Hydropower station reservoir flow prediction and dispatch optimization method

Abstract

The application provides a hydropower station reservoir flow prediction and scheduling optimization method, which separates different scale hydrological characteristics through multi-time scale processing of multi-source data, extracts reservoir inflow characteristics by cooperating with gray correlation analysis, avoids time structure aliasing, and makes the hydrological law clearer; the prediction interval is corrected by the deviation of the observed value of the reservoir inflow, the prediction result is dynamically adjusted with the actual hydrology, the error of the initial prediction is made up, the prediction accuracy is twice improved, and the problem of inaccurate reservoir inflow prediction is solved. In addition, the corrected prediction interval is used as an uncertainty set, a double-layer robust optimization model of safety margin constraint is combined with a confidence parameter adjustment, scheduling safety and benefit are balanced; different risk level strategies are generated by weighting and combining scheduling decision variables through a risk preference parameter, and the scientificity and stability of reservoir scheduling are improved.

Description

Technical Field

[0001] This invention relates to the field of water resources management technology, and in particular to a method for predicting and optimizing the flow of hydropower station reservoirs. Background Technology

[0002] Hydropower station reservoir operation is a core component of optimal water resource allocation and efficient clean energy utilization. The operational effectiveness of hydropower station reservoirs directly impacts the achievement of multiple objectives, including maximizing power generation efficiency, ensuring flood control in the basin, and guaranteeing ecological baseflow. The scientific and rational nature of reservoir operation decisions is primarily achieved by predicting the spatiotemporal evolution of inflow and optimizing outflow and power generation control strategies in advance. This effectively avoids the risk of water level exceeding limits, reduces water wastage losses, and ensures stable unit operation. However, with intensifying climate change and diversified water demand in basins, inflow is increasingly influenced by multiple factors such as rainfall, evaporation, and runoff, exhibiting significant non-stationarity and multi-timescale characteristics. This presents a severe challenge to the accurate prediction and optimization of reservoir operation.

[0003] Currently, methods for predicting and scheduling inflows to hydropower station reservoirs mainly rely on two technical approaches: one is based on time-series statistical models, and the other is based on regression analysis using empirical formulas. Traditional methods can provide basic prediction accuracy under relatively stable meteorological and hydrological conditions. The core idea is to establish a linear relationship between input variables and inflows by mining statistical patterns in historical data at a single time scale, thereby guiding scheduling decisions. At the scheduling optimization level, existing technologies mostly employ single-objective or simple multi-objective optimization models with fixed constraints, treating the prediction results as deterministic inputs. They fail to fully consider the uncertainty caused by prediction errors, merely addressing fluctuations in hydrological processes by setting fixed safety margins.

[0004] However, existing technologies have significant shortcomings in adaptability and reliability under complex watershed conditions. They lack the ability to decompose and analyze features across multiple time scales. Traditional methods often use data input at a single time scale or within a fixed time window, making it difficult to separate the combined effects of long-term trends, periodic changes, and short-term fluctuations. This results in severe temporal structure aliasing among input variables, and the predictive models are insufficiently responsive to hydrological characteristics at different scales. Furthermore, they fail to effectively explore the nonlinear relationships between meteorological, precipitation, and evapotranspiration data and inflow, relying only on a small number of linearly correlated variables for modeling. This leads to weak anti-interference capabilities in the predictive models, and a significant drop in prediction accuracy during periods of drastic fluctuations in hydrological processes. Summary of the Invention

[0005] This invention provides a method for predicting and optimizing the flow of water in a hydropower station reservoir, which solves the problem of insufficient accuracy in predicting inflow in the prior art.

[0006] On the one hand, the present invention provides a method for predicting and optimizing the flow of a hydropower station reservoir, comprising: Multi-source data, including meteorological data, precipitation data, evapotranspiration data, and historical inflow data, are collected and processed at multiple time scales to obtain multi-scale time series. By using grayscale correlation analysis, the inflow characteristics of the multi-scale time series are extracted and input into the inflow prediction model, which outputs the predicted inflow value and prediction interval. Obtain inbound flow observations, and update the confidence parameters of the inbound flow prediction model and correct the prediction interval based on the deviation between the inbound flow observations and the predicted inbound flow. Using the prediction interval as the set of uncertainties as input, a two-layer robust optimization model is established, which includes an upper-layer multi-objective optimization function and a lower-layer operational constraint. The safety margin of the lower-layer operational constraint is adjusted in combination with the confidence parameter, and the scheduling decision variables for each time period are obtained by solving. The scheduling decision variables include outflow and power generation. The scheduling decision variables are weighted and adjusted according to preset risk preference parameters to generate scheduling strategies corresponding to different risk levels; the scheduling strategies include executable outbound and power generation control commands.

[0007] Optionally, the step of performing multi-time-scale processing on multi-source data to obtain multi-scale time series includes: Sliding windows of different time lengths are used to sample each component of the multi-source data to generate time series at different scales; Based on the scale of the time series, match the sliding window length corresponding to the scale; Using the specified sliding window length, the time series is processed by central moving average and normalization to map the data of the time series to a unified numerical range, thus obtaining a multi-scale time series.

[0008] Optionally, the step of extracting the inflow characteristics of the multi-scale time series through grayscale correlation analysis includes: The historical inbound traffic data is sorted by time to obtain the target reference sequence; Calculate the difference between the target reference sequence and the multi-scale time series at the same time point to obtain the difference sequence; Based on the overall distribution of all difference sequences, calculate the grey correlation coefficient sequence between the multi-scale time series and the target reference sequence; Calculate the mean of the grey relational coefficient sequence to obtain the comprehensive relational degree; The variables of the multi-scale time series are sorted from high to low according to the comprehensive correlation degree, and a preset number of variables are selected from the top to obtain the inbound flow characteristics.

[0009] Optionally, obtaining inbound flow observations and correcting the prediction interval based on the deviation between the inbound flow observations and the predicted inbound flow includes: Calculate the residual sequence between the predicted inbound flow rate and the observed inbound flow rate; The quantiles of the residual sequence are calculated to obtain the upper quantile and the lower quantile; The upper quantile is added to the observed inflow rate to obtain the upper limit of the prediction interval; The lower quantile is added to the observed inflow rate to obtain the lower limit of the prediction interval. The prediction interval is adjusted based on the upper limit and the lower limit.

[0010] Optionally, the confidence parameter of the inbound flow prediction model is updated based on the deviation between the observed inbound flow and the predicted inbound flow, including: Calculate the absolute deviation between the observed inbound flow rate and the predicted inbound flow rate; If the observed inflow rate falls within the predicted range, a coverage indication signal is generated; The absolute deviation is input into the error reduction function to calculate the first influence factor on the confidence parameter; The coverage indication signal is input into the coverage increment function to calculate the second influence factor on the confidence parameter; The confidence parameter, the first influence factor, and the second influence factor are weighted and summed to obtain the updated confidence parameter.

[0011] Optionally, the prediction interval is used as the input uncertainty set to establish a two-layer robust optimization model containing an upper-layer multi-objective optimization function and a lower-layer operational constraint. The safety margin of the lower-layer operational constraint is adjusted based on the confidence parameter, and the scheduling decision variables for each time period are obtained. These scheduling decision variables include outflow and power generation, including: Construct a multi-objective optimization function; wherein the multi-objective optimization function is: ; in, The total objective function value; The weighting coefficients are non-negative and satisfy the following conditions: ; This represents the penalty function for deviations in power generation from the target output; This represents the penalty function for water levels deviating from the target water level. A smoothing penalty function representing changes in outbound flow rate; This represents the penalty function for water wastage. This represents the amount of water discarded in the current period. Construct lower-level operational constraints, including water balance constraints, water level constraints, outflow constraints, and power constraints; The safety margin of the lower-level operational constraints is adjusted in conjunction with the confidence level parameter. Based on the multi-objective optimization function and the lower-level operational constraints, a two-layer robust optimization model is established. The uncertainty set, which is the prediction interval, is input into the two-layer robust optimization model. Using the scenario generation method, the model optimizes the solution by minimizing the total objective function value while satisfying the lower-level operational constraints, and outputs the scheduling decision variables for each time period.

[0012] Optionally, the scheduling decision variables are weighted and adjusted according to preset risk preference parameters to generate scheduling strategies corresponding to different risk levels, including: Introducing risk preference parameters Calculate the weighted combination; wherein the weighted combination is: ; in, This is the upper limit of the prediction interval. This represents the lower limit of the prediction interval; Will Input the two-layer robust optimization model and generate corresponding... conservative scheduling strategy Neutral scheduling strategy and The proactive scheduling strategy.

[0013] Optional, also includes: Based on the predicted inflow rate, a predicted water level is generated. Obtain the reservoir water level and calculate the deviation between the predicted water level and the reservoir water level; When the deviation exceeds a preset deviation threshold, the residual parameters and confidence parameters of the inbound flow prediction model are corrected according to the direction of the deviation.

[0014] Optionally, when the deviation exceeds a preset deviation threshold, the residual parameters and confidence parameters of the inbound flow prediction model are corrected according to the direction of the deviation, including: When the direction of the deviation is positive, the upper quantile of the residual is shifted in the positive direction; when the direction of the deviation is negative, the lower quantile of the residual is shifted in the negative direction. The confidence parameter value is reduced proportionally based on the magnitude of the deviation.

[0015] Optional, also includes: Obtain the reservoir's water storage capacity; When the reservoir's water storage capacity exceeds the first preset water storage capacity, the risk preference parameter will be... Increase the preset value; When the reservoir's water storage capacity is less than the second preset water storage capacity, the risk preference parameter will be... Lower the preset value.

[0016] On the other hand, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the hydropower station reservoir flow prediction and scheduling optimization method as described above.

[0017] On the other hand, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the hydropower station reservoir flow prediction and scheduling optimization method as described above.

[0018] On the other hand, the present invention also provides a computer program product, including a computer program that, when executed by a processor, implements the hydropower station reservoir flow prediction and scheduling optimization method as described above.

[0019] The method for predicting and optimizing reservoir flow in hydropower stations provided by this invention processes multi-source data at multiple time scales to separate hydrological features at different scales. It then uses gray-scale correlation analysis to extract inflow characteristics, avoiding temporal structure overlap and making hydrological patterns clearer. By correcting the prediction interval through deviations in inflow observations, the prediction results are dynamically adjusted according to actual hydrological changes, compensating for initial prediction errors and achieving a secondary improvement in prediction accuracy, thus solving the problem of inaccurate inflow prediction. Furthermore, the corrected prediction interval is used as an uncertainty set, combined with a two-layer robust optimization model that adjusts the safety margin constraint using confidence parameters, balancing scheduling safety and efficiency. Finally, by weighting and combining scheduling decision variables using risk preference parameters, strategies with different risk levels are generated, improving the scientific nature and stability of reservoir scheduling. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating the method for predicting and optimizing the flow of a hydropower station reservoir provided in an embodiment of the present invention. Figure 2This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0023] Figure 1 This is a flowchart illustrating the method for predicting and optimizing the flow of a hydropower station reservoir provided in an embodiment of the present invention.

[0024] like Figure 1 As shown in the figure, the method for predicting and optimizing the flow of a hydropower station reservoir provided in this embodiment of the invention mainly includes the following steps: 101. Collect multi-source data including meteorological data, precipitation data, evapotranspiration data and historical inflow data, and process the multi-source data at multiple time scales to obtain multi-scale time series.

[0025] Among these methods, multi-timescale processing of multi-source data can yield feature information at different time granularities, such as data sequences at hourly, daily, and monthly scales. This multi-timescale processing approach helps to comprehensively capture the changing patterns of data across different time dimensions, providing richer and more accurate data support for water level prediction and scheduling optimization.

[0026] Specifically, multi-source data is processed at multiple time scales to obtain multi-scale time series, including: Sliding windows of different time lengths are used to sample each component of multi-source data to generate time series at different scales; Based on the scale of the time series, match the sliding window length corresponding to the scale; By using the sliding window length, the time series is processed by central moving average and normalization, mapping the time series data to a unified numerical range to obtain a multi-scale time series.

[0027] The selection of the sliding window length needs to be dynamically adjusted according to the data characteristics and prediction requirements. The central moving average processing can effectively reduce random interference by calculating the arithmetic mean of the data within the window, while preserving the periodic characteristics of the original sequence. The normalization processing maps data of different dimensions to the interval [0,1] or [-1,1] through linear transformation, eliminating the model training bias caused by differences in data scale, and providing standardized input for subsequent feature extraction and model fusion.

[0028] Taking historical inflow data and meteorological data as examples, let the sequence of historical inflow data be: ; in, Indicates time step The inflow observation value, This represents the total observation time.

[0029] Suppose the meteorological observation data sequence is as follows: ; in, Indicates time step The meteorological element vector includes indicators such as precipitation, evaporation, and temperature.

[0030] The multi-source data includes historical inflow data, meteorological data, precipitation data, and evapotranspiration data. Its multi-source data components include long-term trend components, periodic components, and short-term fluctuation components. Based on the time-scale attributes of hydrological characteristics, sets of sliding window lengths corresponding to different scales are selected. These correspond to long-term, periodic, and short-term scales, respectively. The smoothing operation expression is: ; in, Indicates the window length is time The result is a smooth one.

[0031] For short-term time series, a shorter sliding window length is used to preserve short-term fluctuations; for periodic time series, a medium-length sliding window is used to capture periodic changes; and for long-term time series, a longer sliding window is used to extract long-term trends. This precise matching of window length and scale ensures that each scale's time series maximizes the highlighting of the core change features of its corresponding dimension, avoiding interference between features from different scales.

[0032] By performing smoothing on different window lengths, the long-term trend series is obtained. Periodic sequence and short-term fluctuation sequences Finally, the multi-scale decomposition expression of the inflow sequence can be obtained: ; in, Indicates long-term component, Represents periodic components, Indicates the short-term component.

[0033] After obtaining time series at different time scales, Min-Max standardization is used for normalization to map all data to the [0,1] interval, eliminating the influence of dimensional differences and numerical ranges, and finally forming a multi-scale time series with clear structure and prominent features.

[0034] 102. Through grayscale correlation analysis, extract the inflow characteristics of multi-scale time series and input them into the inflow prediction model to output the predicted inflow value and prediction interval.

[0035] The expression for the inbound flow prediction model is as follows: ; in, For the predicted time The predicted inbound flow rate For regression coefficients, For the selected number of features, The constant term represents the prediction baseline level when all input feature variables are zero or take the baseline value. It is obtained by the least squares method and is used to supplement the bias of the inflow forecast model, so that the output of the inflow forecast model is consistent with the historical observation value at the mean level.

[0036] To determine the upper and lower limits of the prediction interval, the historical residual sequence is calculated. And the quantiles of the residuals are calculated.

[0037] Let the lower quantile of the residual be... The upper quantile is Then the upper and lower limits of the prediction interval are respectively: .

[0038] After inputting the inbound flow characteristics into the inbound flow prediction model, the predicted inbound flow value and prediction range can be obtained.

[0039] The characteristics of inbound traffic can be obtained through grayscale correlation analysis. In some embodiments, grayscale correlation analysis is used to extract multi-scale time series inbound traffic characteristics, including: The historical inbound traffic data is sorted by time to obtain the target reference sequence; The difference between the target reference sequence and the multi-scale time series at the same time point is calculated to obtain the difference sequence; Based on the overall distribution of all difference sequences, calculate the grey correlation coefficient sequence between the multi-scale time series and the target reference sequence; Calculate the mean of the grey relational coefficient sequence to obtain the comprehensive relational degree; Sort the variables of the multi-scale time series from high to low according to the comprehensive correlation degree, and select the top preset number of variables to obtain the reservoir inflow characteristics.

[0040] Specifically, sort the collected historical reservoir inflow data in chronological order to construct a target reference sequence; among them, the target reference sequence is: ; Among them, represents the observed value of the reservoir inflow at time step , is the total observation duration. The target reference sequence serves as a benchmark for measuring the correlation degree between other variables and the reservoir inflow.

[0041] Let each variable in the multi-scale time series be a comparison sequence , calculate the difference between the comparison sequence and the target reference sequence at time : ; Define the minimum difference and the maximum difference .

[0042] The gray correlation coefficient is defined as: ; Among them, is the resolution coefficient, satisfying .

[0043] To quantify the overall correlation degree between the comparison sequence and the target reference sequence, calculate the average correlation coefficient of each characteristic variable by averaging the gray correlation coefficient sequences of each comparison sequence: ; Sort all comparison sequences according to the comprehensive correlation degree from high to low, set the preset number according to the actual modeling requirements, select the top variables as the core features, and the top variables are the reservoir inflow characteristics most closely related to the reservoir inflow.

[0044] 103. Obtain the observed value of the reservoir inflow, and update the confidence parameters and correction prediction intervals of the reservoir inflow prediction model according to the deviation of the reservoir inflow prediction value from the observed value of the reservoir inflow.

[0045] After obtaining the reservoir inflow prediction value, it is also necessary to correct the confidence parameters and prediction intervals of the reservoir inflow prediction model according to the deviation between the observed value of the reservoir inflow and the reservoir inflow prediction value.

[0046] Specifically, the process involves obtaining observed inflow rates and updating the confidence parameters of the inflow rate prediction model based on the deviation between the observed and predicted inflow rates. This includes: Calculate the absolute deviation between the observed inflow volume and the predicted inflow volume; If the observed inflow rate falls within the prediction range, a coverage indication signal is generated; The absolute deviation is input into the error reduction function to calculate the first influence factor on the confidence parameter; The coverage indication signal is input into the coverage increment function to calculate the second influence factor on the confidence parameter; The confidence level parameter, the first impact factor, and the second impact factor are weighted and summed to obtain the updated confidence level parameter.

[0047] The confidence parameter describes the reliability of the prediction result, assuming a time interval of 100°. The inbound flow forecast range is Predicted point value The actual observed value is The absolute deviation is: ; in, This represents the absolute deviation between the predicted inflow and the observed inflow.

[0048] After calculating the absolute deviation, determine whether the observed inflow rate falls within the prediction range.

[0049] If the conditions are met Then a coverage indication signal is generated, and the coverage indication signal is defined. ;otherwise, .

[0050] Let the confidence parameter at the previous time step be... The initial confidence parameter is The confidence level parameter takes values ​​within the range of 100%. The confidence parameter is updated using a weighted iterative approach: ; in, The weighting coefficients are non-negative and satisfy the following conditions: .

[0051] This is an error reduction function used to adjust the confidence parameter based on the magnitude of the prediction error. To cover the increasing function, increase the confidence level when the predicted interval covers the observed inflow rate. This is the confidence level term from the previous time step, used to maintain temporal continuity.

[0052] function and It can be expressed in linear or piecewise linear form by substituting the absolute deviation into the function. and The first impact factor and the second impact factor were obtained respectively. For example: when At that time, set when hour, .

[0053] in, To determine the threshold for deviation.

[0054] function Defined as hour, This setting ensures that the confidence parameter automatically decreases when the predicted inflow rate deviates significantly or when the interval does not cover the observed inflow rate.

[0055] Through the above steps, the new confidence parameters are obtained. .

[0056] In addition, inflow observations are obtained, and the forecast interval is corrected based on the deviation between the inflow observations and the predicted inflow values, including: Calculate the residual sequence between the predicted inbound flow and the observed inbound flow; Count the quantiles of the residual sequence to obtain the upper and lower quantiles; Add the upper quantile to the observed inflow rate to obtain the upper limit of the prediction interval; Add the upper and lower quantiles to the inflow flow observations to obtain the lower limit of the prediction interval; The prediction interval is adjusted based on the upper and lower limits.

[0057] Specifically, obtain real-time inbound flow observation values. The corresponding inflow forecast value According to the formula Calculate the single-time residual, where for The time residual is positive, indicating that the predicted value is too small, and negative, indicating that the predicted value is too large.

[0058] The residuals from multiple consecutive time points are arranged in chronological order to form a residual sequence. The residual sequence visually reflects the magnitude and direction distribution of the prediction bias.

[0059] When performing statistical analysis on the constructed residual series, the quantiles are determined based on the updated confidence parameters. For example, the correction coefficient is set as follows: ; in, This is an adjustment factor. When the confidence level is low, Decrease, thereby expanding the interval; when the confidence level is high Approaching 1 causes the interval to shrink. The corrected upper and lower limits of the prediction interval are: ; in, The lower quantile, It represents the upper quantile.

[0060] Calculated and A new prediction interval is constructed to replace the initial prediction interval. The revised prediction interval can dynamically adapt to the actual hydrological changes. When the prediction deviation is large, the interval width is automatically adjusted to improve the reliability of the prediction results and the accuracy of the uncertainty description, providing a more reasonable uncertainty input for the subsequent two-layer robust optimization model.

[0061] 104. Using the prediction interval as the set of uncertainties as input, establish a two-layer robust optimization model that includes an upper-layer multi-objective optimization function and a lower-layer operational constraint. Combine the confidence parameter to adjust the safety margin of the lower-layer operational constraint, and solve for the scheduling decision variables for each time period.

[0062] The scheduling decision variables include outflow and power generation. The upper-level multi-objective optimization function comprehensively considers power generation efficiency, flood control safety, and ecological protection, achieving coordinated optimization of multiple objectives by setting different weight coefficients. The lower-level operational constraints include various constraints. During the solution process, the safety margin of the lower-level operational constraints is dynamically adjusted according to the confidence level parameter. When the confidence level is high, the safety margin is appropriately reduced to improve power generation efficiency; when the confidence level is low, the safety margin is increased to ensure flood control safety and ecological protection. By solving the two-level robust optimization model, the scheduling decision variables such as outflow and power generation for each time period can be obtained.

[0063] Specifically, using the prediction interval as the input uncertainty set, a two-layer robust optimization model is established, comprising an upper-level multi-objective optimization function and a lower-level operational constraint. The safety margin of the lower-level operational constraint is adjusted using a confidence parameter, and the scheduling decision variables for each time period are obtained. These scheduling decision variables include outflow and power generation, including: Construct a multi-objective optimization function; whereby the multi-objective optimization function is: ; in, The total objective function value; The weighting coefficients are non-negative and satisfy the following conditions: ; This represents the penalty function for deviations in power generation from the target output; This represents the penalty function for water levels deviating from the target water level. A smoothing penalty function representing changes in outbound flow rate; This represents the penalty function for water wastage. This represents the amount of water discarded in the current period. Construct lower-level operational constraints, including water balance constraints, water level constraints, outflow constraints, and power constraints; Adjust the safety margin of lower-level operational constraints based on confidence parameters; A two-layer robust optimization model based on multi-objective optimization functions and lower-level operational constraints; The uncertainty set, which is the prediction interval, is input into the two-layer robust optimization model. The scenario generation method is used to minimize the total objective function value and optimize the solution under the condition of satisfying the lower-level operational constraints. The scheduling decision variables for each time period are output.

[0064] Specifically, obtaining the time Inbound flow forecast range and the corresponding confidence parameters At the same time, obtain the current water level. Outbound flow Unit output State variables, etc. Optimization calculations for prediction period. For the time frame, the decision variables include outbound flow rates for each time period. With unit output .

[0065] To simultaneously consider power generation revenue and operational safety, a two-layer robust optimization model is established. The upper-layer objective is a weighted multi-objective optimization function, while the lower-layer includes physical and operational constraints. The upper-layer objective function is defined as follows: The multi-objective optimization function is: ; in, The total objective function value; The weighting coefficients are non-negative and satisfy the following conditions: ; This represents the penalty function for deviations in power generation from the target output; This represents the penalty function for water levels deviating from the target water level. A smoothing penalty function representing changes in outbound flow rate; This represents the penalty function for water wastage. This represents the amount of water discarded during the current period.

[0066] Multi-objective optimization functions can be in piecewise linear or quadratic form to ensure solvability.

[0067] The lower-level constraints include water balance constraints, water level constraints, outflow constraints, and power constraints.

[0068] Water balance constraints are defined as follows: ; in, and Representing time respectively and water level, For time intervals, This represents the average water surface area of ​​the reservoir area.

[0069] Water level constraint is defined as: ; in, The lower limit water level is set as the minimum water level, which is dynamically adjusted based on the confidence level parameter.

[0070] When the confidence level is low, the safety interval is increased to raise the lower limit water level in order to maintain a safety margin; when the confidence level is high, the original lower limit value is maintained.

[0071] Safety margin coefficient Defined as: ; in, To design the minimum operating water level, This is a safety correction term for the confidence level parameter.

[0072] Outbound flow constraint is defined as: ; in, This represents the upper limit of outbound shipments, which varies with the confidence parameter. When the confidence parameter decreases, the upper limit of outbound shipments is appropriately reduced to control the risk of leakage.

[0073] Power constraints are defined as: ; in These are the lower and upper limits of the unit's rated output, respectively: The above constraints ensure the feasibility of the two-layer robust optimization model under the condition of uncertain inflow.

[0074] Robust optimization problems are solved using scenario generation or linear approximation methods. Several representative inflow scenarios within the prediction interval are then used. Input the model and solve the multi-scenario joint optimization problem: ; in, Describing a scenario The objective function value is obtained. By solving this equation, the scheduling decision variables that are robust to uncertain inflow conditions can be obtained.

[0075] 105. Based on preset risk preference parameters, the scheduling decision variables are weighted and adjusted to generate scheduling strategies corresponding to different risk levels; the scheduling strategies include executable outbound and power generation control commands.

[0076] To generate scheduling schemes with different risk preferences, a risk preference parameter is introduced. Specifically, scheduling decision variables are weighted and adjusted based on the preset risk preference parameter to generate scheduling strategies corresponding to different risk levels, including: Introducing risk preference parameters Calculate the weighted combination; where the weighted combination is: ; in, This is the upper limit of the prediction interval. This represents the lower limit of the prediction interval; Will Input a two-layer robust optimization model and generate corresponding... conservative scheduling strategy Neutral scheduling strategy and The proactive scheduling strategy.

[0077] In some embodiments, the hydropower station reservoir flow prediction and scheduling optimization method provided by the present invention further includes: Based on the predicted inflow rate, a predicted water level is generated. Obtain the reservoir water level and calculate the deviation between the predicted water level and the reservoir water level; When the deviation exceeds the preset deviation threshold, the residual parameters and confidence parameters of the inbound flow prediction model are corrected according to the direction of the deviation.

[0078] Specifically, at the current moment reservoir water level Substituting the initial values ​​into the water balance equation: ; in, For predicting time intervals, for Average water surface area of ​​the reservoir at any given time for Forecast value of inbound flow at any time The solution obtained based on the two-layer robust optimization model Real-time outbound flow scheduling decision value for Water volume at all times for Evaporation rate at all times for The leakage rate at any given time; by iteratively calculating the water balance equation over time, the predicted water level for each future time period can be obtained. This forms a complete predicted water level.

[0079] The deviation between the predicted water level and the actual measured water level at the current time can be obtained by subtracting the predicted water level from the actual measured water level at the current time. . when deviation Exceeding the preset deviation threshold , if the deviation If the deviation is greater than 0, then increase the residual parameter of the inbound flow prediction model and decrease the confidence parameter; if the deviation is greater than 0, then increase the residual parameter of the inbound flow prediction model and decrease the confidence parameter. If the value is less than 0, then the residual parameter of the inbound flow prediction model will be reduced and the confidence parameter will be increased.

[0080] Specifically, when the deviation exceeds a preset deviation threshold, the residual parameters and confidence parameters of the inbound flow prediction model are corrected according to the direction of the deviation, including: When the direction of the deviation is positive, the upper quantile of the residual is shifted in the positive direction; when the direction of the deviation is negative, the lower quantile of the residual is shifted in the negative direction. The confidence parameter value is reduced proportionally based on the magnitude of the deviation.

[0081] Among them, when the actual water level is consistently higher than the predicted water level, the upper quantile value will be used. Shift in the positive direction; when the water level remains below the predicted level, adjust the lower quantile value. Shift in the negative direction to make the predicted interval consistent with the actual deviation trend.

[0082] When updating the confidence parameter, it is updated based on the magnitude of the water level deviation. Let the current confidence level be... The corrected confidence level is: ; in, The scaling factor is adjusted to adjust the confidence level. The update rule automatically reduces the confidence level when the prediction error increases, thereby expanding the prediction interval in the next prediction period to increase the safety margin.

[0083] In some embodiments, the hydropower station reservoir flow prediction and scheduling optimization method provided by the present invention further includes: Obtain the reservoir's water storage capacity; When the reservoir's water storage capacity exceeds the first preset water storage capacity, the risk preference parameter will be adjusted. Increase the preset value; When the reservoir's water storage is less than the second preset water storage capacity, the risk preference parameter will be... Lower the preset value.

[0084] The first preset water storage capacity is greater than the second preset water storage capacity, and the preset values ​​can be set and adjusted according to actual conditions. Risk preference parameters are dynamically adjusted based on the reservoir's water storage capacity. This approach allows for more flexible adaptation of hydropower station reservoir flow prediction and scheduling optimization methods to different reservoir water storage conditions, further improving the accuracy and rationality of prediction and scheduling.

[0085] For example, when the reservoir's water storage capacity exceeds the first preset storage capacity, it indicates that the reservoir's current water resources are relatively abundant. In this case, the risk preference parameter... Increasing the preset value allows for a more proactive scheduling strategy in subsequent dispatch decisions, appropriately increasing outflow and power generation to fully utilize the reservoir's water resources and improve power generation efficiency. If the first preset storage capacity is set at 80% of the reservoir's total storage capacity, when the reservoir's storage capacity exceeds this first preset capacity, the risk preference parameter will be adjusted. Increasing the value by 0.1-0.25 will allow for greater consideration of proactive scheduling strategies when solving the two-layer robust optimization model, making scheduling decisions more inclined to improve power generation revenue.

[0086] When the reservoir's water storage capacity is less than the second preset storage capacity, it indicates that the reservoir's current water resources are relatively strained. At this point, the risk preference parameter... Lowering the second preset value requires a more cautious approach in subsequent scheduling decisions, favoring a conservative strategy that reduces outflow and power generation to ensure the reservoir's water level remains within a safe range, safeguarding flood control and ecological protection. If the second preset storage capacity is set at 30% of the total reservoir capacity, when the reservoir's storage falls below this threshold, the risk preference parameter will be adjusted. By lowering the value by 0.1-0.25, when solving the two-layer robust optimization model, more conservative scheduling strategies will be considered, making scheduling decisions more focused on the safe operation of the reservoir.

[0087] Figure 2 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0088] like Figure 2As shown, the electronic device may include a processor 210, a communication interface 220, a memory 230, and a communication bus 240. The processor 210, communication interface 220, and memory 230 communicate with each other via the communication bus 240. The processor 210 can call logical instructions from the memory 230 to execute a method for predicting and optimizing the flow of water in a hydropower station reservoir.

[0089] Furthermore, the logical instructions in the aforementioned memory 230 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0090] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the hydropower station reservoir flow prediction and scheduling optimization method provided by the above methods.

[0091] In another aspect, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to perform the hydropower station reservoir flow prediction and scheduling optimization method provided by the above methods.

[0092] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0093] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0094] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting and optimizing the flow of a hydropower station reservoir, characterized in that, include: Multi-source data, including meteorological data, precipitation data, evapotranspiration data, and historical inflow data, are collected and processed at multiple time scales to obtain multi-scale time series. By using grayscale correlation analysis, the inflow characteristics of the multi-scale time series are extracted and input into the inflow prediction model, which outputs the predicted inflow value and prediction interval. Obtain inbound flow observations, and update the confidence parameters of the inbound flow prediction model and correct the prediction interval based on the deviation between the inbound flow observations and the predicted inbound flow. Using the prediction interval as the set of uncertainties as input, a two-layer robust optimization model is established, which includes an upper-layer multi-objective optimization function and a lower-layer operational constraint. The safety margin of the lower-layer operational constraint is adjusted in combination with the confidence parameter, and the scheduling decision variables for each time period are obtained by solving. The scheduling decision variables include outflow and power generation. The scheduling decision variables are weighted and adjusted according to preset risk preference parameters to generate scheduling strategies corresponding to different risk levels; the scheduling strategies include executable outbound and power generation control commands.

2. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 1, characterized in that, The process of performing multi-time-scale processing on multi-source data to obtain multi-scale time series includes: Sliding windows of different time lengths are used to sample each component of the multi-source data to generate time series at different scales; Based on the scale of the time series, match the sliding window length corresponding to the scale; Using the specified sliding window length, the time series is processed by central moving average and normalization to map the data of the time series to a unified numerical range, thus obtaining a multi-scale time series.

3. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 1, characterized in that, The extraction of inbound flow characteristics from the multi-scale time series through grayscale correlation analysis includes: The historical inbound traffic data is sorted by time to obtain the target reference sequence; Calculate the difference between the target reference sequence and the multi-scale time series at the same time point to obtain the difference sequence; Based on the overall distribution of all difference sequences, calculate the grey correlation coefficient sequence between the multi-scale time series and the target reference sequence; Calculate the mean of the grey relational coefficient sequence to obtain the comprehensive relational degree; The variables of the multi-scale time series are sorted from high to low according to the comprehensive correlation degree, and a preset number of variables are selected from the top to obtain the inbound flow characteristics.

4. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 1, characterized in that, Obtaining inbound flow observations and correcting the prediction interval based on the deviation between the inbound flow observations and the predicted inbound flow, including: Calculate the residual sequence between the predicted inbound flow rate and the observed inbound flow rate; The quantiles of the residual sequence are calculated to obtain the upper quantile and the lower quantile; The upper quantile is added to the observed inflow rate to obtain the upper limit of the prediction interval; The lower quantile is added to the observed inflow rate to obtain the lower limit of the prediction interval. The prediction interval is adjusted based on the upper limit and the lower limit.

5. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 4, characterized in that, Obtain inbound flow observations and update the confidence parameters of the inbound flow prediction model based on the deviation between the observed inbound flow values ​​and the predicted inbound flow values, including: Calculate the absolute deviation between the observed inflow volume and the predicted inflow volume; If the observed inflow rate falls within the predicted range, a coverage indication signal is generated; The absolute deviation is input into the error reduction function to calculate the first influence factor on the confidence parameter; The coverage indication signal is input into the coverage increment function to calculate the second influence factor on the confidence parameter; The confidence parameter, the first influence factor, and the second influence factor are weighted and summed to obtain the updated confidence parameter.

6. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 1, characterized in that, Using the prediction interval as the input uncertainty set, a two-layer robust optimization model is established, comprising an upper-layer multi-objective optimization function and a lower-layer operational constraint. The safety margin of the lower-layer operational constraint is adjusted based on the confidence parameter, and the scheduling decision variables for each time period are obtained. These scheduling decision variables include outflow and power generation, including: Construct a multi-objective optimization function; wherein the multi-objective optimization function is: ； in, The total objective function value; The weighting coefficients are non-negative and satisfy the following conditions: ; This represents the penalty function for deviations in power generation from the target output; This represents the penalty function for water levels deviating from the target water level. A smoothing penalty function representing changes in outbound flow; This represents the penalty function for water wastage. This represents the amount of water discarded in the current period. Construct lower-level operational constraints, including water balance constraints, water level constraints, outflow constraints, and power constraints; The safety margin of the lower-level operational constraints is adjusted in conjunction with the confidence level parameter. Based on the multi-objective optimization function and the lower-level operational constraints, a two-layer robust optimization model is established. The uncertainty set, which is the prediction interval, is input into the two-layer robust optimization model. Using the scenario generation method, the model optimizes the solution by minimizing the total objective function value while satisfying the lower-level operational constraints, and outputs the scheduling decision variables for each time period.

7. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 1, characterized in that, The scheduling decision variables are weighted and adjusted according to preset risk preference parameters to generate scheduling strategies corresponding to different risk levels, including: Introducing risk preference parameters Calculate the weighted combination; wherein the weighted combination is: ； in, This is the upper limit of the prediction interval. This represents the lower limit of the prediction interval; Will Input the two-layer robust optimization model and generate corresponding... conservative scheduling strategy Neutral scheduling strategy and The proactive scheduling strategy.

8. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 4, characterized in that, Also includes: Based on the predicted inflow rate, a predicted water level is generated. Obtain the reservoir water level and calculate the deviation between the predicted water level and the reservoir water level; When the deviation exceeds a preset deviation threshold, the residual parameters and confidence parameters of the inbound flow prediction model are corrected according to the direction of the deviation.

9. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 8, characterized in that, When the deviation exceeds a preset deviation threshold, the residual parameters and confidence parameters of the inbound flow prediction model are corrected according to the direction of the deviation, including: When the direction of the deviation is positive, the upper quantile of the residual is shifted in the positive direction; when the direction of the deviation is negative, the lower quantile of the residual is shifted in the negative direction. The confidence parameter value is reduced proportionally based on the magnitude of the deviation.

10. The method for predicting and optimizing the flow of a hydropower station reservoir according to claim 7, characterized in that, Also includes: Obtain the reservoir's water storage capacity; When the reservoir's water storage capacity exceeds the first preset water storage capacity, the risk preference parameter will be... Increase the preset value; When the reservoir's water storage capacity is less than the second preset water storage capacity, the risk preference parameter will be... Lower the preset value.

Images