervals. These values ​​are used to characterize the significance of the response of various monitoring data in each time interval during the PID control process of the water turbine.

[0035] In this embodiment, the x-th type of monitoring data is used in the... The response characteristic values ​​for each time interval are denoted as Its specific expression is: In the formula, This indicates that the x-th type of monitoring data is in the th... Response characteristic values ​​for each time interval; This indicates that the x-th type of monitoring data is in the th... Relative weights within a time interval; This indicates that the x-th type of monitoring data is in the th... The time interval and the first Euclidean distance between feature vectors in each time interval; This indicates the total number of time intervals.

[0036] Calculated The larger the value, the higher the value of the x-th type of monitoring data in the PID control process of the turbine. The greater the significance of the response in a given time interval.

[0037] Step 3: Based on the differences between the importance of various monitoring data and other types of monitoring data in each time interval and their response characteristic values, as well as the response characteristic values ​​of various monitoring data in each time interval, obtain the adjustment coefficients of various monitoring data. Then, adjust the initial weights assigned to various monitoring data by the Critic network in the DDPG algorithm to obtain the PID parameter adjustment model. Combine the various monitoring data at the current moment and the preset PID parameters to obtain the PID parameter adjustment amount at the current moment, and then adjust the preset PID parameters at the current moment.

[0038] Furthermore, if the relative weights of various monitoring data in each time interval differ more from the weights of the response characteristic values ​​of the corresponding monitoring data in each time interval, it indicates that the corresponding monitoring data plays a greater role in the operation of the turbine in different dimensions, and thus the corresponding monitoring data has more significant and important characteristics in the operation of the turbine.

[0039] Therefore, the absolute difference between the relative weight of the x-th type of monitoring data in the y-th time interval and the normalized response feature value is denoted as the feature difference value of the x-th type of monitoring data in the y-th time interval, which is used to characterize the importance of the x-th type of monitoring data to the turbine operation process compared with other types of monitoring data in the y-th time interval. In this embodiment, the Softmax function is used to normalize the data.

[0040] As a preferred implementation, the adjustment coefficients of various monitoring data are obtained based on the differences between the importance of various monitoring data and other types of monitoring data in each time interval and their response characteristic values, as well as the response characteristic values ​​of various monitoring data in each time interval. These coefficients are used to characterize the importance of various monitoring data in the operation control of the turbine.

[0041] In this embodiment, the first The adjustment coefficient for the monitoring data is denoted as Its specific expression is: In the formula, Indicates the first Adjustment coefficients for monitoring data; This represents the normalized result of the response characteristic value of the x-th type of monitoring data in the y-th time interval; This represents the characteristic difference value of the x-th type of monitoring data in the y-th time interval; Indicates the first The sum of characteristic differences in the monitoring data across all time intervals; This indicates the total number of time intervals.

[0042] The larger the value, the greater the importance of the x-th type of monitoring data in the y-th time interval to the turbine operation process, and the higher the reference value of the x-th type of monitoring data in the y-th time interval when calculating PID parameters; The larger the value, the greater the importance of the x-th type of monitoring data in the turbine operation control, and the higher the reference value of the x-th type of monitoring data when calculating PID parameters.

[0043] Furthermore, in the traditional adaptive PID control process based on the DDPG algorithm, when assigning weights to each type of monitoring data through the Critic network, the differences in response characteristics of different types of monitoring data during turbine regulation are ignored.

[0044] Therefore, as a preferred implementation, the weights assigned to each type of monitoring data in the Critic network of the traditional DDPG algorithm are adjusted according to the adjustment coefficients of various types of monitoring data. A flowchart illustrating the adjustment of the initial weights assigned to each type of monitoring data in the Critic network of the DDPG algorithm is shown below. Figure 2 As shown.

[0045] In this embodiment, the first The adjusted weights of the monitoring data are denoted as follows: The specific adjustment relationship is as follows: In the formula, Indicates the first Adjusted weights for the monitoring data; and These represent the Critic network in the traditional DDPG algorithm as the [number]th [level]. Class and First The initial weights assigned to the monitoring data. and They represent the first Class and First Adjustment coefficients for monitoring data, The total number of categories of monitored data is 4 in this embodiment.

[0046] The larger the adjustment coefficient of each type of monitoring data, the greater the weight of the corresponding category of monitoring data; the denominator in the formula is used to normalize the weight of all categories of monitoring data.

[0047] Based on the above steps, the adjusted weights of various monitoring data under any working condition are obtained.

[0048] Similarly, we obtain the adjusted weights of all types of monitoring data under all operating conditions.

[0049] Furthermore, the full-condition dataset is used as the input to the DDPG algorithm. The adjusted weights of various monitoring data under each condition are used as the weights of various monitoring data in the Critic network during the training of the DDPG algorithm model. The loss function of the Actor network is the policy gradient loss function, the loss function of the Critic network is the mean squared error function, and the optimizer is the Adam optimizer. The DDPG algorithm is trained to obtain the trained model, which is denoted as the PID parameter adjustment model. The training process of the DDPG algorithm is a well-known technique, and the specific process will not be elaborated here. The trained model is deployed to the NGS6100 controller to realize adaptive PID control based on the DDPG algorithm.

[0050] Furthermore, the real-time PID parameters during turbine operation are controlled and adjusted. Specifically, various monitoring data are collected in real time by sensors, and preset PID parameters for each moment are obtained. The monitoring data includes rotational speed, real-time head, grid load, and guide vane opening. The real-time collected monitoring data and preset PID parameters are input into the PID parameter adjustment model, which then outputs the PID parameter adjustment amount for the current moment. Based on the obtained PID parameter adjustment amount, the preset PID parameters for the current moment in the PID control algorithm are adjusted; thus, the turbine is controlled in real time by the PID controller with adjusted parameters. This avoids mechanical losses caused by over-adjustment and eliminates drastic load fluctuations under high new energy penetration.

[0051] Based on the same inventive concept as the above methods, this application also provides a turbine adaptive PID control device based on the DDPG algorithm, including a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, it implements the steps of any one of the above-described turbine adaptive PID control methods based on the DDPG algorithm.

[0052] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, specific embodiments of this specification have been described above. Additionally, the processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are possible or may be advantageous.

[0053] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

[0054] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the principles of this application should be included within the protection scope of this application.

Claims

1. A turbine adaptive PID control method based on the DDPG algorithm, characterized in that, The method includes the following steps: Multiple monitoring data were acquired from the turbine under various operating conditions. The entire data collection time is divided into multiple time intervals based on the degree of anomaly in all types of monitoring data at each time point; based on the data change trends of various types of monitoring data in each time interval under each working condition and various statistical characteristics of the data, feature vectors of various types of monitoring data in each time interval are obtained; based on the importance of various types of monitoring data in each time interval compared with other types of monitoring data, and the distance between the feature vectors of various types of monitoring data in each time interval and the feature vectors of other time intervals, response feature values ​​of various types of monitoring data in each time interval are obtained. Based on the differences between the importance of various monitoring data and other types of monitoring data in each time interval and their response characteristic values, as well as the response characteristic values ​​of various monitoring data in each time interval, the adjustment coefficients of various monitoring data are obtained. Then, the initial weights assigned to various monitoring data by the Critic network in the DDPG algorithm are adjusted to obtain the PID parameter adjustment model. Combining the various monitoring data at the current moment and the preset PID parameters, the PID parameter adjustment amount at the current moment is obtained, and then the preset PID parameters at the current moment are adjusted. The process of obtaining the feature vectors of various monitoring data in each time interval is as follows: The trend statistics and multiple statistical characteristic values ​​of various monitoring data in each time interval are obtained. These multiple statistical characteristic values ​​refer to the mean, variance, and mean deviation of various monitoring data in each time interval. The mean deviation refers to the mean of the absolute differences between all data points and their rated data in each time interval. The vector formed by the trend statistics and all statistical characteristic values ​​of various monitoring data in each time interval is used as the feature vector of various monitoring data in each time interval. The formulas for calculating the response characteristic values ​​of the various monitoring data in each time interval are as follows: In the formula, This indicates that the x-th type of monitoring data is in the th... Response characteristic values ​​for each time interval; This indicates that the x-th type of monitoring data is in the th... Relative weights within a time interval; This indicates that the x-th type of monitoring data is in the th... The time interval and the first Euclidean distance between feature vectors in each time interval; Represents the total number of time intervals; where, the first... The relative weights of various types of monitoring data within a given time interval are obtained as follows: all types of monitoring data within the y-th time interval are used as input to the CRITIC algorithm to obtain the relative weights of each type of monitoring data within the y-th time interval; The formulas for calculating the adjustment coefficients of the various types of monitoring data are as follows: In the formula, Indicates the first Adjustment coefficients for monitoring data; This represents the normalized result of the response characteristic value of the x-th type of monitoring data in the y-th time interval; This represents the characteristic difference value of the x-th type of monitoring data in the y-th time interval; Indicates the first The sum of characteristic differences in the monitoring data across all time intervals; Indicates the total number of time intervals; The characteristic difference values ​​of various monitoring data in different time intervals refer to the absolute difference between the relative weights of various monitoring data in different time intervals and the normalized response characteristic values; The calculation formula for adjusting the initial weights assigned to various monitoring data by the Critic network in the DDPG algorithm is as follows: In the formula, Indicates the first Adjusted weights for the monitoring data; and These represent the Critic network in the traditional DDPG algorithm as the [number]th [level]. Class and First The initial weights assigned to the monitoring data. and They represent the first Class and First Adjustment coefficients for monitoring data, This represents the total number of categories of monitored data.

2. The adaptive PID control method for water turbines based on the DDPG algorithm as described in claim 1, characterized in that, The specific process of dividing the entire data acquisition time into multiple time intervals is as follows: the sequence of various monitoring data according to the order of acquisition time is used as the monitoring data sequence of various monitoring data; the LOF value of each monitoring data in each monitoring data sequence is obtained; the mean of the LOF values ​​corresponding to all types of monitoring data collected at each time is used as the abnormal significance value at each time; the time when the abnormal significance value is greater than or equal to 1 is recorded as the characteristic time; and all characteristic times are used as dividing points to divide the entire data acquisition time into multiple time intervals.

3. The adaptive PID control method for water turbines based on the DDPG algorithm as described in claim 1, characterized in that, The process of obtaining the PID parameter adjustment model is as follows: The dataset consisting of monitoring data at all times and corresponding PID parameters is denoted as the full-condition dataset. The full-condition dataset is used as the input to the DDPG algorithm. The adjusted weights of various monitoring data under each condition are used as the weights of various monitoring data in the Critic network during the training of the DDPG algorithm model, thereby completing the training of the model and obtaining the PID parameter adjustment model.

4. The adaptive PID control method for water turbines based on the DDPG algorithm as described in claim 3, characterized in that, The specific process for obtaining the PID parameter adjustment amount at the current moment is as follows: input various monitoring data and preset PID parameters at the current moment into the PID parameter adjustment model, and output the PID parameter adjustment amount at the current moment.

5. A turbine adaptive PID control device based on the DDPG algorithm, 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 adaptive PID control method for water turbines based on the DDPG algorithm as described in any one of claims 1-4.

Images