CN-121984388-A - Prediction control method and system for guide vane opening and exciting current of hydroelectric generating set

CN121984388ACN 121984388 ACN121984388 ACN 121984388ACN-121984388-A

Abstract

The invention discloses a predictive control method and a predictive control system for the opening degree of a guide vane and exciting current of a hydroelectric generating set. The method comprises the steps of collecting operation signals of a water turbine, a generator, an excitation system and a power grid system, carrying out multi-time scale feature decomposition by adopting VMD and WPT, embedding a water turbine energy conversion equation, a generator electromagnetic equation and an excitation system dynamic equation into a physical information neural network, introducing a four-component loss function comprising structural loss, boundary loss, recursion loss and measured data loss, constructing an improved PhyPINN-TI physical constraint prediction model, processing time-varying input through an SSF algorithm, optimizing SSF time window parameters by adopting a chaotic evolution algorithm, rolling and predicting future evolution tracks of guide vane opening and exciting current based on the prediction model, generating a control instruction after control constraint mapping of a prediction result, and outputting the control instruction to an electrohydraulic servo system and the excitation system for execution, thereby realizing closed-loop prediction control of the guide vane opening and the exciting current of the hydroelectric generating set.

Inventors

CHEN PENG
LIU ZHIJUN
WENG HAOPING
LEI YE
XU KAI

Assignees

浙江松阳谢村源水利水电开发有限责任公司
浙江浙能华光潭水力发电有限公司
龙泉市岩樟溪流域水电开发有限公司

Dates

Publication Date: 20260505
Application Date: 20260121

Claims (9)

1. The method for predicting and controlling the opening degree and exciting current of the guide vane of the hydroelectric generating set is characterized by comprising the following steps of: step 1, collecting original signals of a water turbine, a generator, an excitation system and a power grid system, wherein the original signals comprise static input variables and dynamic input variables; Step 2, carrying out multi-time scale feature extraction on dynamic input variables of original signals by adopting a VMD and WPT combined decomposition method, and combining a WPT low-frequency trend component and a VMD high-frequency IMF component; Embedding physical constraints such as a water turbine energy conversion equation, a generator electromagnetic equation, an excitation system dynamic equation and the like into a physical information neural network to construct PINN-TI prediction model; Step 4, introducing a four-component loss function comprising differential equation structure loss, boundary condition loss, recursion loss and measured data loss based on the PINN-TI prediction model constructed in the step 3, and forming an improved PhyPINN-TI physical constraint prediction model; Step 5, optimizing time window parameters of a sectional state propagation SSF algorithm by adopting a chaotic evolution algorithm, segmenting time-varying input according to a time window by using the optimized sectional state propagation SSF algorithm, wherein the optimized time window parameters are kept fixed in a real-time prediction control process; And 6, based on the time-varying input processed in the step 5 in a sectionalized way, executing rolling prediction based on real-time state information by using an improved PhyPINN-TI prediction model in the running process, outputting the prediction result of the guide vane opening degree and the exciting current in real time, generating a control instruction after the prediction result is subjected to control constraint mapping, and outputting the control instruction to an electrohydraulic servo system and an exciting system for execution, so that closed-loop prediction control of the guide vane opening degree and the exciting current of the hydroelectric generating set is realized.
2. The method for predicting and controlling the opening degree and exciting current of the guide vane of the hydroelectric generating set according to claim 1, wherein the multi-physical-field dynamic data in the step 1 comprises water turbine parameters, generator parameters, exciting system parameters and power grid system parameters, wherein the water turbine parameters comprise rotating speed of a rotating wheel, water head, flow, efficiency of the water turbine, vacuum degree of a tail water pipe, operating moment of the guide vane, rotating speed change rate of the rotating wheel, flow change rate and acceleration of the opening degree of the guide vane, the generator parameters comprise an active power reference value, reactive power reference value, generator frequency, power factor, generator end voltage, stator current, stator temperature and rotor exciting flux linkage, the exciting system parameters comprise exciting current, exciting voltage, de-excitation resistance, exciting current change rate and exciting action time, and the power grid system parameters comprise power grid frequency deviation, power grid voltage deviation, harmonic distortion rate and power grid frequency change rate.
3. The method for predicting and controlling the opening degree and exciting current of a guide vane of a hydroelectric generating set according to claim 2, wherein a water turbine input data matrix is constructed for the collected dynamic data of multiple physical fields, the input data is divided into a static input variable and a dynamic input variable, the dynamic input variable comprises a rotating wheel rotating speed change rate, a flow change rate, a guide vane opening acceleration, an exciting current change rate, an exciting action time, a grid frequency change rate and a grid harmonic distortion rate, and the balance is the static input variable.
4. The method for predicting and controlling the opening degree and exciting current of the guide vane of the hydroelectric generating set according to claim 1, wherein the implementation process of the step 2 is as follows: step 2.1 dynamically adapting the WPT layer number according to the dominant frequency of each time-varying input physical quantity ; Step 2.11 extracting dominant frequencies of each physical quantity Dynamically determining the optimal number of layers of WPT Calculating an autocorrelation function, the first significant peak of which corresponds to the dominant period The dominant frequencies are: ; In the formula, Is the sampling interval; Is the sampling frequency; Step 2.12 calculating the optimal number of layers of the WPT WPT (WPT) The frequency range of the low-frequency component of the layer is Is covered by Constraint conditions: ) +1 after rounding down, ensure low frequency component coverage ; Step 2.2 optimal number of layers based on each physical quantity Performing WPT decomposition to obtain WPT low-frequency trend component And WPT high frequency trend component ; Step 2.3 high frequency trend component output to WPT Dynamic optimization through energy and complexity assessment Iterative optimization Is a function of the objective function of: ; In the formula, Is redundancy; Sample entropy for the kth IMF component; step 2.4 based on For a pair of Performing VMD decomposition, obtaining VMD high-frequency IMF components, and combining the WPT low-frequency trend components with the VMD high-frequency IMF components: 。
5. The method for predicting and controlling the opening degree and exciting current of the guide vane of the hydroelectric generating set according to claim 1, wherein the PINN-TI prediction model in the step 3 is implemented as follows: the physical information neural network PINN reforms the initial conditions of the input layer Representing Time of day guide vane opening Exciting current Water head of water turbine Frequency deviation of power grid And the first derivative, ellipsis Then other time-varying input initial conditions are implied, time variations in the time domain are predicted Time-varying input discrete sets Representing The hidden layer adopts a fully connected neural network MLP, the activation function is tanh, and the output layer predicts the opening degree of the guide vane And exciting current As a proxy model for the true values.
6. The method for predicting and controlling the opening degree and exciting current of the guide vane of the hydroelectric generating set according to claim 1, wherein the four-component loss function in the step 4 is specifically: the four-component loss function of the PINN-TI predictive model includes structural losses And boundary loss And new recursive losses And measurement loss with time weighting Forming a four-component loss function: ; In the formula, Is the structure loss weight; is the boundary loss weight; is a recursive loss weight; is the measure loss weight; Structural loss The method comprises the following steps: , is the loss of the dynamic transition process of the water turbine, Is the structural loss of the electromagnetic link of the generator, Structural losses for the excitation system; Boundary loss The method comprises the following steps: , Is a hydraulic boundary constraint loss, Loss of constraint for electromechanical boundaries; the recursive loss function is: ; In the formula, Is the total time step; The flow rate change predicted at the moment; the terminal voltage change rate is predicted at the moment; to the change rate of the opening degree of the guide vane And rate of head change As the flow rate change mapping function of the input water turbine link, To the change rate of exciting current And rate of change of rotational speed A terminal voltage change rate mapping function of an input generator link; the measured loss function with time weight is: ; Wherein, the Is that A measurement of time of day; Is that The predicted value of the time of day, To take into account both the "time decay characteristic" and the "measurement uncertainty" weighting function.
7. The method for predicting and controlling the opening degree and exciting current of the guide vane of the hydroelectric generating set according to claim 1, wherein in the step 5, the time window parameters of the segment state propagation SSF algorithm are optimized by adopting a chaotic evolution algorithm, and the specific process is as follows: step 5.1 optimizing Using the chaotic evolutionary Algorithm Demarcating Is a constraint range of (2) , wherein, And Respectively representing the highest effective cut-off frequency obtained after VMD decomposition and the lowest fundamental frequency obtained after WPT decomposition; step 5.2 sampling with Latin hypercube LHS Internal generation Initial candidates A value; and 5.3, constructing a two-component fitness function, wherein the formula is as follows: ; In the formula, The recursive loss function is represented as a function of the recursive loss, , In order to be able to predict the value, The actual measurement value of the opening degree and exciting current of the guide vane; And Is a weighting coefficient; step 5.4 introducing chaotic map pair candidates And (3) performing a variation formula: ; In the formula, Is the variation step length; Is a chaotic sequence, generates traversals Is a random sequence of (a); step 5.5 after mutation Is still under In the range, otherwise, cut off or regenerate And after variation Gene exchange was performed, formula: ; In the formula, Balancing the influence of parent and variant solutions for the crossover factor; Step 5.6, reserving a better solution by using greedy criteria: ; Repeating the above steps until reaching convergence times or accuracy requirement, and outputting the minimum fitness F The embedded SSF algorithm achieves optimal segmentation of the time-varying input.
8. The method for predicting and controlling the opening degree and exciting current of the guide vane of the hydroelectric generating set according to claim 7, wherein in the step 5, the time-varying input is segmented according to a time window by an optimized segmentation state propagation SSF algorithm, specifically: SSF algorithm discretizes continuous time into time window By using Discrete set of time-varying inputs Cut into Corresponding to Recursively updating the initial conditions and inputting the initial conditions 、 Prediction of Predicting time status First, the Step input 、 Prediction of Time of day state The recursive formula: ; Splicing the multi-step prediction results to obtain a long time domain Is of the guide vane opening degree Exciting current A curve, wherein, the curve is a curve, Indicating that the system is experiencing a first optimized time window The obtained full state set is not only the output result of the first-step prediction, but also the initial condition input as the second-step recursion prediction; And (3) with Respectively represent The opening degree of the guide vane and the first derivative thereof at the moment, reflecting the control state of the water inflow of the water turbine; Respectively represent Exciting current and its first derivative at moment reflecting the regulation state of electromagnetic field of generator Then implicate the basis Other system states which need to be synchronously updated are defined to form the complete physical boundary condition at the next moment of the system.
9. A system based on the method for predictive control of the opening degree and exciting current of the guide vane of the hydroelectric generating set according to any one of claims 1 to 8, comprising: The data acquisition module is used for acquiring original signals of the water turbine, the generator, the excitation system and the power grid system and covering multi-physical-field dynamic data of the hydroelectric generating set; the signal decomposition processing module is used for extracting multi-time scale characteristics of the original signal by adopting a VMD and WPT combined decomposition method, and combining the WPT low-frequency trend component and the VMD high-frequency IMF component; The model construction module is used for constructing PINN-TI prediction model and introducing four-component loss functions of differential equation structure loss, boundary condition loss, recursion loss and measured data loss; The segmentation state propagation module is used for optimizing time window parameters of a segmentation state propagation SSF algorithm by adopting a chaos evolution algorithm, and segmenting time-varying input according to a time window by the optimized segmentation state propagation SSF algorithm; The model prediction module is used for performing rolling prediction based on real-time state information in the running process of the prediction model based on the time-varying input after the segmentation processing and outputting the prediction results of the opening degree of the guide vane and the exciting current in real time; And the closed-loop control module is used for generating a control instruction after control constraint mapping based on a prediction result, outputting the control instruction to the electrohydraulic servo system and the excitation system for execution, and realizing closed-loop prediction control of the opening degree of the guide vane and the excitation current of the hydroelectric generating set.

Description

Prediction control method and system for guide vane opening and exciting current of hydroelectric generating set Technical Field The invention relates to a prediction control method and a prediction control system for the opening degree of a guide vane and exciting current of a hydroelectric generating set, and belongs to the field of pumped storage. Background The hydroelectric generating set is used as an indispensable key device for peak regulation, valley filling, frequency modulation and phase modulation in an electric power system, has extremely complex running state, relates to multi-physical field coupling of a water turbine, a generator, an excitation system and a power grid, and has strong nonlinearity, time-varying characteristics and multi-time-scale dynamic characteristics. In the prior art, the control of the opening degree of a guide vane and exciting current of a hydroelectric generating set faces a plurality of challenges. From the control model perspective, the traditional PID control is mostly dependent on an empirical formula and fixed parameter setting, and is difficult to effectively integrate physical constraints such as energy conversion of a water turbine, electromagnetic induction of a generator and the like with actual operation data. The method has the advantages that when the model faces complex and changeable working conditions, fitting accuracy of complex dynamic characteristics of the system is seriously insufficient, and the change of the opening degree of the guide vane and exciting current cannot be accurately predicted. In the aspect of signal processing, a time-varying signal generated by multi-physical field coupling contains multiple components such as high-frequency noise, low-frequency trend and the like, and the existing method has limited processing capacity. The simple filtering mode is difficult to effectively remove noise interference while keeping key features, so that the key features related to guide vane vibration, runner rotation and the like are difficult and heavy to extract, and further the subsequent control decision is influenced. For long-time-domain prediction, existing methods have significant drawbacks in dealing with the dynamics of time-varying inputs. Over time, errors in the recursive calculation process can be accumulated continuously, so that the predicted result gradually deviates from the actual value, and the requirements of the hydroelectric generating set on long-term stable operation cannot be met. In addition, in the aspect of control parameter optimization, the traditional method mostly adopts a single objective function, only focuses on a certain aspect such as control precision or response speed, and the effective balance among the control precision, the response speed and the energy consumption is difficult to realize. In actual operation, the system may have problems of high control precision, excessive energy consumption, high response speed, insufficient control precision and the like, and the overall operation efficiency and the economy of the hydroelectric generating set are seriously affected. In order to improve stability, economy and electric energy quality of the hydroelectric generating set, a predictive control method and a predictive control system which can integrate physical constraint and data driving, accurately process time-varying signals and realize multi-objective optimization are urgently needed. Disclosure of Invention Aiming at the problems pointed out in the background technology, the invention discloses a method and a system for predicting and controlling the opening degree of a guide vane and exciting current of a hydroelectric generating set, and the running efficiency of the hydroelectric generating set is effectively improved. The invention discloses a prediction control method for the opening degree and exciting current of a guide vane of a hydroelectric generating set, which comprises the following steps: step 1, collecting original signals of a water turbine, a generator, an excitation system and a power grid system, wherein the original signals comprise static input variables and dynamic input variables; Step 2, carrying out multi-time scale feature extraction on dynamic input variables of original signals by adopting a VMD and WPT combined decomposition method, and combining a WPT low-frequency trend component and a VMD high-frequency IMF component; Embedding physical constraints such as a water turbine energy conversion equation, a generator electromagnetic equation, an excitation system dynamic equation and the like into a physical information neural network to construct PINN-TI prediction model; Step 4, introducing a four-component loss function comprising differential equation structure loss, boundary condition loss, recursion loss and measured data loss based on the PINN-TI prediction model constructed in the step 3, and forming an improved PhyPINN-TI physical constraint prediction model; Step