A wind power hierarchical dynamic optimization scheduling method based on high reliability guidance
By improving the Crossformer model and dynamic path adaptive selection, and combining frequency domain feature analysis and physical constraint neural networks, the contradiction between prediction consistency and power balance in the hierarchical dispatching of distribution networks is resolved, the accuracy of wind power dispatching and system stability are improved, and low-carbon and efficient power grid optimization is achieved.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- JIANGSU UNIV
- Filing Date
- 2026-03-10
- Publication Date
- 2026-06-05
AI Technical Summary
Existing hierarchical dispatching methods for distribution networks struggle to maintain consistency between the system and node levels when faced with the strong randomness and volatility of wind power, leading to a contradiction between prediction accuracy and power balance constraints. Traditional methods neglect the dynamic changes in inter-level error accumulation and prediction reliability, failing to effectively balance node accuracy with system physical constraints.
A hierarchical dynamic optimization scheduling method for wind power based on high reliability guidance is adopted. By improving the Crossformer model to generate node-level and system-level prediction benchmarks in parallel, and combining frequency domain feature analysis and dynamic path adaptive selection, a physical constraint neural network is used to optimize node prediction, thereby achieving dynamic coordination and adaptive correction.
It significantly improved system voltage stability and renewable energy absorption efficiency, reduced network losses, ensured the physical feasibility of the scheduling scheme, and achieved low-carbon and efficient scheduling of the system.
Smart Images
Figure CN122159261A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to power grid dispatching technology, specifically to a hierarchical dynamic optimization dispatching method for wind power based on high reliability guidance. Background Technology
[0002] With the global energy transition, new energy sources, represented by wind power, are being integrated into distribution networks on a large scale. The strong randomness and volatility of wind power significantly increase the difficulty for Active Distribution Networks (ADNs) to maintain power balance and safe operation. Especially under a hierarchical dispatch framework, system-level and node-level forecast information must be consistent, i.e., meet strict power balance constraints. However, traditional forecasting methods often sacrifice the accuracy of local nodes in pursuit of global power balance, leading to decreased dispatch efficiency or even making the solution infeasible.
[0003] Specifically, existing hierarchical coordination forecasting and scheduling methods for distribution networks face the following technical bottlenecks when addressing these challenges: 1) While bottom-up aggregation strategies can fully preserve the independent characteristics of each node, this simple superposition often ignores the cumulative effect of errors between levels, making it difficult to meet system consistency requirements; 2) While top-down decomposition methods prioritize system-level global power balance, their strict boundary constraints force the total power to be allocated proportionally, easily disrupting the inherent output characteristics of distributed nodes; 3) Existing coordination optimization strategies typically employ fixed coordination paths, ignoring the dynamic changes in the reliability of different level prediction models under different operating conditions, and lacking real-time quantification and dynamic utilization of prediction reliability. This is the fundamental reason why existing methods cannot effectively balance node prediction accuracy and system physical constraints. In summary, existing technologies have significant shortcomings in terms of the limitations of fixed coordination paths and the balance between system balance and node accuracy. Therefore, it is necessary to design a new method that can dynamically coordinate multi-level prediction information and balance node prediction accuracy and physical constraints to meet practical needs. Summary of the Invention
[0004] This invention addresses the problem that multi-level forecasting in active distribution network hierarchical scheduling often has to sacrifice node accuracy to meet power balance constraints. To address the dynamic evolution of forecast model reliability caused by the strong randomness of wind power output, a dynamic coordination scheme is designed to achieve optimal synergy between physical constraints, system balance, and node accuracy.
[0005] This invention employs the following technical means to achieve a hierarchical dynamic optimization scheduling method for wind power based on high reliability guidance, comprising the following steps:
[0006] Step 1: Construct a wind power prediction model. Based on historical power and meteorological multi-source data, use an improved Crossformer model to generate initial power prediction benchmarks at the bottom node level and the upper system level in parallel.
[0007] Step 2: Prediction credibility assessment. A frequency domain feature analysis module is introduced, and the prediction results obtained in Step 1 are subjected to spectrum analysis using Discrete Fourier Transform to quantify the predictability and credibility index of prediction results at each level in real time.
[0008] Step 3: Dynamic path adaptive selection, constructing a dynamic hybrid coordination mechanism, adaptively switching between bottom-up aggregation and top-down decomposition coordination paths based on the relative strength of system-level and node-level credibility;
[0009] Step 4: Perform dynamic optimization scheduling. When the bottom-up aggregation path is selected, high-reliability nodes are used to guide and correct the deviation of low-reliability nodes. When the top-down decomposition path is selected, a physical constraint neural network (ConsNN) is introduced to reconstruct node predictions while strictly satisfying power balance constraints.
[0010] The beneficial effects of adopting the above technical solution in this invention are:
[0011] 1. This invention resolves the conflict between global balance and local accuracy in hierarchical scheduling. The dynamic hybrid coordination strategy proposed in this invention breaks through the limitations of traditional fixed paths. It can adaptively select the optimal coordination path based on changes in model reliability under different weather conditions, thereby improving the prediction accuracy of each node while maintaining system-level power balance consistency.
[0012] 2. Significantly improves system voltage stability and power quality. Through dynamic path optimization, the method of this invention can effectively improve voltage distribution during periods of severe system voltage fluctuations. Simulation results show that, compared with the traditional top-down decomposition method, the voltage deviation of the method of this invention is further reduced by 1.98%, and the average voltage of the entire network is maintained at 0.998 pu, which is closer to the standard value, effectively reducing the risk of node voltage exceeding limits.
[0013] 3. Reduced network losses and improved renewable energy utilization efficiency. The method of this invention demonstrates excellent performance in reducing system active power losses, with the total daily network loss reduced to 490.7 kW, significantly lower than the bottom-up aggregation method (537.2 kW) and the top-down decomposition method (545.0 kW). Simultaneously, the renewable energy utilization rate increased to 89.4%, approximately 14.5% higher than the bottom-up aggregation method, achieving low-carbon and efficient system scheduling.
[0014] 4. Ensures the physical feasibility of the scheduling scheme. The physical constraint neural network (ConsNN) introduced in the top-down decomposition path overcomes the defect of pure data-driven methods that are prone to violating the physical laws of the power grid. By embedding hard constraints in the network structure and adding physical constraint penalties to the loss function, it is ensured that the final generated node power reconfiguration scheme strictly satisfies mathematical equality constraints while not violating line capacity and node voltage limits. Attached Figure Description
[0015] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:
[0016] Figure 1 This is a schematic diagram of the process architecture of a hierarchical dynamic optimization scheduling method for wind power based on high reliability guidance.
[0017] Figure 2 This is a comparison diagram of the voltage distribution of the entire distribution network during peak load periods using the method of this invention and a comparative method.
[0018] Figure 3 This is a comparison chart of the active power loss of the power distribution network throughout the day using the method of this invention and the comparative method.
[0019] Figure 4 This is a comparison chart of typical node wind power output curves optimized using the method of this invention. Detailed Implementation
[0020] The following will refer to the appendices in the embodiments of the present invention. Figure 1-4 The technical solutions in the embodiments of the present invention will be clearly and completely described.
[0021] The aforementioned hierarchical dynamic optimization scheduling method for wind power based on high reliability guidance includes the following process:
[0022] Figure 1 This is a schematic diagram of the process architecture of a hierarchical dynamic optimization scheduling method for wind power based on high reliability guidance.
[0023] The infrastructure computation for the improved Crossformer model described in step 1 includes:
[0024] 1) Utilize a Dimensionally Segmented Embedding Layer (DSW) to segment each dimension into segments of length 1. The data points are divided into lengths of The segments are divided, and each segment is embedded into a vector through linear projection and positional embedding:
[0025] (1);
[0026] In the formula, It is a two-dimensional vector with dimension . The Time series; For learnable projection matrix; yes The length of the dimension is The Segment vector; for Weizhongdi Learnable embedding positions of segment vectors.
[0027] 2) Capture cross-temporal and cross-dimensional dependencies using a two-stage attention layer (TSA), with its cross-dimensional stage output... The calculation is as follows:
[0028] (2);
[0029] In the formula, This is the output of the multilayer perceptron; , These are the input and output of the TSA, respectively; This represents the TSA function.
[0030] 3) Prediction is performed using a hierarchical encoder-decoder (HED), and the final prediction result is... The sum of the prediction results for each layer:
[0031] (3);
[0032] In the formula, ; It is the first The prediction results for the layer; The current moment; To predict the length of the time domain.
[0033] The Crossformer model is improved by introducing a TRPCA layer to construct the ITSA, which will output the results across time phases. Decomposed into a low-rank clean matrix and sparse noise matrix The optimization problem of the decomposition process is expressed as:
[0034] (4);
[0035] In the formula, f is the objective function of the optimization problem of the decomposition process; Indicated by and For variable pairs The objective of minimizing the rank is to solve the problem. Represents the 0 norm; yes The weights.
[0036] The improved Crossformer model introduces a closed-loop feedback mechanism, utilizing cross-dimensional stage outputs. Output across time stages Perform adaptive calibration, and correct the features. The calculation formula is:
[0037] (5);
[0038] In the formula, For activation functions; This is the feedback intensity coefficient; Represents element-wise multiplication; A learnable mapping matrix; corrected features It is substituted into the TRPCA optimization model to replace the original input.
[0039] The prediction reliability assessment described in step 2 first uses Discrete Fourier Analysis to decompose the wind power output sequence:
[0040] , k=1,…,K (6);
[0041] In the formula, Represents the decomposition of the first... One component; It is the sequence length; It refers to the quantity of the ingredients; This is the original wind power time series; It is the imaginary unit.
[0042] Using relative error Evaluate the prediction accuracy of each component and determine the cutoff frequencies that the model can effectively capture before overfitting. :
[0043] (7);
[0044] In the formula, For true value components; These are the components of the predicted value.
[0045] (8);
[0046] In the formula, This was the earliest round to show overfitting; It is the interval between rounds; and These represent the relative errors corresponding to the rounds in which the kth and (k+1)th components first show overfitting, respectively. and These represent the intervals between rounds. and .
[0047] Define the proportion of the total number of predictable components. To quantify the strength of predictability, i.e., the prediction reliability index:
[0048] (9);
[0049] In the formula, It is the proportion of each frequency component; For time series indexing; For the first spectral components of each sample; For true value components; It is the total energy of the k-th component currently under consideration; This represents the sum of the (k+1)th component to the Kth component.
[0050] The dynamic path adaptive selection mechanism described in step 3 first calculates the weighted average of the credibility of all nodes. :
[0051] (10);
[0052] In the formula, This represents the total number of wind power nodes in the system. Let be the weighting coefficient, satisfying ; This represents the credibility of the i-th node.
[0053] Set switching threshold Select a route:
[0054] (1) Bottom-up aggregation (BU-Agg) path: when hour;
[0055] (2) Top-down decomposition (TD-Dec) path: when hour;
[0056] (3) Keep the current path: when hour.
[0057] In step 4, when selecting the bottom-up aggregation path, a set of nodes with significantly lower-than-average credibility is identified. By utilizing a highly reliable guided correction mechanism, the predicted values of all nodes are aggregated to obtain a coordinated predicted value of the total system power. :
[0058] (11);
[0059] In the formula, This is the corrected prediction value adjusted based on high-reliability neighborhood nodes; These are the initial predicted values.
[0060] When choosing a top-down decomposition path, the Physically Constrained Neural Network (ConsNN) generates a normalized allocation ratio vector for the i-th node using the Softmax function. , so that:
[0061] (12);
[0062] Thus satisfying the power balance equation constraint The training objective of ConsNN is to minimize the total loss function. :
[0063] (13);
[0064] In the formula, As a balance factor; The weighted reconstruction loss.
[0065] Among them, weighted reconstruction loss Utilize node trustworthiness Weighting the deviations:
[0066] (14);
[0067] Physical constraint loss Calculate line power flow using embedded power flow models and node voltage :
[0068] (15);
[0069] In the formula, and These are line capacity limits and node voltage limits, respectively. A collection of distribution network lines; The calculated line power flow value; The calculated node voltage values.
[0070] To verify the effectiveness of the method of this invention, simulation analysis was performed on an IEEE 33-node radial distribution system. This system contains 32 branches, a voltage level of 12.66 kV, a total active power load of 3715 kW, and a total reactive power load of 2547 kvar. Considering that the distributed generation (DG) connected to the medium- and low-voltage distribution network is mainly wind power, to ensure that DG participates in dispatching at all times, wind power is also included as a grid-connected DG in the optimized dispatching process.
[0071] Figure 2 This is a comparison chart of the overall network voltage distribution during peak load periods using different optimization methods. Analysis Figure 2 It can be seen that the voltage fluctuation of the traditional bottom-up aggregation (BU-Agg) method is relatively significant, with an average voltage of 0.995 pu, and some nodes are at risk of exceeding voltage limits. The average voltage of the top-down decomposition (TD-Dec) method is 0.997 pu. In contrast, the average voltage of the proposed method (Dyn-Mix) is 0.998 pu, which is closer to the standard value. This indicates that the proposed method has significant advantages in improving system and global voltage stability and effectively reducing voltage deviation.
[0072] Figure 3 Comparison curves of network loss over the entire day for different methods. Figure 3 It can be seen that the method of the present invention exhibits a significant optimization effect on network active power loss during peak power consumption periods. The overall network loss of the system shows a downward trend, with a total network loss of only 490.7 kW, which is significantly lower than that of the BU-Agg method (537.2 kW) and the TD-Dec method (545.0 kW).
[0073] Figure 4 The optimized wind power output curves for typical nodes are shown. Analysis. Figure 4 It can be seen that during the period from 10:00 to 16:00, when wind power output is relatively concentrated, the wind power utilization rate of the method of the present invention is significantly improved, especially during the period from 12:00 to 15:00.
[0074] As shown in Table 1, the average voltage deviation of Method 2 is 2.27% lower than that of Method 1, indicating that considering system-level constraints can improve stability. The voltage deviation of the method of this invention (Method 3) is further reduced by 1.98% compared to Method 2, demonstrating that the dynamic coordination mechanism can more effectively improve system response performance. In terms of economy and environmental friendliness, the method of this invention has the lowest network loss (490.7 kW) and a renewable energy utilization rate of 89.4%, a significant increase of approximately 14.5% compared to Method 1 and 3.6% compared to Method 2, achieving efficient and low-carbon system scheduling. Although the scheduling cost of Method 3 is basically the same as that of Method 2, its overall performance in terms of physical constraint satisfaction and renewable energy consumption is the best.
[0075] Table 1 compares the different performance metrics of the three methods:
[0076] ;
[0077] The results show that the method proposed in this invention can accurately predict and dynamically coordinate the output range of distributed power sources and the voltage variation range of the system, and achieve better distribution network optimization results than traditional methods while ensuring the safe operation of the system.
Claims
1. A hierarchical dynamic optimization scheduling method for wind power based on high reliability guidance, characterized in that, Includes the following steps: Step 1: Construct a wind power prediction model. Based on historical power and meteorological multi-source data, use the improved Crossformer model to generate initial power prediction benchmarks at the bottom node level and the upper system level in parallel. Step 2: Prediction credibility assessment. A frequency domain feature analysis module is introduced. Discrete Fourier transform is used to perform spectrum analysis on the prediction results obtained in Step 1, and the predictability and credibility index of the prediction results at each level are quantified in real time. Step 3: Dynamic path adaptive selection, constructing a dynamic hybrid coordination mechanism, adaptively switching between bottom-up aggregation and top-down decomposition coordination paths based on the relative strength of system-level and node-level credibility; Step 4: Perform dynamic optimization scheduling. When the bottom-up aggregation path is selected, high-reliability nodes are used to guide and correct the deviation of low-reliability nodes. When the top-down decomposition path is selected, a physical constraint neural network (ConsNN) is introduced to reconstruct node predictions while strictly satisfying power balance constraints.
2. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance according to claim 1, characterized in that: The infrastructure computation for the improved Crossformer model described in step 1 includes: 1) Utilize a dimensional segmentation embedding layer (DSW) to segment each dimension into segments of length 1. The data points are divided into lengths of The segments are divided, and each segment is embedded into a vector through linear projection and positional embedding: (1)； In the formula, It is a two-dimensional vector with dimension . The Time series; For learnable projection matrix; yes The length of the dimension is The Segment vector; for Weizhongdi Learnable embedding positions of segment vectors; 2) Utilize a two-stage attention layer TSA to capture cross-temporal and cross-dimensional dependencies, with its cross-dimensional stage output... The calculation is as follows: (2)； In the formula, This is the output of the multilayer perceptron; , These are the input and output of the TSA, respectively; Represents the TSA function; 3) Prediction is performed using a hierarchical encoder-decoder, and the final prediction result is obtained. The sum of the prediction results for each layer: (3)； In the formula, ; It is the first The prediction results for the layer; The current moment; To predict the length of the time domain.
3. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance according to claim 1, characterized in that: The improved Crossformer model described in step 1 introduces a total variation regularized robust principal component analysis (TRPCA) layer and an improved two-stage attention layer (ITSA) to the Crossformer model, thus controlling the output across time stages. Decomposed into a low-rank clean matrix and sparse noise matrix The optimization problem of the decomposition process is expressed as: (4)； In the formula, f is the objective function of the optimization problem in the decomposition process; Indicates and For variable pairs The objective of minimizing the rank is to solve the problem. Represents the 0 norm; yes The weights.
4. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance according to claim 1, characterized in that: The improved Crossformer model described in step 1 introduces a closed-loop feedback mechanism, utilizing cross-dimensional stage outputs. Output across time phases Perform adaptive calibration, and correct the features. The calculation formula is: (5)； In the formula, For activation functions; This is the feedback intensity coefficient; Represents element-wise multiplication; A learnable mapping matrix; corrected features It is substituted into the TRPCA optimization model to replace the original input.
5. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance as described in claim 1, characterized in that: The prediction reliability assessment described in step 2 first uses Discrete Fourier Analysis to decompose the wind power output sequence: , k=1,…,K (6); In the formula, Represents the decomposition of the first... One component; It is the sequence length; It refers to the quantity of the ingredients; This is the original wind power time series; It is the imaginary unit.
6. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance according to claim 5, characterized in that: Step 2 utilizes relative error Evaluate the prediction accuracy of each component and determine the cutoff frequencies that the model can effectively capture before overfitting. : (7)； In the formula, For true value components; For the predicted value components; (8)； In the formula, This was the earliest round to show overfitting; It is the interval between rounds; and These represent the relative errors corresponding to the rounds in which the kth and (k+1)th components first show overfitting, respectively. and These represent the intervals between rounds. and .
7. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance according to claim 1, characterized in that: Step 2 defines the proportion of the total number of predictable components. To quantify the strength of predictability, i.e., the prediction reliability index: (9)； In the formula, It is the proportion of each frequency component; For time series indexing; For the first spectral components of each sample; For true value components; It is the total energy of the k-th component currently under consideration; This represents the sum of the (k+1)th component to the Kth component.
8. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance according to claim 1, characterized in that: The dynamic path adaptive selection described in step 3 first calculates the weighted average of the credibility of all nodes. : (10)； In the formula, This represents the total number of wind power nodes in the system. These are the weighting coefficients; satisfying... ; This represents the credibility of the i-th node; Set switching threshold Select a route: (1) Bottom-up aggregation of BU-Agg paths: when hour; (2) Decompose the TD-Dec path from top to bottom: when hour; (3) Keep the current path: when hour.
9. The wind power hierarchical dynamic optimization scheduling method based on high reliability guidance according to claim 1, characterized in that: In step 4, when selecting the bottom-up aggregation path, a set of nodes with significantly lower-than-average credibility is identified. By utilizing a highly reliable guided correction mechanism, the predicted values of all nodes are aggregated to obtain a coordinated predicted value of the total system power. : (11)； In the formula, This is the corrected prediction value adjusted based on high-reliability neighborhood nodes; These are the initial predicted values.
10. A hierarchical dynamic optimization scheduling method for wind power based on high reliability guidance as described in claim 9, characterized in that: In step 4, when a top-down decomposition path is selected, the Physically Constrained Neural Network (ConsNN) generates a normalized allocation ratio vector for the i-th node using the Softmax function. , so that: (12)； Thus satisfying the power balance equation constraint The training objective of ConsNN is to minimize the total loss function. : (13)； In the formula, As a balance factor; For weighted reconstruction loss; Among them, weighted reconstruction loss Utilize node trustworthiness Weighting the deviations: (14)； Physical constraint loss Calculate line power flow using embedded power flow models and node voltage : (15)； In the formula, and These are line capacity limits and node voltage limits, respectively. A collection of distribution network lines; The calculated line power flow value; The calculated node voltage values.