A power grid daily scale scheduling method, device and medium

Abstract

The application relates to the field of power systems, and discloses a power grid daily scale scheduling method, equipment and medium. The method comprises the following steps: acquiring future time sequence data of a plurality of new energy stations; obtaining a Pearson correlation coefficient matrix according to the future time sequence data, so as to reflect the output relationship between the new energy stations; obtaining a new energy operation scenario set according to the future time sequence data and taking the Pearson correlation coefficient matrix as a constraint; and sequentially performing multi-day cycle scheduling optimization and single-day scheduling optimization according to the new energy operation scenario set, so as to obtain a plurality of single-day scheduling plans, thereby obtaining a multi-day scheduling plan. The application accurately depicts the multi-day evolution law, and leads to inaccuracy of the generated multi-day collaborative scheduling plan.

Description

Technical Field

[0001] This invention relates to the field of power systems, and more particularly to a daily-scale dispatching method, equipment, and medium for power grids. Background Technology

[0002] With the increasing penetration of high-proportion renewable energy sources in the power grid, power system dispatch and operation face new challenges. Traditional dispatch methods are significantly insufficient in dealing with the strong uncertainty and complex spatiotemporal correlations of renewable energy sources: the output correlation between renewable energy power plants is highly nonlinear and time-varying, and is significantly affected by weather system evolution, geographical factors, etc. Traditional linear statistical models are unable to accurately characterize its multi-day evolution patterns, resulting in inaccurate multi-day coordinated dispatch plans. Summary of the Invention

[0003] This invention aims to at least solve one of the technical problems existing in the prior art. To this end, this invention proposes a daily-scale scheduling method for power grids, which can accurately characterize its multi-day evolution pattern, thus preventing the generation of inaccurate multi-day coordinated scheduling plans.

[0004] The present invention also proposes equipment and media having the above-mentioned daily-scale power grid scheduling method.

[0005] A power grid daily-scale scheduling method according to a first aspect of the present invention includes: Obtain future time-series data for several new energy power plants; Based on the future time series data, the Pearson correlation coefficient matrix is ​​obtained to reflect the power output relationship between various new energy power plants. Using the Pearson correlation coefficient matrix as a constraint, a set of new energy operation scenarios is obtained based on the future time series data; Based on the set of new energy operation scenarios, multi-day cycle scheduling optimization and single-day scheduling optimization are performed sequentially to obtain several single-day scheduling plans, thereby obtaining a multi-day scheduling plan.

[0006] According to an embodiment of the present invention, a daily-scale grid dispatching method has at least the following beneficial effects: The present application essentially generates a grid dispatching plan for a grid with a high proportion of renewable energy generation. First, it acquires future time-series data of renewable energy power plants to obtain highly correlated renewable energy power plants, integrating complex spatiotemporal correlations into the dispatching process. Then, based on this, it generates multiple renewable energy operation output scenarios to accurately simulate the renewable energy output situation. Finally, based on the generated renewable energy output scenarios, it performs dispatching optimization to obtain a daily dispatching plan, thereby completing grid dispatching on multiple consecutive daily scales. The present invention accurately depicts the multi-day evolution pattern, leading to inaccuracies in the generated multi-day collaborative dispatching plan.

[0007] According to some embodiments of the present invention, obtaining the Pearson correlation coefficient matrix based on the future time-series data to reflect the power output relationship among various renewable energy power plants includes: Several data windows are continuously extracted from the future time series data to obtain the multi-day power generation points of each new energy power station corresponding to a certain data window; The intraday power output points are encoded into a two-dimensional matrix image; Feature extraction is performed on the two-dimensional matrix image to obtain the Pearson correlation coefficient matrix.

[0008] According to some embodiments of the present invention, the step of extracting features from the two-dimensional matrix image to obtain the Pearson correlation coefficient matrix includes: The two-dimensional matrix images corresponding to each new energy power station in the same data window are stitched together; The pseudo-covariance matrix is ​​obtained by sequentially mapping and outer productting the spliced ​​feature vectors of multiple stations. The pseudo-covariance matrix is ​​normalized to obtain the Pearson correlation coefficient matrix.

[0009] According to some embodiments of the present invention, encoding the intraday power output points into a two-dimensional matrix image includes: Normalize the intraday output points to obtain a normalized sequence; Mapping the normalized sequence to the polar coordinate system yields a set of polar coordinates; Calculate the angles between the polar coordinates in the polar coordinate set to obtain the Gram matrix, which is a two-dimensional matrix image.

[0010] According to some embodiments of the present invention, obtaining a set of new energy operation scenarios based on the future time-series data, constrained by the Pearson correlation coefficient matrix, includes: Using the Pearson correlation coefficient matrix as a constraint, an inner-loop conditional generation network is trained based on historical meteorological data from the acquired historical time-series data; The network is trained to optimize the outer loop conditions based on the historical daily profiles generated by the network output under the inner loop conditions. The future time series numbers are sequentially input into the inner loop condition generation network and the outer loop condition optimization network to obtain a set of new energy operation scenarios.

[0011] According to some embodiments of the present invention, the inner loop condition generation network includes: an inner loop generator, which adopts an encoder-decoder structure, inputs random noise vector, meteorological data, and Pearson correlation coefficient matrix into the inner loop generator, and outputs a daily profile; The inner discriminator receives the daily profile output by the inner loop generator and determines whether the daily profile conforms to historical new energy scenarios, thereby driving the optimization of the inner loop generator.

[0012] According to some embodiments of the present invention, the outer loop condition optimization network includes: an outer loop generator for performing time-series correction on the daily profile output by the inner loop condition generation network; The outer discriminator is used to extract the temporal features of the outer loop generator's output and determine whether the outer loop generator's output conforms to historical new energy scenarios.

[0013] According to some embodiments of the present invention, the step of sequentially performing multi-day cycle scheduling optimization and single-day scheduling optimization based on the new energy operation scenario set to obtain several single-day scheduling plans includes: With the goal of minimizing the total cost over a multi-day period, the daily boundary conditions for a certain day are obtained based on the set of new energy operation scenarios and the initial state of each new energy power station. Based on the daily boundary conditions, and with the goal of minimizing the daily operating cost, the daily scheduling plan and its operating results are optimized. Update the daily boundary conditions for the next day based on the daily scheduling plan and its operation results.

[0014] An electronic device according to a second aspect of the present invention includes: Memory, used to store programs; A processor for executing a program stored in the memory, wherein when the processor executes the program stored in the memory, the processor is configured to perform the method as described in any one of the first aspects.

[0015] According to a third aspect of the present invention, a storage medium stores computer-executable instructions for performing the method as described in any one of the first aspects.

[0016] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing 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 description, claims, and drawings. Attached Figure Description

[0017] The accompanying drawings are provided to further understand the technical solutions of the present invention and constitute a part of the specification. They are used together with the embodiments of the present invention to explain the technical solutions of the present invention, and do not constitute a limitation on the technical solutions of the present invention.

[0018] Figure 1 This is a flowchart of a daily-scale power grid scheduling method provided in an embodiment of the present invention. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0020] It should be understood that in the description of the embodiments of the present invention, "multiple" (or "amounts") means two or more, "greater than," "less than," and "exceeding" are understood to exclude the stated number, while "above," "below," and "within" are understood to include the stated number. If "first," "second," etc., are used in the description, they are only for the purpose of distinguishing technical features and should not be construed as indicating or implying relative importance, or implicitly indicating the number of indicated technical features, or implicitly indicating the order of the indicated technical features.

[0021] like Figure 1 As shown, this embodiment of the invention provides a daily-scale power grid scheduling method, including: Step S100: Obtain future time-series data of several new energy power stations; Step S200: Based on future time series data, obtain the Pearson correlation coefficient matrix to reflect the power output relationship between various new energy power plants; Step S300: Using the Pearson correlation coefficient matrix as a constraint, obtain the set of new energy operation scenarios based on future time series data; Step S400: Based on the set of new energy operation scenarios, perform multi-day cycle scheduling optimization and single-day scheduling optimization in sequence to obtain several single-day scheduling plans, thereby obtaining a multi-day scheduling plan.

[0022] This application essentially generates a grid dispatch plan for a power grid with a high proportion of renewable energy generation. First, it acquires future time-series data of renewable energy power plants to identify highly correlated plants, integrating complex spatiotemporal correlations into the dispatch process. Then, based on this data, it generates multiple renewable energy operation output scenarios, accurately simulating the renewable energy output conditions. Finally, based on these generated renewable energy output scenarios, it optimizes the dispatch to obtain a daily dispatch plan, thereby completing grid dispatch on multiple consecutive daily scales. This invention accurately depicts the multi-day evolution pattern, leading to inaccuracies in the generated multi-day collaborative dispatch plan.

[0023] It should be noted that the future time series data is obtained through prediction using existing technologies.

[0024] In one embodiment, the future time-series data includes: new energy output data: the actual active power output of each wind farm and photovoltaic power station; meteorological data: numerical weather forecast data of the location of each station, including wind speed, wind direction, irradiance, temperature, humidity, and cloud cover; and spatiotemporal context data: calendar information (season, month, weekday / weekend / holiday) and station geographical coordinates (latitude and longitude). After acquiring future time-series data, the data is cleaned, including handling missing values ​​and removing obvious outliers (using time series linear interpolation to fill in continuous missing data of no more than 2 hours; for missing data of longer periods, the data for that day is directly removed; data points with power values ​​exceeding 120% of rated capacity or below -5% in reverse power transmission are removed); all multi-source data are aligned according to a unified timestamp and resampled to a common time resolution (e.g., 1 hour) to form a multi-dimensional spatiotemporal dataset for subsequent processing.

[0025] In one embodiment, in step S200, a Pearson correlation coefficient matrix is ​​obtained based on future time-series data to reflect the power output relationship between various renewable energy power plants, including: By continuously extracting several data windows from future time-series data, the multi-day power generation points of each new energy power station corresponding to a certain data window can be obtained. Encode the intraday power output points into a two-dimensional matrix image; Feature extraction is performed on the two-dimensional matrix image to obtain the Pearson correlation coefficient matrix.

[0026] Specifically, based on future time series data, a Pearson correlation coefficient matrix is ​​obtained to reflect the power output relationship between various new energy power plants. This includes: sliding out a data window of D consecutive days (D≥3) from the preprocessed future time series data, with each window constituting a sample; for each sample, extracting the power output sequence of each new energy power plant for the next D days (D+1 to D+D days) to form a set of multi-day power output curves. The Gramian Angular Field (GAF) method is used to encode each value point (intraday output point, such as the 24-hour output value of a single station) in the one-dimensional time series data, i.e., the output series, into a two-dimensional matrix image. This transformation maps the time series data to the polar coordinate system and calculates the angle and / or cosine values, transforming the temporal dependence and dynamic features of the time series into spatial texture and structural information in the image, providing a suitable input format for subsequent convolutional neural network processing.

[0027] In one embodiment, feature extraction of a two-dimensional matrix image includes: The two-dimensional matrix image is input into a pre-trained parametric gaze model for feature weighting and denoising; the parametric gaze model is a trainable neural network layer, the core of which is a parameter matrix. (For example: for a 72x72 two-dimensional matrix image, Wa is a trainable 72x72 matrix); for a two-dimensional matrix image I, the output after the parameter staring layer is a weighted feature map.

[0028]

[0029] in This indicates element-wise multiplication. For example, the activation function (such as the Sigmoid function); Through training, the parameter matrix The model learns to focus on different image regions, thereby guiding it to automatically focus on key patterns most relevant to multi-day correlations (e.g., patterns that characterize the coordinated increase or decrease in output of multiple stations under the influence of a specific weather system), while suppressing unimportant or noisy information. The image output from the parametric gaze model is input into a pre-trained convolutional neural network (CNN). The CNN's basic architecture employs a network structure with four convolutional blocks, ensuring depth of feature extraction while controlling computational complexity. Each convolutional block contains: a convolutional layer (using a 3x3 kernel), a ReLU activation function layer, and a 2x2 max-pooling layer. After feature extraction from the four convolutional blocks, the network is connected to a fully connected layer with 512 neurons for advanced feature integration. In one embodiment, the following settings are adopted for efficient training of a convolutional neural network (CNN): Optimizer: The adaptive moment estimation optimizer (Adam) is used, which can effectively handle sparse gradients and adjust the learning rate; Initial learning rate: set to 0.001, and dynamically reduce the learning rate during the plateau period of training loss using the "Reduce LR On Plateau" strategy, for example, by multiplying it by a factor of 0.1; Batch size: Set to 32, which means that 32 samples are used to calculate the gradient each time the parameters are updated, in order to strike a balance between training stability and efficiency; Loss function: The mean squared error loss function is used to directly optimize the difference between the correlation matrix predicted by the model and the true correlation matrix; Training cycles: Typically, training is conducted for 100 to 200 cycles, and the criterion for early stopping is that the loss on the validation set no longer decreases significantly.

[0030] In one embodiment, feature extraction is performed on a two-dimensional matrix image to obtain a Pearson correlation coefficient matrix, including: The two-dimensional matrix images corresponding to each new energy power station in the same data window are stitched together; The pseudo-covariance matrix is ​​obtained by sequentially mapping and outer productting the spliced ​​feature vectors of multiple stations. The pseudo-covariance matrix is ​​normalized to obtain the Pearson correlation coefficient matrix.

[0031] Specifically, the image feature vectors of different renewable energy power stations under the same data window are concatenated, and the fused multi-station features are mapped into a standard Pearson correlation coefficient matrix. The specific steps are as follows: Let the fused multi-site feature vector be... Where L is the dimension of the feature vector. First, the vector z is mapped to a new vector space with a dimension equal to the number of renewable energy power stations N, using a trainable linear weight matrix with no bias term:

[0032] in, The purpose of this transformation is to project the high-dimensional fusion features into a low-dimensional representation that is directly related to the number of stations, which contains the potential correlation information between stations, as a trainable weight matrix. Subsequently, the transformed vector Perform its own outer product operation to generate a matrix :

[0033] At this time, the matrix elements in It can be regarded as the unnormalized correlation strength between station i and station j; The pseudo-covariance matrix Standardization transforms it into a standard Pearson correlation coefficient matrix where the main diagonal elements are 1 and the off-diagonal elements range from -1 to 1. Specifically, this includes: Pearson correlation coefficient The definition is the product of the covariance of two variables divided by their respective standard deviations. In a matrix... In the middle, the elements on the main diagonal This represents the variance estimate for the i-th station. Therefore, the standard deviation for each station... ; Calculate a standard deviation product matrix D, where each element... Finally, the Pearson correlation coefficient matrix R is obtained by dividing matrix S by matrix D element by element.

[0034] Matrix R is the Pearson correlation matrix; Output a Dynamic Pearson correlation coefficient matrix R Where N is the number of renewable energy power stations, and D is the number of look-ahead days (i.e., the number of days scheduled); the elements in matrix R... The strength and direction of the linear correlation between the power output of station i and station j on the d-th look-ahead day (D+d) were quantified. The mean squared error (MSE) is used as the loss function to minimize the correlation coefficient matrix predicted by the model. Compared with the label matrix calculated based on historical real data The differences between them.

[0035] It is easy to understand that the data used for training and validating the above model is historical time-series data, that is, historical time-series data of each new energy power station corresponding to the future time-series data.

[0036] In one embodiment, encoding intraday power output points into a two-dimensional matrix image includes: Normalize the intraday output points to obtain a normalized sequence; Mapping the normalized sequence to the polar coordinate system yields a set of polar coordinates; Calculate the angles between the polar coordinates in the polar coordinate set to obtain the Gram matrix, which is a two-dimensional matrix image.

[0037] Specifically, the intraday power output points are encoded into a two-dimensional matrix image, including: Preprocessing and normalization of new energy power output time series data: From historical databases or forecasting systems, extract the active power output sequence of the target new energy power station over several consecutive days (D days), denoted as the original one-dimensional time series X={x_1,x_2,...,x_n}, where n is the total length of the sequence (e.g., if D=3 days and the time resolution is 1 hour, then n=72); To eliminate the differences in the output dimensions of different power stations and meet the requirements for Gram angle field calculation, perform minimum-maximum normalization on the sequence X, scaling it to the interval [-1,1]:

[0038] in, The value is the normalized value; Mapping the normalized sequence to polar coordinates aims to transform one-dimensional time series points into geometric points in polar coordinates, achieving a conversion from the "time-power" domain to the "radius-angle" domain: each normalized power value... Mapped to polar angle ,

[0039] in ∈[0,π]; This mapping ensures that each power value uniquely corresponds to an angle, and the larger the power value, the smaller the angle; Timestamp Mapping to Radius: Map the time index i of each data point to the polar radius. To preserve timing information:

[0040] Where n is the total length of the sequence, this operation makes the polar radius increase over time, thereby encoding the temporal order in the radial dimension of polar coordinates; Calculate the Gram matrix to generate a 2D image: based on a set of polar coordinates By calculating the trigonometric function values ​​between each pair of angles, a Gram matrix (i.e., a two-dimensional matrix image) that can characterize the intrinsic relationship of the time series is constructed: the cosine value of the angle sum is calculated to generate the matrix GASF.

[0041] The elements on the main diagonal (i=j) of this matrix are It can effectively preserve the absolute numerical information of time series; The final GASF matrix is (For example, if D=3 days and the time resolution is 1 hour, then n=72) is a symmetric real matrix. This matrix can be viewed as a grayscale image, where: the image coordinates (i, j) correspond to the original time points i and j, and the pixel values ​​correspond to the angular relationship strength of their time points in polar coordinates.

[0042] This image transforms the temporal dependence and numerical correlation of new energy output into a spatial structural feature, thereby transforming the time series problem, which was originally suitable for processing by recurrent neural networks (RNN), into an image recognition problem that is more suitable for processing by parametric staring-convolutional neural networks (CNN) in this invention. This lays a key technical foundation for subsequent in-depth mining of multi-day correlations of new energy.

[0043] In one embodiment, in step S300, using the Pearson correlation coefficient matrix as a constraint, a set of new energy operation scenarios is obtained based on future time-series data, including: Using the Pearson correlation coefficient matrix as a constraint, an inner-loop conditional generative network is trained based on historical meteorological data from the acquired historical time-series data. The network is trained to optimize the outer loop conditions based on the historical daily profiles generated by the network output under the inner loop conditions. By sequentially inputting future time series numbers into the inner loop condition generation network and the outer loop condition optimization network, a set of new energy operation scenarios is obtained.

[0044] In one embodiment, the inner loop condition generation network includes: an inner loop generator, which adopts an encoder-decoder structure, inputs random noise vector, meteorological data, and Pearson correlation coefficient matrix into the inner loop generator, and outputs a daily profile; The inner discriminator receives the daily profile output by the inner loop generator and determines whether the daily profile conforms to historical new energy scenarios, thereby driving the optimization of the inner loop generator.

[0045] Specifically, the inner-loop conditional generative adversarial network (cGAN) is the inner-loop conditional generative network architecture design: a. Inner loop generator (G) inner Design: The generator adopts an encoder-decoder structure, and its input includes: a random noise vector z sampled from a standard normal distribution to ensure the diversity of generated scenes; and a meteorological condition vector. Meteorological data, including numerical weather forecast information such as wind speed, irradiance, and temperature for the target day; and a relevant context vector. The Pearson correlation coefficient matrix is ​​used to embed the spatiotemporal correlation features between power stations; the generator output is a complete daily profile of new energy output, load demand and energy storage status, with a time resolution of 15 minutes.

[0046] b. Internal discriminator (D) inner Design: The discriminator is a convolutional neural network used to determine whether the input daily profile data comes from a real historical dataset or is generated by the generator G. inner The goal is to accurately distinguish between true and false data, thereby driving generator optimization. The training process of the inner-loop conditional generation network includes: An adversarial training mechanism is used to optimize the following loss function:

[0047] in, This represents the mathematical expectation of the variable x. For historical new energy scenarios in real historical time-series data, x is the sampled value of the real data distribution, and C is the condition vector. , This represents the estimated probability that the internal discriminator determines input data x to be true data under given condition C. Similarly, z represents the mathematical expectation, derived from a simple prior distribution. The noise vector obtained by random sampling in the middle, The inner loop generator fabricates a single-day scene based on random noise z and condition C. The internal discriminator performs a authenticity check on this fake scene and outputs the probability that it considers it "real". Through the game between the inner loop generator and the inner discriminator, G is made inner It learns the ability to generate statistically realistic single-day scenes given meteorological and related backgrounds.

[0048] It is easy to understand that meteorological data includes future meteorological data and historical meteorological data.

[0049] In one embodiment, the outer loop condition optimization network includes: an outer loop generator for performing time-series correction on the daily profile output by the inner loop condition generation network; The outer discriminator is used to extract the temporal features of the outer loop generator's output and determine whether the outer loop generator's output conforms to historical new energy scenarios.

[0050] Specifically, the inner loop generator G inner Independently generated, consecutive D daily scene profiles By piecing together the beginning and end in chronological order, an initial and complete draft of the Day D scene is formed. .

[0051] Outer-loop adversarial coherence optimization networks, also known as outer-loop conditional optimization networks, include: a. Outer Loop Generator (G) outer Design: This generator acts as a "timing corrector," taking the concatenated draft as its input. The system is fine-tuned through learning using a one-dimensional convolutional network. The focus is on optimizing the power curve at the daily transition points, eliminating unreasonable jumps, and ensuring that the multi-day series conforms to physical laws in its overall dynamic evolution (such as changes in energy storage SOC and load cycle patterns). b. External discriminator (D) outer Design: This discriminator has a larger temporal receptive field to distinguish whether the entire D-day long sequence is real long-term historical data or generated by the generator G. outer The corrected results. It focuses on assessing the long-term temporal coherence of the series, such as the natural transition of weather processes and the smooth switching between weekend and weekday sunshine load patterns.

[0052] The discrimination mechanism is as follows: First, D outer The input sequence is processed through a temporal feature extraction module (typically composed of a multi-layer one-dimensional dilated convolutional network or a Transformer encoder). This structure enables it to capture long-range dependencies spanning multiple days. Next, the extracted deep features are fed into a discriminative scoring module, which comprehensively evaluates the sequence's coherence across multiple dimensions, including but not limited to: the natural smoothness of weather system evolution (such as wind speed and irradiance), the inter-diurnal spatiotemporal correlation of renewable energy output fluctuations, and the regularity and smoothness of load pattern switching (especially between weekends and weekdays). Finally, D... outer Output a discrimination probability, quantifying the likelihood that the sequence is real historical data.

[0053] The processing flow for the discrimination results is as follows: Loss Calculation and Feedback: D outer The discriminant output is compared with the real label (real data is 1, generated data is 0), and the adversarial loss (such as standard cross-entropy loss or Wasserstein distance) is calculated. Gradient backpropagation: The gradient generated by this loss function propagates in both directions simultaneously. Update D outer It itself: updates its parameters through backpropagation to improve its ability to distinguish between genuine and fake sequences; Guide G outer Optimization: Gradient through G outer Further backpropagation provides a clear direction for optimization; G outer The goal is to generate a generator capable of "deceiving" D outer The sequence is such that the generated sequence is forced to continuously approximate the real data in terms of long-term coherence; Closed-loop iterative optimization: The above process constitutes an adversarial training loop. Through repeated iterations, G... outer When correcting the draft of a single-day profile, long-term consistency constraints will be increasingly consciously incorporated to output statistically realistic and time-series-reasonable multi-day scenarios; ultimately, when D outer When it becomes difficult to distinguish between the real sequence and the generated sequence, the system reaches a dynamic equilibrium, indicating that G... outer It already has the capability to generate high-quality multi-day scenes.

[0054] It should be noted that the outer loop uses adversarial training, and its objective function is similar to that of the inner loop, but the object of the operation is a multi-day long sequence. The outer discriminator D... outer Drive external generator G outer Generate scenes that are equally realistic and coherent across multiple day scales.

[0055] In one embodiment, the training process of the outer-loop conditional optimization network and the inner-loop conditional generation network includes: Inner loop pre-training phase: First, fix the outer loop network, and use only a large amount of daily historical data to train the inner loop's G. inner and D inner Train it thoroughly until it can consistently generate high-quality daily profiles; Outer loop optimization phase: Fix the pre-trained G inner Using multi-day continuous historical data sequences, the outer ring G outer and D outer Train it to focus on learning multi-day splicing and coherence optimization; Global fine-tuning phase: Remove all fixes and fine-tune the four networks of inner and outer loops together to achieve global optimization of daily generation quality and multi-day consistency.

[0056] After training, inputting random noise vectors and future weather forecast data into the aforementioned dual-ring adversarial network can efficiently generate a large number of multi-day new energy operation scenarios that conform to both short-term statistical distribution and long-term temporal continuity.

[0057] In one embodiment, in step S400, based on the new energy operation scenario set, multi-day cycle scheduling optimization and single-day scheduling optimization are performed sequentially to obtain several single-day scheduling plans, including: With the goal of minimizing the total cost over a multi-day period, the daily boundary conditions for a given day are obtained based on the set of new energy operation scenarios and the initial state of each new energy power station. Based on the daily boundary conditions, and with the goal of minimizing the daily operating cost, the daily scheduling plan and its operating results are optimized. Update the daily boundary conditions for the next day based on the daily scheduling plan and its operation results.

[0058] Specifically, in step S400, based on the new energy operation scenario set, multi-day cycle scheduling optimization and single-day scheduling optimization are performed sequentially to obtain several single-day scheduling plans, including: (1) Hierarchical optimization modeling: The multi-day joint optimization problem is decomposed into two levels: the outer multi-day slow dynamic optimization and the inner single-day fast response optimization. The balance between computational complexity and optimization accuracy is achieved through hierarchical iteration.

[0059] (1.1) Outer multi-day slow dynamic optimization layer a. Optimization Objective: The core objective is to minimize the total operating cost over the entire scheduling cycle (e.g., 3-7 days). The total cost mainly includes the fuel cost of thermal power units, the start-up and shutdown costs of the units, the penalty costs for curtailment of wind and solar power, and the compensation costs for interruptible loads.

[0060] b. Decision Variables and Cycle: Optimizing "slow-dynamic" resources where decisions change slowly, with a decision cycle of "day" granularity. Specific variables include: Hydropower plant: Reservoir water level at the end of each day or planned water consumption for power generation per day.

[0061] Thermal power units: Daily coal storage and weekly start-up and shutdown plans (for units with long start-up and shutdown cycles).

[0062] Cross-day energy storage: The state of charge (SOC) of an energy storage device (such as pumped hydro storage) at the end of each day.

[0063] c. Constraints: These include dynamic balance constraints for slow-dynamic resources (such as reservoir water balance and coal yard inventory balance) and total resource constraints on a weekly or monthly basis (such as total power generation and total coal consumption).

[0064] (1.2) Inner layer daily rapid response optimization layer A. Optimization objective: Under the given daily boundary conditions of the outer layer, perform refined intraday economic scheduling with the goal of minimizing daily operating costs.

[0065] B. Decision Variables and Cycle: Optimizing fast-responding "fast-dynamic" resources, with a decision cycle of 15 minutes. Specific variables include: Unit Combination and Economic Dispatch: Start-up and shutdown of conventional units (for units with rapid start-up and shutdown) and real-time output planning.

[0066] Energy storage systems: intraday charge and discharge plans for chemical energy storage and pumped hydro storage.

[0067] Load response: The daily response of interruptible and adjustable loads.

[0068] C. Constraints: These include real-time power balance constraints, unit ramp-up rate constraints, network power flow security constraints, and energy storage intraday circulation constraints.

[0069] (2) Adaptive stitching and boundary smoothing mechanism (2.1) Iterative Interaction Process A. Outer layer initialization and information transmission: The outer layer performs the first multi-day optimization based on the multi-day new energy scenario set and the initial state of the system (such as the initial water level of the reservoir), and obtains the first round of daily boundary condition sequence B^outer={B_1,B_2,...,B_D} (for example, the water level target and coal storage target at the end of each day on day D), and transmits it to the inner layer.

[0070] B. Inner Layer Daily Optimization and Feedback: The inner layer receives the boundary conditions from the outer layer and performs detailed intraday optimization for each day, starting from the first day. After optimization, the inner layer feeds back the actual operating results of the day (including: actual total cost of the day, actual operating status of the unit, and actual SOC curve of the energy storage) to the outer layer.

[0071] C. Outer Layer Correction and Re-optimization: The outer layer compares the actual operating data fed back from the inner layer with its own expectations. Based on the differences, the outer layer adaptively corrects the boundary conditions for subsequent days (for example, if the actual hydropower generation on the first day is less than expected, the lower limit of thermal power output needs to be increased on the second day, and the subsequent dispatch lines of the reservoir need to be adjusted accordingly), and performs a new round of multi-day optimization. This process constitutes a closed-loop iteration.

[0072] (2.2) State memory and boundary smoothing A. State Memory Unit: The algorithm sets up a memory unit to store the decision history of the inner and outer layers, the sequence of boundary conditions, and the objective function value in each iteration.

[0073] B. Adaptive splicing strategy: When initializing a new round of iterations, the initial points of inner and outer layer optimizations are intelligently set using historical information in the memory unit. In particular, for the transition points between days, by introducing smoothing terms or constraints (such as adding a secondary penalty term to the objective function for changes in unit output or energy storage SOC at the transition point between adjacent days), it is ensured that the power output and energy storage status of the optimization plans for the two days do not undergo drastic changes at the transition point (such as midnight of each day), thus achieving "seamless splicing".

[0074] (3) Convergence determination and scheduling plan output (3.1) Convergence criterion: Set a dual convergence criterion to ensure the stability and quality of the solution.

[0075] Objective function change criterion: In two consecutive iterations, the absolute value of the change in the total cost of the outer optimization is less than a preset threshold. _cost; Boundary condition stability criterion: In two consecutive iterations, the norm of the change in the key boundary conditions (such as the daily water level and coal storage) output by the outer layer is less than a preset threshold. The algorithm is considered converged when both of the above criteria are met.

[0076] (3.2) Plan output: After the algorithm terminates, all the daily detailed plans obtained by the inner layer optimization in the last iteration are adaptively spliced ​​in chronological order to output a complete, globally near-optimal daily-scale look-ahead scheduling plan that takes into account the multi-day correlation of new energy sources.

[0077] This invention also provides an electronic device, which includes, but is not limited to: Memory, used to store programs; The processor is used to execute programs stored in memory. When the processor executes programs stored in memory, it is used to execute the above-mentioned daily-scale scheduling method for the power grid.

[0078] The processor and memory can be connected via a bus or other means.

[0079] Memory, as a non-transitory computer-readable storage medium, can be used to store non-transitory software programs and non-transitory computer-executable programs, such as the method described in the embodiments of the present invention. The processor implements the above method by running the non-transitory software program and instructions stored in the memory.

[0080] The memory may include a program storage area and a data storage area, wherein the program storage area may store the operating system and application programs required for at least one function; the data storage area may store data for executing the methods described above. Furthermore, the memory may include high-speed random access memory and may also include non-transitory memory, such as at least one disk storage device, flash memory device, or other non-transitory solid-state storage device. In some embodiments, the memory may optionally include memory remotely located relative to the processor, which can be connected to the processor via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0081] The non-transitory software program and instructions required to implement the above terminal selection method are stored in memory and are executed by one or more processors.

[0082] This invention also provides a storage medium storing computer-executable instructions for performing the above-described methods.

[0083] In one embodiment, the storage medium stores computer-executable instructions that are executed by one or more control processors.

[0084] The embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0085] It will be understood by those skilled in the art that all or some of the steps and systems in the methods disclosed above can be implemented as software, firmware, hardware, and suitable combinations thereof. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, digital signal processor, or microprocessor, or as hardware, or as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include computer storage media (or non-transitory media) and communication media (or transient media). As is known to those skilled in the art, the term computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disc (DVD) or other optical disc storage, magnetic cartridges, magnetic tape, disk storage or other magnetic storage devices, or any other medium that can be used to store desired information and is accessible to a computer. Furthermore, as is known to those skilled in the art, communication media typically include computer-readable instructions, data structures, program modules, or other data in modulated data signals such as carrier waves or other transmission mechanisms, and may include any information delivery medium.

[0086] This document describes embodiments of the invention, including preferred embodiments known to the inventors for carrying out the invention. Variations of these embodiments will become apparent to those skilled in the art upon reading the foregoing description. The inventors encourage those skilled in the art to adopt such variations as appropriate, and the inventors intend to practice embodiments of the invention in ways other than those specifically described herein. Therefore, the scope of the invention includes all modifications and equivalents of the subject matter set forth in the appended claims, as permitted by applicable law. Furthermore, the scope of the invention covers any combination of the foregoing elements in all possible variations thereof, unless otherwise indicated herein or otherwise clearly contradicted by the context.

Claims

1. A daily-scale power grid dispatching method, characterized in that, include: Obtain future time-series data for several new energy power plants; Based on the future time series data, the Pearson correlation coefficient matrix is ​​obtained to reflect the power output relationship between various new energy power plants. Using the Pearson correlation coefficient matrix as a constraint, a set of new energy operation scenarios is obtained based on the future time series data; Based on the set of new energy operation scenarios, multi-day cycle scheduling optimization and single-day scheduling optimization are performed sequentially to obtain several single-day scheduling plans, thereby obtaining a multi-day scheduling plan.

2. The power grid daily-scale dispatching method according to claim 1, characterized in that, The step of obtaining the Pearson correlation coefficient matrix based on the future time-series data to reflect the power output relationship between various renewable energy power plants includes: Several data windows are continuously extracted from the future time series data to obtain the multi-day power generation points of each new energy power station corresponding to a certain data window; The intraday power output points are encoded into a two-dimensional matrix image; Feature extraction is performed on the two-dimensional matrix image to obtain the Pearson correlation coefficient matrix.

3. The power grid daily-scale dispatching method according to claim 2, characterized in that, The step of extracting features from the two-dimensional matrix image to obtain the Pearson correlation coefficient matrix includes: The two-dimensional matrix images corresponding to each new energy power station in the same data window are stitched together; The pseudo-covariance matrix is ​​obtained by sequentially mapping and outer productting the spliced ​​feature vectors of multiple stations. The pseudo-covariance matrix is ​​normalized to obtain the Pearson correlation coefficient matrix.

4. The power grid daily-scale dispatching method according to claim 2, characterized in that, Encoding the intraday power output points into a two-dimensional matrix image includes: Normalize the intraday output points to obtain a normalized sequence; Mapping the normalized sequence to the polar coordinate system yields a set of polar coordinates; Calculate the angles between the polar coordinates in the polar coordinate set to obtain the Gram matrix, which is a two-dimensional matrix image.

5. The power grid daily-scale dispatching method according to claim 1, characterized in that, The set of new energy operation scenarios, constrained by the Pearson correlation coefficient matrix and based on the future time-series data, includes: Using the Pearson correlation coefficient matrix as a constraint, an inner-loop conditional generation network is trained based on historical meteorological data from the acquired historical time-series data; The network is trained to optimize the outer loop conditions based on the historical daily profiles generated by the network output under the inner loop conditions. The future time series numbers are sequentially input into the inner loop condition generation network and the outer loop condition optimization network to obtain a set of new energy operation scenarios.

6. The power grid daily-scale dispatching method according to claim 5, characterized in that, The inner loop condition generation network includes: an inner loop generator, which adopts an encoder-decoder structure, inputting random noise vectors, meteorological data, and Pearson correlation coefficient matrix into the inner loop generator, and outputting a daily profile; The inner discriminator receives the daily profile output by the inner loop generator and determines whether the daily profile conforms to historical new energy scenarios, thereby driving the optimization of the inner loop generator.

7. A daily-scale power grid dispatching method according to claim 5, characterized in that, The outer loop condition optimization network includes: an outer loop generator, which is used to perform time-series correction on the daily profile output by the inner loop condition generation network; The outer discriminator is used to extract the temporal features of the outer loop generator's output and determine whether the outer loop generator's output conforms to historical new energy scenarios.

8. The power grid daily-scale dispatching method according to claim 1, characterized in that, Based on the set of new energy operation scenarios, multi-day cycle scheduling optimization and single-day scheduling optimization are performed sequentially to obtain several single-day scheduling plans, including: With the goal of minimizing the total cost over a multi-day period, the daily boundary conditions for a certain day are obtained based on the set of new energy operation scenarios and the initial state of each new energy power station. Based on the daily boundary conditions, and with the goal of minimizing the daily operating cost, the daily scheduling plan and its operating results are optimized. Update the daily boundary conditions for the next day based on the daily scheduling plan and its operation results.

9. An electronic device, characterized in that, include: Memory, used to store programs; A processor for executing a program stored in the memory, wherein when the processor executes the program stored in the memory, the processor is configured to perform the method as described in any one of claims 1 to 8.

10. A storage medium, characterized in that, The device stores computer-executable instructions for performing the method as described in any one of claims 1 to 8.

Images