Water power cluster generation scheduling optimization method and system based on water regime prediction

Abstract

The present application relates to the technical field of water and electricity dispatching, and particularly relates to a water and electricity cluster power generation dispatching optimization method and system based on water regime prediction. Multi-source prediction data of water, wind and light are loaded first, multi-dimensional joint scene trees are obtained through multi-source uncertainty joint scene generation, and joint distribution fuzzy sets are constructed accordingly; meanwhile, a multi-element dynamic constraint set is formed by loading a cascade water and electricity fine coupling model and various constraint data, a multi-time scale nested distribution robust rolling optimization model is constructed and executed based on the two, unit load distribution instructions are obtained, and finally a dispatching decision scheme is output, so that multi-dimensional optimization of water and electricity cluster dispatching is realized.

Description

Technical Field

[0001] This invention relates to the field of hydropower dispatching technology, and in particular to a method and system for optimizing hydropower cluster generation dispatching based on hydrological forecasting. Background Technology

[0002] The combined power generation system, consisting of cascade hydropower stations and new energy sources such as wind and solar power, is an important form of achieving efficient consumption of clean energy. However, the optimal scheduling of this system faces a dual challenge: on the one hand, hydropower scheduling is highly dependent on natural runoff that is difficult to predict accurately; on the other hand, the intermittency and volatility of wind and solar power output are transmitted in reverse through the power grid, requiring hydropower to undertake more frequent regulation tasks.

[0003] Existing scheduling methods have the following shortcomings: While patent CN114529075A uses bibliometric optimization to handle wind and solar forecasting errors, its modeling object is a comprehensive energy system containing multiple loads including cooling, heating, and electricity, failing to consider runoff uncertainty and the hydraulic coupling constraints unique to cascade hydropower. Patent CN120033771A proposes a conceptual framework for wind-solar-hydro complementary scheduling, but it only focuses on aggregation characteristic analysis and flow matching modes, without disclosing specific uncertainty modeling methods and multi-timescale optimization algorithms. Furthermore, traditional methods often optimize medium- and long-term reservoir water level control and short-term load allocation separately, making it difficult to adapt to the rapid fluctuations of new energy sources, resulting in both head benefit losses and water abandonment risks.

[0004] Therefore, there is an urgent need for a scheduling method that can simultaneously handle the uncertainties of runoff and wind-solar interaction and achieve multi-timescale collaborative optimization. Summary of the Invention

[0005] This invention addresses the problem in existing technologies that cannot simultaneously handle the uncertainties of runoff and wind-solar combined processes and achieve multi-timescale collaborative optimization by providing a method and system for optimizing the scheduling of hydropower clusters based on hydrological forecasting.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: In a first aspect, the present invention provides a method for optimizing the scheduling of hydropower cluster generation based on hydrological forecasting, including: Load multi-source runoff prediction data of cascade hydropower station groups, wind power prediction data of wind farms, and photovoltaic power prediction data of photovoltaic power stations; Based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data, a multi-source uncertainty joint scenario generation is performed to obtain a multi-dimensional joint scenario tree, wherein the multi-dimensional joint scenario tree includes several runoff scenarios, several wind power output scenarios, and several photovoltaic power output scenarios. Based on the multidimensional joint scene tree, a joint distributed fuzzy set is constructed using the Wasserstein distance metric, wherein the joint distributed fuzzy set is a Wasserstein sphere centered on the empirical distribution of the multidimensional joint scene tree and measured by the dynamic fuzziness radius. Load the refined coupling model of the cascade hydropower station group, and load the power grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data and new energy consumption constraint data, and configure it as a multi-dimensional dynamic constraint set; Based on the joint distributed fuzzy set and the multivariate dynamic constraint set, a multi-time-scale nested sub-Bruker rolling optimization model is constructed and executed to obtain the automatic generation control unit load allocation command; Extract the load allocation instructions of the automatic generation control unit and configure them as the output of the dispatch decision scheme.

[0007] Optionally, load multi-source runoff prediction data for the cascade hydropower station group, including: Numerical weather forecast data for the basin where the cascade hydropower station group is located is obtained, including rainfall forecast data, temperature forecast data, and evaporation forecast data; Load the measured data of the hydrological stations in the watershed, which includes measured data of upstream inflow, measured data of inter-regional runoff, and measured data of downstream water level; Based on the numerical weather forecast data and the measured data from the hydrological stations, a deep neural network quantile regression model is used to predict runoff probability, obtaining short-term runoff prediction data, medium-term runoff prediction data and long-term runoff prediction data, which are then configured as the multi-source runoff prediction data.

[0008] The data includes wind power prediction data for wind farms and photovoltaic power prediction data for photovoltaic power plants, including: Obtain numerical weather forecast data for the area where the wind farm is located, the numerical weather forecast data including wind speed forecast data and wind direction forecast data; Load the historical power output data of the wind farm, and based on the wind speed forecast data, the wind direction forecast data and the historical power output data, use a temporal convolutional network to perform wind power probability prediction to obtain the wind power prediction data; Numerical weather forecast data for the area where the photovoltaic power station is located is obtained, including solar radiation forecast data, temperature forecast data, and cloud cover forecast data; The historical output data of the photovoltaic power station is loaded, and based on the solar radiation forecast data, temperature forecast data, cloud cover forecast data, and historical output data, a temporal convolutional network is used to perform photovoltaic power probability prediction to obtain the photovoltaic power prediction data.

[0009] Optionally, based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data, a multi-source uncertainty joint scenario generation is performed to obtain a multi-dimensional joint scenario tree, which includes: Historical runoff data of the basin where the cascade hydropower station group is located, historical wind power output data of the wind farm, and historical photovoltaic power output data of the photovoltaic power station are collected. A first spatiotemporal correlation coefficient matrix between the historical runoff data and the historical wind power output data is calculated, a second spatiotemporal correlation coefficient matrix between the historical runoff data and the historical photovoltaic power output data is calculated, and a third spatiotemporal correlation coefficient matrix between the historical wind power output data and the historical photovoltaic power output data is calculated. Using the first spatiotemporal correlation coefficient matrix, the second spatiotemporal correlation coefficient matrix, and the third spatiotemporal correlation coefficient matrix as inputs, a Copula function is constructed to describe the joint distribution characteristics among runoff, wind power output, and photovoltaic power output. Based on the Copula function, several independent scenarios sampled from the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data are coupled and associated to generate the multi-dimensional joint scenario tree that includes the runoff scenario, the wind power output scenario, and the photovoltaic power output scenario.

[0010] Optionally, based on the multidimensional joint scene tree, a joint distributed fuzzy set is constructed using the Wasserstein distance metric, including: The multidimensional joint scene tree is treated as a number of discrete sample points, and an empirical probability distribution of the multidimensional joint scene tree is constructed. Historical meteorological forecast error data of the basin where the cascade hydropower station group is located were collected, and the distribution characteristics of forecast error under different forecast lead times were statistically analyzed. Based on the forecast error distribution characteristics, the dynamic ambiguity radius that matches the current forecast lead time is dynamically calculated, wherein the dynamic ambiguity radius is positively correlated with the degree of dispersion of the forecast error distribution characteristics; With the empirical probability distribution as the center and the dynamic ambiguity radius as the measurement range, a Wasserstein sphere is constructed in the probability distribution space, which is defined as the joint distributed ambiguity set.

[0011] Optionally, a refined coupled model of the cascade hydropower stations is loaded, along with grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data, and new energy consumption constraint data, configured as a multi-dimensional dynamic constraint set, including: Load the reservoir capacity curve data, downstream water level-flow relationship curve data, head change influence coefficient data and tailrace backwater influence coefficient data of the cascade hydropower station group, and configure them as the refined coupling model of the cascade hydropower. Load the power grid safety operation constraint data, which includes power grid frequency safety constraint data, voltage stability constraint data, and line transmission capacity constraint data; Load the reservoir operation constraint data, which includes upper limit constraint data for reservoir water level, lower limit constraint data for reservoir water level, upper limit constraint data for outflow, lower limit constraint data for outflow, water level fluctuation constraint data, and water level fluctuation rate constraint data. Load the ecological flow constraint data, which includes minimum ecological outflow constraint data and ecological flow process line constraint data; Load the new energy consumption constraint data, which includes the minimum utilization rate constraint data for new energy and the upper limit constraint data for wind and solar curtailment rates; The power grid safety operation constraint data, the reservoir operation constraint data, the ecological flow constraint data, and the new energy consumption constraint data are merged and configured into the multi-dimensional dynamic constraint set.

[0012] Optionally, construct and execute a multi-timescale nested sub-Blule rolling optimization model, prior to which: Collect historical net load data of the power grid in the area where the cascade hydropower station group is located. The historical net load data is the difference between the original load data of the power grid and the output data of new energy sources. Based on the historical net load data, calculate the net load ramp-up rate and net load fluctuation amplitude of the regional power grid; Based on the net load ramp rate and the net load fluctuation amplitude, the regional power grid's regulation capacity requirement for the cascade hydropower station group is quantified, configured as flexibility requirement constraint data, and added to the multivariate dynamic constraint set.

[0013] Optionally, based on the joint distributed fuzzy set and the multivariate dynamic constraint set, a multi-time-scale nested sub-Bruker rolling optimization model is constructed and executed to obtain automatic generation control unit load allocation instructions, including: Construct a multi-timescale nested sub-Blule rolling optimization model that includes a medium-to-long-term optimization layer, a short-term optimization layer, and a real-time optimization layer; The medium- and long-term optimization layer aims to maximize the expected energy storage value of the basin, solves the reservoir water level control strategy that satisfies the multivariate dynamic constraint set under all probability distributions covered by the joint fuzzy set, and outputs the medium- and long-term water level control trajectory of each cascade hydropower station to the short-term optimization layer. The short-term optimization layer aims to maximize the expected power generation revenue. Based on the medium- and long-term water level control trajectory, it solves the daily power output plan of the cascade hydropower station group that satisfies the multivariate dynamic constraint set under several scenarios included in the multidimensional joint scenario tree. It then outputs the hourly power output plan and unit start-up and shutdown combination scheme of each cascade hydropower station to the real-time optimization layer. The real-time optimization layer aims to minimize the real-time output tracking error. It employs a model predictive control method to continuously refresh the optimization problem within each preset time window based on the hourly output plan and the unit start-stop combination scheme, thereby solving for the automatic generation control unit load allocation command that satisfies the multivariate dynamic constraint set.

[0014] The method employs model predictive control to continuously refresh and optimize the problem within each preset time window based on the hourly output plan and the unit start-stop combination scheme, solving for automatic generation control unit load allocation instructions that satisfy the multivariate dynamic constraint set, including: The optimization objective is to minimize the sum of squared deviations between the real-time output of the cascade hydropower station group and the hourly output plan at each time within a preset time domain from the current time to the future. At the arrival of each preset control cycle, the automatic generation control unit load allocation instructions at each time within the current control cycle are obtained by rolling the solution based on the latest measured operating status data of the cascade hydropower station group.

[0015] Secondly, the present invention provides a hydropower cluster generation scheduling optimization system based on hydrological forecasting, comprising: The multi-source prediction data acquisition module is used to load multi-source prediction data of runoff from cascade hydropower station groups, wind power prediction data of wind farms, and photovoltaic power prediction data of photovoltaic power stations. The multi-dimensional joint scenario tree acquisition module is used to perform multi-source uncertainty joint scenario generation based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data to obtain a multi-dimensional joint scenario tree, wherein the multi-dimensional joint scenario tree includes several runoff scenarios, several wind power output scenarios, and several photovoltaic power output scenarios. The joint distribution fuzzy set acquisition module is used to construct a joint distribution fuzzy set based on the multidimensional joint scene tree using the Wasserstein distance metric, wherein the joint distribution fuzzy set is a Wasserstein sphere centered on the empirical distribution of the multidimensional joint scene tree and measured by the dynamic fuzziness radius. The multi-dimensional dynamic constraint set acquisition module is used to load the cascade hydropower fine coupling model of the cascade hydropower station group, and load power grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data and new energy consumption constraint data, and configure them as a multi-dimensional dynamic constraint set; The load allocation instruction acquisition module is used to construct and execute a multi-time-scale nested sub-Bruker rolling optimization model based on the joint distributed fuzzy set and the multivariate dynamic constraint set to obtain the load allocation instruction of the automatic generation control unit; The scheduling decision acquisition module is used to extract the load allocation instructions of the automatic generation control unit. By implementing this invention, it is possible to load multi-source runoff prediction data of cascade hydropower station groups, wind power prediction data of wind farms, and photovoltaic power prediction data of photovoltaic power stations; integrate multi-dimensional prediction data of water, wind, and solar power to achieve unified collection of multi-energy data and avoid the one-sidedness of single-energy data supporting scheduling; and acquire data by adopting multi-source runoff prediction and wind and solar power probability prediction methods to improve the comprehensiveness and accuracy of basic data, laying a data foundation for subsequent accurate analysis of energy output uncertainty.

[0016] By implementing this invention, it is possible to generate multi-source uncertainty joint scenarios based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data, thereby obtaining a multi-dimensional joint scenario tree. This multi-dimensional joint scenario tree includes several runoff scenarios, several wind power output scenarios, and several photovoltaic power output scenarios. By using the Copula function to characterize the joint distribution characteristics among water, wind, and solar power, the limitations of single-energy uncertainty analysis are overcome, accurately reflecting the coupling and correlation laws of multiple energy outputs. The generated multi-scenario tree can comprehensively cover all possible situations of water, wind, and solar power outputs, avoiding scheduling analysis biases caused by ignoring the correlation between energy sources, and providing comprehensive scenario support for subsequent robust optimization.

[0017] By implementing this invention, it is possible to construct a joint distribution fuzzy set based on the multidimensional joint scene tree using the Wasserstein distance metric. The joint distribution fuzzy set is a Wasserstein sphere centered on the empirical distribution of the multidimensional joint scene tree and measured by a dynamic fuzziness radius. Using the Wasserstein distance metric improves the accuracy of uncertainty distribution characterization, better reflecting actual probability distribution characteristics compared to traditional methods. The dynamic fuzziness radius is dynamically adjusted according to the error distribution during the forecast period, making the definition of uncertainty boundaries more closely match actual forecast conditions. This avoids overly conservative or overly aggressive optimization problems caused by fixed fuzziness, enhancing the adaptive characterization capability for uncertainty.

[0018] By implementing this invention, a refined coupling model of the cascade hydropower station group can be loaded, along with grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data, and renewable energy consumption constraint data, configured as a multi-dimensional dynamic constraint set. The refined coupling model of the cascade hydropower considers key hydraulic factors such as reservoir capacity, head, and tailrace backwater, breaking through the limitations of traditional simplified models and improving the accuracy of the hydropower dispatching model. The multi-dimensional constraint set comprehensively covers various requirements for grid safety, reservoir operation, ecological protection, and renewable energy consumption, taking into account the economy, safety, and ecology of dispatching, while also incorporating grid flexibility demand constraints to adapt to hydropower regulation requirements under renewable energy fluctuations.

[0019] By implementing this invention, a multi-time-scale nested sub-Brussels bar rolling optimization model can be constructed and executed based on the joint distributed fuzzy set and the multivariate dynamic constraint set to obtain automatic generation control unit load allocation instructions. The multi-time-scale nested design realizes the coordinated optimization of medium- and long-term reservoir water level control and short-term, real-time load allocation, solving the problem of spatiotemporal scale separation in traditional methods and reducing head benefit loss and water abandonment risk. The sub-Brussels bar optimization combined with the rolling solution method not only ensures the model's anti-interference ability against multi-source uncertainties, but also dynamically refreshes the optimization results based on real-time operating data, improving the real-time performance and robustness of scheduling decisions. The hierarchical optimization has clear objectives, taking into account both the long-term planning and short-term execution of hydropower scheduling.

[0020] By implementing this invention, it is possible to extract the load allocation instructions of the automatic power generation control unit and configure them as a dispatching decision scheme output; directly output a standardized dispatching decision scheme, reduce the conversion cost of model results to actual dispatching operations, and improve the engineering practicality of the dispatching optimization method; the instructions accurately correspond to the load allocation of the automatic power generation control unit, which can directly guide the actual operation of the hydropower unit, realize the implementation of the optimization results, and ensure the actual manifestation of the dispatching optimization effect.

[0021] In summary, by implementing this invention, the ability of hydropower dispatch to adjust and adapt to the intermittent and fluctuating output of new energy sources is improved, ensuring the safe operation of the power grid and the efficient consumption of new energy sources. At the same time, the head benefits of cascade hydropower, the safety of reservoir operation, and the requirements of watershed ecological protection are taken into account. This effectively reduces the loss of head benefits and the risk of water abandonment, and significantly improves the dispatch economy, safety, robustness, and engineering practicality of the clean energy combined power generation system. Attached Figure Description

[0022] Figure 1 A flowchart illustrating the hydropower cluster generation scheduling optimization method based on hydrological prediction provided by this invention; Figure 2 This is a schematic diagram of the structure of the hydropower cluster generation scheduling optimization system based on hydrological prediction provided by the present invention.

[0023] In the attached diagram, the components represented by each number are as follows: The module includes: 11 for acquiring multi-source prediction data, 12 for acquiring multi-dimensional joint scenario tree, 13 for acquiring joint distributed fuzzy set, 14 for acquiring multi-dimensional dynamic constraint set, 15 for acquiring load allocation instruction, and 16 for acquiring scheduling decision scheme. Detailed Implementation

[0024] Example 1, as Figure 1 As shown, this embodiment of the invention provides a method for optimizing the scheduling of hydropower cluster generation based on hydrological forecasting, including: S100: Load multi-source runoff prediction data of cascade hydropower station groups, wind power prediction data of wind farms, and photovoltaic power prediction data of photovoltaic power stations. S200: Based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data, perform multi-source uncertainty joint scenario generation to obtain a multi-dimensional joint scenario tree, wherein the multi-dimensional joint scenario tree includes several runoff scenarios, several wind power output scenarios, and several photovoltaic power output scenarios. S300: Based on the multidimensional joint scene tree, construct a joint distributed fuzzy set using the Wasserstein distance metric, wherein the joint distributed fuzzy set is a Wasserstein sphere centered on the empirical distribution of the multidimensional joint scene tree and measured by the dynamic fuzziness radius. S400: Load the refined coupling model of the cascade hydropower station group, and load the power grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data and new energy consumption constraint data, and configure them as a multi-dimensional dynamic constraint set; S500: Based on the joint distributed fuzzy set and the multivariate dynamic constraint set, construct and execute a multi-time-scale nested sub-Bruker rolling optimization model to obtain the automatic generation control unit load allocation command; S600: Extract the load allocation instructions of the automatic generation control unit and configure them as a dispatch decision scheme output.

[0025] In step S100 of this application embodiment, loading multi-source runoff prediction data of the cascade hydropower station group includes: Numerical weather forecast data for the basin where the cascade hydropower station group is located is obtained, including rainfall forecast data, temperature forecast data, and evaporation forecast data; Load the measured data of the hydrological stations in the watershed, which includes measured data of upstream inflow, measured data of inter-regional runoff, and measured data of downstream water level; Based on the numerical weather forecast data and the measured data from the hydrological stations, a deep neural network quantile regression model is used to predict runoff probability, obtaining short-term runoff prediction data, medium-term runoff prediction data and long-term runoff prediction data, which are then configured as the multi-source runoff prediction data.

[0026] In step S100 of this application embodiment, the purpose of the above steps is to obtain multi-source runoff prediction data of the cascade hydropower station group at all time scales, so as to provide accurate and comprehensive runoff data foundation for the subsequent generation of multi-source uncertainty joint scenarios, solve the problem that natural runoff is difficult to predict accurately in hydropower scheduling, and support the multi-time scale collaborative optimization of subsequent hydropower cluster power generation scheduling.

[0027] To achieve the above objectives, it is first necessary to obtain numerical weather forecast data for the basin where the cascade hydropower station group is located. The numerical weather forecast data includes rainfall forecast data, temperature forecast data, and evaporation forecast data. This step involves obtaining basic data on the influencing factors of runoff formation from a meteorological perspective, providing raw meteorological input for runoff forecasting. For example, it involves obtaining hourly rainfall in millimeters, daily average temperature in degrees Celsius, and daily evaporation in millimeters for a certain watershed over the next 7 days.

[0028] Next, the measured data from the hydrological stations in the basin need to be loaded. The measured data from the hydrological stations include measured data of upstream inflow, measured data of inter-regional runoff, and measured data of downstream water level. This step involves obtaining real data on the actual operation of the watershed's hydrology, providing a measured benchmark and data verification basis for runoff forecasting. For example, it involves loading the measured data of upstream inflow in cubic meters per second, measured data of inter-regional runoff in cubic meters per second, and measured data of downstream water level in meters for each hydrological station within the watershed.

[0029] Furthermore, based on the numerical weather forecast data and the measured data from the hydrological stations, a deep neural network quantile regression model is used to predict runoff probability, thereby obtaining short-term runoff prediction data, medium-term runoff prediction data, and long-term runoff prediction data, which are then configured as the multi-source runoff prediction data.

[0030] This step involves integrating meteorological and measured data from both dimensions, and using specialized models to predict runoff probability at different time scales, resulting in multi-source, full-cycle runoff predictions. For example, by integrating meteorological forecast data and hydrological measured data from the aforementioned watershed, a deep neural network quantile regression model is used to predict the short-term runoff cubic meters per second for the next day, the medium-term runoff cubic meters per second for the next 15 days, and the long-term runoff cubic meters per second for the next 3 months. These three types of data are then integrated into multi-source runoff prediction data for the cascade hydropower station group.

[0031] The deep neural network quantile regression model mentioned above can be adapted to the hydrological field by obtaining open-source quantile regression model code from platforms such as GitHub and Kaggle. The network parameters and loss function weights can be adjusted according to the characteristics of the runoff data from cascade hydropower stations, such as the input dimension and the number of quantiles, and then validated for use. Specific validation and adaptation methods for such mature models are existing technologies and will not be elaborated here.

[0032] In step S100 of this application embodiment, loading wind power prediction data for wind farms and photovoltaic power prediction data for photovoltaic power plants includes: Obtain numerical weather forecast data for the area where the wind farm is located, the numerical weather forecast data including wind speed forecast data and wind direction forecast data; Load the historical power output data of the wind farm, and based on the wind speed forecast data, the wind direction forecast data and the historical power output data, use a temporal convolutional network to perform wind power probability prediction to obtain the wind power prediction data; Numerical weather forecast data for the area where the photovoltaic power station is located is obtained, including solar radiation forecast data, temperature forecast data, and cloud cover forecast data; The historical output data of the photovoltaic power station is loaded, and based on the solar radiation forecast data, temperature forecast data, cloud cover forecast data, and historical output data, a temporal convolutional network is used to perform photovoltaic power probability prediction to obtain the photovoltaic power prediction data.

[0033] In step S100 of this application embodiment, the purpose of the above steps is to obtain wind power prediction data of wind farms and photovoltaic power prediction data of photovoltaic power plants, so as to provide accurate new energy output prediction data foundation for subsequent multi-source uncertainty joint scenario generation, solve the scheduling problem caused by the intermittency and volatility of wind power and photovoltaic output, support the adjustment and adaptation of hydropower clusters to new energy output fluctuations, and ensure the effectiveness of subsequent multi-time-scale nested sub-Bruker rolling optimization model construction.

[0034] To achieve the above objectives, it is first necessary to obtain numerical weather forecast data for the area where the wind farm is located. The numerical weather forecast data includes wind speed forecast data and wind direction forecast data. This step involves acquiring data on core environmental factors affecting wind power output from a meteorological perspective, providing raw meteorological input for wind power probability prediction. For example, it involves obtaining forecast data on wind speed per second at a height of 10 meters and wind direction angle for the next 48 hours in the area where a wind farm is located from a meteorological data platform.

[0035] Next, the historical power output data of the wind farm is loaded. Based on the wind speed forecast data, the wind direction forecast data, and the historical power output data, a temporal convolutional network is used to perform wind power probability prediction to obtain the wind power prediction data. This step involves integrating meteorological forecast data with historical wind power operation data, using professional models to make probabilistic predictions of wind power output, and generating accurate wind power output prediction results. For example, by loading the historical hourly megawatt output data of a wind farm over the past year, and combining it with the wind speed and direction forecast data for the region, a temporal convolutional network is used to predict the hourly megawatt wind power output of the wind farm for the next 48 hours.

[0036] The temporal convolutional network mentioned above does not require additional acquisition; it is an existing mature deep learning network model that can be built and implemented using a Python deep learning framework. First, a network structure including causal convolution and dilated convolution is constructed based on mainstream frameworks such as TensorFlow and PyTorch, with residual connection modules added to improve feature extraction capabilities. Then, considering the requirements of wind power and photovoltaic power prediction scenarios, the network is trained, validated, and optimized using historical meteorological data and historical power output data to determine hyperparameters such as convolutional kernel size, dilation coefficient, and number of neurons, adapting it to the power probability prediction task in this application, thus obtaining a usable temporal convolutional network. The aforementioned temporal convolutional network and the subsequent methods for obtaining it are all existing technologies and will not be elaborated upon here.

[0037] Furthermore, numerical weather forecast data for the area where the photovoltaic power station is located is obtained, including solar radiation forecast data, temperature forecast data, and cloud cover forecast data; This step involves acquiring data on key environmental factors affecting photovoltaic output from a meteorological perspective, providing raw meteorological input for photovoltaic power probability prediction. For example, it involves acquiring forecast data on irradiance per square meter, module surface temperature in degrees Celsius, and cloud cover percentage for the next 48 hours in the area where a photovoltaic power station is located.

[0038] Furthermore, the historical output data of the photovoltaic power station is loaded, and based on the solar radiation forecast data, the temperature forecast data, the cloud cover forecast data, and the historical output data, a temporal convolutional network is used to perform photovoltaic power probability prediction to obtain the photovoltaic power prediction data.

[0039] This step involves integrating meteorological forecast data with historical photovoltaic (PV) operation data, using specialized models to make probabilistic predictions of PV power, and generating accurate PV output prediction results. For example, by loading the historical hourly megawatt output data of a PV power station for the past year, and combining it with the region's solar radiation, temperature, and cloud cover forecast data, a temporal convolutional network is used to predict the hourly megawatt PV power of the PV power station for the next 48 hours.

[0040] Finally, in step S200 of this embodiment, based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data, multi-source uncertainty joint scenario generation is performed to obtain a multi-dimensional joint scenario tree, which includes the following: Historical runoff data of the basin where the cascade hydropower station group is located, historical wind power output data of the wind farm, and historical photovoltaic power output data of the photovoltaic power station are collected. A first spatiotemporal correlation coefficient matrix between the historical runoff data and the historical wind power output data is calculated, a second spatiotemporal correlation coefficient matrix between the historical runoff data and the historical photovoltaic power output data is calculated, and a third spatiotemporal correlation coefficient matrix between the historical wind power output data and the historical photovoltaic power output data is calculated. Using the first spatiotemporal correlation coefficient matrix, the second spatiotemporal correlation coefficient matrix, and the third spatiotemporal correlation coefficient matrix as inputs, a Copula function is constructed to describe the joint distribution characteristics among runoff, wind power output, and photovoltaic power output. Based on the Copula function, several independent scenarios sampled from the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data are coupled and associated to generate the multi-dimensional joint scenario tree that includes the runoff scenario, the wind power output scenario, and the photovoltaic power output scenario.

[0041] In step S200 of this application embodiment, the purpose of the above steps is to explore the spatiotemporal correlation characteristics among runoff, wind power output, and photovoltaic power output, and to construct a Copula function that can accurately describe the joint distribution law of the three. This provides theoretical and model support for the subsequent effective coupling and correlation of independent scenarios of the three types of prediction data, ensuring that the generated multi-dimensional joint scenario tree can truly reflect the multi-source uncertainty coupling characteristics of runoff and new energy power output. It also provides a realistic scenario data foundation for the subsequent construction of joint distribution fuzzy sets and the execution of multi-timescale nested sub-Blule rolling optimization models.

[0042] To achieve the above objectives, it is first necessary to collect historical runoff data of the basin where the cascade hydropower station group is located, historical wind power output data of the wind farm, and historical photovoltaic power output data of the photovoltaic power station. Then, a first spatiotemporal correlation coefficient matrix between the historical runoff data and the historical wind power output data, a second spatiotemporal correlation coefficient matrix between the historical runoff data and the historical photovoltaic power output data, and a third spatiotemporal correlation coefficient matrix between the historical wind power output data and the historical photovoltaic power output data are calculated. This step quantifies the spatiotemporal correlation between runoff, wind power output, and photovoltaic output using historical data, providing quantitative correlation parameters for the subsequent construction of a joint distribution function. For example, historical data on daily runoff cubic meters per second for the past 5 years, historical data on daily wind power output (megawatts) for the supporting wind farms for the past 5 years, and historical data on daily photovoltaic power output (megawatts) for the supporting photovoltaic power plants for the past 5 years are collected. Statistical calculations yield a first spatiotemporal correlation coefficient matrix with values ​​of [-0.2, 0.1; 0.05, 0.3], a second spatiotemporal correlation coefficient matrix with values ​​of [0.15, 0.08; -0.03, 0.22], and a third spatiotemporal correlation coefficient matrix with values ​​of [0.25, 0.12; 0.09, 0.3].

[0043] The spatiotemporal correlation coefficients are obtained by performing spatiotemporal correlation statistical calculations on historical data of runoff and wind power output, runoff and photovoltaic power output, and wind power and photovoltaic power output respectively, thereby obtaining a first spatiotemporal correlation coefficient matrix, a second spatiotemporal correlation coefficient matrix, and a third spatiotemporal correlation coefficient matrix. The values ​​in the matrices are the spatiotemporal correlation coefficients between the corresponding elements.

[0044] For example, the spatiotemporal correlation statistical calculation can be specifically performed by aligning the three types of historical data to the same time granularity, such as a daily scale, according to the time dimension; and then dividing the data into different regional units according to the spatial dimension, such as watershed zones, wind farm turbine locations, and photovoltaic power station arrays. Based on the aligned spatiotemporal sequence data, the Pearson correlation coefficient method is used to calculate the correlation coefficients between runoff and wind power output, runoff and photovoltaic power output, and wind power and photovoltaic power output in each spatiotemporal unit. The results of each unit are then integrated to form the corresponding spatiotemporal correlation coefficient matrix, thus completing the statistical calculation.

[0045] Next, using the first spatiotemporal correlation coefficient matrix, the second spatiotemporal correlation coefficient matrix, and the third spatiotemporal correlation coefficient matrix as inputs, a Copula function is constructed to describe the joint distribution characteristics among runoff, wind power output, and photovoltaic power output. This step is based on the quantitative results of the spatiotemporal correlation between the three factors, and constructs a function model that can characterize the overall joint distribution law of the three factors. This enables a mathematical expression of the joint uncertainty characteristics of runoff and new energy power output from multiple sources. For example, the first spatiotemporal correlation coefficient matrix, the second spatiotemporal correlation coefficient matrix, and the third spatiotemporal correlation coefficient matrix obtained above are input into the Copula function construction model to construct a Gumbel-Copula function applicable to the runoff and power output of the supporting wind and solar power stations in the basin. This Gumbel-Copula function can accurately describe the joint distribution characteristics among the three factors.

[0046] The Copula function construction model can be described as follows: taking the first spatiotemporal correlation coefficient matrix of runoff and wind power output, the second spatiotemporal correlation coefficient matrix of runoff and photovoltaic power output, and the third spatiotemporal correlation coefficient matrix of wind power and photovoltaic power output as core inputs, combining the marginal distribution characteristics of the three, selecting an appropriate Copula function type, determining the function parameters through parameter estimation and goodness-of-fit test, and completing the construction of a Copula function that can accurately describe the joint distribution characteristics of runoff, wind power output, and photovoltaic power output. This overall modeling process is the Copula function construction model.

[0047] The constructed Copula function can characterize the spatiotemporal correlation among the three factors, providing a mathematical expression for the joint uncertainty of runoff and renewable energy output from multiple sources. Its core function is to couple and correlate the independent scenarios sampled from the three types of prediction data, providing a model foundation for generating a realistic multidimensional joint scenario tree. The Gumbel-Copula function, a type of Copula function, excels at characterizing the upper-tail dependence of variables and can be adapted to variables such as runoff, wind power, and photovoltaic output, which are prone to extreme value linkages, as described in this application.

[0048] Furthermore, based on the Copula function, several independent scenarios sampled from the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data are coupled and associated to generate the multi-dimensional joint scenario tree that includes the runoff scenario, the wind power output scenario, and the photovoltaic power output scenario.

[0049] This step utilizes the constructed Copula function to fuse and correlate the independent sampling scenarios of the three types of prediction data, generating a multi-dimensional joint scenario that reflects the actual coupling relationship among the three, forming a hierarchical scenario tree structure. For example, based on the constructed Gumbel-Copula function, 50 independent scenarios of runoff cubic meters per second sampled from runoff multi-source prediction data, 50 independent scenarios of wind power output megawatts sampled from wind power prediction data, and 50 independent scenarios of photovoltaic power output megawatts sampled from photovoltaic power prediction data are coupled and correlated to generate a multi-dimensional joint scenario tree containing 100 coupled scenarios. Each scenario simultaneously contains the corresponding runoff scenario, wind power output scenario, and photovoltaic power output scenario.

[0050] The runoff scenario is generated by sampling multi-source predicted runoff data from a cascade hydropower station group, reflecting different natural runoff change states in the basin and including core information such as runoff values ​​for different forecast periods. The wind power output scenario is generated by sampling wind power predicted data, reflecting different output fluctuation states of wind farms and covering wind power values ​​at different times. The photovoltaic output scenario is generated by sampling photovoltaic power predicted data, presenting different output change states of photovoltaic power stations and including photovoltaic power values ​​at different times. All three scenarios are components of a multi-dimensional joint scenario tree.

[0051] In step S300 of this application embodiment, based on the multidimensional joint scene tree, a joint distributed fuzzy set is constructed using the Wasserstein distance metric, including: The multidimensional joint scene tree is treated as a number of discrete sample points, and an empirical probability distribution of the multidimensional joint scene tree is constructed. Historical meteorological forecast error data of the basin where the cascade hydropower station group is located were collected, and the distribution characteristics of forecast error under different forecast lead times were statistically analyzed. Based on the forecast error distribution characteristics, the dynamic ambiguity radius that matches the current forecast lead time is dynamically calculated, wherein the dynamic ambiguity radius is positively correlated with the degree of dispersion of the forecast error distribution characteristics; With the empirical probability distribution as the center and the dynamic ambiguity radius as the measurement range, a Wasserstein sphere is constructed in the probability distribution space, which is defined as the joint distributed ambiguity set.

[0052] In step S300 of this application embodiment, the purpose of the above steps is to construct a joint distributed fuzzy set that fits the actual forecast error characteristics based on a multi-dimensional joint scenario tree and using Wasserstein distance as a metric, to quantify the multi-source joint uncertainty of runoff, wind power output, and photovoltaic output, and to provide a scientific uncertainty constraint basis for the subsequent multi-time-scale nested sub-Bruker rolling optimization model, thereby improving the robustness of scheduling optimization.

[0053] To achieve the above objectives, it is first necessary to treat the several scenarios contained in the multidimensional joint scene tree as discrete sample points and construct the empirical probability distribution of the multidimensional joint scene tree. This step transforms the scenario-based multi-source data into a probability distribution form, providing a core distribution benchmark for the subsequent construction of the Wasserstein sphere. For example, each scenario in a multidimensional joint scenario tree containing 200 coupled scenarios is used as a discrete sample point. The frequency of occurrence of each combination scenario of runoff, wind power output, and photovoltaic output is counted, and an empirical probability distribution of the multidimensional joint scenario tree corresponding to the probability value of each scenario is constructed. Among them, the empirical probability of a certain runoff scenario combined with a wind power and photovoltaic output scenario is 0.008.

[0054] Next, historical meteorological forecast error data of the basin where the cascade hydropower station group is located were collected, and the distribution characteristics of forecast errors under different forecast lead times were statistically analyzed. This step involves exploring the distribution patterns of weather forecast errors at different time scales to provide data support and characteristic basis for calculating the dynamic ambiguity radius. For example, historical weather forecast error data such as rainfall, temperature, and wind speed for the past three years were collected for a certain cascade hydropower station group basin. Statistical analysis revealed that short-term forecast errors are normally distributed with low dispersion, medium-term forecast errors are skewed with moderate dispersion, and long-term forecast errors are uniformly distributed with high dispersion.

[0055] The length of the forecast lead time directly affects the dispersion of forecast errors. The longer the forecast lead time, the more difficult it is to predict meteorological elements related to runoff and wind and solar activity, the wider the deviation of the prediction results, the stronger the dispersion of data distribution, and the greater the dispersion of forecast errors. Conversely, the shorter the forecast lead time, the higher the accuracy of meteorological element prediction, the narrower the deviation, the more concentrated the data distribution, and the smaller the dispersion of forecast errors.

[0056] Then, based on the forecast error distribution characteristics, the dynamic ambiguity radius that matches the current forecast lead time is dynamically calculated, wherein the dynamic ambiguity radius is positively correlated with the degree of dispersion of the forecast error distribution characteristics; This step involves determining the metric range parameters of the Wasserstein sphere based on the forecast error fluctuations over different forecast periods, so that the construction of the fuzzy set can adapt to the current forecast error level.

[0057] The dynamic ambiguity radius is strictly positively correlated with the dispersion of the forecast error. The greater the dispersion of the forecast error, the larger the calculated dynamic ambiguity radius; the smaller the dispersion of the forecast error, the smaller the calculated dynamic ambiguity radius.

[0058] For example, based on the aforementioned statistical distribution characteristics of forecast errors, for the current short-term (1-day) forecast period, due to the small dispersion of forecast errors, the calculated dynamic ambiguity radius is 0.05. For the current long-term (30-day) forecast period, due to the large dispersion of forecast errors, the calculated dynamic ambiguity radius is 0.3. Specific methods to make the dynamic ambiguity radius positively correlated with the dispersion of the aforementioned forecast error distribution characteristics can be implemented mathematically, and the methods are not unique; they will not be elaborated here.

[0059] Furthermore, with the empirical probability distribution as the center and the dynamic fuzziness radius as the measurement range, a Wasserstein sphere is constructed in the probability distribution space, which is defined as the joint distributed fuzzy set.

[0060] This step combines the baseline distribution and the dynamic measurement range to complete the construction of a fuzzy set for multi-source uncertainty, thereby effectively defining the fluctuation range of the actual probability distribution. For example, using the constructed multidimensional joint scenario tree empirical probability distribution as the center and the dynamic fuzziness radius of 0.05 corresponding to the current short-term forecast as the measurement range, a corresponding Wasserstein sphere is constructed in the probability distribution space. This Wasserstein sphere is directly defined as a joint distribution fuzzy set, which can cover the actual probability distribution fluctuation range of runoff and wind and solar power output under the short-term forecast.

[0061] In step S400 of this application embodiment, the refined coupling model of the cascade hydropower station group is loaded, and grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data, and new energy consumption constraint data are loaded and configured as a multi-dimensional dynamic constraint set, including: Load the reservoir capacity curve data, downstream water level-flow relationship curve data, head change influence coefficient data and tailrace backwater influence coefficient data of the cascade hydropower station group, and configure them as the refined coupling model of the cascade hydropower. Load the power grid safety operation constraint data, which includes power grid frequency safety constraint data, voltage stability constraint data, and line transmission capacity constraint data; Load the reservoir operation constraint data, which includes upper limit constraint data for reservoir water level, lower limit constraint data for reservoir water level, upper limit constraint data for outflow, lower limit constraint data for outflow, water level fluctuation constraint data, and water level fluctuation rate constraint data. Load the ecological flow constraint data, which includes minimum ecological outflow constraint data and ecological flow process line constraint data; Load the new energy consumption constraint data, which includes the minimum utilization rate constraint data for new energy and the upper limit constraint data for wind and solar curtailment rates; The power grid safety operation constraint data, the reservoir operation constraint data, the ecological flow constraint data, and the new energy consumption constraint data are merged and configured into the multi-dimensional dynamic constraint set.

[0062] In step S400 of this application embodiment, the purpose of the above steps is to construct a refined coupled model of cascade hydropower and integrate multi-dimensional operational constraint data to form a multi-dimensional dynamic constraint set. This provides a model foundation that fits the actual operating characteristics of the cascade hydropower cluster for the construction and solution of the subsequent multi-time-scale nested sub-Bruker rolling optimization model, as well as constraint boundaries covering the entire dimensions of power grid, reservoir, ecology, and new energy consumption. This ensures that the scheduling optimization results meet the actual engineering operation requirements and take into account the technical feasibility of hydropower scheduling, power grid security, ecological environmental protection, and maximization of new energy consumption.

[0063] To achieve the above objectives, it is first necessary to load the reservoir capacity curve data, downstream water level-flow relationship curve data, head change influence coefficient data on unit output data, and tailrace backwater influence coefficient data of the cascade hydropower station group, and configure them as the refined coupling model of the cascade hydropower. This step involves extracting core characteristic data on the hydraulics and unit operation of a cascade hydropower station group, and building a refined coupling model of the cascade hydropower that accurately reflects the coupling relationship between various links in the cascade hydropower. This provides core model support for scheduling optimization. For example, loading curve data on the relationship between reservoir capacity and water level of a certain cascade hydropower station group, curve data on the relationship between downstream water level and downstream flow, coefficient data on the change in unit output for every 1 meter change in head (0.02 MW / m), and coefficient data on the change in unit output for every 0.5 meter increase in tailrace support height (-0.01 MW / 0.5m), and integrating these data into a refined coupling model of the cascade hydropower station group.

[0064] For example, the reservoir capacity curve data, the downstream water level-flow relationship curve data, the head change-affected unit output coefficient data, and the tailrace backwater influence coefficient data are standardized and verified. After removing outliers, the four types of data are integrated and matched according to the actual operating logic of hydraulic coupling and unit output correlation of cascade hydropower. Through mathematical formulas, parameter correlation, and logical modeling, a model that can accurately reflect the coupling characteristics of each link of cascade hydropower is constructed, which is the refined coupling model of cascade hydropower.

[0065] Load the power grid safety operation constraint data, which includes power grid frequency safety constraint data, voltage stability constraint data, and line transmission capacity constraint data; This step involves obtaining the core operational constraints on the power grid side to define the boundaries for safe operation of the power grid for dispatch optimization. For example, the frequency safety constraints for a certain region's power grid are 50±0.2Hz, the voltage stability constraints are 35kV±5%, and the transmission capacity constraints for the regional transmission lines are 800MW.

[0066] Load the reservoir operation constraint data, which includes upper limit constraint data for reservoir water level, lower limit constraint data for reservoir water level, upper limit constraint data for outflow, lower limit constraint data for outflow, water level fluctuation constraint data, and water level fluctuation rate constraint data. This step involves obtaining the hard technical constraints for reservoir operation to ensure the safety and stability of reservoir scheduling. For example, the upper limit constraint data for a certain cascade reservoir is 280 meters, the lower limit constraint data is 250 meters, the upper limit constraint data for outflow is 1500 cubic meters per second, the lower limit constraint data for outflow is 200 cubic meters per second, the daily water level fluctuation constraint data is 5 meters, and the water level fluctuation rate constraint data is 0.5 meters per hour.

[0067] Load the ecological flow constraint data, which includes minimum ecological outflow constraint data and ecological flow process line constraint data; This step involves obtaining the flow constraints for watershed ecological protection to ensure that hydropower scheduling takes into account ecological and environmental needs. For example, the minimum ecological discharge flow constraint data for the downstream river channel of a certain cascade hydropower station is 150 cubic meters per second, and the ecological flow process line constraint data from the wet season to the dry season is a gradual curve of 150-300 cubic meters per second.

[0068] Load the new energy consumption constraint data, which includes the minimum utilization rate constraint data for new energy and the upper limit constraint data for wind and solar curtailment rates; This step involves obtaining quantitative constraints on the consumption of new energy sources to promote the efficient consumption of clean energy through hydropower dispatch. For example, the minimum utilization rate constraint for new energy sources in a certain region is 95%, and the upper limit constraint for wind and solar curtailment rate is 5%.

[0069] The power grid safety operation constraint data, the reservoir operation constraint data, the ecological flow constraint data, and the new energy consumption constraint data are merged and configured into the multi-dimensional dynamic constraint set.

[0070] This step involves integrating constraint data from all dimensions to form a unified constraint set, providing an integrated constraint basis for solving the subsequent sub-Bruker rolling optimization model. For example, the four types of constraint data loaded above—grid safety operation, reservoir operation, ecological flow, and new energy consumption—are integrated, and after eliminating conflicting terms between data, they are configured into a multi-dimensional dynamic constraint set applicable to the power generation scheduling optimization of this cascade hydropower cluster.

[0071] In step S500 of this embodiment, the construction and execution of a multi-timescale nested sub-Blule rolling optimization model includes the following prior steps: Collect historical net load data of the power grid in the area where the cascade hydropower station group is located. The historical net load data is the difference between the original load data of the power grid and the output data of new energy sources. Based on the historical net load data, calculate the net load ramp-up rate and net load fluctuation amplitude of the regional power grid; Based on the net load ramp rate and the net load fluctuation amplitude, the regional power grid's regulation capacity requirement for the cascade hydropower station group is quantified, configured as flexibility requirement constraint data, and added to the multivariate dynamic constraint set.

[0072] In step S500 of this application embodiment, the purpose of the above steps is to quantify the regional power grid's demand for regulation capacity of the cascade hydropower station group and transform it into flexibility demand constraint data, which is then added to the multivariate dynamic constraint set. This allows the subsequently constructed multi-time-scale nested sub-Bruker rolling optimization model to fit the actual regulation needs of the power grid, ensuring that the scheduling optimization results of the cascade hydropower station group can match the ramp-up and fluctuation characteristics of the power grid's net load, improving the adaptability of hydropower to the power grid, and ensuring the safe and stable operation of the power grid.

[0073] To achieve the above objectives, it is first necessary to collect historical net load data of the power grid in the area where the cascade hydropower station group is located. The historical net load data is the difference between the original load data of the power grid and the output data of new energy sources. This step involves acquiring core historical data from the power grid side that reflects actual load regulation needs, providing a data foundation for subsequent calculations of regulation capacity-related indicators. For example, collecting hourly historical data of raw power grid load, wind power and photovoltaic output from the power grid in the region where a cascade hydropower station group is located for the past year, and calculating the difference to obtain the hourly historical data of net load of the power grid in the region for the past year.

[0074] Then, based on the historical net load data, the net load ramp-up rate and net load fluctuation amplitude of the regional power grid are calculated. This step involves extracting core dynamic characteristic indicators of the power grid load from historical net load data, quantifying the rate of change and the degree of fluctuation of the power grid load, and providing a quantitative basis for subsequent calculation of regulation capacity requirements. For example, based on the hourly net load historical data of the power grid in the above-mentioned region, the net load ramp rate of the power grid is calculated to be 50 MW / hour and the net load fluctuation range is 200 MW.

[0075] The method for calculating the net load ramp-up rate is as follows: take net load data for consecutive time periods, calculate the ratio of the net load difference between adjacent time periods to the time interval, obtain the ramp-up rate for a single time period, and then statistically analyze the maximum absolute value within the study period as the net load ramp-up rate.

[0076] The method for calculating the net load fluctuation range is as follows: first, calculate the average net load within the study period, then calculate the absolute value of the deviation between the net load and the average value for each period, and take the maximum value of the deviation as the net load fluctuation range.

[0077] Furthermore, based on the net load ramp rate and the net load fluctuation amplitude, the regional power grid's regulation capacity requirement for the cascade hydropower station group is quantified, configured as flexibility requirement constraint data, and added to the multivariate dynamic constraint set.

[0078] This step transforms the dynamic characteristics of the power grid load into rigid regulation constraints for the cascade hydropower station group, enriching the content of the multivariate dynamic constraint set and making the constraints of the subsequent optimization model more realistic. For example, based on the calculated net load ramp rate of 50 MW / h and net load fluctuation range of 200 MW, the flexibility requirement constraint data of the power grid in this region requiring the cascade hydropower station group to have a minimum regulation rate of 50 MW / h and a minimum regulation capacity of 200 MW is quantified, and this data is added to the configured multivariate dynamic constraint set.

[0079] In step S500 of this application embodiment, based on the joint distributed fuzzy set and the multivariate dynamic constraint set, a multi-time-scale nested sub-Bruker rolling optimization model is constructed and executed to obtain automatic generation control unit load allocation instructions, including: Construct a multi-timescale nested sub-Blule rolling optimization model that includes a medium-to-long-term optimization layer, a short-term optimization layer, and a real-time optimization layer; The medium- and long-term optimization layer aims to maximize the expected energy storage value of the basin, solves the reservoir water level control strategy that satisfies the multivariate dynamic constraint set under all probability distributions covered by the joint fuzzy set, and outputs the medium- and long-term water level control trajectory of each cascade hydropower station to the short-term optimization layer. The short-term optimization layer aims to maximize the expected power generation revenue. Based on the medium- and long-term water level control trajectory, it solves the daily power output plan of the cascade hydropower station group that satisfies the multivariate dynamic constraint set under several scenarios included in the multidimensional joint scenario tree. It then outputs the hourly power output plan and unit start-up and shutdown combination scheme of each cascade hydropower station to the real-time optimization layer. The real-time optimization layer aims to minimize the real-time output tracking error. It employs a model predictive control method to continuously refresh the optimization problem within each preset time window based on the hourly output plan and the unit start-stop combination scheme, thereby solving for the automatic generation control unit load allocation command that satisfies the multivariate dynamic constraint set.

[0080] In step S500 of this application embodiment, the purpose of the above steps is to construct and execute a multi-timescale nested sub-Bruker rolling optimization model. By combining the multi-source uncertainty representation of the joint distributed fuzzy set and the full-dimensional operation constraints of the multi-dimensional dynamic constraint set, the scheduling optimization solution of the cascade hydropower station group is realized in a hierarchical manner. Finally, the load allocation instructions of the automatic generation control unit adapted to the real-time operation requirements of the power grid are obtained, so that the hydropower cluster generation scheduling can take into account both medium- and long-term energy storage and power generation benefits, accurately track real-time load demand, and adapt to the multi-source uncertainty of runoff and new energy output, thus ensuring the robustness and practicality of scheduling optimization.

[0081] To achieve the above objectives, it is first necessary to construct a multi-timescale nested sub-Bluer rolling optimization model that includes a medium-to-long-term optimization layer, a short-term optimization layer, and a real-time optimization layer. This step involves building a hierarchical, nested optimization model framework that connects and supports scheduling objectives and solution strategies at different time scales, adapting to the full-cycle needs of hydropower cluster scheduling. For example, for the scheduling needs of a certain cascade hydropower cluster, a multi-time-scale nested sub-Blu-shaped rolling optimization model is constructed, with a medium-to-long-term optimization layer at the monthly time scale, a short-term optimization layer at the daily time scale, and a real-time optimization layer at the 15-minute time scale. Each layer is nested with the others and data is transmitted bidirectionally.

[0082] The medium- and long-term optimization layer aims to maximize the expected energy storage value of the basin, solves the reservoir water level control strategy that satisfies the multivariate dynamic constraint set under all probability distributions covered by the joint fuzzy set, and outputs the medium- and long-term water level control trajectory of each cascade hydropower station to the short-term optimization layer. This step involves long-term optimization from the perspective of overall watershed energy storage, determining the core control strategy for reservoir water levels, and defining water level boundaries for short-term scheduling. For example, the medium-to-long-term optimization layer of a certain cascade hydropower cluster aims to maximize the expected quarterly watershed energy storage value. Under all probability distributions covered by the joint distributed fuzzy set, it solves the reservoir water level control strategy that satisfies the multivariate dynamic constraint set, and outputs the results of the quarterly water level control trajectory of reservoir A in the cascade hydropower cluster as a gradual change from 260 meters to 275 meters and the quarterly water level control trajectory of reservoir B as a gradual change from 220 meters to 230 meters to the short-term optimization layer.

[0083] Maximizing the expected energy storage value means maximizing the mathematical expectation of the overall energy storage of the cascade reservoirs by optimizing the reservoir water level control strategy under all probability distributions covered by the joint distributed fuzzy set and in combination with multivariate dynamic constraints.

[0084] The short-term optimization layer aims to maximize the expected power generation revenue. Based on the medium- and long-term water level control trajectory, it solves the daily power output plan of the cascade hydropower station group that satisfies the multivariate dynamic constraint set under several scenarios included in the multidimensional joint scenario tree. It then outputs the hourly power output plan and unit start-up and shutdown combination scheme of each cascade hydropower station to the real-time optimization layer. This step involves optimizing short-cycle power generation revenue under medium- to long-term water level constraints, formulating specific output plans and unit operation schemes, and providing execution basis for real-time dispatch. For example, the short-term optimization layer of a certain cascade hydropower cluster aims to maximize the expected daily power generation revenue. Based on the medium- to long-term water level control trajectory, it solves the daily power generation plan that satisfies the multivariate dynamic constraint set under a multidimensional joint scenario tree containing 100 coupled scenarios. The hourly output plan of reservoir A is 800MW during the morning peak, 500MW during the flat period, and 900MW during the evening peak, as well as the start-up and shutdown combination scheme of 3 operating and 1 standby for 4 units. The hourly output plan and unit start-up and shutdown combination scheme of reservoir B are sent to the real-time optimization layer.

[0085] Maximizing the expected power generation revenue refers to maximizing the expected hydropower potential energy stored in the cascade reservoir group as a whole by optimizing the reservoir water level control strategy under all probability distributions covered by the joint distributed fuzzy set and in combination with multivariate dynamic constraints.

[0086] The real-time optimization layer aims to minimize the real-time output tracking error. It employs a model predictive control method to continuously refresh the optimization problem within each preset time window based on the hourly output plan and the unit start-stop combination scheme, thereby solving for the automatic generation control unit load allocation command that satisfies the multivariate dynamic constraint set.

[0087] The model mentioned here, "using model predictive control method," refers to the real-time optimization layer model of the aforementioned multi-timescale nested sub-Bruker rolling optimization model. Minimizing the real-time output tracking error means minimizing the deviation between the actual real-time output of the cascade hydropower station group and the hourly output plan issued by the short-term optimization layer.

[0088] In step S500 of this embodiment, a model predictive control method is used to solve the automatic generation control unit load allocation command that satisfies the multivariate dynamic constraint set by continuously refreshing the optimization problem based on the hourly output plan and the unit start-up and shutdown combination scheme in each preset time window. This includes: The optimization objective is to minimize the sum of squared deviations between the real-time output of the cascade hydropower station group and the hourly output plan at each time within a preset time domain from the current time to the future. At the arrival of each preset control cycle, the automatic generation control unit load allocation instructions at each time within the current control cycle are obtained by rolling the solution based on the latest measured operating status data of the cascade hydropower station group.

[0089] In step S500 of this application embodiment, the purpose of the above steps is to accurately solve the load allocation instructions of the automatic generation control units that satisfy the multivariate dynamic constraint set through the specific implementation of the model predictive control method, so that the real-time output of the cascade hydropower station group can closely match the hourly output plan, minimize the output tracking deviation, and at the same time, combine the latest hydropower station operation status data for rolling optimization, so that the load allocation instructions can be adapted to the real-time operation status of the hydropower cluster, ensuring the timeliness, accuracy and executability of the dispatch instructions, and realizing the real-time and accurate adjustment of the grid load by the hydropower cluster.

[0090] To achieve the above objectives, the first optimization goal is to minimize the sum of squared deviations between the real-time output of the cascade hydropower station group and the hourly output plan at each time point within the preset future time domain. This step clarifies the core optimization direction of the model predictive control method in solving the load allocation command. By quantifying the calculation standard of output tracking error, the solution can accurately match the hourly output plan formulated by the short-term optimization layer. For example, if the preset time domain is set to the next hour, the optimization goal is to minimize the sum of squared deviations between the real-time output of the cascade hydropower station group every 15 minutes and the corresponding hourly output plan value. If the output of a certain 15-minute period in the hourly output plan is 500MW, then the sum of squared deviations between the real-time output of that period and 500MW is included in the overall calculation and the minimum value is pursued.

[0091] Next, at the arrival of each preset control cycle, the latest measured operating status data of the cascade hydropower station group needs to be used to solve the problem dynamically to obtain the automatic generation control unit load allocation instructions for each moment within the current control cycle. This step relies on the preset control cycle to achieve dynamic rolling solution of the optimization problem, and combines the real-time operating status data of the hydropower station to correct the scheduling instructions, ensuring that the instructions meet the actual operating conditions and can adapt to the grid demand in real time. For example, if a 15-minute preset control cycle is set, at the arrival of each 15-minute cycle, the optimization problem is solved dynamically based on the latest measured operating status data of the cascade hydropower station group, such as the reservoir water level of 265 meters, the inflow of 300 cubic meters per second, and the current unit output of 480MW. Finally, the automatic generation control unit load allocation instructions for each 5 minutes within the 15-minute control cycle are obtained, such as 490MW for the unit load allocation in minutes 1-5, 495MW for minutes 6-10, and 500MW for minutes 11-15.

[0092] In step S600 of this application embodiment, it is necessary to extract the load allocation instruction of the automatic power generation control unit and configure it as a dispatch decision scheme output.

[0093] In step S600 of this application embodiment, the purpose of the above step is to extract and standardize the obtained automatic power generation control unit load allocation instructions, form a hydropower cluster power generation dispatch decision scheme that can be directly implemented and output it as the execution basis for the actual power generation operation of each unit in the cascade hydropower station group, realize the transformation from solving the sub-Bruker rolling optimization model to actual dispatch application, and ensure the actual operability of the hydropower cluster power generation dispatch optimization results.

[0094] To achieve the above objectives, the first step is to extract the load allocation instructions of the automatic generation control units. This step involves filtering the load allocation-related instruction data for each automatic generation control unit in a cascade hydropower station group from the solution results of the multi-time-scale nested sub-Bruker rolling optimization model, thus completing the extraction and aggregation of effective data. For example, the load allocation instructions for all automatic generation control units in a certain cascade hydropower station group, such as Unit 1 (200MW), Unit 2 (250MW) of Reservoir 1, Unit 1 (180MW), and Unit 2 (220MW) of Reservoir 2, can be extracted from the solution results of the sub-Bruker rolling optimization model.

[0095] Example 2, as Figure 2 As shown, based on the same inventive concept as the hydropower cluster generation scheduling optimization method based on hydrological prediction provided in Embodiment 1, this embodiment of the invention also provides a hydropower cluster generation scheduling optimization system based on hydrological prediction, including: The multi-source prediction data acquisition module 11 is used to load multi-source prediction data of runoff from the cascade hydropower station group, prediction data of wind power from the wind farm, and prediction data of photovoltaic power from the photovoltaic power station. The multi-dimensional joint scenario tree acquisition module 12 is used to perform multi-source uncertainty joint scenario generation based on the runoff multi-source prediction data, the wind power prediction data and the photovoltaic power prediction data, and obtain a multi-dimensional joint scenario tree, wherein the multi-dimensional joint scenario tree includes several runoff scenarios, several wind power output scenarios and several photovoltaic power output scenarios. The joint distribution fuzzy set acquisition module 13 is used to construct a joint distribution fuzzy set based on the multidimensional joint scene tree using the Wasserstein distance metric, wherein the joint distribution fuzzy set is a Wasserstein sphere centered on the empirical distribution of the multidimensional joint scene tree and measured by the dynamic fuzziness radius. The multi-dimensional dynamic constraint set acquisition module 14 is used to load the cascade hydropower fine coupling model of the cascade hydropower station group, and load power grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data and new energy consumption constraint data, and configure them as a multi-dimensional dynamic constraint set; The load allocation instruction acquisition module 15 is used to construct and execute a multi-time-scale nested sub-Bruker rolling optimization model based on the joint distributed fuzzy set and the multivariate dynamic constraint set to obtain the load allocation instruction of the automatic generation control unit; The scheduling decision scheme acquisition module 16 is used to extract the load allocation instructions of the automatic generation control unit and configure them as the scheduling decision scheme output.

[0096] Furthermore, the multi-source prediction data acquisition module 11 includes the following execution steps: Numerical weather forecast data for the basin where the cascade hydropower station group is located is obtained, including rainfall forecast data, temperature forecast data, and evaporation forecast data; Load the measured data of the hydrological stations in the watershed, which includes measured data of upstream inflow, measured data of inter-regional runoff, and measured data of downstream water level; Based on the numerical weather forecast data and the measured data from the hydrological stations, a deep neural network quantile regression model is used to predict runoff probability, obtaining short-term runoff prediction data, medium-term runoff prediction data and long-term runoff prediction data, which are then configured as the multi-source runoff prediction data.

[0097] Obtain numerical weather forecast data for the area where the wind farm is located, the numerical weather forecast data including wind speed forecast data and wind direction forecast data; Load the historical power output data of the wind farm, and based on the wind speed forecast data, the wind direction forecast data and the historical power output data, use a temporal convolutional network to perform wind power probability prediction to obtain the wind power prediction data; Numerical weather forecast data for the area where the photovoltaic power station is located is obtained, including solar radiation forecast data, temperature forecast data, and cloud cover forecast data; The historical output data of the photovoltaic power station is loaded, and based on the solar radiation forecast data, temperature forecast data, cloud cover forecast data, and historical output data, a temporal convolutional network is used to perform photovoltaic power probability prediction to obtain the photovoltaic power prediction data.

[0098] Furthermore, the multi-dimensional joint scene tree acquisition module 12 includes the following execution steps: Historical runoff data of the basin where the cascade hydropower station group is located, historical wind power output data of the wind farm, and historical photovoltaic power output data of the photovoltaic power station are collected. A first spatiotemporal correlation coefficient matrix between the historical runoff data and the historical wind power output data is calculated, a second spatiotemporal correlation coefficient matrix between the historical runoff data and the historical photovoltaic power output data is calculated, and a third spatiotemporal correlation coefficient matrix between the historical wind power output data and the historical photovoltaic power output data is calculated. Using the first spatiotemporal correlation coefficient matrix, the second spatiotemporal correlation coefficient matrix, and the third spatiotemporal correlation coefficient matrix as inputs, a Copula function is constructed to describe the joint distribution characteristics among runoff, wind power output, and photovoltaic power output. Based on the Copula function, several independent scenarios sampled from the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data are coupled and associated to generate the multi-dimensional joint scenario tree that includes the runoff scenario, the wind power output scenario, and the photovoltaic power output scenario.

[0099] Furthermore, the joint distributed fuzzy set acquisition module 13 includes the following execution steps: The multidimensional joint scene tree is treated as a number of discrete sample points, and an empirical probability distribution of the multidimensional joint scene tree is constructed. Historical meteorological forecast error data of the basin where the cascade hydropower station group is located were collected, and the distribution characteristics of forecast error under different forecast lead times were statistically analyzed. Based on the forecast error distribution characteristics, the dynamic ambiguity radius that matches the current forecast lead time is dynamically calculated, wherein the dynamic ambiguity radius is positively correlated with the degree of dispersion of the forecast error distribution characteristics; With the empirical probability distribution as the center and the dynamic ambiguity radius as the measurement range, a Wasserstein sphere is constructed in the probability distribution space, which is defined as the joint distributed ambiguity set.

[0100] Furthermore, the multivariate dynamic constraint set acquisition module 14 includes the following execution steps: Load the reservoir capacity curve data, downstream water level-flow relationship curve data, head change influence coefficient data and tailrace backwater influence coefficient data of the cascade hydropower station group, and configure them as the refined coupling model of the cascade hydropower. Load the power grid safety operation constraint data, which includes power grid frequency safety constraint data, voltage stability constraint data, and line transmission capacity constraint data; Load the reservoir operation constraint data, which includes upper limit constraint data for reservoir water level, lower limit constraint data for reservoir water level, upper limit constraint data for outflow, lower limit constraint data for outflow, water level fluctuation constraint data, and water level fluctuation rate constraint data. Load the ecological flow constraint data, which includes minimum ecological outflow constraint data and ecological flow process line constraint data; Load the new energy consumption constraint data, which includes the minimum utilization rate constraint data for new energy and the upper limit constraint data for wind and solar curtailment rates; The power grid safety operation constraint data, the reservoir operation constraint data, the ecological flow constraint data, and the new energy consumption constraint data are merged and configured into the multi-dimensional dynamic constraint set.

[0101] Furthermore, the load allocation instruction acquisition module 15 includes the following execution steps: Collect historical net load data of the power grid in the area where the cascade hydropower station group is located. The historical net load data is the difference between the original load data of the power grid and the output data of new energy sources. Based on the historical net load data, calculate the net load ramp-up rate and net load fluctuation amplitude of the regional power grid; Based on the net load ramp rate and the net load fluctuation amplitude, the regional power grid's regulation capacity requirement for the cascade hydropower station group is quantified, configured as flexibility requirement constraint data, and added to the multivariate dynamic constraint set.

[0102] Construct a multi-timescale nested sub-Blule rolling optimization model that includes a medium-to-long-term optimization layer, a short-term optimization layer, and a real-time optimization layer; The medium- and long-term optimization layer aims to maximize the expected energy storage value of the basin, solves the reservoir water level control strategy that satisfies the multivariate dynamic constraint set under all probability distributions covered by the joint fuzzy set, and outputs the medium- and long-term water level control trajectory of each cascade hydropower station to the short-term optimization layer. The short-term optimization layer aims to maximize the expected power generation revenue. Based on the medium- and long-term water level control trajectory, it solves the daily power output plan of the cascade hydropower station group that satisfies the multivariate dynamic constraint set under several scenarios included in the multidimensional joint scenario tree. It then outputs the hourly power output plan and unit start-up and shutdown combination scheme of each cascade hydropower station to the real-time optimization layer. The real-time optimization layer aims to minimize the real-time output tracking error. It employs a model predictive control method to continuously refresh the optimization problem within each preset time window based on the hourly output plan and the unit start-stop combination scheme, thereby solving for the automatic generation control unit load allocation command that satisfies the multivariate dynamic constraint set.

[0103] The method employs model predictive control to continuously refresh and optimize the problem within each preset time window based on the hourly output plan and the unit start-stop combination scheme, solving for automatic generation control unit load allocation instructions that satisfy the multivariate dynamic constraint set, including: The optimization objective is to minimize the sum of squared deviations between the real-time output of the cascade hydropower station group and the hourly output plan at each time within a preset time domain from the current time to the future. At the arrival of each preset control cycle, the automatic generation control unit load allocation instructions at each time within the current control cycle are obtained by rolling the solution based on the latest measured operating status data of the cascade hydropower station group.

Claims

1. A method for optimizing the scheduling of hydropower cluster generation based on hydrological forecasting, characterized in that, include: Load multi-source runoff prediction data of cascade hydropower station groups, wind power prediction data of wind farms, and photovoltaic power prediction data of photovoltaic power stations; Based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data, a multi-source uncertainty joint scenario generation is performed to obtain a multi-dimensional joint scenario tree, wherein the multi-dimensional joint scenario tree includes several runoff scenarios, several wind power output scenarios, and several photovoltaic power output scenarios. Based on the multidimensional joint scene tree, a joint distributed fuzzy set is constructed using the Wasserstein distance metric, wherein the joint distributed fuzzy set is a Wasserstein sphere centered on the empirical distribution of the multidimensional joint scene tree and measured by the dynamic fuzziness radius. Load the refined coupling model of the cascade hydropower station group, and load the power grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data and new energy consumption constraint data, and configure it as a multi-dimensional dynamic constraint set; Based on the joint distributed fuzzy set and the multivariate dynamic constraint set, a multi-time-scale nested sub-Bruker rolling optimization model is constructed and executed to obtain the automatic generation control unit load allocation command; Extract the load allocation instructions of the automatic generation control unit and configure them as the output of the dispatch decision scheme.

2. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 1, characterized in that, Load multi-source runoff prediction data for the cascade hydropower station group, including: Numerical weather forecast data for the basin where the cascade hydropower station group is located is obtained, including rainfall forecast data, temperature forecast data, and evaporation forecast data; Load the measured data of the hydrological stations in the watershed, which includes measured data of upstream inflow, measured data of inter-regional runoff, and measured data of downstream water level; Based on the numerical weather forecast data and the measured data from the hydrological stations, a deep neural network quantile regression model is used to predict runoff probability, obtaining short-term runoff prediction data, medium-term runoff prediction data and long-term runoff prediction data, which are then configured as the multi-source runoff prediction data.

3. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 2, characterized in that, Load wind power forecast data for wind farms and photovoltaic power forecast data for photovoltaic power plants, including: Obtain numerical weather forecast data for the area where the wind farm is located, the numerical weather forecast data including wind speed forecast data and wind direction forecast data; Load the historical power output data of the wind farm, and based on the wind speed forecast data, the wind direction forecast data and the historical power output data, use a temporal convolutional network to perform wind power probability prediction to obtain the wind power prediction data; Numerical weather forecast data for the area where the photovoltaic power station is located is obtained, including solar radiation forecast data, temperature forecast data, and cloud cover forecast data; The historical output data of the photovoltaic power station is loaded, and based on the solar radiation forecast data, temperature forecast data, cloud cover forecast data, and historical output data, a temporal convolutional network is used to perform photovoltaic power probability prediction to obtain the photovoltaic power prediction data.

4. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 1, characterized in that, Based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data, a multi-source uncertainty joint scenario generation is performed to obtain a multi-dimensional joint scenario tree, which includes: Historical runoff data of the basin where the cascade hydropower station group is located, historical wind power output data of the wind farm, and historical photovoltaic power output data of the photovoltaic power station are collected. A first spatiotemporal correlation coefficient matrix between the historical runoff data and the historical wind power output data is calculated, a second spatiotemporal correlation coefficient matrix between the historical runoff data and the historical photovoltaic power output data is calculated, and a third spatiotemporal correlation coefficient matrix between the historical wind power output data and the historical photovoltaic power output data is calculated. Using the first spatiotemporal correlation coefficient matrix, the second spatiotemporal correlation coefficient matrix, and the third spatiotemporal correlation coefficient matrix as inputs, a Copula function is constructed to describe the joint distribution characteristics among runoff, wind power output, and photovoltaic power output. Based on the Copula function, several independent scenarios sampled from the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data are coupled and associated to generate the multi-dimensional joint scenario tree that includes the runoff scenario, the wind power output scenario, and the photovoltaic power output scenario.

5. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 1, characterized in that, Based on the aforementioned multidimensional joint scene tree, a joint distributed fuzzy set is constructed using the Wasserstein distance metric, including: The multidimensional joint scene tree is treated as a number of discrete sample points, and an empirical probability distribution of the multidimensional joint scene tree is constructed. Historical meteorological forecast error data of the basin where the cascade hydropower station group is located were collected, and the distribution characteristics of forecast error under different forecast lead times were statistically analyzed. Based on the forecast error distribution characteristics, the dynamic ambiguity radius that matches the current forecast lead time is dynamically calculated, wherein the dynamic ambiguity radius is positively correlated with the degree of dispersion of the forecast error distribution characteristics; With the empirical probability distribution as the center and the dynamic ambiguity radius as the measurement range, a Wasserstein sphere is constructed in the probability distribution space, which is defined as the joint distributed ambiguity set.

6. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 1, characterized in that, Load the refined coupling model of the cascade hydropower station group, and load grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data, and new energy consumption constraint data, configuring them as a multi-dimensional dynamic constraint set, including: Load the reservoir capacity curve data, downstream water level-flow relationship curve data, head change influence coefficient data and tailrace backwater influence coefficient data of the cascade hydropower station group, and configure them as the refined coupling model of the cascade hydropower. Load the power grid safety operation constraint data, which includes power grid frequency safety constraint data, voltage stability constraint data, and line transmission capacity constraint data; Load the reservoir operation constraint data, which includes upper limit constraint data for reservoir water level, lower limit constraint data for reservoir water level, upper limit constraint data for outflow, lower limit constraint data for outflow, water level fluctuation constraint data, and water level fluctuation rate constraint data. Load the ecological flow constraint data, which includes minimum ecological outflow constraint data and ecological flow process line constraint data; Load the new energy consumption constraint data, which includes the minimum utilization rate constraint data for new energy and the upper limit constraint data for wind and solar curtailment rates; The power grid safety operation constraint data, the reservoir operation constraint data, the ecological flow constraint data, and the new energy consumption constraint data are merged and configured into the multi-dimensional dynamic constraint set.

7. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 1, characterized in that, Construct and execute a multi-timescale nested sub-Blule rolling optimization model, which previously included: Collect historical net load data of the power grid in the area where the cascade hydropower station group is located. The historical net load data is the difference between the original load data of the power grid and the output data of new energy sources. Based on the historical net load data, calculate the net load ramp-up rate and net load fluctuation amplitude of the regional power grid; Based on the net load ramp rate and the net load fluctuation amplitude, the regional power grid's regulation capacity requirement for the cascade hydropower station group is quantified, configured as flexibility requirement constraint data, and added to the multivariate dynamic constraint set.

8. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 1, characterized in that, Based on the joint distributed fuzzy set and the multivariate dynamic constraint set, a multi-timescale nested sub-Bruker rolling optimization model is constructed and executed to obtain automatic generation control unit load allocation instructions, including: Construct a multi-timescale nested sub-Blule rolling optimization model that includes a medium-to-long-term optimization layer, a short-term optimization layer, and a real-time optimization layer; The medium- and long-term optimization layer aims to maximize the expected energy storage value of the basin, solves the reservoir water level control strategy that satisfies the multivariate dynamic constraint set under all probability distributions covered by the joint fuzzy set, and outputs the medium- and long-term water level control trajectory of each cascade hydropower station to the short-term optimization layer. The short-term optimization layer aims to maximize the expected power generation revenue. Based on the medium- and long-term water level control trajectory, it solves the daily power output plan of the cascade hydropower station group that satisfies the multivariate dynamic constraint set under several scenarios included in the multidimensional joint scenario tree. It then outputs the hourly power output plan and unit start-up and shutdown combination scheme of each cascade hydropower station to the real-time optimization layer. The real-time optimization layer aims to minimize the real-time output tracking error. It employs a model predictive control method to continuously refresh the optimization problem within each preset time window based on the hourly output plan and the unit start-stop combination scheme, thereby solving for the automatic generation control unit load allocation command that satisfies the multivariate dynamic constraint set.

9. The hydropower cluster generation scheduling optimization method based on hydrological prediction according to claim 8, characterized in that, A model predictive control method is employed to continuously refresh the optimization problem within each preset time window based on the hourly output plan and the unit start-up / shutdown combination scheme. This problem solves for the automatic generation control unit load allocation command that satisfies the multivariate dynamic constraint set, including: The optimization objective is to minimize the sum of squared deviations between the real-time output of the cascade hydropower station group and the hourly output plan at each time within a preset time domain from the current time to the future. At the arrival of each preset control cycle, the automatic generation control unit load allocation instructions at each time within the current control cycle are obtained by rolling the solution based on the latest measured operating status data of the cascade hydropower station group.

10. A hydropower cluster generation scheduling optimization system based on hydrological forecasting, characterized in that, The system is used to implement the hydropower cluster generation scheduling optimization method based on hydrological prediction as described in any one of claims 1-9, and the system includes: The multi-source prediction data acquisition module is used to load multi-source prediction data of runoff from cascade hydropower station groups, wind power prediction data of wind farms, and photovoltaic power prediction data of photovoltaic power stations. The multi-dimensional joint scenario tree acquisition module is used to perform multi-source uncertainty joint scenario generation based on the multi-source runoff prediction data, the wind power prediction data, and the photovoltaic power prediction data to obtain a multi-dimensional joint scenario tree, wherein the multi-dimensional joint scenario tree includes several runoff scenarios, several wind power output scenarios, and several photovoltaic power output scenarios. The joint distribution fuzzy set acquisition module is used to construct a joint distribution fuzzy set based on the multidimensional joint scene tree using the Wasserstein distance metric, wherein the joint distribution fuzzy set is a Wasserstein sphere centered on the empirical distribution of the multidimensional joint scene tree and measured by the dynamic fuzziness radius. The multi-dimensional dynamic constraint set acquisition module is used to load the cascade hydropower fine coupling model of the cascade hydropower station group, and load power grid safety operation constraint data, reservoir operation constraint data, ecological flow constraint data and new energy consumption constraint data, and configure them as a multi-dimensional dynamic constraint set; The load allocation instruction acquisition module is used to construct and execute a multi-time-scale nested sub-Bruker rolling optimization model based on the joint distributed fuzzy set and the multivariate dynamic constraint set to obtain the load allocation instruction of the automatic generation control unit; The scheduling decision scheme acquisition module is used to extract the load allocation instructions of the automatic generation control unit and configure them as the output of the scheduling decision scheme.

Images