Smart Grid Power Dispatching Optimization Method

Abstract

The present invention relates to the field of smart grid technology and discloses a smart grid power dispatch optimization method, which first obtains grid operation data and uses federated learning to deal with data missing and noise problems; and constructs a load forecasting model through a dynamic time warping algorithm and a specific network. A multi-energy coupling dispatching model and a demand response game model are constructed, and a multi-time scale rolling optimization framework is established. A sensitivity analysis is performed on the dispatching parameters, a hierarchical collaborative optimization mechanism is designed, and a robust optimization model is constructed to deal with power flow uncertainty. A dispatch instruction verification module is integrated and an online incremental learning mechanism is deployed. The present invention can effectively process data, accurately predict load, optimize multi-energy dispatching, guide demand response, deal with uncertainty, verify dispatch instructions and update models in real time, thereby improving the safety, reliability and economy of smart grid power dispatching and achieving optimal allocation of power resources.

Description

Technical Field

[0001] The present invention relates to the technical field of smart grids, and in particular to a smart grid power dispatch optimization method. Background Art

[0002] With energy transitions and growing electricity demand, smart grid construction has become a key development direction for the power industry. Traditional grid power dispatching methods face numerous challenges in adapting to the characteristics of new power systems. Regarding data processing, smart grid data comes from a wide range of sources and is complex, including data from generators, distribution networks, energy storage devices, renewable energy, and user loads. This data is prone to missing or noisy data. Traditional centralized data processing methods are not only inefficient but also pose privacy risks. For example, when power data from multiple regions is processed centrally, data privacy in each region is difficult to guarantee. Furthermore, large-scale data transmission can cause communication congestion, impacting the timeliness of data processing and failing to meet the real-time requirements of smart grids.

[0003] Power load forecasting is a crucial basis for power dispatch. Traditional forecasting methods struggle to accurately capture load fluctuation characteristics and time-series correlations, resulting in low forecast accuracy. In smart grids, meteorological factors significantly impact load, but traditional methods fail to fully account for the spatiotemporal coupling between meteorological factors and load. For example, during extreme weather conditions, traditional forecasting models are unable to quickly and accurately reflect load fluctuations, leading to a mismatch between power dispatch and actual demand, increasing uncertainty and costs in power supply.

[0004] Coordinated multi-energy scheduling is a key feature of smart grids. When coordinating multiple energy sources—thermal power, wind power, photovoltaics, and energy storage—combined, complex power flow constraints between energy nodes exist, making traditional scheduling models ineffective. Energy transmission losses across voltage levels are also difficult to accurately calculate, leading to inefficient energy utilization. For example, in regions rich in wind and solar resources, curtailment of wind and solar power is common due to a lack of effective scheduling models, resulting in energy waste.

[0005] In terms of demand response, users' load adjustment behavior is complex under real-time electricity price fluctuations. Traditional electricity pricing strategies fail to fully consider the game-playing nature of user behavior, making it difficult to effectively guide users to rationally adjust their electricity usage. This makes it difficult to achieve a balanced supply and demand through flexible demand-side adjustments, reducing the overall operational efficiency of the power system.

[0006] Furthermore, the integration of distributed generation increases the uncertainty of grid power flows, making traditional deterministic optimization methods incapable of addressing this. Furthermore, the lack of effective verification of the security and reliability of dispatch instructions creates potential risks for grid operations. Traditional power dispatch optimization models are not updated in a timely manner and are unable to adapt to the grid's rapidly changing operational state. Summary of the Invention

[0007] The purpose of the present invention is to provide a smart grid power dispatch optimization method to solve the problems raised in the above background technology.

[0008] To achieve the above objectives, the present invention provides the following technical solution: a smart grid power dispatch optimization method, the method comprising:

[0009] Step 1: Obtain grid operation data, including generator output curves, distribution network topology, energy storage device capacity, renewable energy forecast output, and user load time series data. When data is missing or noisy, a federated learning method is used to combine multi-region edge nodes for distributed data cleaning and interpolation.

[0010] Step 2: Based on the dynamic time warping algorithm, historical load data and meteorological factor time series are aligned to build a spatiotemporal coupled power load prediction model. A deep residual network enhanced with an attention mechanism is used to extract load fluctuation characteristics. The LSTM network is integrated to capture time series correlations and generate regional load forecast results.

[0011] Step 3: Build a multi-energy coupling scheduling model, using thermal power, wind power, photovoltaic power, and energy storage as decision-making units. A two-layer graph convolutional network is used to model power flow constraints between energy nodes. The node balance equation is defined as the tensor product of the topological adjacency matrix and the energy conversion efficiency. Virtual nodes are introduced to address energy transmission losses across voltage levels.

[0012] Step 4: Establish a demand response game model based on real-time electricity price fluctuation signals, modeling user load adjustment behavior as a non-cooperative game process under incomplete information. Use quantum particle swarm optimization to solve the Nash equilibrium point and generate a time-of-use electricity price incentive strategy.

[0013] Step 5: Establish a multi-timescale rolling optimization framework, decomposing the dispatch cycle into three phases: day-ahead, intraday, and real-time. Use a deep reinforcement learning algorithm to construct a dynamic decision-making strategy, defining the state space as grid operating parameters and the action space as unit start and stop instructions and energy storage charge and discharge rates. Update the strategy network parameters using a double-delayed deep deterministic policy gradient algorithm.

[0014] Step 6: Conduct multi-objective sensitivity analysis on the scheduling parameters, use the improved Hilbert-Schmidt independence criterion to screen key variables, construct a surrogate model based on sparse polynomial chaos expansion to replace the high-dimensional optimization problem, and use adaptive sparse grid sampling to improve model accuracy;

[0015] Step 7: Design a hierarchical collaborative optimization mechanism. The outer layer uses mixed integer programming to optimize the unit commitment and network topology. The inner layer uses the alternating direction multiplier method to solve the power allocation problem. The boundary conditions are transferred by Lagrange multipliers to achieve iterative convergence of cross-layer coupling constraints.

[0016] Step 8: To address the power flow uncertainty caused by the integration of distributed generation, a robust optimization model is constructed. Interval-valued fuzzy sets are used to describe the fluctuation range of wind and solar power output. The robustness index is defined as the system stability threshold under the worst-case scenario. The main problem and sub-problems are decoupled and solved based on the Benders decomposition algorithm.

[0017] Step 9: Integrate the scheduling instruction verification module, use smart contract technology to perform topological security verification and N-1 fault rehearsal on the optimization results, generate scheduling instruction execution certificates through formal verification methods, and implement cross-domain instruction synchronization under privacy protection based on zero-knowledge proof;

[0018] Step 10: Deploy an online incremental learning mechanism, use an online sequential extreme learning machine to update the load forecasting model parameters in real time, combine the sliding window technology to dynamically adjust the optimization model weight coefficients, and use the Markov chain Monte Carlo method to evaluate the model drift to trigger retraining conditions.

[0019] Preferably, the federated learning adopts a dynamic weight aggregation strategy to calculate the aggregation weight based on the edge node data quality evaluation value. The data quality evaluation indicators include data integrity, time series continuity and KL divergence with the central node distribution, and a gradient transmission protocol based on Paillier homomorphic encryption is designed to prevent the leakage of original data.

[0020] Preferably, the deep residual network includes a residual block stacking structure, each residual block is composed of a gated convolution layer, a batch normalization layer and a jump connection, the activation function of the gated convolution layer adopts an adaptive gated activation function, and the network output end is connected to the self-attention module to calculate the spatial correlation weight matrix of the load feature.

[0021] Preferably, the particle position encoding in the quantum particle swarm algorithm is represented by quantum bit phase angle, the quantum rotating gate update strategy is defined as the phase difference function between the particle historical optimum and the global optimum, and quantum entanglement operation is introduced to realize particle swarm collaborative search. The algorithm termination condition is that the group diversity is lower than a preset threshold or the maximum number of iterations is reached.

[0022] Preferably, the basis function of the sparse polynomial chaos expansion proxy model selects Legendre orthogonal polynomials, an adaptive forward-backward selection algorithm is used to determine the polynomial order and cross terms, and the model complexity is optimized based on the Akaike information criterion.

[0023] Preferably, the main problem of the Benders decomposition algorithm is modeled as a mixed integer quadratic programming problem, and the subproblems are converted into semidefinite programming forms to solve the extreme value of system stability, and an accelerated cutting plane generation strategy is designed, and effective constraints are screened using dual variable sensitivity analysis.

[0024] Preferably, the formal verification method uses temporal logic formulas to describe power grid safety constraints, including bus voltage over-limit, line overload and frequency deviation indicators, and verifies the compliance of scheduling instructions by traversing state transition paths through symbolic model detection technology.

[0025] Preferably, the dynamic expansion strategy of the hidden layer nodes of the online sequence extreme learning machine adopts the error sensitivity criterion, defines the node addition threshold as the moving average variance of the prediction residual, and designs a forgetting factor mechanism to attenuate the influence weight of historical data on model updating.

[0026] Preferably, the smart contract verification module includes a topology connectivity verification submodule, which uses a depth-first search algorithm to detect island operation risks, and calculates the minimum breakpoint set based on the improved Floyd-Warschel shortest path algorithm to ensure the N-1 criterion satisfaction rate.

[0027] Preferably, the Markov Chain Monte Carlo method uses a sampler to generate the posterior distribution of model parameters, defines the drift indicator as the distance of the parameter distribution, and triggers the full model retraining process when the drift exceeds a preset threshold.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] Federated learning is used to integrate distributed data cleaning and interpolation with multi-regional edge nodes. A dynamic weight aggregation strategy is employed to improve data processing accuracy. A gradient transmission protocol using Paillier homomorphic encryption prevents the leakage of raw data. This approach not only ensures data privacy but also effectively processes large-scale, complex data, improving data quality and providing a reliable basis for subsequent scheduling decisions. For example, power data from different regions can be cleaned and interpolated locally, with only encrypted gradients transmitted, preventing the leakage of sensitive data while improving the accuracy of data fusion.

[0030] Based on a dynamic time warping algorithm, the model for spatiotemporal coupled power load forecasting, which aligns historical load data with meteorological factor time series, employs a deep residual network and LSTM network enhanced by an attention mechanism. This model can more accurately extract load fluctuation characteristics and capture temporal correlations, significantly improving the accuracy of regional load forecasts. In practical applications, it can effectively reduce power supply and demand imbalances caused by load forecast errors and lower power system operating costs.

[0031] The multi-energy coupling scheduling model uses thermal power, wind power, photovoltaic power, and energy storage as decision-making units. It utilizes a two-layer graph convolutional network to model power flow constraints between energy nodes and introduces virtual nodes to handle energy transmission losses across voltage levels. This enables efficient coordinated scheduling of multiple energy sources and improves energy efficiency. For example, by accurately calculating power flows and losses, it can reduce energy waste, improve the grid's ability to accommodate renewable energy, and promote a green transformation of the energy structure.

[0032] The demand response game model established based on real-time electricity price fluctuation signals models the user load adjustment behavior as a non-cooperative game process under incomplete information. The quantum particle swarm algorithm is used to solve the Nash equilibrium point to generate a time-of-use electricity price incentive strategy, which can effectively guide users to reasonably adjust their electricity consumption behavior, achieve a dynamic balance between electricity supply and demand, and improve the flexibility and stability of the power system.

[0033] A multi-timescale rolling optimization framework decomposes the dispatch cycle into three phases: day-ahead, intraday, and real-time. A deep reinforcement learning algorithm is used to construct a dynamic decision-making strategy. The policy network parameters are updated using a double-delayed deep deterministic policy gradient algorithm. This framework allows for rapid adjustment of dispatch policies based on the real-time operating status of the power grid, improving the real-time and adaptability of dispatch decisions and ensuring stable grid operation at different timescales. Multi-objective sensitivity analysis of dispatch parameters is performed, and an improved Hilbert-Schmidt independence criterion is used to screen key variables. A surrogate model based on sparse polynomial chaos expansion is constructed to replace the high-dimensional optimization problem. Adaptive sparse grid sampling is used to improve model accuracy, significantly reducing optimization computational complexity and improving efficiency. This provides an efficient computational approach for optimal dispatch of complex power systems. A hierarchical collaborative optimization mechanism is designed. The outer layer uses mixed integer programming to optimize unit commitment and network topology, while the inner layer uses the alternating direction multiplier method to solve the power allocation problem. Lagrange multipliers are used to propagate boundary conditions to achieve iterative convergence of cross-layer coupling constraints. This effectively coordinates optimization decisions at different levels of the power system, improving overall optimization effectiveness and reducing system operating costs.

[0034] To address the power flow uncertainty caused by the integration of distributed power sources, a robust optimization model was constructed that uses interval-valued fuzzy sets to describe the fluctuation range of wind and solar power output. The robustness index is defined as the system stability threshold under the worst-case scenario, and a decoupling solution is obtained based on the Benders decomposition algorithm. This enhances the grid's ability to cope with uncertainty and improves the safety and reliability of grid operation. An integrated dispatch instruction verification module uses smart contract technology for topological security verification and N-1 fault rehearsal. Dispatch instruction execution certificates are generated through formal verification methods, and cross-domain instruction synchronization is achieved under privacy protection based on zero-knowledge proofs. This ensures the security and reliability of dispatch instructions while protecting data privacy in cross-domain dispatching.

[0035] An online incremental learning mechanism is deployed, and an online sequential extreme learning machine is used to update the load forecasting model parameters in real time. The sliding window technology is combined to dynamically adjust the optimization model weight coefficients, and the Markov chain Monte Carlo method is used to evaluate the model drift to trigger the retraining conditions. This enables the model to track changes in the power grid operation status in real time, continuously maintain good prediction and optimization performance, and improve the accuracy and adaptability of power dispatch. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1This is a working principle diagram of the smart grid power dispatch optimization method according to the present invention;

[0037] Figure 2 Schematic diagram of federated learning data processing and privacy protection;

[0038] Figure 3 Schematic diagram of the Benders decomposition algorithm for solving a robust optimization model. DETAILED DESCRIPTION

[0039] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0040] See also Figure 1-3 The present invention provides a smart grid power dispatch optimization method, the overall implementation scheme of which is as follows:

[0041] Step 1: Acquire key data on grid operations, such as generator output curves, distribution network topology, energy storage device capacity, predicted renewable energy output, and user load time series data. When data is missing or noisy, federated learning methods are used to coordinate distributed data cleaning and interpolation across multiple regional edge nodes.

[0042] Step 2: Using a dynamic time warping algorithm, historical load data is aligned with meteorological factor time series. Regional load forecasting is achieved by constructing a spatiotemporal coupled power load forecasting model. This model uses a deep residual network enhanced with an attention mechanism to extract load fluctuation characteristics and integrates an LSTM network to capture temporal correlations.

[0043] Step 3: Construct a multi-energy coupling scheduling model with thermal power, wind power, photovoltaic power, and energy storage as decision-making units. A two-layer graph convolutional network is used to model the power flow constraints between energy nodes. The node balance equation is defined as the tensor product of the topological adjacency matrix and the energy conversion efficiency. Virtual nodes are introduced to account for energy transmission losses across voltage levels.

[0044] Step 4: Build a demand response game model based on real-time electricity price fluctuation signals, treating user load adjustment behavior as a non-cooperative game process under incomplete information. Use a quantum particle swarm algorithm to solve for the Nash equilibrium point and generate a time-of-use electricity price incentive strategy.

[0045] Step 5: Construct a multi-timescale rolling optimization framework, dividing the scheduling cycle into three phases: day-ahead, intraday, and real-time. A dynamic decision-making strategy is constructed using a deep reinforcement learning algorithm. The state space is defined as grid operating parameters, and the action space is defined as unit start and stop commands and energy storage charge and discharge rates. The policy network parameters are updated using a double-delayed deep deterministic policy gradient algorithm.

[0046] Step 6: Conduct a multi-objective sensitivity analysis of the scheduling parameters and use the improved Hilbert-Schmidt independence criterion to screen key variables. Construct a surrogate model based on sparse polynomial chaos expansion to replace the high-dimensional optimization problem, and improve model accuracy through adaptive sparse grid sampling.

[0047] Step 7: Design a hierarchical collaborative optimization mechanism. The outer layer uses mixed integer programming to optimize the unit combination and network topology. The inner layer uses the alternating direction multiplier method to solve the power allocation problem. The boundary conditions are transferred with the help of Lagrange multipliers to achieve iterative convergence of cross-layer coupling constraints.

[0048] Step 8: Construct a robust optimization model to address the power flow uncertainty caused by the integration of distributed generation. Use interval-valued fuzzy sets to describe the fluctuation range of wind and solar power output. Define the robustness index as the system stability threshold under the worst-case scenario. Decouple the main problem from the subproblems using the Benders decomposition algorithm.

[0049] Step 9: Integrate the scheduling instruction verification module and use smart contract technology to perform topological security verification and N-1 fault rehearsal on the optimization results. Generate scheduling instruction execution certificates through formal verification methods, and implement cross-domain instruction synchronization under privacy protection based on zero-knowledge proof.

[0050] Step 10: Deploy an online incremental learning mechanism, using an online sequential extreme learning machine to update the load forecasting model parameters in real time. A sliding window technique is used to dynamically adjust and optimize the model weight coefficients, and a Markov Chain Monte Carlo method is used to evaluate model drift to trigger retraining conditions.

[0051] The implementation of the present invention will be further described below with reference to Examples 1 to 5.

[0052] Example 1:

[0053] In step 1, data acquisition and processing, when there is missing or noisy data, the federated learning method is used to combine multi-region edge nodes for distributed data cleaning and interpolation.

[0054] Federated learning adopts a dynamic weight aggregation strategy. Suppose there are n edge nodes. For the i-th edge node, its data quality evaluation value Q i By Data Integrity I i , time series continuity C iand the KL divergence of the central node distribution Jointly determined, through a weighted calculation:

[0055]

[0056] Among them, w1, w2, and w3 are weight coefficients, and w1 + w2 + w3 = 1. These three weight coefficients can be set according to the actual data situation and experience. For example, if the current data integrity has a significant impact on the overall data quality, the value of w1 can be appropriately increased.

[0057] Aggregation weight ω i Calculated according to the data quality assessment value, the formula is:

[0058]

[0059] In the process of gradient transmission, in order to prevent the leakage of original data, a gradient transmission protocol based on Paillier homomorphic encryption is designed. Assume that the gradient calculated by edge node i is g i , first use the Paillier encryption algorithm to encrypt the gradient. The public key of the Paillier encryption algorithm is (N, g), the private key is λ, and the encryption function is E. Then the encrypted gradient is E(g i ).

[0060] The central node receives the encrypted gradient {E(g1), E(g2), …, E(g n )}, the aggregation operation is performed. Due to the homomorphism of the Paillier encryption algorithm, the aggregation operation can be performed in the ciphertext space, that is, the calculation Finally, the central node uses the private key λ to decrypt the aggregated ciphertext and obtain the aggregated gradient Used for model updates.

[0061] Through the above-mentioned dynamic weight aggregation strategy and encrypted gradient transmission protocol, the data quality differences of each edge node are fully considered, the accuracy of data aggregation is improved, and the leakage of original data during transmission is effectively prevented, thus ensuring data privacy and security.

[0062] Example 2:

[0063] In step 2, when constructing the spatiotemporal coupling prediction model of power load, the deep residual network plays a key role.

[0064] The deep residual network consists of a stacked residual block structure. Each residual block consists of a gated convolution layer, a batch normalization layer, and a jump connection. Let the input feature map be x. The convolution operation of the gated convolution layer performs convolution calculations on the input feature map through the convolution kernel W to obtain the convolution result y. conv :

[0065] y conv =W*x

[0066] Among them, * represents the convolution operation.

[0067] The activation function of the gated convolutional layer adopts the adaptive gated activation function (AGAF). Let the adaptive gated activation function be σ, and its calculation formula is:

[0068]

[0069] Through the adaptive gated activation function, the gated convolutional layer can adaptively adjust information transmission according to the input features and effectively filter out valuable information for load fluctuation feature extraction.

[0070] After the gated convolution layer, the data enters the batch normalization layer. The batch normalization layer normalizes the data to make it have a stable distribution, which helps accelerate and converge the model training. Let the input of the batch normalization layer be y conv , the output is y bn , the calculation formula for batch normalization is:

[0071]

[0072] Among them, μ and σ 2 are the mean and variance of the current batch of data, ∈ is a small constant to prevent the denominator from being zero, and γ and β are learnable parameters.

[0073] The skip connection connects the input x of the residual block directly to the output y after the gated convolution layer and batch normalization layer. bn Add together to get the output y of the residual block res :

[0074] y res =y bn +x

[0075] This skip connection structure can effectively solve the gradient vanishing problem in deep networks, allowing the network to learn more complex features.

[0076] Multiple residual blocks are stacked to form a deep residual network, and the network output is connected to the self-attention module. Let the final output feature map of the deep residual network be F. The self-attention module first calculates the similarity between different positions in the feature map F. For each position i and j in the feature map F, its similarity s is calculated. ij :

[0077]

[0078] Among them, f and g are two different linear transformation functions, F i and Fj They represent the feature vectors of positions i and j in the feature map F respectively, and N is the total number of positions in the feature map F.

[0079] The spatial association weight matrix S is obtained by the above calculations = (s ij ), the self-attention module performs weighted summation on the feature map F according to the weight matrix to obtain a feature representation with spatial correlation information, thereby more accurately capturing the spatial correlation relationship of load characteristics and improving the accuracy of load forecasting.

[0080] Example 3:

[0081] In step 4, when establishing a demand response game model based on the real-time electricity price fluctuation signal, the quantum particle swarm algorithm is used to solve the Nash equilibrium point.

[0082] In the quantum particle swarm algorithm, the particle position is encoded using the quantum bit phase angle. Let the particle position be X = (x1, x2, ..., x D ), where D is the dimension of the problem, and for each dimension d, the position of the particle is expressed by the phase angle θ d Indicates that x d =cos(2θ d ).

[0083] The quantum rotating gate update strategy is defined as the phase difference function between the particle's historical optimum and the global optimum. Assume that the phase angle of particle i at the tth iteration is The phase angle of the historical optimal position of particle i is The phase angle of the global optimal position is The update formula of the quantum rotation gate is:

[0084]

[0085] Where Δθ i,d is the phase angle adjustment, calculated as follows:

[0086]

[0087] Here, S(a) is the sign function. When a > 0, S(a) = 1; when a = 0, S(a) = 0; and when a < 0, S(a) = -1. k is a parameter that controls the rotation step size and can be adjusted based on the actual problem.

[0088] In order to realize the collaborative search of particle swarm, quantum entanglement operation is introduced. Assume that the phase angles of particles i and j in a certain dimension d are θ i,d and θ j,d , when performing quantum entanglement operation, randomly select an entanglement coefficient α∈[0,1], and then update the phase angle of the two particles:

[0089]

[0090] Through quantum entanglement operations, particles can share information, enhance collaborative search capabilities, and prevent the algorithm from falling into local optimality.

[0091] The algorithm terminates when the population diversity is lower than the preset threshold or the maximum number of iterations is reached. The population diversity can be measured by calculating the standard deviation of the particle positions. Let the position vectors of the particle group be X1, X2,…, X N , the calculation formula of group diversity is:

[0092]

[0093] in, Is the average value of the positions of all particles in dimension d. When Diversity is less than the preset threshold τ, or the number of algorithm iterations reaches the maximum number of iterations T max When , the algorithm stops iterating and outputs the optimal solution as the Nash equilibrium point, and then generates a time-of-use electricity price incentive strategy.

[0094] Example 4:

[0095] In step 6, the sparse polynomial chaos expansion method is used to perform multi-objective sensitivity analysis on scheduling parameters and build the proxy model.

[0096] The basis function of the sparse polynomial chaos expansion agent model is the Legendre orthogonal polynomial. For a d-dimensional random variable z=(z1,z2,…,z d ), the general form of its sparse polynomial chaos expansion is:

[0097]

[0098] Among them, y(z) is the function to be approximated, a i is the expansion coefficient, Ψ i (z) is a multidimensional polynomial consisting of Legendre orthogonal polynomials, and P is the total number of terms in the polynomial.

[0099] Adaptive forward-backward selection algorithm is used to determine the polynomial order and cross terms. In the adaptive forward selection stage, starting from an initial low-order polynomial model, a new term that can minimize the model error is selected each time to be added to the model. Let the current model be M k , the model error is expressed as mean square error MSE k Measure, for candidate item j, calculate the model error after adding this item Select The smallest item is added to the model, that is:

[0100]

[0101] In the adaptive backward selection phase, the selected items are evaluated and those that contribute less to the reduction of model error are removed. The model error after removing each item is calculated. if MSE with the current model error k If the increase is less than a preset threshold ∈, the item is removed.

[0102] The model complexity is optimized based on the Akaike Information Criterion (AIC). The calculation formula of AIC is:

[0103] AIC=nln(MSE)+2p

[0104] Here, n is the number of samples, MSE is the mean squared error of the model, and p is the number of parameters in the model (including the coefficients and order of the polynomials). During the model construction process, the order and cross terms of the polynomials are continuously adjusted, and the model with the lowest AIC value is selected as the final sparse polynomial chaos expansion proxy model. In this way, while ensuring model accuracy, it effectively avoids model overfitting and improves the model's generalization ability, thus enabling more accurate replacement of high-dimensional optimization problems and providing efficient model support for subsequent optimization calculations.

[0105] Example 5:

[0106] In step 8, when constructing a robust optimization model for the power flow uncertainty caused by the integration of distributed generation, the Benders decomposition algorithm is used to decouple the main problem and sub-problems for solution.

[0107] The main problem of the Benders decomposition algorithm is modeled as a mixed integer quadratic programming problem. Let the decision variables be x (including discrete variables such as unit commitment and continuous variables such as power allocation), the objective function be f(x), and the constraints include power balance constraints and equipment capacity constraints. The general form of the main problem can be expressed as:

[0108]

[0109] stAx≤b

[0110] x∈X

[0111] Where A is the constraint matrix, b is the constraint vector, and X is the feasible region of the decision variables.

[0112] The subproblem is transformed into a semidefinite programming form to solve the system stability extreme value. Let the decision variable of the subproblem be y, which is related to the main problem through the Lagrange multiplier λ. The objective function of the subproblem is to maximize the system stability index g(y,x,λ) given the decision variable x and Lagrange multiplier λ of the main problem. Constraints include power flow equation constraints and voltage stability constraints. The subproblem can be expressed as:

[0113]

[0114] stCy≤d(x)

[0115] y∈Y

[0116] Where C is the constraint matrix of the subproblem, d(x) is the constraint vector associated with the decision variable x of the main problem, and Y is the feasible region of the decision variable of the subproblem.

[0117] In order to improve the solution efficiency, an accelerated cutting plane generation strategy is designed, and the dual variable sensitivity analysis is used to screen effective constraints. According to the duality theory, the dual variable of the sub-problem reflects the degree of influence of the constraint on the objective function. The sensitivity index of the dual variable to the decision variable of the main problem is calculated. Let the dual variable be μ and the sensitivity index S be ij Defined as:

[0118]

[0119] Where i is the number of the subproblem constraint, and j is the number of the decision variable in the main problem. By comparing the sensitivity index, we filter out the constraints that have a significant impact on the main problem's objective function and only add these effective constraints to the main problem, thereby reducing the size of the main problem and accelerating the algorithm's convergence.

[0120] In step 9, the integrated dispatch instruction verification module uses formal verification methods to verify the compliance of the dispatch instruction. The formal verification method uses temporal logic formulas to describe the power grid security constraints. For example, the bus voltage limit constraint can be expressed as:

[0121]

[0122] Where V(t) represents the bus voltage at time t, V min and V max are the lower and upper limits of the bus voltage respectively. The line overload constraint can be expressed as:

[0123]

[0124] Where, I(t) represents the line current at time t, I max is the maximum allowable current of the line. The frequency deviation index constraint can be expressed as:

[0125]

[0126] Where f(t) represents the grid frequency at time t, f0 is the rated frequency, and Δf is the allowable frequency deviation range.

[0127] The compliance of dispatch instructions is verified by traversing state transition paths using symbolic model checking technology. Symbolic model checking represents the system's state space as a Boolean formula and uses efficient Boolean operations to verify whether the system satisfies a given temporal logic formula. Starting from the initial state, the verification process gradually explores all possible state transition paths based on the system's state transition relationships, checking for violations of grid safety constraints. If the constraints are met in all reachable states, a dispatch instruction execution certificate is generated, indicating compliance. Otherwise, the dispatch instruction poses a risk and requires readjustment.

[0128] In step 10, the online incremental learning mechanism is deployed, and the load forecasting model parameters are updated in real time using an online sequence extreme learning machine. The dynamic expansion strategy of the hidden layer nodes of the online sequence extreme learning machine adopts the error sensitivity criterion and defines the node addition threshold as the moving average variance of the prediction residual. Let the prediction residual be e(t), and the moving average variance MSE be ma The calculation formula for (t) is:

[0129]

[0130] Where w is the size of the moving window. ma When (t) is greater than the preset node addition threshold δ, the hidden layer node expansion operation is triggered to add new nodes to improve the model fitting ability.

[0131] At the same time, a forgetting factor mechanism is designed to attenuate the influence of historical data on model updates. Let the forgetting factor be β∈(0,1). During the model update process, the weight of historical data is adjusted. For the training data at time t, its weight is β t-i , where i is the time when the data is collected. In this way, as time goes by, the influence of new data on model updates gradually increases, while the influence of old data gradually decreases, enabling the model to better adapt to changes in the operating status of the power grid.

[0132] The smart contract verification module includes a topology connectivity verification submodule that uses a depth-first search algorithm to detect islanding risks. Starting from a node in the power grid, the module marks visited nodes and recursively visits unvisited nodes adjacent to the current node. If any unvisited nodes remain after the search, the power grid is at risk of islanding. A modified Floyd-Warschel shortest path algorithm is used to calculate the minimum breakpoint set to ensure compliance with the N-1 criterion. The Floyd-Warschel algorithm calculates the shortest path between any two points in a graph. The improvement lies in recording path information during the calculation process. The minimum breakpoint set is determined by analyzing how the shortest path changes under different node failure conditions. The minimum breakpoint set is the minimum set of nodes that, after removing these nodes, allows the power grid to maintain connectivity under N-1 failure conditions. During the calculation of the minimum breakpoint set, different node combinations are repeatedly removed to evaluate the grid's connectivity and find the minimum breakpoint set that meets the N-1 criterion, thereby ensuring the grid's reliability in the event of a single component failure.

[0133] When evaluating the model drift based on the Markov Chain Monte Carlo method, a sampler is used to generate the posterior distribution of the model parameters. Let the model parameter be θ, and the sample set {θ1,θ2,…,θ N The drift indicator is defined as the distance of the parameter distribution. For example, the Kullback-Leibler divergence (KL divergence) is used to measure the difference between the current model parameter distribution P(θ) and the initial model parameter distribution Q(θ):

[0134]

[0135] When the drift D KL When (P||Q) exceeds the preset threshold γ, the full model retraining process is triggered, and the model is retrained using the latest power grid operation data to ensure the accuracy and adaptability of the model.

[0136] Through the above embodiments, the smart grid power dispatching optimization method involved in the present invention has been implemented and verified in detail in various key technical links, which can effectively improve the optimization level of smart grid power dispatching and ensure the safe, stable and economical operation of the power grid.

[0137] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "includes," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0138] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A smart grid power dispatch optimization method, characterized in that: The following steps are involved: Step 1: Obtain grid operation data, including generator output curves, distribution network topology, energy storage device capacity, renewable energy forecast output, and user load time series data. When data is missing or noisy, a federated learning method is used to combine multi-region edge nodes for distributed data cleaning and interpolation. Step 2: Based on the dynamic time warping algorithm, historical load data and meteorological factor time series are aligned to build a spatiotemporal coupled power load prediction model. A deep residual network enhanced with an attention mechanism is used to extract load fluctuation characteristics. The LSTM network is integrated to capture time series correlations and generate regional load forecast results. Step 3: Build a multi-energy coupling scheduling model, using thermal power, wind power, photovoltaic power, and energy storage as decision-making units. A two-layer graph convolutional network is used to model power flow constraints between energy nodes. The node balance equation is defined as the tensor product of the topological adjacency matrix and the energy conversion efficiency. Virtual nodes are introduced to address energy transmission losses across voltage levels. Step 4: Establish a demand response game model based on real-time electricity price fluctuation signals, modeling user load adjustment behavior as a non-cooperative game process under incomplete information. Use quantum particle swarm optimization to solve the Nash equilibrium point and generate a time-of-use electricity price incentive strategy. Step 5: Establish a multi-timescale rolling optimization framework, decomposing the dispatch cycle into three phases: day-ahead, intraday, and real-time. Use a deep reinforcement learning algorithm to construct a dynamic decision-making strategy, defining the state space as grid operating parameters and the action space as unit start and stop instructions and energy storage charge and discharge rates. Update the strategy network parameters using a double-delayed deep deterministic policy gradient algorithm. Step 6: Conduct multi-objective sensitivity analysis on the scheduling parameters, use the improved Hilbert-Schmidt independence criterion to screen key variables, construct a surrogate model based on sparse polynomial chaos expansion to replace the high-dimensional optimization problem, and use adaptive sparse grid sampling to improve model accuracy; Step 7: Design a hierarchical collaborative optimization mechanism. The outer layer uses mixed integer programming to optimize the unit commitment and network topology. The inner layer uses the alternating direction multiplier method to solve the power allocation problem. The boundary conditions are transferred by Lagrange multipliers to achieve iterative convergence of cross-layer coupling constraints.

2. The smart grid power dispatch optimization method according to claim 1, characterized in that: The federated learning adopts a dynamic weight aggregation strategy and calculates the aggregation weight according to the edge node data quality evaluation value. The data quality evaluation indicators include data integrity, time series continuity and KL divergence with the central node distribution. A gradient transmission protocol based on Paillier homomorphic encryption is designed to prevent the leakage of original data.

3. The smart grid power dispatch optimization method according to claim 1, characterized in that: The deep residual network includes a residual block stacking structure, each residual block consists of a gated convolution layer, a batch normalization layer and a skip connection. The activation function of the gated convolution layer adopts an adaptive gated activation function, and the network output end is connected to the self-attention module to calculate the spatial correlation weight matrix of the load feature.

4. The smart grid power dispatch optimization method according to claim 1, characterized in that: In the quantum particle swarm algorithm, particle position encoding is represented by quantum bit phase angle, and the quantum rotating gate update strategy is defined as the phase difference function between the particle's historical optimum and the global optimum. Quantum entanglement operation is introduced to realize particle swarm collaborative search. The algorithm termination condition is that the group diversity is lower than the preset threshold or the maximum number of iterations is reached.

5. The smart grid power dispatch optimization method according to claim 1, characterized in that: The basis function of the sparse polynomial chaos expansion agent model is Legendre orthogonal polynomial, an adaptive forward-backward selection algorithm is used to determine the polynomial order and cross terms, and the model complexity is optimized based on the Akaike information criterion.

6. The smart grid power dispatch optimization method according to claim 1, characterized in that: The method further comprises: To address the power flow uncertainty caused by the integration of distributed power sources, a robust optimization model is constructed. Interval fuzzy sets are used to describe the fluctuation range of wind and solar power output. The robustness index is defined as the system stability threshold under the worst-case scenario. The main problem and sub-problems are decoupled and solved based on the Benders decomposition algorithm. The main problem of the Benders decomposition algorithm is modeled as a mixed integer quadratic programming problem. The subproblems are transformed into semidefinite programming forms to solve the extreme value of system stability. An accelerated cutting plane generation strategy is designed, and dual variable sensitivity analysis is used to screen effective constraints.

7. The smart grid power dispatch optimization method according to claim 1, characterized in that: The method further comprises: The integrated scheduling instruction verification module uses smart contract technology to perform topological security verification and N-1 fault rehearsal on the optimization results. It generates scheduling instruction execution certificates through formal verification methods and implements cross-domain instruction synchronization under privacy protection based on zero-knowledge proof. The proposed formal verification method uses sequential logic formulas to describe grid safety constraints, including bus voltage over-limit, line overload, and frequency deviation indicators, and verifies the compliance of scheduling instructions by traversing state transition paths through symbolic model detection technology.

8. The smart grid power dispatch optimization method according to claim 1, characterized in that: The method further comprises: An online incremental learning mechanism is deployed, and the parameters of the power load spatiotemporal coupling prediction model are updated in real time using an online sequence extreme learning machine. The sliding window technology is combined to dynamically adjust the optimization model weight coefficients, and the model drift is evaluated based on the Markov chain Monte Carlo method to trigger the retraining conditions.

9. The smart grid power dispatch optimization method according to claim 8, characterized in that: The dynamic expansion strategy of hidden layer nodes of the online sequence extreme learning machine adopts the error sensitivity criterion, defines the node addition threshold as the moving average variance of the prediction residual, and designs a forgetting factor mechanism to attenuate the influence weight of historical data on model updating.

10. The smart grid power dispatch optimization method according to claim 9, characterized in that: The Markov Chain Monte Carlo method uses a sampler to generate the posterior distribution of model parameters, defines the drift indicator as the distance of the parameter distribution, and triggers the full model retraining process when the drift exceeds a preset threshold.

Images