Wind-light-water storage double-layer robust optimization scheduling strategy construction method and system

Abstract

The invention provides a wind, light and water storage double-layer robust optimization scheduling strategy construction method and system, and relates to the technical field of energy management, and the method comprises the steps: firstly obtaining multi-source data of a wind power plant, a photovoltaic power station and a cascade hydropower station; performing feature extraction and integrated prediction based on the multi-source data to generate a wind-solar power prediction result; combining hydrodynamic coupling modeling analysis to obtain a hydropower station power generation capacity and adjustable dynamic water volume sequence; the method comprises the following steps: constructing a multi-layer energy coupling graph and performing spectral domain structure optimization to form a power coupling network representing system space-time correlation and ecological sensitivity; and finally, adopting double-layer distributed robust optimization, synchronously optimizing the economic target and the ecological risk constraint, and generating a collaborative scheduling strategy. According to the method, the time sequence matching problem between the wind and light volatility and the hydroelectric time lag is effectively solved, the renewable energy consumption capability and the water resource utilization efficiency are improved, and the ecological operation risk is reduced.

Description

Technical Field

[0001] This invention relates to the field of energy management technology, and more specifically, to a method and system for constructing a two-layer robust optimization scheduling strategy for wind, solar, hydro, and storage. Background Technology

[0002] With the large-scale grid connection of renewable energy sources such as wind and solar power, the power grid faces significant challenges of volatility and uncertainty. In scenarios where cascade hydropower and wind and solar power are operated in conjunction, current dispatching methods are mainly based on daily or hourly power forecasting and economic dispatching models, focusing on medium- and long-term power balance and economic efficiency optimization. However, these methods often overlook the temporal coupling problem between short-term drastic fluctuations in wind and solar power output at the minute to hourly scale and the response lag of the hydropower system: that is, short-term peaks in wind and solar power output may fall precisely during ecologically sensitive periods in downstream river sections, or overlap with the propagation delay time window of water release from cascade reservoirs, forcing the dispatching system to respond by frequently starting and stopping units, additional water curtailment, or releasing water during ecologically unfavorable periods, thereby leading to decreased water resource utilization efficiency, increased curtailment rate, and increased ecological risks.

[0003] Therefore, there is an urgent need for a method and system for constructing a two-layer robust optimization scheduling strategy for wind, solar, hydro, and storage systems to solve the above-mentioned technical problems. Summary of the Invention

[0004] The purpose of this invention is to provide a method and system for constructing a two-layer robust optimization scheduling strategy for wind, solar, hydro, and storage systems, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides a method for constructing a two-layer robust optimization scheduling strategy for wind, solar, hydro, and storage systems, including: Acquire first information, second information and third information. The first information includes wind speed and unit operation status data of wind farms in the target area, and irradiance and module temperature data of photovoltaic power stations. The second information includes inflow runoff data, reservoir capacity-water level data, historical water release-water level response observation data and downstream ecological flow constraint data of cascade hydropower stations. Based on the first information, feature extraction and integrated predictive analysis are performed to generate power prediction results for wind power generation and photovoltaic power generation. Based on the power prediction results of wind power generation and photovoltaic power generation and the second information, hydrodynamic coupling modeling analysis is performed to obtain the power generation prediction sequence and the adjustable water volume sequence of the hydropower station. Based on the power prediction results of wind power generation and photovoltaic power generation, the power generation prediction sequence of hydropower station and the adjustable water volume sequence, a multi-layer energy coupling diagram is constructed, and a spectral domain structure optimization analysis is performed based on the multi-layer energy coupling diagram to generate a power coupling network. Based on the power coupling network, a two-layer distributed robust optimization analysis is performed to generate a comprehensive scheduling strategy for wind, solar, hydro, and storage.

[0005] Secondly, this application also provides a two-layer robust optimization scheduling strategy construction system for wind, solar, hydro, and storage, including: The acquisition unit is used to acquire first information, second information and third information. The first information includes wind speed and unit operation status data of wind farms in the target area, and irradiance and module temperature data of photovoltaic power stations. The second information includes inflow runoff data, reservoir capacity-water level data, historical water release-water level response observation data and downstream ecological flow constraint data of cascade hydropower stations. The prediction unit is used to perform feature extraction and integrated prediction analysis based on the first information to generate power prediction results for wind power generation and photovoltaic power generation. The analysis unit is used to perform hydrodynamic coupling modeling analysis based on the power prediction results of wind power generation and photovoltaic power generation and the second information to obtain the power generation prediction sequence and the adjustable water volume sequence of the hydropower station. The modeling unit is used to construct a multi-layer energy coupling diagram based on the power prediction results of wind power generation and photovoltaic power generation, the power generation prediction sequence of hydropower station and the adjustable water volume sequence, and to perform spectral domain structure optimization analysis based on the multi-layer energy coupling diagram to generate a power coupling network. The optimization unit is used to perform two-layer distributed robust optimization analysis based on the power coupling network to generate a comprehensive scheduling strategy for wind, solar, hydro, and storage. The beneficial effects of this invention are as follows: This invention establishes a wind and solar power output characterization model and couples reservoir capacity-water level data with downstream ecological flow constraint data to construct a spatiotemporal heterogeneous energy map. This effectively avoids operational risks during ecologically sensitive periods, significantly improves water resource utilization efficiency and renewable energy absorption capacity, and simultaneously considers the economic efficiency and safety of system operation. It solves the problems of power curtailment, increased water consumption, and ecological disturbance caused by time mismatch in traditional methods. This invention also adopts a two-layer distributed robust optimization architecture to simultaneously optimize long-term operational economics, cascade water consumption, and ecological risk probability, ultimately generating a coordinated scheduling strategy across hourly and daily scales. This achieves a balance between system operational economics and efficient water resource utilization while ensuring ecological security constraints.

[0006] Other features and advantages of the invention will be set forth in the following description, and will be apparent in part from the description, or may be learned by practicing embodiments of the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings. Attached Figure Description

[0007] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0008] Figure 1 This is a schematic diagram of the process for constructing a two-layer robust optimization scheduling strategy for wind, solar, hydro, and storage as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the system structure for constructing a two-layer robust optimization scheduling strategy for wind, solar, hydro, and storage as described in an embodiment of the present invention.

[0009] The diagram is labeled as follows: 701, acquisition unit; 702, prediction unit; 703, analysis unit; 704, modeling unit; 705, optimization unit. Detailed Implementation

[0010] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0011] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance. Example

[0012] This embodiment provides a method for constructing a two-layer robust optimization scheduling strategy for wind, solar, hydro, and storage.

[0013] See Figure 1 The figure shows that the method includes steps S1, S2, S3, S4 and S5.

[0014] Step S1: Obtain first information, second information and third information. The first information includes wind speed and unit operation status data of wind farms in the target area, and irradiance and module temperature data of photovoltaic power stations. The second information includes inflow runoff data, reservoir capacity-water level data, historical water release-water level response observation data and downstream ecological flow constraint data of cascade hydropower stations. Understandably, this step, by acquiring multi-source heterogeneous data, provides comprehensive and reliable data support for subsequent high-precision modeling and optimization decisions. Unlike traditional scheduling systems that only collect conventional data such as power and water level, this step specifically collects minute-level / high temporal resolution raw meteorological data (such as wind speed and irradiance) and equipment status data (such as unit operating status and module temperature), rather than simply using readily available power data. This allows subsequent models to capture the short-term fluctuation characteristics of wind and solar power output and the impact of equipment performance degradation from the source. This step also acquires dynamic coupling data of the hydropower link, including not only the static reservoir capacity-water level relationship, but more importantly, historical water release-water level response observation data. This provides the only data foundation for subsequent accurate modeling of hydrodynamic lag response. This step uploads the time-series data of downstream ecological flow constraints, along with system operation data such as electricity demand and power curtailment records, to the database, laying the data foundation for introducing ecological dimension constraints into scheduling decisions.

[0015] Specifically, for example, in a cascade hydropower and wind-solar hybrid power base in Southwest China, monitoring equipment such as meteorological stations, water level gauges, and flow sensors, as well as an automatic water situation monitoring system, are deployed to collect real-time data on wind speed per second at the wind farm, irradiance and backplane temperature of the photovoltaic array per minute, reservoir water level changes at the hydropower station every 5 minutes, and ecological flow monitoring values ​​of the downstream river every 15 minutes.

[0016] Step S2: Based on the first information, perform feature extraction and integrated prediction analysis to generate power prediction results for wind power generation and photovoltaic power generation; It is understandable that this step utilizes a variational autoencoder (VAE) to fuse and reduce the dimensionality of the high-dimensional heterogeneous data in the first information. Its unique advantage lies in its ability to learn the complex nonlinear coupling relationship between wind speed, irradiance and equipment operating status in an unsupervised manner, and generate a low-dimensional "wind-solar joint feature representation" containing rich spatiotemporal correlation information. This provides an unprecedented decision-making basis for downstream hydropower stations to determine whether to start the units and in what way (such as continuous full power generation or intermittent peak shaving) to balance this fluctuation. In this step, step S2 includes steps S21 and S22.

[0017] Step S21: Extract the spatiotemporal correlation features from the wind speed and unit operating status data of the wind farm and the irradiance and module temperature data of the photovoltaic power station through the variational autoencoder, and generate a joint wind-solar feature representation. Understandably, this step employs a variational autoencoder (VAE) model to fuse and compress the features of wind and solar power, two different renewable energy sources, thereby extracting a low-dimensional latent variable (i.e., a joint wind-solar feature representation) that can uniformly characterize their spatiotemporal dynamics. This is fundamentally different from traditional methods that simply stitch together data or process wind and solar data separately. A VAE can learn, unsupervised, complex nonlinear spatiotemporal relationships in the data, such as: the spatial correlation of wind speed (wind speed differences at different turbine locations), the temporal coupling relationship between irradiance and module temperature (temperature increases may lead to a decrease in photovoltaic panel efficiency), and the potential impact of wind turbine operating status (such as shutdown for maintenance) on overall output.

[0018] Step S22: Establish a nonlinear mapping relationship between wind and solar power output and meteorological elements based on the wind-solar joint feature representation and temporal convolutional network, and correct the prediction deviation based on the preset historical power curtailment record data to obtain the power prediction results of wind power generation and photovoltaic power generation.

[0019] It is understandable that this step uses a Temporal Convolutional Network (TCN) to learn the complex nonlinear mapping between meteorological elements and power output from the joint wind and solar feature representation. It also innovatively introduces historical power curtailment records as system operation feedback data to specifically correct prediction biases, ultimately generating high-precision power prediction results that fit actual operational constraints. This step does not isolate the prediction model from the actual system operation; instead, it embeds historical power curtailment records as prior knowledge into the model correction process. The TCN includes causal dilated convolutional layers, residual blocks, and a final output layer. The causal dilated convolutional layers ensure that the model uses only information from a preset time and earlier when making predictions at time steps, avoiding data leakage. Through an exponential increase in the dilation factor, the number of layers required to cover the same length of historical sequence is significantly reduced, while the receptive field is dramatically expanded. This allows the model to efficiently capture long-range dependencies ranging from minutes to hours, such as identifying the impact of meteorological elements (e.g., cloud movement, continuously increasing wind speed) on wind and solar power output over several hours. Then, each residual block contains multiple causal dilated convolutional layers, weight normalization, ReLU activation functions, and Dropout layers. Residual connections not only alleviate the vanishing gradient problem in deep networks but also allow the network to learn the residual mapping between input and output, improving training stability and model performance. Finally, the final output layer contains two parallel convolutional layers that output the parameters of the probability distribution (mean and variance of a Gaussian distribution in this step). The input in this step is a sequence of "wind and light joint feature representations," typically with a time resolution of minutes or 15 minutes, and a historical time window length of 2 days. The output is the probability distribution parameters (mean and variance) of the power values ​​at each prediction time point (every 15 minutes) within the next 2 days. The dilation factor sequence uses an exponentially growing sequence with a base of 2 to ensure that the network's receptive field can cover a sufficiently long historical period. The convolutional kernel size is 3 or 5, and this invention uses Dropout and weight normalization to prevent overfitting.

[0020] Step S3: Based on the power prediction results of wind power generation and photovoltaic power generation and the second information, perform hydrodynamic coupling modeling analysis to obtain the power generation prediction sequence and the adjustable water volume sequence of the hydropower station. Understandably, this step first uses a gated recurrent unit to fuse the probability distribution characteristics (mean and variance) of the predicted wind and solar power with the inflow runoff time series data in the second information, generating a "water inflow-power generation joint input feature". This feature not only includes water volume information but also incorporates the uncertainty brought about by wind and solar fluctuations. Subsequently, this feature is input into an encoder-decoder network with an attention mechanism. This network learns from a large amount of historical water release-water level response observation data in the second information to establish a nonlinear mapping relationship between "water release instructions" and "downstream water level changes in multiple future time periods". This step generates a dynamic capability sequence that can be directly used for optimized scheduling at the hydropower station level and quantifies the impact of wind and solar fluctuations. It transforms unstable wind and solar resources into a clear and reliable regulation capability for the hydropower system. In this step, step S3 includes steps S31, S32, and S33.

[0021] Taking a cascade hydropower station in Southwest China responding to a surge in wind power as an example: This invention receives information that wind power output may increase significantly in the next two hours (probability prediction) and simulates how reducing water release from the hydropower station at this time will lead to a rise in reservoir water level (increased head) and how long this will take (time lag) to affect the inflow to downstream power stations. The model will ultimately output the upper limit of the hydropower station's power generation every 15 minutes (e.g., after one hour, due to the rise in water level, the power generation can increase to 105% of the rated capacity) and the range of water volume that can be dynamically adjusted while meeting the downstream ecological flow requirements.

[0022] Step S31: Based on the power prediction results of wind power generation and photovoltaic power generation and the inflow runoff data of cascade hydropower stations, perform hydrodynamic-electric coupling feature extraction processing, wherein the output probability prediction results and inflow runoff time series data are fused by a gated circulation unit to generate inflow runoff-power generation joint input features. Understandably, the gated recurrent unit in this step reads two input sequences simultaneously: one is a wind and solar power output prediction sequence containing the power probability distribution parameters (mean and variance) for each future time point, which quantifies the volatility and uncertainty of renewable energy; the other is the historical inflow runoff time series data provided in the second information. Through its gating mechanism, the gated recurrent unit can identify complex patterns such as "how high-frequency fluctuations in wind and solar power output probability prediction affect the reliability of current runoff prediction" or "how a sustained wind power surge event overlaps with seasonal dry season runoff," ultimately outputting a low-dimensional, dense "inflow runoff-power generation joint input feature" vector. This joint feature not only contains raw data information, but more importantly, it encodes the coupling relationship between wind and solar uncertainty and hydrological dynamics. Its form is very suitable as input for downstream time series models. The gated loop unit includes a reset gate control step, which determines how many past states should be combined with the current input to capture short-term dependencies or abrupt changes (such as sudden changes in wind and solar power or a sharp increase in runoff); a candidate hidden state calculation step, which generates a candidate new state based on the past information filtered by the reset gate and the current input; an update gate control step, which determines how many past states should be retained and how many new candidate states should be injected to effectively capture long-term dependencies (such as the continuous characteristics of wind and solar power output and the daily / seasonal cycle of runoff); and a current hidden state update step, which is the output of the feature that integrates historical and current information.

[0023] Step S32: Based on the combined input characteristics of the inflow runoff and power generation, perform time-delay response modeling processing, use an encoder-decoder network to learn the mapping relationship between historical water release observation data and water level response observation data, and output the head evolution sequence; The encoder in this step compresses the input joint feature sequence (containing wind, solar, and runoff information) into a context vector containing the entire sequence information. Its core function is to capture pattern features such as "a sustained high wind power generation event superimposed on runoff during the dry season." The decoder, based on this context vector, progressively generates head values ​​for various future time points. In practical applications, such as at a cascade hydropower and wind-solar hybrid power plant in Southwest China, when the model receives a joint input feature indicating "continuously high wind power output for the next 6 hours," the decoder will progressively deduce: if the water release from hydropower station A is reduced to absorb wind power, then after 1 hour, the reservoir water level at station A will begin to rise (head increase). After 2.5 hours, the inflow to its downstream station B will decrease, leading to a drop in head at station B after 3 hours. The model can output the minute-level or hour-level change sequence of head for each power station throughout this dynamic process. This step clearly quantifies the lagging impact of scheduling decisions, providing crucial dynamic boundary conditions for the final calculation of the actual power generation capacity and adjustable water volume of the power stations.

[0024] Step S33: Perform collaborative calculation and processing of power generation and adjustable water volume based on the head evolution sequence. Specifically, by simulating the maximum power generation in all preset time periods and coupling reservoir capacity-water level data with downstream ecological flow constraint data, the range of dispatchable water volume is calculated, and finally, the power generation prediction sequence and adjustable water volume sequence of the hydropower station are obtained.

[0025] It is understandable that this step is based on the head evolution sequence and calculates the maximum technical output that the hydropower station can generate at each moment according to the dynamic characteristic curve of the turbine (a nonlinear head-output-flow relationship). By solving over all preset time periods, the results are ensured to meet engineering constraints such as unit vibration zone and output ramp-up rate. For example, when the head rises, the maximum technical output may exceed the rated capacity. The model of this invention accurately captures this characteristic. This invention closely integrates physical and ecological constraints with power generation potential, providing clear, reliable, and directly executable boundary conditions for subsequent multi-energy complementary optimization, avoiding the risk of dispatch instructions violating system safety or ecological constraints. For example, in a cascade hydropower and wind-solar complementary base in Southwest China, a dynamic head prediction sequence for the next 6 hours was received. The model calculates that, to balance wind power fluctuations, the power station can rapidly increase its output to 22 million kilowatts within the next hour (based on the maximum capacity under the current water head). However, to meet the constraint that the ecological flow in the downstream Jingjiang section should not be less than 5,000 cubic meters per second, the water consumption for power generation during this period is limited to a specific range (the water release must be between 8,000 and 10,000 cubic meters per second). Ultimately, it outputs a precise sequence that provides both the maximum power generation capacity and the water flow regulation boundary.

[0026] Step S4: Based on the power prediction results of wind power generation and photovoltaic power generation, the power generation prediction sequence of hydropower station and the adjustable water volume sequence, construct a multi-layer energy coupling diagram, and perform spectral domain structure optimization analysis based on the multi-layer energy coupling diagram to generate a power coupling network. Understandably, this step breaks through the traditional point-to-point connection method that only considers power balance in energy dispatch models. For the first time, it constructs a three-dimensional spatiotemporal heterogeneous graph that includes multiple time scales of minutes, hours, and days, and whose edge attributes integrate time lag, power transmission capacity, and ecological sensitivity. The construction process is as follows: First, wind farms, photovoltaic power stations, various cascade hydropower stations, and energy storage units are mapped as nodes in the graph. The characteristics of the nodes are defined by the predicted wind and solar power, the hydropower generation capacity, and the adjustable water volume range. Subsequently, the edges between nodes are defined according to physical connections (grid lines, waterways) and operational constraints. The weight attributes of the edges integrate the maximum rate of energy transmission (power transmission capacity), the time required for energy to be transmitted from one end to the other (time lag), and the potential impact of this transmission behavior on the downstream ecology (ecological sensitivity). In this step, step S4 includes steps S41, S42, and S43.

[0027] Step S41: The graph structure modeling method maps wind farms, photovoltaic power stations, hydropower stations and energy storage units as heterogeneous nodes to construct an initial multi-layer energy coupling graph containing multiple time scales. It is understandable that this step assigns dynamic, multi-dimensional attribute states to each node: the attributes of wind power and photovoltaic nodes are derived from power prediction results, including not only the expected output value but also probability distribution parameters characterizing their fluctuation uncertainty; the attributes of hydropower station nodes are jointly defined by the power generation prediction sequence and the adjustable water volume sequence, including both their power generation capacity and the upper and lower bound constraints of their regulation capacity; energy storage nodes are preset, and their attributes include their rated power, capacity, and current state of charge. These node attributes are dynamically updated as the prediction time window rolls.

[0028] The edges between nodes are assigned three quantitative attributes: first, time lag, which quantifies the time required for energy to be transferred from one end to the other based on historical observations and water flow propagation data (e.g., the delay between upstream reservoir release and downstream power station inflow); second, power transmission capacity, determined by the rated capacity of the power grid or the water conveyance capacity of the river, defining the physical transmission limit; and third, ecological sensitivity, which quantifies the degree of ecological impact of the transmission behavior represented by the edge during a specific period based on downstream ecological flow constraints (e.g., the sensitivity coefficient to drastic fluctuations in downstream flow increases significantly during fish spawning season). The final constructed initial multi-layer energy coupling graph actually contains topological structures of multiple time layers such as minutes, hours, and days, thereby describing the spatiotemporal coupling relationship between wind farms, photovoltaic power stations, hydropower stations, and energy storage units.

[0029] For example, in a cascade hydropower and wind-solar hybrid base in Southwest China, wind farms, photovoltaic power stations, hydropower stations, and energy storage units are constructed as an initial multi-layer graph. In this graph, the edge attributes from a wind farm node to a hydropower station node are set as follows: time lag = 45 minutes (considering water flow conduction time), power transmission capacity = 2000MW (based on the limit of the power grid line), and ecological sensitivity = 0.8 (the current period is the dry season and is within the ecologically sensitive window).

[0030] Step S42: Based on the initial multilayer energy coupling graph, perform spectral domain structure optimization processing, and perform feature decomposition and redundant edge pruning on the connection relationship between nodes in the initial multilayer energy coupling graph to generate an optimized energy coupling graph. Understandably, this step first involves performing a graph Laplacian transform on the initial multi-layer energy coupling graph, mapping the graph's connectivity in the spatial domain to the spectral domain (frequency domain). By calculating the graph's Laplacian matrix and performing eigenvalue decomposition, a set of eigenvalues ​​and eigenvectors characterizing the smoothness of the graph's connectivity is obtained. The eigenvectors corresponding to smaller eigenvalues ​​represent the low-frequency components (main energy transport paths) in the graph that are smoothly changing and tightly connected, while the high-frequency components corresponding to larger eigenvalues ​​often represent connections in the graph that are drastically changing, redundant, or noisy. Subsequently, based on this spectral decomposition result, spectral filtering and redundant edge pruning are performed. By setting a preset threshold, the connection edges corresponding to the low-frequency components that contribute the most to the characterization system are retained, while redundant edges corresponding to the high-frequency components that have a weak impact on the overall energy transmission and balance (e.g., some physically existing but rarely used backup lines or long-distance couplings that have no practical significance) are pruned. This step significantly reduces the complexity of the subsequent optimization model and improves the computational efficiency. At the same time, it not only does not lose key information, but also strengthens and highlights the most important and essential spatiotemporal energy coupling relationship in the system through mathematical methods, laying the most reasonable structural foundation for the subsequent generation of a high-quality multi-energy complementary power coupling network.

[0031] In the initial diagram of a cascade hydropower and wind-solar hybrid power base in Southwest China, dozens of wind farms, photovoltaic power stations, and hydropower stations are connected, with complex relationships between nodes. After spectral domain analysis, it was found that the connection edge between a wind farm group and a large downstream reservoir power station (despite the long transmission distance and a time delay of 1 hour) exhibits a low-frequency characteristic in the spectral domain. This reveals that this path is a key distant hydropower compensation channel for smoothing wind power fluctuations, and therefore it was retained. On the other hand, the edge between a pumped storage power station and a photovoltaic power station, which is closer in distance, exhibits high-frequency noise in the spectral domain due to its extremely small capacity and infrequent use in historical scheduling. Therefore, it was identified as a redundant connection and pruned.

[0032] Step S43: The power prediction results of wind power generation and photovoltaic power generation, the power generation prediction sequence of hydropower station and the adjustable water volume sequence are embedded into the network nodes through graph neural network, and the edge weights are calculated based on the downstream ecological flow constraint data to generate a multi-energy power coupling network including wind energy, electric energy and hydropower.

[0033] Understandably, this step uses the optimized energy coupling graph as the initial topology, and then uses the power prediction results of wind and solar power generation, the power generation prediction sequence of hydropower stations, and the adjustable water volume sequence as dynamic input features for each corresponding node. The graph neural network, through its message passing mechanism, aggregates information from neighboring nodes and combines it with its own dynamic attributes at each prediction time step (e.g., rolling execution every 15 minutes) to update the embedded representation of all nodes. This process ensures that the state of each node in the graph is no longer isolated, but rather a deep representation that integrates the spatiotemporal correlation information of the entire network. This step provides a dynamic mathematical model that represents both physical connections and embedded risk costs for subsequent two-layer optimization, ensuring that the final generated scheduling strategy, while pursuing economic efficiency, can proactively avoid ecologically sensitive operations.

[0034] Step S5: Perform a two-layer distributed robust optimization analysis based on the power coupling network to generate a comprehensive scheduling strategy for wind, solar, hydro, and storage.

[0035] It is understandable that the upper-level optimization in this step is a slow-time-scale problem on a "weekly" or "monthly" scale. It is based on the entire coupled network, with the joint objective function being the minimization of the system's long-term expected operating cost and the minimization of ecological risk metrics. It utilizes stochastic programming to handle the long-term uncertainty of wind and solar power output, outputting pre-defined capacity allocation and risk control strategies for multiple future scenarios on a monthly basis. For example, to cope with the autumn windy season, more regulation capacity reserves are allocated to hydropower, and stricter scheduling margins are set for ecologically sensitive periods. The lower-level optimization, on the other hand, is a time-scale problem on a "dayly" or "hourly" scale. It receives the upper-level strategy as boundary conditions and, based on updated short-term wind and solar power probability predictions, uses chance-constrained programming to construct a real-time scheduling sub-problem with the objectives of minimizing periodic scheduling probability risks (such as curtailment risk) and cascade water consumption. It is responsible for generating refined, daily-scale pre-scheduling plans, such as specifically arranging the output curves of each power station during the peak photovoltaic output period from 2 PM to 4 PM tomorrow to ensure compliance with the ecological flow fluctuation limits given by the upper level. This step decomposes long-term ecological risk prevention and control goals and embeds them into short-term economic operations, transforming system scheduling from a passive, responsive operation to a proactive, preventative optimization. Step S5 in this step includes steps S51, S52, and S53.

[0036] Step S51: Perform upper-level optimization modeling based on the power coupling network. In this process, a planning model with the objective function of joint optimization of expected operating cost and ecological risk measurement is constructed by stochastic programming method. Based on the preset historical power curtailment records and ecological sensitivity indicators as constraints, a preset time period capacity configuration and risk control strategy on a monthly basis is generated. It is understood that this step uses a multi-energy complementary power coupling network as the topology basis. The node attributes (power plant capacity, maximum output) and edge attributes (transmission capacity, time delay, ecological sensitivity) of this network define the physical constraints of the system. The objective function aims to minimize the weighted sum of the system's long-term expected operating costs (including fuel costs, maintenance costs, curtailment penalties, etc.) and ecological risk metrics. The ecological risk metrics are preset values, based on corresponding historical data. For example, during fish migration seasons, the average of the allowed daily flow variation thresholds from historical data is used. The objective function is as follows:

[0037] in, Let be the objective function. To minimize selection, The expected total operating cost (including fuel, maintenance, and curtailment costs), For ecological risk measurement, and These are preset weighting coefficients used to balance economic and ecological objectives.

[0038] Step S52: Based on the preset time period capacity configuration and risk control strategy on a monthly basis, construct the lower-level scheduling scheme, and use the opportunity constraint programming method to construct a real-time scheduling sub-problem with the goal of minimizing scheduling probability risk and cascade water consumption, and generate a wind, solar, hydro and storage pre-scheduling scheme on a daily basis. It is understandable that this step constructs a two-objective optimization problem: minimizing scheduling probabilistic risk and minimizing cascade water consumption. Scheduling probabilistic risk is a comprehensive indicator that quantifies the probability and severity of various adverse scenarios (such as severe power curtailment or violation of ecological flow) that may occur when executing the daily plan. Minimizing cascade water consumption reflects the pursuit of economical utilization of water resources. Opportunity-constrained programming is used to handle the stochasticity in wind and solar power forecasting. It allows constraints to be violated at a certain confidence level, such as "the probability of curtailment exceeding 5% must not be higher than 10%" or "the probability of downstream flow exceeding the ecological threshold must not be higher than 5%". Its objective optimization formula is shown below:

[0039] The formulas for the curtailment rate constraint and the downstream ecological flow constraint are shown below:

[0040] in, For scheduling decision variables, and For multi-objective weighting coefficients, As a comprehensive indicator of scheduling probability risk, For the set of scheduling decision variables, This serves as an index variable, used to iterate through and refer to each specific hydropower station in the cascade. For the first The water consumption of the hydroelectric power station, The probability of an event occurring. for Power curtailment during specific time periods for Available renewable energy power during the period The threshold for the curtailment rate, As for the confidence level of power curtailment risk, for Discharge flow during specific time periods Within the permitted range of ecological flow, Confidence level for ecological overreach risk. A set of time periods in which the events occurred. This is a collection of ecologically sensitive periods.

[0041] Step S53: Based on the two-layer collaborative optimization process of the preset time period capacity configuration and risk control strategy on a monthly basis and the pre-scheduling scheme of wind, solar, hydro and storage on a daily basis, a comprehensive scheduling strategy for the wind, solar, hydro and storage system is generated.

[0042] Understandably, this step first "reports" the daily-scale pre-scheduling scheme generated by the lower layer to the upper-layer model. The upper-layer model then verifies whether the scheme fully complies with its established monthly capacity and risk control strategies (e.g., whether the maximum water release in the daily plan exceeds the flow fluctuation limit set for the ecologically sensitive period in the monthly plan). If the verification fails, the upper layer will pass its long-term objectives (such as ecological risk measurement) to the lower layer, which will then adjust its objective function accordingly (e.g., increase the penalty weight for violating ecological constraints) and resolve the daily plan. This process will iterate until the upper and lower layer schemes reach a "consensus": that is, the daily plan can meet both short-term economic and operational requirements and fully comply with long-term strategic risk management boundaries. The final output strategy is a set of strategies across time scales that includes multiple scenarios and corresponding scheduling instructions. Example

[0043] like Figure 2As shown, this embodiment provides a two-layer robust optimization scheduling strategy construction system for wind, solar, hydro, and storage. The system includes an acquisition unit 701, a prediction unit 702, an analysis unit 703, a modeling unit 704, and an optimization unit 705.

[0044] The acquisition unit 701 is used to acquire first information, second information and third information. The first information includes wind speed and unit operation status data of wind farms in the target area, and irradiance and module temperature data of photovoltaic power stations. The second information includes inflow runoff data, reservoir capacity-water level data, historical water release-water level response observation data and downstream ecological flow constraint data of cascade hydropower stations. The prediction unit 702 is used to perform feature extraction and integrated prediction analysis based on the first information to generate power prediction results for wind power generation and photovoltaic power generation. Analysis unit 703 is used to perform hydrodynamic coupling modeling analysis based on the power prediction results of wind power generation and photovoltaic power generation and the second information to obtain the power generation prediction sequence and the adjustable water volume sequence of the hydropower station. Modeling unit 704 is used to construct a multi-layer energy coupling diagram based on the power prediction results of wind power generation and photovoltaic power generation, the power generation prediction sequence of hydropower station and the adjustable water volume sequence, and to perform spectral domain structure optimization analysis based on the multi-layer energy coupling diagram to generate a power coupling network. The optimization unit 705 is used to perform a two-layer distributed robust optimization analysis based on the power coupling network to generate a comprehensive scheduling strategy for wind, solar, hydro, and storage.

[0045] It should be noted that the specific methods by which each module performs operations in the system described in the above embodiments have been described in detail in the embodiments related to the method, and will not be elaborated here.

[0046] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

[0047] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for constructing a robust optimization scheduling strategy of wind-solar-water storage dual-layer, characterized in that, The method comprises the following steps: acquiring first information, second information and third information, the first information comprising wind speed and unit operating state data of a wind farm in a target area, and irradiance and module temperature data of a photovoltaic power station, the second information comprising reservoir inflow data of a cascade hydropower station, reservoir capacity-water level data, historical discharge-water level response observation data and downstream ecological flow constraint data; performing feature extraction and integrated prediction analysis based on the first information to generate power prediction results of wind power generation and photovoltaic power generation; performing water power coupling modeling analysis based on the power prediction results of wind power generation and photovoltaic power generation and the second information to obtain a power prediction sequence and a movable water volume sequence of the hydropower station; constructing a multi-layer energy coupling graph based on the power prediction results of wind power generation and photovoltaic power generation, the power prediction sequence and the movable water volume sequence of the hydropower station, and performing spectral domain structure optimization analysis based on the multi-layer energy coupling graph to generate a power coupling network; performing double-layer distributed robust optimization analysis based on the power coupling network to generate a wind-solar-hydro-storage comprehensive dispatching strategy. 2.The method of claim 1, wherein The feature extraction and integrated prediction analysis based on the first information comprises: extracting spatiotemporal correlation features in the wind speed and unit operating state data of the wind farm and the irradiance and module temperature data of the photovoltaic power station through a variational autoencoder to generate a wind-solar combined feature representation; establishing a nonlinear mapping relationship between wind-solar output and meteorological elements according to the wind-solar combined feature representation and a time series convolution network, and correcting prediction bias based on preset historical curtailed power record data to obtain the power prediction results of wind power generation and photovoltaic power generation.

3. The method of claim 1, wherein The water power coupling modeling analysis based on the power prediction results of wind power generation and photovoltaic power generation and the second information comprises: performing water power-electricity coupling feature extraction processing according to the power prediction results of wind power generation and photovoltaic power generation and the reservoir inflow data of the cascade hydropower station, wherein an output probability prediction result and reservoir inflow time series data are fused through a gated recurrent unit to generate reservoir inflow-power combined input features; performing time lag response modeling processing according to the reservoir inflow-power combined input features, and learning a mapping relationship between historical discharge observation data and water level response observation data using an encoder-decoder network to output a water head evolution sequence; performing power generation power and adjustable water volume collaborative calculation processing according to the water head evolution sequence, wherein the maximum power generation power in all preset time periods is simulated, and the reservoir capacity-water level data and downstream ecological flow constraint data are coupled to calculate the adjustable water volume range, and finally the power prediction sequence and the movable water volume sequence of the hydropower station are obtained.

4. The method of claim 1, wherein The multi-layer energy coupling graph is constructed based on the power prediction results of wind power generation and photovoltaic power generation, the power prediction sequence and the movable water volume sequence of the hydropower station, and the spectral domain structure optimization analysis is performed based on the multi-layer energy coupling graph, which comprises: mapping the wind farm, the photovoltaic power station, the hydropower station and the energy storage unit into heterogeneous nodes based on a graph structure modeling method to construct an initial multi-layer energy coupling graph containing multiple time scales; Performing spectral domain structure optimization processing based on the initial multi-layer energy coupling graph, decomposing the connection relationship between nodes of the initial multi-layer energy coupling graph, and pruning redundant edges to generate an optimized energy coupling graph; The power prediction results of wind power generation and photovoltaic power generation, the power generation prediction sequence of the hydropower station, and the adjustable water volume sequence are embedded into network nodes through a graph neural network, edge weights are calculated based on downstream ecological flow constraint data, and a multi-energy power coupling network containing wind energy, electric energy, and water energy is generated.

5. The method of claim 1, wherein Based on the power coupling network, a double-layer distributed robust optimization analysis is performed to generate a wind-solar-hydro-storage comprehensive scheduling strategy, including: Based on the power coupling network, an upper-layer optimization modeling process is performed, in which a planning model with an expected operation cost and ecological risk measurement joint optimization as an objective function is constructed by a stochastic programming method, and a preset time period capacity configuration and risk control strategy in months is generated based on preset historical off-grid records and ecological sensitivity indicators as constraint conditions; According to the preset time period capacity configuration and risk control strategy in months, a lower-layer scheduling scheme is constructed, an opportunity constraint programming method is used to construct a real-time scheduling sub-problem with scheduling probability risk and minimum cascade water consumption minimization as an objective, and a wind-solar-hydro-storage pre-scheduling scheme in days is generated; According to the double-layer collaborative optimization processing of the preset time period capacity configuration in months, the risk control strategy, and the wind-solar-hydro-storage pre-scheduling scheme in days, a comprehensive scheduling strategy of the wind-solar-hydro-storage system is generated.

6. A wind-solar-water storage double-layer robust optimization scheduling strategy construction system, characterized in that, including: An acquisition unit is configured to acquire first information, second information, and third information, the first information includes wind speed and unit operation state data of a wind farm in a target region, irradiance and module temperature data of a photovoltaic power station, the second information includes inflow runoff data, reservoir capacity-water level data, historical water release-water level response observation data, and downstream ecological flow constraint data of a cascade hydropower station; A prediction unit is configured to perform feature extraction and integrated prediction analysis based on the first information to generate power prediction results of wind power generation and photovoltaic power generation; An analysis unit is configured to perform hydrodynamic coupling modeling analysis based on the power prediction results of wind power generation and photovoltaic power generation and the second information to obtain a power generation prediction sequence and an adjustable water volume sequence of a hydropower station; A modeling unit is configured to construct a multi-layer energy coupling graph based on the power prediction results of wind power generation and photovoltaic power generation, the power generation prediction sequence and the adjustable water volume sequence of the hydropower station, and perform spectral domain structure optimization analysis based on the multi-layer energy coupling graph to generate a power coupling network; An optimization unit is configured to perform a double-layer distributed robust optimization analysis based on the power coupling network to generate a wind-solar-hydro-storage comprehensive scheduling strategy.

7. The wind-solar-water storage double-layer robust optimization scheduling strategy construction system according to claim 6, characterized in that, The prediction unit includes: A first prediction sub-unit is configured to extract spatio-temporal correlation features in wind speed and unit operation state data of a wind farm, irradiance and module temperature data of a photovoltaic power station through a variational autoencoder to generate wind-solar joint feature representation; The second prediction unit is configured to establish a nonlinear mapping relationship between wind and light output and meteorological elements according to the wind and light combined feature representation and the time sequence convolution network, and correct prediction deviation based on preset historical power curtailment record data, to obtain power prediction results of wind power generation and photovoltaic power generation.

8. The wind-solar-water dual-layer robust optimization scheduling strategy construction system according to claim 6, characterized in that, The analysis unit comprises: The first analysis subunit is configured to perform hydro-power coupling feature extraction processing according to the power prediction results of wind power generation and photovoltaic power generation and reservoir inflow data of the cascade hydropower station, wherein the output probability prediction results and the inflow time series data are fused through a gating cycle unit to generate inflow-runoff-power combined input features; The second analysis subunit is configured to perform time lag response modeling processing according to the inflow-runoff-power combined input features, learn a mapping relationship between historical water release observation data and water level response observation data by using an encoder-decoder network, and output a water head evolution sequence; The third analysis subunit is configured to perform power generation and adjustable water volume collaborative calculation processing according to the water head evolution sequence, wherein the maximum power generation in all preset time periods is simulated, and the reservoir capacity-water level data and downstream ecological flow constraint data are coupled to calculate the adjustable water volume range, so as to finally obtain a power prediction sequence and an adjustable water volume sequence of the hydropower station. 9.The system according to claim 6, wherein, The modeling unit comprises: The first modeling subunit is configured to map wind farms, photovoltaic power stations, hydropower stations and energy storage units into heterogeneous nodes based on a graph structure modeling method, and construct an initial multi-layer energy coupling graph containing multiple time scales; The second modeling subunit is configured to perform spectral domain structure optimization processing based on the initial multi-layer energy coupling graph, decompose the connection relationship between nodes of the initial multi-layer energy coupling graph, and prune redundant edges to generate an optimized energy coupling graph; The third modeling subunit is configured to embed the power prediction results of wind power generation and photovoltaic power generation, the power prediction sequence and the adjustable water volume sequence of the hydropower station into network nodes by using a graph neural network, calculate edge weights based on downstream ecological flow constraint data, and generate a multi-energy power coupling network containing wind energy, electrical energy and water energy.

10. The system according to claim 6, wherein, The optimization unit comprises: The first optimization subunit is configured to perform upper-layer optimization modeling processing based on the power coupling network, wherein a planning model with an expected operation cost and ecological risk metric joint optimization as an objective function is constructed by using a stochastic programming method, and preset historical power curtailment records and ecological sensitivity indicators are used as constraint conditions to generate a preset time period capacity configuration and risk control strategy in months; The second optimization subunit is configured to construct a lower-layer scheduling scheme according to the preset time period capacity configuration and risk control strategy in months, construct a real-time scheduling sub-problem with scheduling probability risk and minimum cascade water consumption minimization as an objective by using an opportunity constraint programming method, and generate a wind, light, water and storage pre-scheduling scheme in days; The third optimization subunit is configured to perform double-layer collaborative optimization processing according to the preset time period capacity configuration in months, the risk control strategy, and the wind, light, water and storage pre-scheduling scheme in days, to generate a comprehensive scheduling strategy of the wind, light, water and storage system.

Images