Output optimization method, system and device under complementary scheduling and medium

Abstract

The invention discloses an output optimization method, system and device under complementary scheduling and a medium. The method comprises the following steps: acquiring data of cascade hydropower, a wind field and a photoelectric field; a graph neural network is constructed based on the geographic position data, the weight of an edge is configured according to historical power generation data, and nodes of the graph neural network represent fans, photovoltaic arrays and hydropower stations at different geographic positions; carrying out meteorological prediction according to the current short-term meteorological data based on the constructed graph neural network; and constructing an evaluation model based on the double-layer Monte Carlo-conditional risk, outputting a probability density function of output prediction according to the evaluation model, and executing output optimization. According to the method, the propagation condition of possible meteorological elements in the non-uniform observation network is captured by establishing the graph neural network; and comprehensive evaluation of water, wind and light is realized by constructing an evaluation model, and a reference direction of short-term complementary output optimization is given through a probability density function, so that the scheduling effect of complementary optimization is improved.

Description

Technical Field

[0001] This invention relates to the field of power output dispatching technology, and in particular to a method, system, equipment and medium for power output optimization under complementary dispatching. Background Technology

[0002] Developing clean, low-carbon, and renewable energy sources such as wind power and photovoltaics (hereinafter referred to as wind and solar) has become a major strategic initiative for alleviating the global energy crisis, addressing climate change, and improving the ecological environment. The core idea of ​​wind, solar, and hydropower multi-energy complementarity is that the random and intermittent electricity generated by wind and solar power plants is first transmitted to a joint control center connected to the hydropower plant. After being tracked and compensated in real time by nearby hydropower units, the electricity is bundled together and sent to the power grid to mitigate the impact of unstable wind and solar power output on the grid.

[0003] However, wind and solar power are affected by many natural factors such as wind speed, solar radiation, and temperature, resulting in significant randomness, fluctuation, and intermittency in power output. This increases the pressure on the power grid for peak shaving and frequency regulation, which is not conducive to the safe, economical, and stable operation of the power system. It also limits the grid's absorption of wind and solar power, making it difficult to achieve the goal of simultaneous growth in new energy power generation and installed capacity. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method and system for optimizing power output under complementary scheduling to solve the problem that the current power output has significant randomness, fluctuation and intermittency due to many influencing factors, making it difficult to accurately determine the specific power that wind and solar power can provide at subsequent times, increasing the frequency regulation pressure and peak shaving difficulty, and making it difficult to achieve synchronous growth of new energy power generation and installed capacity.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution:

[0007] In a first aspect, the present invention provides a method for optimizing output under complementary scheduling, comprising:

[0008] Obtain geographical location data, corresponding historical meteorological data, and historical power generation data for cascade hydropower, wind farms, and photovoltaic power plants;

[0009] A graph neural network is constructed based on geographic location data, and the weights of the edges are configured according to historical power generation data. The nodes of the graph neural network represent wind turbines, photovoltaic arrays, and hydropower stations in different geographic locations.

[0010] Based on the constructed graph neural network, weather forecasts are made using current short-term meteorological data;

[0011] A two-layer Monte Carlo-conditional risk assessment model is constructed, wherein the outer layer of the assessment model samples the results of meteorological forecasts, and the inner layer models the random variables of water, wind, and solar power.

[0012] The output power prediction probability density function is output based on the evaluation model, and output power optimization is performed based on the probability density function.

[0013] As a preferred embodiment of the power output optimization method under complementary scheduling described in this invention, the step of constructing a graph neural network based on geographical location data and configuring edge weights according to historical power generation data includes:

[0014] In the graph neural network, nodes with related relationships are grouped into a node group, and classification is performed based on the node group.

[0015] Based on the classification results, calculate the terrain similarity between the terrain relationships covered by each group of nodes in any node group;

[0016] The weights of the graph neural network edges are configured based on the terrain similarity between nodes, the geographical distance between nodes, and historical relevance.

[0017] The advantages of this preferred solution are: the graph structure can effectively model the potential mutual influences between power plants scattered in different geographical locations, and the weights can be configured using historical power generation data, rather than relying on human assumptions, which can improve the model's ability to express real physical and operational relationships.

[0018] As a preferred embodiment of the power output optimization method under complementary scheduling described in this invention, the method further includes, before the weather forecast:

[0019] A dual-clock system consisting of a hardware clock and a logic clock is established, wherein the hardware clock is used to record the timestamp of the original data acquisition.

[0020] The logical clock establishes a confidence interval based on the delay distribution corresponding to the graph neural network; and, using a spatial encoder based on a convolutional neural network, the discrete meteorological observation data is upgraded to the same dimension as the satellite meteorological cloud image.

[0021] As a preferred embodiment of the output optimization method under complementary scheduling described in this invention, it further includes:

[0022] Upgraded meteorological observation data, satellite meteorological cloud images, reservoir water levels, and geographical location data are mapped to a unified spatiotemporal coordinate system;

[0023] In this process, geographic location data and hardware clocks are mapped to a high-dimensional feature space through a learnable hash function.

[0024] As a preferred embodiment of the output optimization method under complementary scheduling described in this invention, the logical clock establishes a confidence interval based on the delay distribution corresponding to the graph neural network, including:

[0025] Physical connection markers are added based on graph neural networks; these physical connection markers are used to identify the data transmission relationships between nodes.

[0026] Based on the physical connection markers, a cross-layer influence factor is established, and the cross-layer influence factor is used to describe the intensity of the impact of the delay in meteorological observation data on the reservoir water level.

[0027] The established cross-layer influence factors are assigned corresponding weights and used as confidence levels.

[0028] As a preferred embodiment of the output optimization method under complementary scheduling described in this invention, the forward prediction compensation of the satellite meteorological cloud image specifically includes:

[0029] Sequentially record the acquired satellite meteorological cloud image data;

[0030] Configure the delay step size for satellite meteorological cloud image data, and use the delay step size as a window to capture the acquired satellite meteorological cloud image data;

[0031] Using captured satellite meteorological cloud image data, a satellite meteorological cloud image with a delay step size is predicted based on an LSTM model to perform forward compensation.

[0032] As a preferred embodiment of the power output optimization method under complementary scheduling described in this invention, the method includes: outputting a probability density function for power output prediction based on the evaluation model, comprising:

[0033] The market electricity price in the random variable is modeled using mean regression, and the failure rate in the random variable is randomly generated based on the failure state.

[0034] A power aggregation model for water, wind, and solar power is established based on the random generation of fault states and mean regression modeling.

[0035] Based on the power output aggregation model, the probability density function for power output prediction is determined using kernel density estimation.

[0036] The advantages of this preferred scheme are: the kernel density function can provide complete probability distribution information of the total possible output of the system in the future. By using the PDF information, it is possible to select the output scheme that achieves the best balance between expected cost and risk exposure.

[0037] Secondly, the present invention provides an output optimization system under complementary scheduling, comprising:

[0038] The acquisition module is used to acquire geographical location data, corresponding historical meteorological data, and historical power generation data of cascade hydropower, wind farms, and photovoltaic power plants;

[0039] The first construction module is used to construct a graph neural network based on geographic location data and configure the weights of the edges according to historical power generation data. The nodes of the graph neural network represent wind turbines, photovoltaic arrays and hydropower stations in different geographic locations.

[0040] The prediction module is used to make weather forecasts based on the constructed graph neural network and current short-term meteorological data.

[0041] The second building module is used to build an assessment model based on a two-layer Monte Carlo-conditional risk. The outer layer of the assessment model samples the results of meteorological forecasts, and the inner layer models the random variables of water, wind, and solar power.

[0042] The output execution module is used to output a probability density function of the predicted output based on the evaluation model, and to perform output optimization based on the probability density function.

[0043] Thirdly, the present invention provides a computer device, comprising:

[0044] Memory and processor;

[0045] The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the output optimization method under complementary scheduling.

[0046] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the output optimization method under complementary scheduling.

[0047] Compared with the prior art, the beneficial effects of the present invention are as follows: The present invention establishes a graph neural network to capture the propagation of possible meteorological elements in a non-uniform observation network; and realizes the comprehensive assessment of water, wind and solar power by constructing an assessment model based on two-layer Monte Carlo-conditional risk, and provides a reference direction for short-term complementary power output optimization through the probability density function of power output prediction, thereby improving the scheduling effect of complementary optimization. Attached Figure Description

[0048] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a schematic diagram of the overall process of the output optimization method under complementary scheduling according to an embodiment of the present invention. Detailed Implementation

[0050] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0051] Example 1, referring to Figure 1 As an embodiment of the present invention, a method for optimizing output under complementary scheduling is provided, comprising:

[0052] S100: Acquire geographical location data, corresponding historical meteorological data, and historical power generation data of cascade hydropower, wind farms, and photovoltaic power plants;

[0053] S200: A graph neural network is constructed based on geographic location data, and the weights of the edges are configured according to historical power generation data. The nodes of the graph neural network represent wind turbines, photovoltaic arrays and hydropower stations in different geographic locations.

[0054] S300: Based on the constructed graph neural network, it makes weather forecasts according to the current short-term meteorological data;

[0055] S400: Construct an assessment model based on a two-layer Monte Carlo-conditional risk model. The outer layer of the assessment model samples the results of meteorological forecasts, while the inner layer models the random variables of water, wind, and solar power.

[0056] S500: Outputs the probability density function of the predicted output based on the evaluation model, and performs output optimization based on the probability density function.

[0057] It should be noted that wind and solar power output cannot be precisely predicted or controlled, relying entirely on instantaneously changing natural conditions such as wind speed, solar irradiance, and temperature. This results in a highly uncertain power output curve, unlike traditional thermal or hydropower which can be dispatched on demand. This randomness makes it difficult for power system dispatching agencies to accurately determine the specific power that wind and solar power can provide at the next moment, increasing the difficulty and risk of balancing power supply and demand. At the same time, the large amplitude and rapid speed of these changes mean that output power may drop sharply from near-rated value to zero or rise rapidly from zero in a short period of time, exacerbating the peak-valley difference in the system. Furthermore, it cannot provide a continuous and stable base load, thus limiting the increase in the actual power generation share.

[0058] Therefore, by constructing a graph neural network based on geographic data through steps S100-S500, meteorological elements are propagated in a non-uniform observation network. Furthermore, by establishing an evaluation model and providing a probability density function for output prediction, short-term complementary output optimization is achieved.

[0059] Example 2, refer to Figure 1 As an embodiment of the present invention, based on the above embodiment, a method for optimizing output under complementary scheduling is provided.

[0060] In this embodiment of the application, step S100 involves acquiring the geographical location data of the cascade hydropower, wind farm, and photovoltaic power plant, as well as the corresponding historical meteorological data and historical power generation data.

[0061] Specifically, geographical location can be represented using latitude and longitude.

[0062] Historical meteorological data can include historical meteorological cloud images and meteorological observation data, such as wind force level, rainfall, and sunshine intensity.

[0063] Historical power generation data can include power generation information from various units in hydropower, wind farms, and photovoltaic power plants.

[0064] In this embodiment of the application, step S200, which involves constructing a graph neural network based on geographic location data and configuring edge weights according to historical power generation data, includes the following steps A1-A3:

[0065] A1: In a graph neural network, nodes with relationships are grouped together and classified according to the node groups.

[0066] For example, nodes that are connected to each other can be grouped together and classified according to the connection relationships between the nodes, such as hydropower node-hydropower node, hydropower node-wind power node, photovoltaic node-wind power node, etc.

[0067] A2: Based on the classification results, calculate the terrain similarity between the terrain relationships covered by each group of nodes in any node group;

[0068] In one alternative implementation, terrain data between nodes can be determined based on the geographic location data of the nodes, and terrain similarity can be calculated. Specifically, terrain undulation levels can be set, and terrain data between nodes can be divided into multiple levels according to the terrain undulation relationship, thereby reducing the amount of similarity calculation.

[0069] A3: Configure the weights of the graph neural network edges based on the terrain similarity between nodes, the geographical distance between nodes, and historical relevance.

[0070] For example, the higher the terrain similarity between nodes, the shorter the distance, and the stronger the correlation with historical power generation, the greater the weight of the edge.

[0071] Specifically, the historical relevance in this application refers to the power output under historical meteorological data.

[0072] For example, if, under the same meteorological conditions, the wind power and photovoltaic output of a set of nodes both exceed 80% for a period of time exceeding a preset value, then the two nodes are considered to have a strong historical correlation.

[0073] It should be noted that wind and solar power fields are significantly affected by meteorological conditions. Therefore, predicting the propagation of meteorological elements is crucial for the complementary scheduling of hydropower, wind, and solar power fields. The above design enables the propagation of meteorological elements within a non-uniform observation network using a constructed graph neural network, thereby allowing for the prediction of short-term meteorological activities in hydropower, wind, and solar power fields based on the acquired current meteorological data.

[0074] In this embodiment of the application, the nodes of the graph neural network in step S200 represent wind turbines, photovoltaic arrays, and hydropower stations in different geographical locations.

[0075] For example, an undirected graph neural network can be constructed; specifically, the propagation of possible meteorological elements in a non-uniform observation network can be captured through the information transmission between nodes in the graph neural network.

[0076] It should be noted that graph structures can effectively model the potential interactions between power plants scattered across different geographical locations, such as regional meteorological pattern correlations, which is also key to complementary optimization. Furthermore, by using historical power generation data to configure weights, the graph can automatically learn the strength of relationships between nodes (power plants) based on historical actual operations, rather than relying on human assumptions, thus enhancing the model's ability to represent real physical and operational relationships.

[0077] In this embodiment of the application, before performing weather forecasting based on the constructed graph neural network and current short-term meteorological data in step S300, the following steps B1-B3 are also included:

[0078] B1: Establish a dual clock system consisting of a hardware clock and a logic clock, where the hardware clock is used to record the timestamps of the raw data acquisition.

[0079] B2: The logic clock establishes a confidence interval based on the delay distribution corresponding to the graph neural network;

[0080] Specifically, the hardware clock is the recording time of the raw data, while the logical clock is used to describe the network latency of the nodes in the graph neural network and establish a confidence interval.

[0081] B3: Additionally, by using a spatial encoder based on a convolutional neural network, discrete meteorological observation data can be upscaled to the same dimension as satellite meteorological cloud images.

[0082] For example, discrete meteorological observation data can be non-continuously collected data such as wind force level and rainfall, and the discrete data can be upgraded.

[0083] In this embodiment of the application, step S300 further includes step B4:

[0084] B4: Map the upgraded meteorological observation data, satellite meteorological cloud images, reservoir water levels, and geographical location data to a unified spatiotemporal coordinate system;

[0085] In this process, geographic location data and hardware clocks are mapped to a high-dimensional feature space through a learnable hash function.

[0086] For example, if the reservoir water level is collected discretely, it can be fitted and then upgraded to the same dimension as the satellite meteorological cloud image before the data is mapped to a unified spatiotemporal coordinate system.

[0087] In this embodiment of the application, step S300 establishes a confidence interval for the logic clock based on the delay distribution corresponding to the graph neural network, including the following steps C1-C3:

[0088] C1: Add physical connection tags based on graph neural networks. Physical connection tags are used to identify the data transmission relationship between nodes.

[0089] For example, physical connection markers can be added to the graph neural network for physical nodes such as reservoir water level sensors, meteorological observation stations, and data relay aggregation nodes that may exist.

[0090] C2: Based on the physical connection markers, establish cross-layer influence factors and use cross-layer influence factors to describe the intensity of the impact of the delay in meteorological observation data on reservoir water level;

[0091] It should be noted that the logical clock is mainly used to analyze the impact of meteorological station data delays on reservoir water level prediction.

[0092] For example, the cross-layer impact factor can be calculated based on the physical connection marker: ω = 1 / (transmission distance + 0.1 * hop count).

[0093] C3: Assign corresponding weights based on the established cross-layer influence factors and use them as confidence levels.

[0094] In this embodiment of the application, the forward prediction compensation of the satellite meteorological cloud image in step S300 specifically includes steps D1-D3:

[0095] D1: Sequentially record the acquired satellite meteorological cloud image data;

[0096] D2: Configure the delay step size for satellite meteorological cloud image data, and use the delay step size as a window to capture the acquired satellite meteorological cloud image data;

[0097] D3: Using the captured satellite meteorological cloud image data, predict the satellite meteorological cloud image after the delay step based on the LSTM model to perform forward compensation.

[0098] For example, if the current time is t and the cloud map data has a delay of Δ steps, then from k = t - Δ to k = t, the state prediction of Δ steps is performed. That is, the predicted cloud map is generated using the LSTM model at the target time t. When the actual cloud map data arrives, the residual is calculated and updated.

[0099] In an optional implementation, steps D1-D3 can also project satellite cloud image raster data onto the graph neural network nodes constructed in S200, generate corresponding local cloud image features for each wind, solar and water node, construct a spatiotemporal map sequence with a delay step size T as the window, and perform forward compensation by predicting the node-level cloud image features of the next TT steps through ST-GCN for use in Monte Carlo sampling in step S400.

[0100] In another alternative implementation, the forward compensation is extended to a probabilistic prediction model, directly outputting the distribution of future cloud maps to support the outer Monte Carlo sampling of the S400. Specifically, while recording historical cloud map sequences, their sources of uncertainty can be labeled, a conditional variational autoencoder (CVAE) can be trained, and N sets of possible future cloud maps can be sampled from the CVAE decoder, directly serving as the input scene for the outer Monte Carlo sampling.

[0101] It should be noted that the GNN model in step S300 utilizes the spatial correlations learned between nodes in step S200 for information transmission and aggregation. This allows it to capture regional meteorological evolution patterns more effectively than traditional single-point prediction methods, improving the prediction accuracy of key input variables such as wind speed, irradiance, and precipitation. By adding physical connection markers and establishing cross-layer influence factors, data latency, such as the uncertainty brought about by meteorological station data to reservoir scheduling decisions, is quantified. Confidence intervals are also established in the prediction, ensuring that the prediction results include not only point estimates but also reliability assessments.

[0102] In this embodiment of the application, step S400 constructs an assessment model based on a two-layer Monte Carlo-conditional risk. The outer layer of the assessment model samples the results of meteorological forecasts, while the inner layer models the random variables of water, wind, and solar power.

[0103] Specifically, the random variables used in the inner layer to model water, wind, and solar power include: equipment failure rates of water, wind, and solar power units and market electricity prices.

[0104] It should be noted that many units of equipment, especially those designed for wind turbines and photovoltaic fields, are greatly affected by meteorological factors. By designing a two-layer Monte Carlo-conditional risk assessment model, indicators such as equipment availability (failure rate) can be introduced for comprehensive evaluation.

[0105] In this embodiment of the application, step S500, which outputs the probability density function of the predicted output force based on the evaluation model, includes steps E1-E3:

[0106] E1: The market electricity price in the random variable is modeled using mean regression, and the failure rate in the random variable is randomly generated to determine the failure state.

[0107] For example, the stochastic process of market electricity prices can be described using a mean regression model as follows:

[0108] dλ t =κ(θ) t -λ t )d t +σ λ dW t

[0109] Where, θ t κ represents the long-term trend of electricity prices, and σ represents the regression rate. λ This represents volatility.

[0110] E2: Establish a power aggregation model for water, wind, and solar power based on the random generation of fault conditions and mean regression modeling;

[0111] Specifically, in the two-layer Monte Carlo framework, step E2 can be interacted with in the following ways:

[0112] Outer circulation (meteorological sampling): Input: meteorological ensemble forecast; Output: theoretical output of wind, solar and hydropower;

[0113] Inner loop (equipment and electricity price):

[0114] The equipment failure model is represented as follows:

[0115]

[0116] in, This indicates the available output of node k at time t. This represents the theoretical output at time t based on meteorological component i (e.g., wind and sunlight are different meteorological components) without considering the impact of faults. Let represent the failure rate of device j at node k at time t.

[0117] The stochastic process of electricity prices is represented as follows:

[0118] dλt =κ(θ) t -λ t )d t +σ λ dW t (k)

[0119] Where, λ t Let θ be the market electricity price at time t. t Let be the long-run equilibrium electricity price at time t, κ be the mean regression rate parameter, and σ be the mean. λ For electricity price volatility, dW t (k) Let be the Wiener process increment at node k at time t.

[0120] The power aggregation model of water, wind, and solar power satisfies:

[0121]

[0122] in, This represents the optimal scheduling output of node k in meteorological component i at time t. Let C be the actual electricity price at node k at time t. curt For the cost of energy curtailment, Let P be the available output of node k under meteorological component i, and let P be the dispatched output.

[0123] E3: Based on the power output aggregation model, the probability density function of the power output prediction is determined using kernel density estimation.

[0124] Specifically, a stochastic optimization model can be constructed based on the predicted probability density function PDF, and then optimization can be performed.

[0125] In an alternative implementation, step S500 may take into account the randomness of electricity prices and the uncertainty of power output to maximize expected revenue.

[0126] In another alternative implementation, step S500 can also control the conditional risk value under extreme scenarios to minimize risk loss, and perform output optimization by introducing a probability density function.

[0127] It should be noted that the PDF provides a complete probability distribution of the system's total future output, including not only expected values ​​but also fluctuation ranges and risk probabilities, such as the probabilities of low and high output, which is an important basis for optimization. Furthermore, using PDF information, it is possible to select an output scheme that achieves the optimal balance between expected cost and risk exposure. For example, risk-averse investors may require a sufficiently low probability of the left tail (low output) of the PDF.

[0128] In summary, this invention combines the spatial modeling capabilities of Graph Neural Networks (GNNs), the comprehensive uncertainty sampling of Monte Carlo simulations, and the information condensation capabilities of Kernel Density Estimation (KDE). It comprehensively quantifies the risks of hierarchical modeling of meteorological forecast errors, power output randomness, market fluctuations, and equipment failures. This allows the optimization to focus on controlling the probability and impact of extreme adverse situations. Ultimately, the output probability density function provides scheduling decision-makers with the most comprehensive risk and opportunity information, supporting more robust and better-matched optimized scheduling schemes that match decision-makers' risk preferences, thereby improving the reliability and economy of hydro-wind-solar hybrid systems.

[0129] Example 3 illustrates a schematic scheme for an output optimization method under complementary scheduling. It should be noted that the technical solution of this output optimization system under complementary scheduling is based on the same concept as the technical solution of the output optimization method under complementary scheduling described above. Details not described in detail in this example can be found in the description of the technical solution of the output optimization method under complementary scheduling described above.

[0130] This embodiment also provides another output optimization system under complementary scheduling, including:

[0131] The acquisition module is used to acquire geographical location data, corresponding historical meteorological data, and historical power generation data of cascade hydropower, wind farms, and photovoltaic power plants;

[0132] The first building module is used to construct a graph neural network based on geographic location data and configure the weights of the edges according to historical power generation data. The nodes of the graph neural network represent wind turbines, photovoltaic arrays and hydropower stations in different geographic locations.

[0133] The prediction module is used to make weather forecasts based on the constructed graph neural network and current short-term meteorological data.

[0134] The second building module is used to construct an assessment model based on a two-layer Monte Carlo-conditional risk model. The outer layer of the assessment model samples the results of meteorological forecasts, while the inner layer models the random variables of water, wind, and solar power.

[0135] The output execution module is used to output the probability density function of the power output prediction based on the evaluation model, and to perform power output optimization based on the probability density function.

[0136] This embodiment also provides a computer device suitable for output optimization under complementary scheduling, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the output optimization method under complementary scheduling as proposed in the above embodiment.

[0137] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the output optimization method under complementary scheduling as proposed in the above embodiments.

[0138] The storage medium proposed in this embodiment and the output optimization method under complementary scheduling proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0139] Based on the above description of the implementation methods, those skilled in the art will clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.

[0140] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for optimizing output under complementary scheduling, characterized in that, include: Obtain geographical location data, corresponding historical meteorological data, and historical power generation data for cascade hydropower, wind farms, and photovoltaic power plants; A graph neural network is constructed based on geographic location data, and the weights of the edges are configured according to historical power generation data. The nodes of the graph neural network represent wind turbines, photovoltaic arrays, and hydropower stations in different geographic locations. Based on the constructed graph neural network, weather forecasts are made using current short-term meteorological data; A two-layer Monte Carlo-conditional risk assessment model is constructed, wherein the outer layer of the assessment model samples the results of meteorological forecasts, and the inner layer models the random variables of water, wind, and solar power. The output power prediction probability density function is output based on the evaluation model, and output power optimization is performed based on the probability density function.

2. The output optimization method under complementary scheduling as described in claim 1, characterized in that, The construction of a graph neural network based on geographic location data, and the configuration of edge weights according to historical power generation data, includes: In the graph neural network, nodes with related relationships are grouped into a node group, and classification is performed based on the node group. Based on the classification results, calculate the terrain similarity between the terrain relationships covered by each group of nodes in any node group; The weights of the graph neural network edges are configured based on the terrain similarity between nodes, the geographical distance between nodes, and historical relevance.

3. The output optimization method under complementary scheduling as described in claim 2, characterized in that, Prior to the aforementioned weather forecast, the following is also included: A dual-clock system consisting of a hardware clock and a logic clock is established, wherein the hardware clock is used to record the timestamp of the original data acquisition. The logical clock establishes a confidence interval based on the delay distribution corresponding to the graph neural network; and, using a spatial encoder based on a convolutional neural network, the discrete meteorological observation data is upgraded to the same dimension as the satellite meteorological cloud image.

4. The output optimization method under complementary scheduling as described in claim 3, characterized in that, Also includes: Upgraded meteorological observation data, satellite meteorological cloud images, reservoir water levels, and geographical location data are mapped to a unified spatiotemporal coordinate system; In this process, geographic location data and hardware clocks are mapped to a high-dimensional feature space through a learnable hash function.

5. The output optimization method under complementary scheduling as described in claim 4, characterized in that, The logical clock establishes a confidence interval based on the delay distribution corresponding to the graph neural network, including: Physical connection markers are added based on graph neural networks; these physical connection markers are used to identify the data transmission relationships between nodes. Based on the physical connection markers, a cross-layer influence factor is established, and the cross-layer influence factor is used to describe the intensity of the impact of the delay in meteorological observation data on the reservoir water level. The established cross-layer influence factors are assigned corresponding weights and used as confidence levels.

6. The output optimization method under complementary scheduling as described in claim 5, characterized in that, The aforementioned forward prediction compensation for satellite meteorological cloud images specifically includes: Sequentially record the acquired satellite meteorological cloud image data; Configure the delay step size for satellite meteorological cloud image data, and use the delay step size as a window to capture the acquired satellite meteorological cloud image data; Using captured satellite meteorological cloud image data, a satellite meteorological cloud image with a delay step size is predicted based on an LSTM model to perform forward compensation.

7. The output optimization method under complementary scheduling as described in claim 6, characterized in that, The probability density function for predicting the output force based on the evaluation model includes: The market electricity price in the random variable is modeled using mean regression, and the failure rate in the random variable is randomly generated based on the failure state. A power aggregation model for water, wind, and solar power is established based on the random generation of fault states and mean regression modeling. Based on the power output aggregation model, the probability density function for power output prediction is determined using kernel density estimation.

8. A power output optimization system under complementary scheduling, using the method described in any one of claims 1-7, characterized in that, include: The acquisition module is used to acquire geographical location data, corresponding historical meteorological data, and historical power generation data of cascade hydropower, wind farms, and photovoltaic power plants; The first construction module is used to construct a graph neural network based on geographic location data and configure the weights of the edges according to historical power generation data. The nodes of the graph neural network represent wind turbines, photovoltaic arrays and hydropower stations in different geographic locations. The prediction module is used to make weather forecasts based on the constructed graph neural network and current short-term meteorological data. The second building module is used to build an assessment model based on a two-layer Monte Carlo-conditional risk. The outer layer of the assessment model samples the results of meteorological forecasts, and the inner layer models the random variables of water, wind, and solar power. The output execution module is used to output a probability density function of the predicted output based on the evaluation model, and to perform output optimization based on the probability density function.

9. A computer device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions, which, when executed by the processor, implement the steps of the output optimization method under complementary scheduling as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the output optimization method under complementary scheduling as described in any one of claims 1 to 7.

Images