Robust power distribution network planning method based on convex deep learning
Through a robust distribution network planning method based on convex deep learning, the problems of limited computing power and insufficient flexibility in the existing technology are solved, efficient and accurate distribution network planning is achieved, and the stability and reliability of the power grid are improved. , adapt to the dynamic changes and uncertainties of the power grid.
Patent Information
- Application Number
- CN202410832815.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-26
- Publication Date
- 2024-11-19
- Estimated Expiration
- Not applicable · inactive patent
AI Technical Summary
The existing distribution network planning methods have limited computing capabilities, lack flexibility, insufficient optimization accuracy, low computing efficiency, and difficult to adapt to dynamic changes and uncertain factors of the power grid, resulting in low efficiency and insufficient reliability of the planning scheme.
Using a robust distribution network planning method based on convex deep learning, the complex nonlinear features of power grid data are automatically extracted by constructing a convex deep learning model, parameter updates are performed in combination with adaptive gradient methods, and topological planning and link optimization are combined to perform reliance. Rod analysis to ensure the effectiveness and safety of the planning scheme.
Improve the accuracy, efficiency and adaptability of distribution network planning, enhance the prediction and analysis capabilities of power grid operation status, reduce link switching, improve overall stability and reliability, reduce operation and maintenance costs, and ensure Effectiveness and safety under uncertainty and changing conditions.
Smart Images

Figure CN118982089A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the technical field of distribution network planning, operation and optimization, and specifically relates to a robust distribution network planning method based on convex deep learning. Background Art
[0002] Distribution network planning is an important part of power system management, which involves the operation efficiency, reliability and cost-effectiveness of the power grid. Existing technologies usually rely on traditional optimization algorithms and empirical rules to plan distribution networks, which have limitations when dealing with complex power grid systems. The dynamic changes and uncertainties in power grid operation cannot be fully considered, resulting in low efficiency or insufficient reliability of planning schemes in practical applications.
[0003] Deficiencies of existing technology:
[0004] Limited processing power: Traditional methods have limited computing power when processing large-scale, high-dimensional power grid data, making it difficult to achieve global optimization.
[0005] Lack of flexibility: Traditional planning methods lack sufficient adaptability and flexibility for dynamic changes and uncertainties in the power grid.
[0006] Insufficient optimization accuracy: Existing technologies may not be able to accurately predict the operating status of the power grid, resulting in deviations between the optimization results and actual needs.
[0007] Low computational efficiency: When performing large-scale calculations, traditional optimization algorithms may take a long time, affecting the timeliness of decision-making.
[0008] To this end, we propose a robust distribution network planning method based on convex deep learning. Summary of the invention
[0009] In view of the above deficiencies in the prior art, in order to overcome the shortcomings and deficiencies of the prior art, the purpose of the present invention is to provide a robust distribution network planning method based on convex deep learning. By introducing advanced convex deep learning technology and robustness analysis, the existing distribution network planning method is significantly improved, the accuracy, efficiency and adaptability of the planning are improved, and the requirements of modern power systems for high efficiency and high reliability are met.
[0010] In order to achieve the above objectives, the technical solution of the present invention is implemented by the following method:
[0011] A robust distribution network planning method based on convex deep learning, comprising:
[0012] S1. Collect the operation data of the distribution network, identify all nodes and branches of the distribution network from the collected data, use graph theory to represent the distribution network to construct an adjacency matrix, model the electrical parameters of each node and branch to construct a node parameter matrix, and use the adjacency matrix and node parameter matrix to construct a mathematical model of the distribution network;
[0013] Furthermore, in step S1, the adjacency matrix is represented by A, where A is an N×N square matrix, where N is the number of nodes in the distribution network, and A in the adjacency matrix A ij Indicates whether there is a connection between node i and node j; the node parameter matrix is represented by M, and the node parameter matrix M is an N×F matrix, where F is the number of electrical parameters of each node and branch;
[0014] The mathematical model of the distribution network constructed using the adjacency matrix A and the node parameter matrix M is:
[0015]
[0016] P ij =V i I j cos(θ i -θ j )+Q ij sin(θ i -θ j );
[0017] In the formula, V i and V j is the voltage between nodes i and j, R ij is the branch parameter from node i to node j, I j is the current flowing through branch node j, A ij Indicates whether there is a connection between node i and node j. If there is a branch between node i and node j in the adjacency matrix A, then A ij =1, otherwise A ij =0;P ij is the active power, Q ij is the reactive power, θ i and θ j is the voltage phase angle between node i and node j.
[0018] S2. Extract features from the mathematical model established in step S1 and construct a convex deep learning model. The convex deep learning model can automatically extract complex nonlinear features in the data and provide powerful prediction and analysis capabilities for the optimization of the distribution network;
[0019] In step S2, the constructed convex deep learning model represents a function f, which is a composite function representing multiple levels of the model. f accepts input x and generates output y:
[0020] y = f(x; β);
[0021] In the formula, x is the input vector, which contains features extracted from the adjacency matrix A and the node parameter matrix M. β is the model parameter, which includes the weight matrix W and the bias vector b of all layers. b is expressed as
[0022]
[0023] Furthermore, in step S2, in the convex deep learning model representation function f, the input value vector z of each layer l of the model representation function f (l) And the final output vector a (L) It is expressed as:
[0024] z (l) =W (l) a (l-1) +b (l) ;
[0025] a (l) =σ (l) (z l );
[0026] a L =σ (L) (W (L) z (L-1) +b (L) );
[0027] z (l) is the input value vector of the lth layer, a (l-1) is the activation output of layer l-1, a (l-1) The activation output is the previous layer a (l) The value processed by the activation function is used as the input of the next layer; W (l) is the weight matrix of the lth layer, which is used to convert the output of the previous layer into the input of the current layer; b (l) is the bias vector of the lth layer, which is used to add a constant bias term to the input of each layer, increasing the flexibility of the model; σ (l) The activation function of the lth layer, σ (l) Represents the input value vector of each layer, which is a nonlinear function;
[0028] a (L) represents the final prediction of the model; W (L) is the weight matrix of the final output layer, σ (L) is the activation function of the final output layer, σ (L) Depends on the specific task of the model; b (L) is the bias vector of the final output layer; L is the number of layers in the model, z(L-1) Represents the last layer input value vector;
[0029] The loss function of the convex deep learning model is defined as:
[0030]
[0031] Where K is the number of training samples, y i is the actual output of the tth sample, is the output of the model predicted for the tth sample.
[0032] S3. Using the convex deep learning model constructed in step S2, the parameters of the constructed convex deep learning model are updated by an adaptive gradient method. The parameter update enables the convex deep learning model to gradually learn the distribution characteristics of the data, so as to improve the efficiency and accuracy of model training. The parameter model after the adaptive gradient update is:
[0033]
[0034] in, is the loss function ζ of the convex deep learning model for the weight matrix W of the lth layer (l) The gradient of (l)' The derivative of the activation function, used to calculate the gradient in back propagation; z (l) represents the input value of the lth layer; σ (l) represents the activation function of the lth layer, usually a nonlinear function; η t is the adaptive learning rate, η t Used to update the model parameters in each iteration.
[0035] S4. Based on the optimization in step S3, topology planning and link optimization are used to further optimize the physical structure of the distribution network, reduce link switching, and improve overall stability. The optimized topology structure reduces link switching, improves the stability and reliability of the distribution network, and reduces operation and maintenance costs.
[0036] Furthermore, in step S4, the method of further optimizing the physical structure of the distribution network by using topology planning and link optimization specifically includes:
[0037] S41. Evaluate the topology of the existing distribution network to identify key nodes, branches and potential bottlenecks, and use graph theory analysis to determine the network connectivity, ring network structure and node degree;
[0038] S42. Determine the optimization goal and set the optimization objective function J to provide a clear goal for link optimization. The objective function can be minimizing cost, maximizing reliability, etc.
[0039] S43. Link traffic analysis: analyze the traffic of each link, determine the optimization direction, and identify overloaded or underutilized links. The link traffic is expressed as:
[0040] F ij =P ij +jQ ij ;
[0041] In the formula, F ij is the link traffic from node i to node j, P ij is the active power between node i and node j, Q ij is the reactive power between node i and node j,
[0042] S44. Topology optimization algorithm design uses linear programming to optimize the network structure, reduce link switching, and improve overall stability. The objective function of linear programming is expressed as:
[0043] J(P,Q) M stA′·P+M′·Q=S;
[0044] A′ and M′ are coefficient matrices obtained according to the adjacency matrix A and the node parameter matrix M. P is the active power vector, Q is the reactive power vector, and S is the node power demand vector, which represents the active and reactive power demand of each node;
[0045] S45. Link optimization is implemented. Links are adjusted according to the topology optimization results. The adjusted link traffic F ij ′ should satisfy the optimized objective function J′, and the adjusted link flow F ij ′ and the optimized objective function J′ are specifically:
[0046] F ij ′=F ij -ΔF ij ;
[0047] J′=∑ i,j c ij ·F ij ′+Σ i,j r ij ·(F ij ′) 2 ;
[0048] In the formula, F ij is the link flow from node i to node j, F ij ′ is the optimized link flow from node i to node j, ΔF ij is the change in traffic from node i to node j, c ij is the unit power transmission cost from node i to node j, r ij is the resistance loss coefficient from node i to node j.
[0049] S5. Based on the results of topology planning in step S4, the effectiveness and safety of the planning scheme under various conditions are ensured through robustness analysis. Robustness analysis helps to identify and mitigate potential risks and ensure the stability of the distribution network in the face of uncertainty and changes.
[0050] Furthermore, in S5, the robustness analysis method is:
[0051] S5.1 Determine robustness indicators, select multiple specific indicators related to robustness, and assign weights to these indicators to reflect their importance in the overall robustness. The robustness indicator R is expressed as:
[0052]
[0053] Where R is the overall robustness index, R i is the i-th specific robustness indicator, which may include load capacity, spare capacity, fault recovery time, etc.; ω i The weight of the i-th indicator, ω i Determined according to its impact on system robustness;
[0054] S5.2 simulates uncertainty factors and uses probability distribution to simulate the uncertainty factors that affect the operation of the distribution network. The specific simulation method is:
[0055] Select the probability model: Use the probability distribution model with the uncertainty factor of the normal distribution,
[0056] Parameter estimation: Estimate the mean μ of the load demand based on the parameters of the probability distribution estimated from historical data P , and the variance of the load demand
[0057] Conduct simulations: Generate random variables in the model using the estimated parameters to simulate changes in uncertainty factors.
[0058] During the simulation, the load demand P is simulated using normal distribution, expressed as: μ P The mean of load demand reflects the long-term average load level; The variance of the load demand, which measures the degree of load fluctuation;
[0059] S5.3. Create different operating scenarios to evaluate the response of the planning scheme, conduct sensitivity analysis for different operating scenarios, evaluate the impact of key parameter changes on robustness indicators, and quantify the impact of parameter changes on overall performance indicators.
[0060] The quantitative formula for sensitivity analysis is: Si Indicates the sensitivity of load demand P, ΔR i is the change in robustness index R, ΔP i Indicates the change in load demand P.
[0061] S6. Implement the planning scheme analyzed and confirmed in step S5 into the actual system, and make feedback adjustments based on the implementation effect. The implementation and feedback mechanism ensures that the planning scheme can adapt to various changes in actual operation and improves the practicality and effectiveness of the planning.
[0062] Furthermore, in step S6, the optimization results and robustness analysis of steps S4 and S5 are used to adjust the physical structure and operation strategy of the distribution network, and the optimized planning scheme after the robustness analysis is applied to the actual distribution network; the performance indicators of the distribution network after implementation are monitored in real time, the performance data after implementation is stored and processed, and compared with the expected results. According to the monitoring and data analysis results, necessary adjustments are made to the planning scheme, the deviation between the expected effect after implementation is identified, and the convex deep learning model constructed in step S2 is used for prediction and analysis to guide the adjustment; according to the robustness analysis of step S5, risk mitigation measures and emergency plans are formulated, and the optimization process is continuously iterated in combination with the loop feedback of steps S1 to S5, a continuous optimization mechanism is established, and the performance of the distribution network is continuously improved.
[0063] In summary, due to the adoption of the above technical solution, the beneficial technical effects of the invention are:
[0064] The present invention proposes a robust distribution network planning method based on convex deep learning. The method uses a convex deep learning model to automatically extract complex nonlinear features in power grid data, and combines an adaptive gradient method to perform efficient parameter updates to improve the efficiency and accuracy of model training.
[0065] Convex deep learning model: By constructing a convex deep learning model, the complex nonlinear features in the power grid data are automatically extracted, enhancing the ability to predict and analyze the power grid operation status.
[0066] Adaptive gradient method: The adaptive gradient method is used to update model parameters, which improves the efficiency and accuracy of model training and enables the model to quickly adapt to the distribution characteristics of power grid data.
[0067] Topology planning and link optimization: Based on model optimization, topology planning and link optimization technologies are further adopted to reduce link switching and improve the overall stability and reliability of the distribution network.
[0068] Robustness analysis: Through robustness analysis, we can ensure the effectiveness and security of planning schemes in the face of uncertainty and change, and identify and mitigate potential risks.
[0069] Implementation and feedback adjustment: Implement the planning scheme into the actual system and make feedback adjustments based on the implementation results to ensure that the planning scheme can adapt to various changes in actual operation and improve the practicality and effectiveness of the plan. BRIEF DESCRIPTION OF THE DRAWINGS
[0070] Figure 1 The present invention is a flowchart of a method for implementing power grid planning. DETAILED DESCRIPTION
[0071] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in combination with the embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and are not used to limit the invention.
[0072] A robust distribution network planning method based on convex deep learning,
[0073] S1. Collect the operation data of the distribution network, identify all nodes and branches of the distribution network from the collected data, use graph theory to represent the distribution network to construct an adjacency matrix, model the electrical parameters of each node and branch to construct a node parameter matrix, and use the adjacency matrix and node parameter matrix to construct a mathematical model of the distribution network;
[0074] In step S1, the adjacency matrix is represented by A, which is an N×N square matrix, where N is the number of nodes in the distribution network. ij Indicates whether there is a connection between node i and node j; the node parameter matrix is represented by M, and the node parameter matrix M is an N×F matrix, where F is the number of electrical parameters of each node and branch;
[0075] The distribution network mathematical model constructed using the adjacency matrix M and the node parameter matrix M is:
[0076]
[0077] P ij =V i I j cos(θ i -θ j )+Q ij sin(θ i -θ j );
[0078] In the formula, V i and V j is the voltage between nodes i and j, R ij is the branch parameter from node i to node j, I j is the current flowing through branch node j, A ijIndicates whether there is a connection between node i and node j. If there is a branch between node i and node j in the adjacency matrix A, then A ij =1, otherwise A ij =0;P ij is the active power, Q ij is the reactive power, θ i and θ j is the voltage phase angle between node i and node j.
[0079] S2. Extract features from the mathematical model established in step S1 and construct a convex deep learning model. The convex deep learning model can automatically extract complex nonlinear features in the data and provide powerful prediction and analysis capabilities for the optimization of the distribution network;
[0080] In step S2, the constructed convex deep learning model represents a function f, which is a composite function representing multiple levels of the model. f accepts input x and generates output y:
[0081] y = f(x; β);
[0082] In the formula, x is the input vector, which contains features extracted from the adjacency matrix A and the node parameter matrix M. β is the model parameter, which includes the weight matrix W and the bias vector b of all layers. b is expressed as
[0083]
[0084] Each layer l of the model can be expressed as:
[0085] z (l) =W (l) a (l-1) +b (l) ;
[0086] a (l) =σ (l) (z l );
[0087] The final output layer a (L) It can be expressed as:
[0088] a L =σ (L) (W (L) z (L-1) +b (L) );
[0089] z (l) is the input value vector of the lth layer, a (l) is the activation output of layer l, a (l-1) is the activation output of the l-1th layer, the activation output a (l-1) The previous layer is activated by function a(l) The processed value is used as the input of the next layer; W (l) is the weight matrix of the lth layer, which is used to convert the output of the previous layer into the input of the current layer; b (l) is the bias vector of the lth layer, which is used to add a constant bias term to the input of each layer, increasing the flexibility of the model; σ (l) The activation function of the lth layer is a nonlinear function;
[0090] a (L) is the activation output of the final output layer, a (L) represents the final prediction of the model; W (L) is the weight matrix of the final output layer, σ (L) is the activation function of the final output layer, σ (L) Depends on the specific task of the model; b (L) is the bias vector of the final output layer; L is the number of layers in the model, z (L-1) Represents the last layer input value vector;
[0091] The loss function of the convex deep learning model is defined as:
[0092]
[0093] Where K is the number of training samples, y i is the actual output of the tth sample, is the output of the model predicted for the tth sample.
[0094] S3. Using the convex deep learning model constructed in step S2, the convex deep learning model constructed by the adaptive gradient method is used to update the parameters. The parameter update enables the convex deep learning model to gradually learn the distribution characteristics of the data, so as to improve the efficiency and accuracy of model training. The parameter model after the adaptive gradient update is:
[0095]
[0096] in, is the loss function ζ of the convex deep learning model for the weight matrix W of the lth layer (l) The gradient of (l)' The derivative of the activation function, used to calculate the gradient in back propagation; z (l) represents the input value of the lth layer; σ (l) represents the activation function of the lth layer, usually a nonlinear function; η t is the adaptive learning rate, η t Used to update the model parameters in each iteration.
[0097] S4. Based on the optimization in step S3, topology planning and link optimization are used to further optimize the physical structure of the distribution network, reduce link switching, and improve overall stability. The optimized topology structure reduces link switching, improves the stability and reliability of the distribution network, and reduces operation and maintenance costs.
[0098] In step S4, the method of further optimizing the physical structure of the distribution network by using topology planning and link optimization specifically includes:
[0099] S41. Evaluate the topology of the existing distribution network to identify key nodes, branches and potential bottlenecks, and use graph theory analysis to determine the network connectivity, ring network structure and node degree;
[0100] S42. Determine the optimization goal and set the optimization objective function J to provide a clear goal for link optimization. The objective function can be minimizing cost, maximizing reliability, etc.
[0101] S43. Link traffic analysis: analyze the traffic of each link, determine the optimization direction, and identify overloaded or underutilized links. The link traffic is expressed as:
[0102] F ij =P ij +jQ ij ;
[0103] In the formula, F ij is the link traffic from node i to node j, P ij is the active power between node i and node j, Q ij is the reactive power between node i and node j,
[0104] S44. Topology optimization algorithm design uses linear programming to optimize the network structure, reduce link switching, and improve overall stability. The objective function of linear programming is expressed as:
[0105] J(P,Q) M stA′·P+M′·Q=S;
[0106] A′ and M′ are coefficient matrices obtained according to the adjacency matrix A and the node parameter matrix M. P is the active power vector, Q is the reactive power vector, and S is the node power demand vector, which represents the active and reactive power demand of each node;
[0107] S45. Link optimization is implemented. Links are adjusted according to the topology optimization results. The adjusted link traffic F ij ′ should satisfy the optimized objective function J′, and the adjusted link flow F ij ′ and the optimized objective function J′ are specifically:
[0108] Fij ′=F ij -ΔF ij ;
[0109] J′=∑ i,j c ij ·F ij ′+Σ i,j r ij ·(F ij ′) 2 ;
[0110] In the formula, F ij is the link flow from node i to node j, F ij ′ is the optimized link flow from node i to node j, ΔF ij is the change in traffic from node i to node j, c ij is the unit power transmission cost from node i to node j, r ij is the resistance loss coefficient from node i to node j.
[0111] S5. Based on the results of topology planning in step S4, the effectiveness and safety of the planning scheme under various conditions are ensured through robustness analysis. Robustness analysis helps to identify and mitigate potential risks and ensure the stability of the distribution network in the face of uncertainty and changes.
[0112] In S5, the robustness analysis method is:
[0113] S5.1 Determine robustness indicators, select multiple specific indicators related to robustness, and assign weights to these indicators to reflect their importance in the overall robustness. The robustness indicator R is expressed as:
[0114]
[0115] Where R is the overall robustness index, R i is the i-th specific robustness indicator, which may include load capacity, spare capacity, fault recovery time, etc.; ω i is the weight of the ith indicator, ω i Determined according to its impact on system robustness;
[0116] S5.2 simulates uncertainty factors and uses probability distribution to simulate the uncertainty factors that affect the operation of the distribution network. The specific simulation method is:
[0117] Select the probability model: Use the probability distribution model with the uncertainty factor of the normal distribution,
[0118] Parameter estimation: Estimate the mean μ of the load demand based on the parameters of the probability distribution estimated from historical data P , and the variance of the load demand
[0119] Conduct simulations: Generate random variables in the model using the estimated parameters to simulate changes in uncertainty factors.
[0120] During the simulation, the load demand P is simulated using normal distribution, expressed as: μ P The mean of load demand reflects the long-term average load level; The variance of the load demand, which measures the degree of load fluctuation;
[0121] S5.3. Create different operating scenarios to evaluate the response of the planning scheme, conduct sensitivity analysis for different operating scenarios, evaluate the impact of key parameter changes on robustness indicators, and quantify the impact of parameter changes on overall performance indicators.
[0122] The quantitative formula for sensitivity analysis is: S i Indicates the sensitivity of load demand P, ΔR i is the change in robustness index R, ΔP i The change in load demand P.
[0123] S6. Implement the planning scheme analyzed and confirmed in step S5 into the actual system, and make feedback adjustments based on the implementation effect. The implementation and feedback mechanism ensures that the planning scheme can adapt to various changes in actual operation and improves the practicality and effectiveness of the planning.
[0124] like Figure 1 As shown, the specific steps of implementing power grid planning in the present invention are: adjusting the physical structure and operation strategy of the distribution network based on the optimization results and robustness analysis of steps S4 and S5, and applying the optimized planning scheme after robustness analysis to the actual distribution network; real-time monitoring of the performance indicators of the distribution network after implementation, storing and processing the performance data after implementation, comparing with the expected results, making necessary adjustments to the planning scheme according to the monitoring and data analysis results, identifying the deviation between the expected effect after implementation, and using the convex deep learning model constructed in step S2 for prediction and analysis to guide the adjustment; formulating risk mitigation measures and emergency plans based on the robustness analysis of step S5, combining the loop feedback of steps S1 to S5, continuously iterating the optimization process, establishing a continuous optimization mechanism, and continuously improving the performance of the distribution network.
[0125] In order to make the purpose, technical solution and advantages of the invention more clear, the invention is further described in detail below in combination with the embodiments. It should be understood that the specific embodiments described here are only used to explain the invention and are not used to limit the invention.
Claims
1. A robust distribution network planning method based on convex deep learning, characterized in that: The following steps are involved: S1. Collect the operation data of the distribution network, identify all nodes and branches of the distribution network from the collected data, use graph theory to represent the distribution network to construct an adjacency matrix, model the electrical parameters of each node and branch to construct a node parameter matrix, and use the adjacency matrix and node parameter matrix to construct a mathematical model of the distribution network; S2. extracting features from the mathematical model established in step S1 and constructing a convex deep learning model; S3. Using the convex deep learning model constructed in step S2, updating the parameters of the constructed convex deep learning model by an adaptive gradient method; S4. Based on step S3, topology planning and link optimization are used to further optimize the physical structure of the distribution network; S5. Based on the result of topology planning in step S4, the effectiveness and security of the planning scheme under various conditions are ensured through robustness analysis; S6. Implement the optimization results of steps S4 and S5 and the planning scheme confirmed by the robustness analysis into the actual system, and make feedback adjustments based on the implementation results; In step S1, the adjacency matrix is represented by A, which is an N×N square matrix, where N is the number of nodes in the distribution network. ij Indicates whether there is a connection between node i and node j; the node parameter matrix is represented by M, and the node parameter matrix M is an N×F matrix, where F is the number of electrical parameters of each node and branch; The mathematical model of the distribution network constructed using the adjacency matrix A and the node parameter matrix M is: P ij =V i I j cos(θ i -θ j )+Q ij sin(θ i -θ j ); In the formula, V i and V j is the voltage between nodes i and j, R ij is the branch parameter from node i to node j, I j is the current flowing through branch node j, A ij Indicates whether there is a connection between node i and node j. If there is a branch between node i and node j in the adjacency matrix A, then A ij =1, otherwise A ij =0;P ij is the active power, Q ij is the reactive power, θ i and θ j is the voltage phase angle between node i and node j.
2. A robust distribution network planning method based on convex deep learning according to claim, characterized in that: In step S2, the constructed convex deep learning model representation function is f, which is a composite function representing multiple levels of the model. f accepts an input vector x and generates an output y, and the formula is expressed as: y = f(x; β); In the formula, x is the input vector, which contains features extracted from the adjacency matrix A and the node parameter matrix M. β is the model parameter, which includes the weight matrix W and the bias vector b of all layers. b is expressed as 3. A robust distribution network planning method based on convex deep learning according to claim 2, characterized in that: The convex deep learning model represents the function f, and the model represents the input value vector z of each layer l of the function f (l) And the final output vector a (L) It is expressed as: z (l) =W (l) a (l-1) +b (l) ; a (l) =σ (l) (z l ); and L =σ (L) (IN (L) With (L-1) +b (L) ); z (l) is the input value vector of the lth layer, a (l) is the activation output of layer l, a (l-1) is the activation output of the l-1th layer, the activation output a (l-1) It is the previous layer a (l) The value processed by the activation function is used as the input of the next layer; W (l) is the weight matrix of the lth layer, which is used to convert the output of the previous layer into the input of the current layer; b (l) is the bias vector of the lth layer, which is used to add a constant bias term to the input of each layer, increasing the flexibility of the model; σ (l) represents the activation function of the lth layer, σ (l) Represents the input value vector of each layer, which is a nonlinear function; a (L) represents the final prediction of the model; W (L) is the weight matrix of the final output layer, σ (L) is the activation function of the final output layer, σ (L) Depends on the specific task of the model; b (L) is the bias vector of the final output layer; L is the number of layers in the model, z (L-1) Represents the last layer input value vector.
4. The robust distribution network planning method based on convex deep learning according to claim 2 is characterized in that: In the convex deep learning model representation function f, the loss function of the convex deep learning model representation function f is defined as: Where K is the number of training samples, y i is the actual output of the tth sample, is the output of the model predicted for the tth sample.
5. The method for robust distribution network planning based on convex deep learning according to claim 1, characterized in that: In step S3, the parameter model after adaptive gradient updating is: Among them, Δ W(l) ζ is the loss function of the convex deep learning model ζ for the weight matrix W of the lth layer (l) The gradient of (l)' The derivative of the activation function, used to calculate the gradient in back propagation; z (l) represents the input value of the lth layer; σ (l) Represents the activation function of the lth layer, which is a nonlinear function; η t is the adaptive learning rate, η t Used to update the model parameters in each iteration.
6. A robust distribution network planning method based on convex deep learning according to claim, characterized in that: In step S4, the method of further optimizing the physical structure of the distribution network by using topology planning and link optimization includes: S41. Evaluate the topology of the existing distribution network to identify key nodes, branches and potential bottlenecks, and use graph theory analysis to determine the network connectivity, ring network structure and node degree; S42. Determine the optimization goal and set the optimization objective function J to provide a clear goal for link optimization. The objective function can be to minimize cost and maximize reliability. S43. Link traffic analysis: analyze the traffic of each link, determine the optimization direction, and identify overloaded or underutilized links. The link traffic is expressed as: F ij =P ij +jQ ij ; In the formula, F ij is the link traffic from node i to node j, P ij is the active power between node i and node j, Q ij is the reactive power between node i and node j, S44. Topology optimization algorithm design uses linear programming to optimize the network structure, reduce link switching, and improve overall stability. The objective function of linear programming is expressed as: J(P,Q) M s.t.A′·P+M′·Q=S; A′ and M′ are coefficient matrices obtained according to the adjacency matrix A and the node parameter matrix M. P is the active power vector, Q is the reactive power vector, and S is the node power demand vector, which represents the active and reactive power demand of each node; S45. Link optimization is implemented. Links are adjusted according to the topology optimization results. The adjusted link traffic F ij ′ should satisfy the optimized objective function J′, and the adjusted link flow F ij ′ and the optimized objective function J′ are specifically: F ij ′=F ij -ΔF ij ; J′=∑ i,j c ij ·F ij ′+∑ i,j r ij ·(F ij ′) 2 ; In the formula, F ij is the link flow from node i to node j, F ij ′ is the optimized link flow from node i to node j, ΔF ij is the change in traffic from node i to node j, c ij is the unit power transmission cost from node i to node j, r ij is the resistance loss coefficient from node i to node j.
7. A robust distribution network planning method based on convex deep learning according to claim, characterized in that: In S5, the robustness analysis method is: S5.1 Determine the robustness index, select multiple specific indicators related to robustness, and assign weights to these indicators. The robustness index R is expressed as: Where R is the overall robustness index, R i is the i-th specific robustness indicator, which may include load capacity, spare capacity, fault recovery time, etc.; ω i is the weight of the ith indicator, ω i Determined according to its impact on system robustness; S5.
2. Simulate uncertainty factors. Use probability distribution to simulate the uncertainty factors that affect the operation of the distribution network, including the following steps: Select the probability model: Use the probability distribution model with the uncertainty factor of the normal distribution, Parameter estimation: Estimate the mean μ of the load demand based on the parameters of the probability distribution estimated from historical data P , and the variance of the load demand Conduct simulations: Generate random variables in the model using the estimated parameters to simulate changes in uncertainty factors; During the simulation, the load demand P is simulated using normal distribution, expressed as: P is the load demand, μ P The mean of load demand reflects the long-term average load level; The variance of the load demand, which measures the degree of load fluctuation; S5.
3. Create different operating scenarios, evaluate the response of the planning scheme, conduct sensitivity analysis for different operating scenarios, evaluate the impact of key parameter changes on robustness indicators, and quantify the impact of parameter changes on overall performance indicators; The quantitative formula for sensitivity analysis is: S i Indicates the sensitivity of load demand P, ΔR i is the change in robustness index R, ΔP i The change in load demand P.
8. The method for robust distribution network planning based on convex deep learning according to claim 1, characterized in that: In the step S6, the optimization results and robustness analysis of steps S4 and S5 are used to adjust the physical structure and operation strategy of the distribution network, and the optimization planning scheme after the robustness analysis is applied to the actual distribution network; Monitor the performance indicators of the distribution network after implementation in real time, store and process the performance data after implementation, compare it with the expected results, make necessary adjustments to the planning scheme based on the monitoring and data analysis results, identify the deviation between the expected effect after implementation, and use the convex deep learning model constructed in step S2 for prediction and analysis to guide the adjustment; formulate risk mitigation measures and emergency plans based on the robustness analysis of step S5, combine the loop feedback of steps S1 to S5, continuously iterate the optimization process, and establish a continuous optimization mechanism.
Citation Information
Cited By
Automobile mechanical simulation model generation method based on graph theory
CN119538419A
A method for generating automobile mechanical simulation model based on graph theory
CN119538419B
Park energy router and coordinated configuration method thereof
CN120450364A
