CN-121998153-A - Water level, water head and reservoir capacity prediction method of pumped storage power station considering time factors

CN121998153ACN 121998153 ACN121998153 ACN 121998153ACN-121998153-A

Abstract

The invention discloses a water level, water head and reservoir capacity prediction method of a pumped storage power station taking time factors into consideration, which comprises the steps of adopting a support vector machine SVM to fit a nonlinear relation between upper reservoir water level and lower reservoir water level and reservoir capacity to generate a corresponding model, taking time information sine and cosine codes, normalized pumped power generation point numbers and the like as inputs, predicting water level variation through a XGBoost model of a Bayesian optimization super parameter, obtaining a predicted water level by combining an initial water level, obtaining reservoir capacity through the water level-reservoir capacity model, and subtracting water head from two reservoir water levels.

Inventors

PENG CHUAN
WANG JIN
TANG ZHENGMAO
CUN WENQUAN
YANG JIANMIAO
QIN MINGWEI
Peng Zhengpeng
ZHANG CHENGYUN

Assignees

长电新能有限责任公司

Dates

Publication Date: 20260508
Application Date: 20251203

Claims (13)

1. A method for predicting the water level, water head and reservoir capacity of a pumped storage power station by considering time factors is characterized by comprising the following steps: Step 1, acquiring corresponding data of water levels and reservoir capacities of an upper reservoir and a lower reservoir of a pumped storage power station, and fitting the corresponding data by adopting a support vector machine to generate water level-reservoir capacity curves of the upper reservoir and the lower reservoir, so as to obtain a corresponding relation model of the water levels and the reservoir capacities of the upper reservoir and the lower reservoir; Step 2, determining input variables of XGBoost models and preprocessing, wherein the input variables comprise time information, pumping power generation points, pumping/power generation mode identification and initial up-and-down water levels, and the preprocessing comprises time information coding, pumping power generation point normalization and initial up-and-down water level normalization to obtain standardized input data; Step 3, randomly extracting part of data from the standardized input data to be used as a Bayesian optimization dataset, optimizing the super parameters of the XGBoost model by using a Bayesian optimization algorithm, and determining an optimal super parameter combination; Step 4, dividing the standardized input data into a training set, a verification set and a test set, training XGBoost models by using the training set and the verification set based on the optimal super parameter combination to obtain a relation model of the number of power generation/pumping points and the water level variation of the upper and lower libraries, and verifying the performance of the model by using the test set; And 5, inputting pumping power generation points, target time information, target pumping/power generation modes and target initial up-down warehouse water levels in a target pumping/power generation plan, obtaining up-down warehouse water level variation through a XGBoost model trained in the step 4, obtaining up-down warehouse predicted water levels by combining the target initial up-down warehouse water levels, obtaining up-down warehouse predicted storage capacities according to the up-down warehouse predicted water levels through a water level-storage capacity corresponding relation model obtained in the step 1, and subtracting the up-down warehouse predicted water levels to obtain a predicted water head.
2. The method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by taking time factors into consideration as claimed in claim 1, wherein in the step 1, the Support Vector Machine (SVM) adopts a Support Vector Regression (SVR) model, and optimal parameters of the SVM are determined through a grid search algorithm, and the optimal parameters comprise a regularization parameter C and a kernel function.
3. A pumped storage power station water level, head and reservoir capacity prediction method taking into account time factors as claimed in claim 2, wherein said kernel function is selected from one of RBF kernel function, linear kernel function, polynomial kernel function or neural network kernel function.
4. The method for predicting water level, water head and reservoir capacity of a pumped-storage power station taking time factors into consideration as set forth in claim 2, wherein the regularization parameter C has a value in the range of [0.001,1000].
5. The method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by taking time factors into consideration as claimed in claim 1, wherein in the step 1, the water level is used as an input parameter of the SVM, the reservoir capacity is used as an output parameter of the SVM, and the corresponding relation model of the water level and the reservoir capacity of the upper reservoir and the lower reservoir is obtained by training the SVM model by using the corresponding data of the water level and the reservoir capacity of the upper reservoir and the lower reservoir in the initial stage of water storage.
6. The method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by taking time factors into consideration as claimed in claim 1, wherein in the step 2, the time information comprises month information, day information and minute information, the time information is encoded by adopting a sine and cosine encoding mode, and an encoding formula is as follows: Sin_Mon=sin(2π×Month/12),Cos_Mon=cos(2π×Month/12); Sin_Day=sin(2π×Day/31),Cos_Day=cos(2π×Day/31); Sin_Min=sin(2π×Minute/1440),Cos_Min=cos(2π×Minute/1440); wherein, month represents the Month, day represents the Day of the Month, minute represents the Minute of the Day, and the time information is encoded in the following format: Time=[(Sin_Mon,Cos_Mon),(Sin_Day,Cos_Day),(Sin_Min,Cos_Min)]。
7. the method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by taking time factors into consideration as set forth in claim 1, wherein in the step 2, a normalization processing formula of the number of pumped power generation points is as follows: d f =D 1 /(a×96) for the number of generation points; for pumping points, D f =D 2 /(a×96); Wherein D 1 is the actual issued power generation point, D 2 is the actual issued pumping point, a is the number of sets of the pumping energy storage power station, 96 is the maximum point of a single set in one day, and D f is the normalized pumping power generation point.
8. The method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by considering the time factors according to claim 1 is characterized in that in the step 2, the definition of the pumping/generating mode identifier is that the pumping mode corresponds to identifier 0 and the generating mode corresponds to identifier 1, the initial upper reservoir water level and the initial lower reservoir water level are normalized and are expressed as H= [ H s ,H x ], wherein H s is the normalized upper reservoir initial water level, and H x is the normalized lower reservoir initial water level.
9. A method for predicting water level, head and capacity of a pumped-storage power plant taking into account time factors as defined in claim 1, wherein in step 3, the bayesian optimized objective function f (p) is defined as: wherein p is a super parameter space, n is the number of test sets in the Bayesian optimization data set, e (p, i) is a predicted value of the ith data of the test set under the super parameter combination, real is a true value of the ith data of the test set, and the smaller the objective function value is, the better the super parameter optimization effect is.
10. The method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by taking time factors into consideration as claimed in claim 1, wherein the Bayesian optimization adopts a GP proxy model, an optimal sampling point is selected by a method of maximizing expected increment EI, and the proxy model is iteratively updated to obtain global optimal super parameters.
11. The method for predicting water level, water head and reservoir capacity of a pumped-storage power station with consideration of time factors according to claim 1, wherein in step 3, super parameters of the XGBoost model include learning rate learning_rate, decision tree number n_ estimators, maximum tree depth_max and minimum node number min_child.
12. The method for predicting water level, water head and reservoir capacity of a pumped storage power station by taking time factors into consideration as set forth in claim 1, wherein in step 4, the training set is used for model parameter fitting, the verification set is used for monitoring the over-fitting condition in the model training process and adjusting parameters, the test set is used for evaluating the prediction accuracy of the model after training, the extraction proportion of the bayesian optimized data set is determined according to the data scale, and the reliability of super-parameter optimizing is ensured.
13. The method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by taking time factors into consideration according to claim 1 is characterized in that in the step 5, the water level variation of the upper reservoir and the lower reservoir is directly output through a XGBoost model, the predicted water level of the upper reservoir and the lower reservoir=the target initial upper reservoir and lower reservoir water level+the corresponding water level variation, the predicted water level of the upper reservoir and the lower reservoir is calculated by inputting the predicted water level of the upper reservoir and the lower reservoir into a water level-reservoir capacity corresponding relation model obtained in the step 1, and the predicted water head=the predicted water level of the upper reservoir and the predicted water level of the lower reservoir.

Description

Water level, water head and reservoir capacity prediction method of pumped storage power station considering time factors Technical Field The invention relates to the technical field of pumped storage power stations, in particular to a method for predicting the water level, water head and reservoir capacity of a pumped storage power station by taking time factors into consideration. Background The daily power generation and pumping plan of the pumped storage power station is regulated by a net to be sent in the previous day, and the power generation and pumping operation can cause the water level and water head change of the upper and lower reservoirs. The upper reservoir and the lower reservoir of the pumped storage power station can be divided into dead reservoir capacity, storage capacity regulation, dead water level and adjustable water level, if the power generation and the pumping plan are improper, the water level is lower than or higher than the adjustable water level, at the moment, the normal power generation or pumping of the unit can be influenced, the stress change of the reservoir bank and the dam body can be aggravated, and the problems of cracks, leakage and the like can be caused for a long time. Therefore, the water level and water head changes of the upper and lower reservoirs must be predicted in advance through the pumping and power generation plan issued by the network, and whether the power generation and pumping plan is reasonable or not is verified. The current mainstream prediction method is to obtain a relation formula of the number of power generation and pumping points and the change of the reservoir capacity through experiments, obtain the change of the reservoir capacities of the upper reservoir and the lower reservoir through the formula, and obtain the change quantity of the water levels of the upper reservoir and the lower reservoir through the corresponding relation between the reservoir capacities and the water levels so as to predict the water head. However, in the practical application process, the corresponding relation between the number of power generation and pumping points and the change of the storage capacity is not constant, errors are easily caused by utilizing a single mathematical formula to correspond, the method does not consider the influence caused by evaporation, precipitation and other environmental factors, and the prediction result is often not quite accurate and cannot be used as a criterion for judging whether the power generation and pumping plan is reasonable or not. Disclosure of Invention The invention aims to overcome the defects and provide a method for predicting the water level, the water head and the storage capacity of a pumped storage power station by considering time factors so as to solve the problems in the background technology. In order to solve the technical problems, the invention adopts the technical scheme that the method for predicting the water level, the water head and the reservoir capacity of the pumped storage power station by considering the time factor comprises the following steps: Step 1, acquiring corresponding data of water levels and reservoir capacities of an upper reservoir and a lower reservoir of a pumped storage power station, and fitting the corresponding data by adopting a support vector machine to generate water level-reservoir capacity curves of the upper reservoir and the lower reservoir, so as to obtain a corresponding relation model of the water levels and the reservoir capacities of the upper reservoir and the lower reservoir; Step 2, determining input variables of XGBoost models and preprocessing, wherein the input variables comprise time information, pumping power generation points, pumping/power generation mode identification and initial up-and-down water levels, and the preprocessing comprises time information coding, pumping power generation point normalization and initial up-and-down water level normalization to obtain standardized input data; Step 3, randomly extracting part of data from the standardized input data to be used as a Bayesian optimization dataset, optimizing the super parameters of the XGBoost model by using a Bayesian optimization algorithm, and determining an optimal super parameter combination; Step 4, dividing the standardized input data into a training set, a verification set and a test set, training XGBoost models by using the training set and the verification set based on the optimal super parameter combination to obtain a relation model of the number of power generation/pumping points and the water level variation of the upper and lower libraries, and verifying the performance of the model by using the test set; And 5, inputting pumping power generation points, target time information, target pumping/power generation modes and target initial up-down warehouse water levels in a target pumping/power generation plan, obtaining up-down warehouse water level variation through a XG