Harbor flexible load multi-level interaction control method oriented to ship-port-network cooperation

Abstract

The invention relates to the technical field of port intellectualization, and discloses a port flexible load multi-level interactive control method for ship-port network collaboration, which comprises the following steps: acquiring a basic data set, and processing based on a graph attention mechanism and a cross-layer attention mechanism to obtain a multi-layer graph feature tensor; probability distribution calculation and tensor decomposition processing are carried out on the multi-layer graph feature tensor and historical time sequence data, and a time sequence feature matrix is obtained; determining time window division points by adopting a dynamic programming algorithm, and processing based on a cross-scale attention mechanism to obtain a multi-scale dynamic feature set; combining the physical constraint data and the logic constraint data to construct a coupling feature matrix; based on the multi-time scale objective function and the system constraint data, an improved NSGA-III algorithm is adopted to obtain a cooperative control strategy set; and carrying out iterative optimization on the control strategy by adopting a reinforcement learning method to generate an optimal control strategy. Therefore, multi-level cooperative control of the port load is realized, and the dynamic response capability and the control performance of the system are improved.

Description

Technical Field

[0001] The present invention belongs to the field of port intelligence, and in particular to a multi-level interactive control method of port flexible loads oriented to ship-port network collaboration. Background Art

[0002] With the deepening of the intelligent and green transformation of ports, the port electricity load has shown a significant flexible feature. How to make full use of the flexible characteristics of port loads and achieve coordinated optimization of ship electricity demand and port power systems is of great significance to improving port energy utilization efficiency and ensuring the stable operation of the power system.

[0003] At present, the research in the field of flexible control of port loads mainly focuses on three aspects: first, the demand response method based on load aggregation, which realizes centralized control of load groups by classifying and aggregating the loads of port equipment; second, the hierarchical control strategy based on model prediction, which uses a hierarchical architecture to decompose and coordinate control objectives of different time scales; third, the multi-objective optimization method based on heuristic algorithms, which balances and optimizes the economy and stability of the system by constructing multi-objective functions. These methods have improved the flexible control capabilities of port loads to a certain extent, but there are still some technical limitations.

[0004] Through in-depth analysis of existing technical solutions, the following main problems are found: First, most of the existing load feature extraction methods adopt independent modeling, ignoring the coupling relationship between ship power consumption, port equipment and power grid operation, and it is difficult to accurately capture the dynamic characteristics of the system; second, the traditional time window division method usually adopts a fixed window size, which cannot adapt to the differences in load changes at different time scales, resulting in insufficient accuracy of feature extraction; third, the existing control strategy optimization method has convergence problems when dealing with high-dimensional constraints, and lacks effective strategy evaluation and optimization mechanisms, making it difficult to ensure the continuous improvement of control effects; finally, when dealing with sudden changes in ship power demand, the existing solutions often have problems such as response lag and control oscillation due to the lack of inter-layer dynamic coupling modeling and multi-time scale collaborative control mechanism. These technical problems seriously restrict the effect of flexible port load regulation. Summary of the invention

[0005] In order to solve the above technical problems, the present invention provides a multi-level interactive control method of port flexible loads oriented to ship-port network collaboration.

[0006] The technical solution adopted by the present invention is as follows: The embodiment of the present invention proposes a multi-level interactive control method for port flexible loads oriented to ship-port-network collaboration, comprising the following steps: step S1, obtaining a pre-stored basic data set, wherein the basic data set comprises a ship data set, a port equipment data set and a power grid topology data set, and constructing adjacency matrices of the ship layer, the port equipment layer and the power grid layer respectively based on the basic data set, and using a graph attention mechanism to process to obtain intra-layer feature weights and inter-layer feature weights, and generating a multi-layer graph feature tensor; step S2, obtaining and based on the multi-layer graph feature tensor and pre-stored historical time series data, calculating the node feature probability distribution of the multi-layer graph feature tensor and the time series change probability distribution of the historical time series data, constructing a multi-dimensional probability tensor, and using a tensor decomposition method to process to obtain the time series feature matrix Array; Step S3, obtain the time series feature matrix, use the dynamic programming algorithm to determine the time window division point, generate a multi-scale time window sequence, and obtain a multi-scale dynamic feature set based on the cross-scale attention mechanism; Step S4, obtain the pre-stored physical constraint data and logical constraint data, combine the multi-scale dynamic feature set, construct the inter-layer static coupling matrix and dynamic coupling matrix, and then generate the coupling feature matrix; Step S5, obtain the coupling feature matrix and the pre-stored system constraint data, construct the multi-time scale objective function and constraint condition set, and use the multi-objective optimization algorithm to obtain the collaborative control strategy set; Step S6, obtain the collaborative control strategy set and the pre-stored evaluation index data, calculate the performance index value and perform strategy optimization, and generate the optimal control strategy.

[0007] Beneficial effects of the present invention: The present invention improves the ability to capture the dynamic characteristics of the system through multi-layer graph feature extraction and probabilistic time series modeling; enhances the ability to respond to load changes at different time scales through adaptive time window division and multi-scale feature fusion; and realizes continuous optimization and performance improvement of the control strategy through the combination of multi-objective optimization and reinforcement learning. The entire solution shows good adaptability and control effect in practical applications, especially when dealing with complex scenarios such as sudden changes in ship power demand, showing significant performance advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 It is a flow chart of a multi-level interactive control method of port flexible load for ship-port network collaboration according to an embodiment of the present invention.

[0009] Figure 2 is a flow chart of step S1 of one embodiment of the present invention.

[0010] Figure 3 is a flow chart of step S2 of one embodiment of the present invention.

[0011] Figure 4 is a flow chart of step S3 of one embodiment of the present invention.

[0012] Figure 5 is a flow chart of step S4 of one embodiment of the present invention.

[0013] Figure 6 is a flow chart of step S5 of one embodiment of the present invention.

[0014] Figure 7 is a flow chart of step S6 of one embodiment of the present invention. DETAILED DESCRIPTION

[0015] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.

[0016] like Figure 1 As shown, a multi-level interactive control method of port flexible load for ship-port network collaboration is provided, comprising the following steps: Step S1, obtaining a pre-stored basic data set, the basic data set includes a ship data set, a port equipment data set and a power grid topology data set, constructing adjacency matrices of the ship layer, the port equipment layer and the power grid layer based on the basic data set, respectively, using a graph attention mechanism to obtain intra-layer feature weights and inter-layer feature weights, and generating a multi-layer graph feature tensor; Step S2, obtaining and calculating the node feature probability distribution of the multi-layer graph feature tensor and the time series change probability distribution of the historical time series data based on the multi-layer graph feature tensor and the pre-stored historical time series data, constructing a multi-dimensional probability tensor, and using a tensor decomposition method to obtain a time series feature matrix; Step S3, obtaining a time series feature matrix, using a dynamic programming algorithm to determine the time window division points, generating a multi-scale time window sequence, and obtaining a multi-scale dynamic feature set based on a cross-scale attention mechanism; Step S4, obtaining pre-stored physical constraint data and logical constraint data, combining with a multi-scale dynamic feature set, constructing an inter-layer static coupling matrix and a dynamic coupling matrix, and then generating a coupling feature matrix; Step S5, obtaining the coupling characteristic matrix and pre-stored system constraint data, constructing a multi-time scale objective function and constraint condition set, and using a multi-objective optimization algorithm to obtain a collaborative control strategy set; Step S6: Obtain the collaborative control strategy set and pre-stored evaluation index data, calculate the performance index value and perform strategy optimization to generate the optimal control strategy.

[0017] In another embodiment of the present application, the basic data set may also include: power grid topology, ship available power, port charging pile equipment power, and power grid demand response power. By constructing a three-layer coordinated flexible load control system for ship-port network, intelligent and precise control of the port power system is achieved. The overall solution adopts the technical route of "multi-layer graph feature extraction-probabilistic time series modeling-multi-scale window division-coupling constraint construction-multi-objective optimization solution-reinforcement learning optimization" to form a complete closed-loop control system. At the feature extraction level, the dynamic association of the three layers of data in the ship-port network is realized through the graph attention mechanism; at the time series modeling level, the dynamic characteristics of the system are captured through probability distribution estimation and tensor decomposition; at the control optimization level, the continuous optimization of the control strategy is achieved through multi-objective optimization and reinforcement learning. This solution not only solves the problems of insufficient load response and poor inter-layer coordination in traditional port power systems, but also significantly improves the overall performance of the system through multi-time scale collaborative control. Especially when dealing with complex power consumption scenarios in large ports, the solution can adaptively adjust the control strategy to ensure the unified coordination of power quality, power balance and economy.

[0018] According to one aspect of the present application, Figure 2 As shown, step S1 is specifically as follows: Step S11: Obtain a pre-stored ship data set, and extract the ship identification, power demand and time window information therefrom; obtain a pre-stored port equipment data set, and extract the equipment identification, capacity and operating status information therefrom; obtain a pre-stored power grid topology data set, and extract the node information and edge information therefrom, and generate an initial feature data set based on the ship identification, power demand and time window information, the equipment identification, capacity and operating status information, and the node information and edge information.

[0019] Step S12: Based on the ship power demand data in the initial feature data set, the ship layer adjacency matrix is ​​calculated; based on the port equipment operation status information in the initial feature data set, the port equipment layer adjacency matrix is ​​calculated; based on the power grid node and edge information in the initial feature data set, the power grid layer adjacency matrix is ​​calculated; according to the connection relationship between the ship and the port equipment, a ship-port association matrix is ​​generated, and according to the connection relationship between the port equipment and the power grid nodes, a port-network association matrix is ​​generated.

[0020] Step S13: The ship layer adjacency matrix, the port equipment layer adjacency matrix and the power grid layer adjacency matrix are processed by the graph attention mechanism respectively to obtain the intra-layer feature weight data; the ship-port association matrix and the port-network association matrix are processed by the cross-layer attention mechanism to obtain the inter-layer feature weight data; the intra-layer feature weight data and the inter-layer feature weight data are combined and processed to generate a multi-layer graph feature tensor.

[0021] Among them, the multi-layer graph attention weight calculation formula is: Among them, the multi-layer graph attention weight calculation formula ; Where: A(i,j) is the attention weight of node i to node j; h_i and h_j are the feature vectors of nodes i and j; W is a trainable weight matrix; b is a bias vector; σ is a softmax activation function; LeakyReLU is a leaky rectified linear unit; cos(h_i, h_j) is the cosine similarity of the feature vector; p_i and p_j are the position codes of the nodes; α, β, γ are adaptive weight coefficients, optimized by gradient descent; || is a vector concatenation operation; is the square of the Euclidean distance. By introducing position encoding and cosine similarity, the modeling ability of topological relationships and feature similarities between nodes is enhanced.

[0022] In another embodiment of the present application, the ship data set may include: basic ship information: ship ID, tonnage level, maximum power demand; shore power demand: charging power curve, expected berthing time, allowed charging time window; energy storage capacity: shipboard energy storage capacity (if any); load characteristics: critical load ratio, interruptible load ratio; historical power consumption mode data; port equipment data set: charging pile information: rated power, operating efficiency, dynamic loss; converter parameters: power capacity, control dead zone, dynamic response characteristics; energy storage system: capacity, charging and discharging efficiency, state of charge (SOC) range; load classification: rigid load, flexible load, transferable load; equipment operating status: online rate, failure rate, maintenance cycle; power grid topology data set; network structure: bus connection relationship, line impedance parameters; electrical parameters: voltage level, power factor, harmonic content; protection settings: overcurrent, overvoltage setting values; operating limits: line capacity, transformer capacity; power quality indicators: voltage deviation, frequency fluctuation.

[0023] By introducing the graph attention mechanism and the cross-layer attention mechanism, the multi-level fusion processing of ship, port equipment and power grid data is carried out, and the adaptive association of the three-layer data of the ship-port network is realized. Specifically, the adjacency matrix is ​​first constructed for the ship power demand data, the port equipment operation status information and the power grid topology data, and the topological relationship within each layer is captured; then the graph attention mechanism is used to adaptively learn the degree of mutual influence between nodes within the layer, and the dynamic association relationship between different levels is established through the cross-layer attention mechanism. The multi-layer graph feature tensor finally generated not only retains the original topological structure information of each layer, but also reflects the coupling relationship between layers. This hierarchical data processing method enables the system to automatically identify and adjust the weights of key nodes within and between layers, thereby providing a comprehensive and accurate feature representation for subsequent collaborative control, and greatly improving the perception of the overall state of the ship-port network system. In addition, through the adaptive weight learning of the graph attention mechanism, the system can dynamically adapt to the changes in port load and the fluctuations in ship power demand, ensuring the real-time and accuracy of feature extraction.

[0024] According to one aspect of the present application, Figure 3 As shown, step S2 is specifically as follows: Step S21: Obtain node feature data in the multi-layer graph feature tensor, and use the probability distribution estimation method to calculate the node feature probability distribution data; obtain pre-stored historical time series data, and use the probability distribution estimation method to calculate the time series change probability distribution data; combine the node feature probability distribution data and the time series change probability distribution data to generate a multi-dimensional probability tensor.

[0025] Among them, the adaptive kernel density probability estimation formula is: ; Where: P(x) is the probability density of feature x; n is the number of samples; K_h is the Gaussian kernel function with adaptive bandwidth, ; u is the input vector, h is the adaptive bandwidth parameter, h=h_0·(1+ η·var(x_i)); w_i is the sample weight coefficient; s_i is the sample sparsity index; λ is the sparsity penalty factor; x_i is the historical sample point; var(x_i) is the local variance; h_0 is the baseline bandwidth; η is the bandwidth adjustment coefficient.

[0026] Step S22: The multidimensional probability tensor is processed by Tucker decomposition method to obtain a core tensor and a factor matrix group; the main characteristic pattern data is extracted based on the core tensor, and combined with the factor matrix group to generate a time series feature matrix.

[0027] Among them, the tensor decomposition model ; Where: X is the original tensor; G is the core tensor; , , is the factor moment matrix, is the tensor product operation of the i-th mode; is the orthogonal constraint term; λ is the regularization coefficient; α, β, γ are the modal weight coefficients; I is the unit matrix.

[0028] Based on the multi-layer graph feature tensor and historical time series data, the dual processing mechanism of probability distribution estimation and tensor decomposition is adopted to realize the efficient extraction and compression of the system dynamic features. Specifically, the probability distribution estimation method is first used to calculate the probability distribution of node features and time series changes respectively, accurately capturing the statistical characteristics of the system state; then the Tucker tensor decomposition method is used to decompose the high-dimensional probability distribution information into core tensors and factor matrix groups, which greatly reduces the data dimension while retaining the key feature patterns. This probabilistic feature extraction method can not only effectively deal with the randomness and uncertainty in the ship-port network system, but also realize data dimensionality reduction and feature compression through tensor decomposition, significantly reducing the computational complexity of subsequent processing. At the same time, due to the use of probabilistic modeling, the system has strong robustness to noise and abnormal data, and can provide stable and reliable feature representation. Especially in the face of complex and changeable load conditions in ports, this method effectively captures the statistical laws of load changes through probabilistic modeling, providing a reliable data basis for subsequent control strategy optimization.

[0029] According to one aspect of the present application, Figure 4 As shown, step S3 is specifically as follows: Step S31: Obtain a time series feature matrix, calculate the time series correlation between features to obtain a correlation matrix; process the correlation matrix using a dynamic programming algorithm to obtain an optimal window partition point set; generate a multi-scale time window sequence based on the optimal window partition point set.

[0030] Among them, the window partition cost function ; Where: C(i,j) is the partition cost of the interval [i,j]; is the characteristic variance term; V(i,j) = |dF(j) / dt - dF(i) / dt| is the rate of change difference term; S(i,j) = exp(-(ji) / τ) is the window scale penalty term; F(i:j) is the characteristic sequence of the interval [i,j]; μ(i:j) is the interval mean; , , is the adaptive weight coefficient; τ is the scale adjustment parameter.

[0031] Step S32: segment the temporal feature matrix based on the multi-scale time window sequence to obtain a local feature matrix set; aggregate the local feature matrix set using a cross-scale attention mechanism to generate a multi-scale dynamic feature set.

[0032] By introducing dynamic programming algorithm and cross-scale attention mechanism for multi-scale division and feature extraction of time windows, adaptive capture of features at different time scales is achieved. Specifically, firstly, the temporal correlation between features is calculated based on the time series feature matrix, and the correlation matrix is ​​constructed, which accurately reflects the correlation pattern of features changing over time; then the dynamic programming algorithm is used to optimally divide the correlation matrix, and the optimal window division point is automatically determined, which effectively avoids the information loss problem caused by artificially fixed window size. Through the cross-scale attention mechanism, the system can adaptively fuse features of different time scales, and can simultaneously handle millisecond-level power quality fluctuations, second-level power balance regulation, and minute-level load scheduling requirements, significantly improving the ability to capture dynamic features at multiple time scales. Especially in the face of sudden changes in ship power load, this method can quickly locate key time points and accurately extract response features of different time scales, providing temporal feature support for subsequent hierarchical control strategies.

[0033] According to one aspect of the present application, Figure 5 As shown, step S4 is specifically as follows: Step S41: obtaining pre-stored physical constraint data and constructing a physical constraint matrix; obtaining pre-stored logical constraint data and constructing a logical constraint matrix; combining the physical constraint matrix and the logical constraint matrix to generate a static coupling matrix.

[0034] Step S42: Obtain a multi-scale dynamic feature set, calculate the inter-layer time-varying coupling strength to obtain a dynamic coupling matrix; fuse the static coupling matrix and the dynamic coupling matrix to generate a coupling feature matrix.

[0035] The coupling strength calculation formula is: ; Where: H(t) is the coupling strength at time t; is the frequency domain feature item, N is the number of sampling points; is the time series feature item; is the modal feature item; w_i is the frequency component weight; fi is the characteristic frequency; φ_i is the phase; h_t and c_t are the LSTM hidden state and memory unit; a_t is the attention weight; s_t is the state vector; α(t), β(t), and γ(t) are time-varying weight coefficients.

[0036] By combining static physical constraints and dynamic logical constraints to construct a coupling feature matrix, the comprehensive modeling of multi-dimensional constraints of the ship-port network system is realized. Specifically, the constraint matrix is ​​first constructed based on the physical constraint data and the logical constraint data, which comprehensively covers static constraints such as equipment operation restrictions and network security boundaries; then the multi-scale dynamic feature set is used to calculate the time-varying coupling strength between layers, accurately describing the dynamic constraint relationship during the system operation process. This modeling method combining static constraints and dynamic constraints not only ensures that the control strategy meets the basic physical and safety constraints, but also adapts to the dynamic changes in the system operation state. Through the coupling feature matrix obtained by fusion processing, the system can achieve the coordinated optimization of ship power demand, port equipment operation and power grid safety while ensuring the satisfaction of various constraints. Especially when dealing with multi-time scale collaborative control problems, this method can respond to the changes in constraint conditions in a timely manner through the dynamic update of the coupling feature matrix, ensuring the real-time effectiveness of the control strategy.

[0037] According to one aspect of the present application, Figure 6 As shown, step S5 is specifically as follows: Step S51: obtain pre-stored frequency regulation index data and construct a millisecond-level objective function; obtain pre-stored voltage regulation index data and construct a second-level objective function; obtain pre-stored peak regulation index data and construct a minute-level objective function; combine and process these three objective functions to generate a multi-scale objective function set.

[0038] Among them, the millisecond-level goal is mainly aimed at the rapid response of power electronic equipment; handling frequency perturbations and rapid power fluctuations; and realizing the primary frequency modulation function. Second-level goals: focus on voltage stability control; coordinate the power distribution of multiple charging piles; realize secondary frequency modulation and voltage regulation. Minute-level goals: optimize load curves and peak-to-valley differences; coordinate ship charging schedules; realize economic scheduling and demand response. Form a hierarchical and time-sharing coordinated control system, each targeting different control objectives, and jointly ensure the stability and economy of the system. Fast response ensures stability, medium-speed response optimizes the operating state, and slow response achieves economic scheduling, cooperating with each other.

[0039] Step S52: Obtain the equipment operation limit values ​​in the pre-stored system constraint data to generate an equipment constraint set; obtain the network security parameters in the pre-stored system constraint data to generate a network constraint set; obtain the timing correlation data in the coupling characteristic matrix to generate a timing constraint set; combine these three constraint sets to generate a comprehensive constraint condition set.

[0040] Step S53: construct a multi-objective optimization model based on the multi-scale objective function set and the comprehensive constraint condition set; use the improved NSGA-III algorithm (non-dominated sorting genetic algorithm) to solve the model and generate a collaborative control strategy set.

[0041] Among them, the improved NSGA-III algorithm is: fitness evaluation function ;in: To control the accuracy target; To control the stability target; is the constraint violation target; w_i is the control error weight; λ_i is the reference point weight; r_i is the position of the current solution in the target space; is the ideal reference point; α_j, β_j are the control cost weights; γ_k, η_k are the constraint violation penalty coefficients; g_k(x) and h_k(x) are the inequality and equality constraint functions, respectively; Δu_j is the control variable change; y_i and are the actual output and the expected output respectively.

[0042] By constructing multi-time scale objective functions and introducing the improved NSGA-III algorithm for multi-objective optimization, the coordinated optimization of different levels of control objectives of the ship-port network system is realized. Specifically, based on the control requirements of different time scales, the millisecond-level frequency regulation objective function, the second-level voltage regulation objective function and the minute-level peak regulation objective function are constructed respectively, and the multi-level control requirements of the system are fully considered; then the comprehensive constraint condition set is generated through the system constraint data to ensure the feasibility of the control strategy. The improved NSGA-III algorithm is used for solving, and through mechanisms such as non-dominated sorting, adaptive crossover and guided mutation, it can efficiently explore the high-dimensional decision space and obtain the Pareto optimal solution set. This multi-level optimization method can not only meet the control requirements of different time scales at the same time, but also ensure the diversity and convergence of the solution through the improved evolutionary algorithm, which significantly improves the quality and reliability of the control strategy. Especially in the face of complex scenarios such as port load fluctuations, this method can quickly generate a collaborative control strategy that meets multiple constraints, effectively ensuring the stable operation of the system.

[0043] According to one aspect of the present application, Figure 7 As shown, step S6 is specifically as follows: Step S61: Obtain the accuracy requirements in the collaborative control strategy set and pre-stored evaluation index data, and calculate the control accuracy index value; obtain the time requirements in the collaborative control strategy set and pre-stored evaluation index data, and calculate the response time index value; obtain the stability requirements in the collaborative control strategy set and pre-stored evaluation index data, and calculate the stability index value; combine these three types of index values ​​to generate a comprehensive evaluation index set.

[0044] Step S62: construct an evaluation function based on the comprehensive evaluation index set; use the reinforcement learning method to iteratively optimize the collaborative control strategy set to generate the optimal control strategy.

[0045] The reinforcement learning method uses a deep deterministic policy gradient algorithm, and the policy gradient update formula is ; Where: J(θ) is the policy objective function; Q(s,a) is the state-action value function; V(s) is the state value function; π_θ(a|s) is the parameterized policy function; is the policy entropy term; is the parameter regularization term; α is the entropy weight coefficient; β is the regularization weight coefficient; is the parameter change penalty coefficient; is the parameter norm penalty coefficient; c is the parameter threshold; is the strategy parameter of the previous round.

[0046] The improved NSGA-III algorithm introduces reference point weights and control stability objectives, and achieves a trade-off between control accuracy and system stability through multi-objective optimization. At the same time, an adaptive penalty mechanism is used to handle constraint violations, which improves the algorithm's solution performance under complex constraints. The improved deep deterministic policy gradient algorithm adds policy entropy terms and parameter regularization terms. The policy entropy term promotes the exploratory nature of the policy and prevents premature convergence to the local optimum; the parameter regularization term limits the range of change of the policy parameters and improves the stability of the learning process. Through these improvements, the algorithm improves the effect of policy optimization while ensuring convergence.

[0047] By introducing reinforcement learning methods to iteratively optimize the control strategy, adaptive optimization and performance improvement of the control strategy are achieved. Specifically, firstly, the control accuracy, response time and stability index values ​​are calculated based on the evaluation index data, and a comprehensive performance evaluation system is constructed; then the control strategy is evaluated and optimized using the deep deterministic policy gradient algorithm, and the performance of the strategy is continuously improved through the experience replay mechanism and soft update algorithm. This optimization method based on reinforcement learning has significant advantages: through continuous interaction with the environment and strategy updates, the system can continuously accumulate experience and improve the control strategy, and has strong adaptability; the experience replay mechanism is used to improve the sample utilization efficiency and accelerate the strategy convergence; the soft update algorithm ensures the stability of the strategy update and avoids drastic fluctuations in the strategy. Especially when dealing with complex dynamic systems such as ship port networks, this method can gradually improve the performance of the control strategy through continuous learning and optimization, and achieve a comprehensive improvement in control accuracy, response speed and system stability.

[0048] According to one aspect of the present application, step S11 is specifically: Step S111, obtain a pre-stored ship data set, use a data segmentation algorithm to encode the ship identification to obtain identification coding data, use a sliding time window method to segment the power demand data to obtain segmented power data, use a timing encoder to map the time window information to obtain time feature data, and fuse the identification coding data, segmented power data and time feature data to obtain initial ship feature data.

[0049] Among them, the segment coding function ; Where: E(x) is the segment code value; is the feature transformation function; is the difference between adjacent segments; w_i is the segment weight coefficient; W_i is the feature transformation matrix; b_i is the bias vector; λ is the difference penalty factor; s_i is the feature mean of the i-th segment; N is the number of segments; sigmoid is the S-type activation function.

[0050] The improved sliding time window method is as follows: power segmentation function ; Where: P(t) is the segment power characteristic at time t: is the weighted mean term; is the weighted standard deviation term; d(t) = (p_t - p_(tL)) / L is the trend term; p_i is the original power value; w_i is the time attenuation weight; L is the window length; α(t), β(t), and γ(t) are adaptive weight coefficients.

[0051] Step S112, obtaining a pre-stored data set of port equipment, using an adaptive quantization method to discretize the equipment capacity data to obtain discrete capacity data, encoding the equipment operation status data to obtain status encoding data, and fusing the discrete capacity data and status encoding data to obtain initial port characteristic data.

[0052] The adaptive quantization method is: capacity quantization function ; Where: Q(x) is the quantized capacity value; is the Gaussian kernel function; is the entropy term; q_i is the quantization center; μ_i is the Gaussian kernel mean; Σ_i is the covariance matrix; p_i is the quantization probability; λ is the entropy weight coefficient; M is the quantization level.

[0053] Step S113, obtaining a pre-stored power grid topology data set, using a depth-first search algorithm to traverse the node information to obtain node sequence data, using a minimum spanning tree algorithm to simplify the edge information to obtain simplified edge set data, and combining the node sequence data and the simplified edge set data to obtain the initial power grid characteristic data.

[0054] Among them, the depth-first search algorithm is specifically: node importance calculation function ; Where: I(v) is the importance of node v; C(v) = |N(v)| / max_u|N(u)| is the node degree centrality; is the betweenness centrality; D(v) = 1 / ∑(u∈V)d(v,u) is the closeness centrality; N(v) is the set of neighbors of node v; σ_st is the number of shortest paths from s to t; σ_st(v) is the number of shortest paths from s to t passing through node v; d(v,u) is the shortest distance from node v to u; α, β, γ are adaptive weight coefficients.

[0055] Among them, the minimum spanning tree algorithm is specifically: edge weight calculation function ;W(e) is the comprehensive weight of edge e; is the feature similarity term; is the traffic weight item; is the time series correlation term; f_i and f_j are the feature vectors of edge-connected nodes; p_i and p_j are the node power values; t_i and t_j are timestamps; α, β, γ are adaptive weight coefficients; σ is the similarity scale parameter; τ is the time series attenuation factor.

[0056] Step S114: input the initial data of ship characteristics, the initial data of port characteristics and the initial data of power grid characteristics into the autoencoder for dimensionality reduction processing to obtain an initial feature data set.

[0057] By introducing multiple data processing mechanisms to standardize and preprocess the basic data, effective integration of heterogeneous data is achieved. Specifically, the ship identification is first encoded using a data segmentation algorithm to ensure the unique identification of ship information; then the power demand data is segmented using the sliding time window method to accurately capture the load change characteristics; finally, the time window information is mapped using a time series encoder to establish a unified time series representation. For port equipment data, the capacity data is discretized through an adaptive quantization method to ensure the standardization of the data; for power grid topology data, depth-first search and minimum spanning tree algorithms are used to process and extract key network structure information. The processed data is reduced in dimension by using an autoencoder, which not only retains the key features of the original data, but also reduces the computational complexity of subsequent processing. This systematic data preprocessing method lays a reliable data foundation for subsequent feature extraction and control optimization.

[0058] According to one aspect of the present application, step S13 is specifically: Step S131: Obtain intra-layer feature data, use a gated graph convolutional network to extract spatial features to obtain spatial feature data, use a recursive neural network to extract temporal features to obtain temporal feature data, and fuse the spatial feature data and the temporal feature data to obtain intra-layer enhanced feature data.

[0059] Among them, the improved algorithm of the gated graph convolutional network is as follows: spatial feature extraction function ; Where: F(X) is the extracted spatial feature; is the normalized adjacency matrix; g(X) = sigmoid(XW_g + b_g) is the gating function; W_f and W_g are learnable weight matrices; D is the degree matrix; λ is the self-loop weight; ⊙ is the Hadamard product; σ is the activation function; X is the input feature matrix; I is the identity matrix.

[0060] Step S132: Obtain inter-layer correlation data, use a bidirectional attention mechanism to calculate inter-layer feature correlation to obtain correlation feature data, and use a residual network to perform feature enhancement processing to obtain inter-layer enhanced feature data.

[0061] Step S133: Obtain the intra-layer enhanced feature data and the inter-layer enhanced feature data, use an adaptive feature fusion algorithm to perform combination processing to obtain fused feature data, use a tensor network to perform high-dimensional feature mapping processing, and generate a multi-layer graph feature tensor.

[0062] In another embodiment of the present application, step S13 may also be: Step S131: Acquire intra-layer feature data, use a node feature embedding algorithm to perform dimension mapping on the node data to obtain node embedding data, use a gated graph convolutional network to extract spatial features from the node embedding data to obtain spatial feature data, and use a recursive neural network to extract temporal features from the spatial feature data to obtain intra-layer enhanced feature data.

[0063] Step S132: Obtain inter-layer correlation data, use a feature embedding network to perform dimension alignment processing on the correlation data to obtain aligned feature data, use a bidirectional attention mechanism to perform inter-layer correlation calculation on the aligned feature data to obtain correlation feature data, and use a residual network to perform feature enhancement processing on the correlation feature data to obtain inter-layer enhanced feature data.

[0064] Step S133: Obtain intra-layer enhanced feature data and inter-layer enhanced feature data, use an adaptive feature fusion algorithm to perform feature combination processing to obtain fused feature data, use a feature consistency verification algorithm to perform data verification processing on the fused feature data to obtain verification feature data, use a tensor network to perform high-dimensional feature mapping processing on the verification feature data, and generate a multi-layer graph feature tensor.

[0065] By introducing feature embedding preprocessing and multi-level attention mechanism, the deep fusion and feature enhancement of the three-layer data of the ship and port network are realized. Specifically, the node feature embedding algorithm is first used to map the dimensions of the original node data to solve the dimension mismatch problem; then the spatial features are extracted by the gated graph convolutional network, and the temporal features are captured by the recurrent neural network, so as to realize the comprehensive extraction of intra-layer features. In terms of inter-layer feature processing, the feature embedding network is used for dimension alignment, the inter-layer correlation is calculated by the bidirectional attention mechanism, and the residual network is used for feature enhancement, so as to ensure the effective fusion of features at different levels. Finally, the reliability and accuracy of the fused features are guaranteed by the adaptive feature fusion algorithm and the feature consistency verification algorithm. This multi-level feature processing method not only solves the dimensional matching and feature loss problems in the traditional method, but also realizes the prominent enhancement of key features through the attention mechanism, which significantly improves the quality of feature representation. Especially when dealing with dynamic scenes such as ship load mutation, this method can quickly capture key features and make adaptive adjustments, providing reliable feature support for subsequent control decisions.

[0066] According to one aspect of the present application, step S21 is specifically: Step S211: Obtain a multi-layer graph feature tensor, use an adaptive kernel density estimation method to perform probability distribution calculation on the node feature data to obtain the node probability initial data, use a Bayesian network to perform conditional probability inference on the node probability initial data to obtain the node conditional probability data, and use the information entropy criterion to optimize the node conditional probability data to obtain the node feature probability distribution data.

[0067] The adaptive kernel density estimation method is specifically: probability density function ;in: is the kernel function; is the adaptive bandwidth; d(x,x_i) is the feature distance; w_i is the sample weight; τ is the temperature parameter; h_0 is the reference bandwidth; σ(x) is the local standard deviation; d is the feature dimension; α is the bandwidth adjustment coefficient.

[0068] Among them, the information entropy criterion is specifically: node probability optimization function ; Where: H(p) is the optimization target; p_i is the node probability distribution; q_i is the prior probability distribution; KL(p||q) is the KL divergence; w_i is the node importance weight; λ is the regularization coefficient; N is the number of nodes.

[0069] Step S212: Obtain pre-stored historical time series data, use a dynamic time warping algorithm to align the time series data to obtain aligned time series data, use a hidden Markov model to calculate the state transition probability of the aligned time series data to obtain transition probability data, and use a sequence prediction model to extract evolutionary features of the transition probability data to obtain time series change probability distribution data.

[0070] The dynamic time warping algorithm is as follows: Alignment cost function , ; Where: D(i,j) is the cumulative alignment cost; d(x_i,y_j) is the local distance metric; γ is the path weight coefficient; x_i and y_j are the i-th and j-th points in the sequence to be aligned respectively; min{} is the minimum value operation.

[0071] Step S213: Obtain node feature probability distribution data and time series change probability distribution data, use a tensor network to perform high-dimensional mapping processing on the probability distribution data to obtain mapping probability data, use an adaptive probability fusion algorithm to integrate the mapping probability data to obtain fused probability data, reconstruct the fused probability data based on the tensor construction criteria, and generate a multidimensional probability tensor.

[0072] By integrating probability distribution estimation and data densification processing, high-quality extraction and representation of system dynamic features are achieved. Specifically, the adaptive kernel density estimation method is used to calculate the preliminary probability distribution of node features, and then conditional probability reasoning is performed through the Bayesian network, and the information entropy criterion is used for optimization to obtain accurate node feature probability distribution. For time series data, the dynamic time warping algorithm is used for alignment processing, the state transition probability is calculated through the hidden Markov model, and the evolution characteristics are extracted using the sequence prediction model. Crucially, the data densification processing mechanism is introduced in the probability distribution calculation process to effectively solve the problem of data sparsity. Through the tensor network for high-dimensional mapping and the adaptive probability fusion algorithm for integration, the multi-dimensional probability tensor finally generated not only maintains the density of the data, but also accurately reflects the dynamic characteristics of the system. This probabilistic feature extraction method can not only effectively deal with the randomness and uncertainty in the system, but also significantly improve the quality of feature representation through data densification, providing a reliable data basis for subsequent dynamic feature analysis.

[0073] In another embodiment of the present application, step S2 may also be: Step S21: Obtain node feature data in the multi-layer graph feature tensor, use the probability distribution estimation method to calculate the initial node probability data, use the data densification algorithm to perform density enhancement processing on the initial node probability data to obtain node density data, and use the probability correction network to optimize the node density data to obtain node feature probability distribution data; obtain pre-stored historical time series data, use the time series probability estimation method to calculate the initial time series probability data, use the data densification algorithm to perform density enhancement processing on the initial time series probability data to obtain time series density data, and use the probability correction network to optimize the time series density data to obtain time series change probability distribution data; combine the node feature probability distribution data and the time series change probability distribution data to generate a multidimensional probability tensor.

[0074] S22: Decompose and extract features from the multidimensional probability tensor to generate a time series feature matrix.

[0075] According to one aspect of the present application, step S31 is specifically: Step S311: Obtain a time series feature matrix, use a local binary pattern algorithm to perform feature encoding processing on the time series data to obtain feature encoding data, use the mutual information criterion to perform correlation measurement processing on the feature encoding data to obtain correlation measurement data, and use an adaptive threshold method to screen the correlation measurement data to obtain a correlation matrix.

[0076] Among them, the local binary pattern algorithm is specifically: feature encoding function ; Where: s(x) is the sign function, which is 1 when x≥0 and 0 otherwise; g_c is the eigenvalue of the center point; g_p is the eigenvalue of the neighborhood point; w_p= exp(-d_p / σ) is the adaptive weight; d_p is the distance from the neighborhood point to the center point; P is the number of neighborhood points; σ is the distance scale parameter.

[0077] Step S312: Obtain the correlation matrix, use the interval graph construction algorithm to convert the correlation data into a graph structure to obtain interval graph data, use the shortest path algorithm to optimize the traversal of the interval graph data to obtain path feature data, and use the dynamic programming algorithm to calculate the partition points of the path feature data to obtain the optimal window partition point set.

[0078] Among them, the shortest path improvement algorithm is specifically: path cost function ; Where: C(p) is the total cost of path p; d(v_i,v_(i+1)) is the distance between nodes; f_i is the node feature vector; s(v_i,v_(i+1)) is the state transfer cost; w_t, w_f, w_s are adaptive weight coefficients; L is the path length.

[0079] The interval graph construction algorithm is as follows: interval similarity function ; Where: S(I_i,I_j) is the similarity between intervals I_i and I_j; T(I_i,I_j) = |I_i∩I_j| / |I_i∪I_j| is the temporal overlap; is the feature overlap; is the duration similarity; f_i and f_j are interval feature vectors; d_i and d_j are interval durations; α, β, γ are adaptive weight coefficients.

[0080] Step S313: Obtain the optimal window division point set, use a multi-scale decomposition algorithm to perform hierarchical division processing on the division point data to obtain hierarchical sequence data, use an adaptive window generation algorithm to perform window construction processing on the hierarchical sequence data to obtain window construction data, use a time series alignment algorithm to perform boundary optimization processing on the window construction data, and generate a multi-scale time window sequence.

[0081] Among them, the multi-scale decomposition algorithm is specifically: scale decomposition function ; Where: M(x,s) is the feature representation at scale s; φ_k(x) is the basis function of the kth scale; w_k is the scale weight; s_k is the reference scale; τ is the scale attenuation parameter; K is the number of decomposition levels.

[0082] The timing alignment algorithm is as follows: Alignment cost function ; Where: A(i,j) is the cumulative alignment cost; is the characteristic distance; is the rate difference term; is the time penalty term; f() is the feature mapping function; r_i and r_j are the change rates; t_i and t_j are the timestamps; λ and μ are weight coefficients. The adaptive window generation algorithm can be implemented using existing technologies and will not be described in detail here.

[0083] In another embodiment of the present application, step S313 may also be: Step S313: Obtain the optimal window division point set, use a multi-scale decomposition algorithm to perform hierarchical division processing on the division point data to obtain hierarchical sequence data, use a control strategy mapping algorithm to perform time scale association processing on the hierarchical sequence data to obtain associated sequence data, use an adaptive window generation algorithm to perform window construction processing on the associated sequence data to obtain window construction data, use a time series alignment algorithm to perform boundary optimization processing on the window construction data, and generate a multi-scale time window sequence.

[0084] By integrating local feature encoding and dynamic programming optimization, adaptive segmentation and multi-scale representation of time series data are realized. Specifically, the local binary pattern algorithm is first used to encode the features of time series data, and the correlation is measured by the mutual information criterion to establish an accurate feature correlation matrix; then the interval graph construction algorithm and the shortest path algorithm are used for optimization traversal, and the optimal partitioning point is calculated by combining the dynamic programming algorithm to realize the adaptive partitioning of the time window. On this basis, the hierarchical partitioning is performed by the multi-scale decomposition algorithm, the window sequence is constructed by the adaptive window generation algorithm, and the boundary is optimized by the time series alignment algorithm, and finally a multi-scale time window sequence is generated. This time window partitioning method based on feature encoding and dynamic optimization can not only accurately capture the feature changes at different time scales, but also improve the accuracy of time series feature extraction through the adaptive partitioning mechanism. Especially when dealing with multi-time scale collaborative control, this method can automatically adjust the size and position of the time window according to the dynamic characteristics of the system, ensuring the accurate capture of key time series features.

[0085] According to one aspect of the present application, step S42 is specifically: Step S421: Acquire a multi-scale dynamic feature set, use a wavelet decomposition algorithm to perform multi-scale decomposition processing on the feature data to obtain decomposed feature data, use a time-varying Fourier transform to perform frequency domain feature extraction on the decomposed feature data to obtain frequency domain feature data, and use an adaptive spectrum analysis algorithm to perform dynamic feature enhancement processing on the frequency domain feature data to obtain initial coupling strength data.

[0086] The wavelet decomposition algorithm in step S421 is specifically: multi-scale feature extraction function ;W(f,s,τ) is the wavelet coefficient; f(t) is the input signal; ψ(t) is the wavelet basis function; s is the scale parameter; τ is the translation parameter; w(t) is the adaptive weight function; is the complex conjugate operation.

[0087] The adaptive spectrum analysis algorithm is specifically: spectrum characteristic function ; Where: F(ω,t) is the time-frequency characteristic; a_k(t) is the instantaneous amplitude; ω_k(t) is the instantaneous frequency; is the frequency weight; j is the imaginary unit; N is the number of components; σ is the frequency band parameter.

[0088] Step S422: Obtain initial coupling strength data, use a recursive neural network to extract the timing coupling features to obtain timing coupling data, use an attention mechanism to perform weight calculation on the timing coupling data to obtain weight feature data, and use a deep residual network to perform feature mapping on the weight feature data to obtain a dynamic coupling matrix.

[0089] Step S423: Obtain a static coupling matrix and a dynamic coupling matrix, use a modal decomposition algorithm to perform feature mode extraction processing on the coupling data to obtain modal feature data, use an adaptive weighted fusion algorithm to perform feature fusion processing on the modal feature data to obtain fused feature data, use a tensor network to perform high-dimensional feature reconstruction processing on the fused feature data, and generate a coupling feature matrix.

[0090] In another embodiment of the present application, Step S421 can also be: obtaining a multi-scale dynamic feature set, using a wavelet decomposition algorithm to perform multi-scale decomposition processing on the feature data to obtain decomposed feature data, using a time-varying Fourier transform to perform frequency domain feature extraction on the decomposed feature data to obtain initial frequency domain feature data, using a spectrum consistency verification algorithm to perform validity verification processing on the initial frequency domain feature data to obtain frequency domain feature data, using a spectrum correction network to perform dynamic feature enhancement processing on the frequency domain feature data to obtain frequency domain enhanced data, and using an adaptive spectrum analysis algorithm to perform coupling feature extraction processing on the frequency domain enhanced data to obtain initial coupling strength data.

[0091] In another embodiment of the present application, step S423 may also be: Step S423: Obtain a static coupling matrix and a dynamic coupling matrix, use a modal decomposition algorithm to perform feature mode extraction processing on the coupling data to obtain modal feature initial data, use a feature correlation analysis algorithm to perform mapping relationship verification processing on the modal feature initial data to obtain modal feature data, use an adaptive weighted fusion algorithm to perform feature fusion processing on the modal feature data to obtain fused feature data, use a consistency verification algorithm to perform data verification processing on the fused feature data to obtain verification feature data, use a tensor network to perform high-dimensional feature reconstruction processing on the verification feature data, and generate a coupling feature matrix.

[0092] Among them, the modal decomposition algorithm is specifically: modal extraction function ; Where: IMF(t) is the intrinsic mode function; x(t) is the original signal; c_i(t) is the component function; is the adaptive weight; H() is the Hilbert transform; M is the modal number; λ is the weight adjustment parameter.

[0093] By combining frequency domain analysis and dynamic feature mapping, accurate modeling and dynamic characterization of system coupling relationship are achieved. Specifically, the wavelet decomposition algorithm and time-varying Fourier transform are used to process multi-scale dynamic features, and the validity of frequency domain features is ensured by the spectrum consistency verification algorithm; then the spectrum correction network is used to enhance dynamic features, and the coupling features are extracted by the adaptive spectrum analysis algorithm to obtain the initial coupling strength data. In terms of dynamic feature processing, a recursive neural network is used to extract time series coupling features, the weights are calculated by the attention mechanism, and the deep residual network is used for feature mapping to generate a dynamic coupling matrix. Finally, the complete expression of coupling features is achieved through the modal decomposition algorithm, adaptive weighted fusion and tensor network reconstruction. The advantages of this dynamic coupling modeling method are that it can consider frequency domain and time domain features at the same time, and ensure the accuracy of feature extraction through multiple verification mechanisms. Especially when dealing with scenarios such as sudden changes in ship power load, it can quickly capture the dynamic coupling changes of the system, providing a reliable basis for real-time adjustment of control strategies.

[0094] According to one aspect of the present application, step S53 is specifically: Step S531: Obtain a multi-scale objective function set and a comprehensive constraint condition set, use the Lagrangian relaxation algorithm to transform the constraints to obtain relaxed constraint data, use the penalty function method to construct an optimization problem on the relaxed constraint data to obtain initial data of the optimization problem, and use the gradient projection method to map the initial data of the optimization problem to a feasible domain to obtain optimization model data.

[0095] Among them, the Lagrangian relaxation algorithm is specifically as follows: Relax the objective function ; Where: L(x,λ) is the relaxed Lagrangian function; f(x) is the original objective function; g_i(x) is the constraint function; λ_i is the Lagrangian multiplier; ρ is the penalty factor; m is the number of constraints.

[0096] The penalty function method is specifically: penalty cost function ; Where: P(x) is the objective function with penalty term; f(x) is the original objective function; g_i(x) is the inequality constraint; h_i(x) is the equality constraint; μ is the penalty factor; α_i and β_i are the constraint weights; p and q are the penalty exponents.

[0097] The gradient projection method is specifically: projection update function ; Where: P_C is the projection operator on the feasible domain C; is the objective function gradient; is the gradient of the constraint function; is the adaptive step size; μ_i is the constraint weight; β is the step size adjustment parameter.

[0098] Step S532: Obtain optimization model data, use a non-dominated sorting algorithm to perform stratification processing on the solution space to obtain stratified solution set data, use an adaptive crossover operator to perform feature recombination processing on the stratified solution set data to obtain recombined feature data, and use a guided mutation algorithm to perform diversity enhancement processing on the recombined feature data to obtain candidate strategy data.

[0099] Among them, the non-dominated sorting algorithm is specifically: the dominance calculation function ; Where: D(x) is the dominance of solution x; is the dominance indicator function, which is 1 when x dominates y and 0 otherwise; d(x,y) is the solution space distance; w_i is the target weight; τ is the distance scale parameter; N is the population size.

[0100] The guided mutation algorithm is specifically: mutation operation function ; Where: M(x) is the solution after mutation; N(0,Σ) is Gaussian noise; is the current optimal solution; is the gradient direction; α, β, γ are adaptive weight coefficients; Σ is the covariance matrix.

[0101] Step S533: Obtain candidate strategy data, use the reference point adaptive update algorithm to dynamically divide the strategy space to obtain divided strategy data, use the local search algorithm to perform neighborhood optimization on the divided strategy data to obtain optimized strategy data, use the elite retention strategy to screen the optimized strategy data, and generate a collaborative control strategy set.

[0102] Among them, the reference point adaptive update algorithm is specifically: reference point update function ; Among them: r_t is the reference point at time t; is the ideal point; z_c is the current optimal solution; z_n is the nearest neighbor non-dominated solution; w_1, w_2, w_3 are adaptive weights; λ is the learning rate.

[0103] The local search algorithm is specifically: search direction function ; Where: d(x) is the search direction; is the objective function gradient; is the optimal solution direction; is the neighborhood direction; is the local optimal solution; x_i is the neighborhood solution; w_i is the neighborhood weight; α, β, γ are adaptive coefficients.

[0104] In another embodiment of the present application, step S53 may also be: Step S531: Obtain a multi-scale objective function set and a comprehensive constraint condition set, use the Lagrangian relaxation algorithm to transform the constraints to obtain relaxed constraint data, use the penalty function method to construct an optimization problem on the relaxed constraint data to obtain initial optimization problem data, use the convergence analysis algorithm to perform stability assessment on the initial optimization problem data to obtain stability assessment data, and use the gradient projection method to perform feasible domain mapping on the stability assessment data to obtain optimization model data.

[0105] Step S532: Obtain optimization model data, use a non-dominated sorting algorithm to perform stratification processing on the solution space to obtain stratified solution set data, use a convergence monitoring algorithm to perform stability verification processing on the stratified solution set data to obtain verification solution set data, use an adaptive crossover operator to perform feature recombination processing on the verification solution set data to obtain recombined feature data, and use a guided mutation algorithm to perform diversity enhancement processing on the recombined feature data to obtain candidate strategy data.

[0106] Step S533: Obtain candidate strategy data, use a reference point adaptive update algorithm to dynamically divide the strategy space to obtain divided strategy data, use a convergence compensation algorithm to perform stability enhancement processing on the divided strategy data to obtain stable strategy data, use a local search algorithm to perform neighborhood optimization processing on the stable strategy data to obtain optimized strategy data, use an elite retention strategy to screen the optimized strategy data, and generate a collaborative control strategy set.

[0107] By integrating multi-objective optimization and convergence guarantee mechanism, the stable solution and performance optimization of high-dimensional control strategies are achieved. Specifically, the constraints are first transformed by Lagrangian relaxation algorithm and penalty function method, and stability is evaluated by convergence analysis algorithm, and an optimization model with convergence guarantee is established; then the solution space is stratified by non-dominated sorting algorithm, stability is verified by convergence monitoring algorithm, and the diversity of solutions is enhanced by adaptive crossover operator and guided mutation algorithm. In the strategy optimization stage, the reference point adaptive update algorithm is used for dynamic partitioning, the convergence compensation algorithm is used to enhance stability, and the local search algorithm and elite retention strategy are combined to continuously improve the quality of the solution. This optimization method with convergence guarantee not only solves the convergence problem in high-dimensional multi-objective optimization, but also improves the solution efficiency through multiple optimization mechanisms. Especially when dealing with the optimization of control strategies under complex constraints, it can stably obtain high-quality Pareto optimal solution sets.

[0108] According to one aspect of the present application, step S62 is specifically: Step S621: Obtain a comprehensive evaluation index set, use a fuzzy evaluation algorithm to quantify the index data to obtain index quantification data, use a hierarchical analysis method to perform weight calculation on the index quantification data to obtain weight distribution data, and use a comprehensive evaluation model to perform index fusion processing on the weight distribution data to obtain evaluation function data.

[0109] Among them, the fuzzy evaluation algorithm is specifically as follows: fuzzy evaluation function ; Where: E(x) is the comprehensive evaluation value; μ_i(x) is the membership function of the i-th indicator; d_i(x) is the indicator deviation; w_i is the indicator weight; σ_i is the fuzzy scale parameter; n is the number of indicators.

[0110] The specific analytic hierarchy process is: weight calculation function ;W(A) is the eigenvector; A is the judgment matrix; e_i is the i-th eigenvector component; is the geometric mean; is the consistency deviation; a_ij is the judgment matrix element; λ_max is the maximum eigenvalue; n is the matrix order.

[0111] Step S622: Obtain evaluation function data and collaborative control strategy set, use deep deterministic policy gradient algorithm to perform state evaluation processing on strategy data to obtain state value data, use advantage function method to perform action evaluation processing on state value data to obtain action value data, and use double temporal difference algorithm to perform value update processing on action value data to obtain strategy update data.

[0112] Among them, the policy network optimization method is specifically: parameter update function ; Where: L(θ_t) is the strategy loss function; is the regularization term; η is the learning rate; β is the regularization weight; α is the momentum coefficient; γ is the norm penalty coefficient; c is the parameter threshold; θ_t is the strategy parameter at time t.

[0113] Advantage function method is as follows: Advantage evaluation function ; Where: A(s,a) is the advantage value; Q(s,a) is the state-action value function; V(s) is the state value function; is the strategy entropy; is the reference action item; π(a|s) is the strategy function; is the expert demonstration action; α and β are adaptive weights; σ is the similarity parameter.

[0114] The double temporal difference algorithm is as follows: value function update function ; Where: ΔQ(s,a) is the Q value update amount; r is the instant reward; is the state value function; γ is the discount factor; λ is the mixing coefficient; α is the learning rate; For the next state; is the maximum action value.

[0115] Step S623: Acquire strategy update data, use the experience playback mechanism to sample historical data to obtain sampling strategy data, use the strategy network to optimize the parameters of the sampling strategy data to obtain optimized strategy data, use the soft update algorithm to enhance the stability of the optimized strategy data, and generate the optimal control strategy.

[0116] The specific soft update algorithm is: parameter update function ;in: is the target network parameter; θ is the online network parameter; θ is the historical optimal parameter; θ_e is the integrated parameter; τ is the soft update coefficient; w_1, w_2, w_3 are the adaptive weight coefficients; t is the update time.

[0117] By integrating deep reinforcement learning and adaptive policy optimization, continuous improvement and performance enhancement of control strategies are achieved. Specifically, the evaluation indicators are quantified and weighted using fuzzy evaluation algorithms and hierarchical analysis methods to construct a comprehensive evaluation function; then, the state is evaluated using a deep deterministic policy gradient algorithm, the action value is evaluated using the advantage function method, and the policy value is updated using a double temporal difference algorithm. In the policy optimization process, the experience replay mechanism is used to sample historical data, the parameters are optimized through the policy network, and the soft update algorithm is used to enhance the stability of the policy. This policy optimization method based on deep reinforcement learning has significant advantages: through continuous interactive learning and policy updates, the system can continuously accumulate experience and improve the control strategy; the experience replay and soft update mechanisms are used to effectively improve the learning efficiency and policy stability; especially when dealing with complex port load scenarios, this method can continuously improve the adaptability and performance of the control strategy through real-time learning and optimization.

[0118] In summary, the present invention proposes a multi-level interactive port flexible load control method for the problem of coordinated control of ship-port network, which builds a complete technical framework of "feature extraction-time series modeling-window division-coupling modeling-strategy optimization-performance improvement". At the feature extraction level, the graph attention mechanism and cross-layer attention mechanism are introduced to realize the dynamic association and feature fusion of the three layers of data: ship layer, port equipment layer and power grid layer. Through node feature embedding preprocessing and multi-level attention calculation, the problem of inter-layer feature mismatch and information loss is effectively solved.

[0119] In terms of time series modeling, the solution combines probability distribution estimation and tensor decomposition technology to perform probabilistic modeling of the system's dynamic characteristics. In particular, the introduction of a data densification processing mechanism effectively solves the data sparsity problem in traditional methods. In the time window division process, a dynamic programming algorithm is used to calculate the optimal division point, combined with a cross-scale attention mechanism, to achieve adaptive capture of features at different time scales, breaking through the limitations of traditional fixed window methods.

[0120] In terms of control strategy optimization, the solution comprehensively considers the control requirements of millisecond-level frequency regulation, second-level voltage regulation, and minute-level peak regulation by constructing a multi-time scale objective function. Multi-objective optimization is performed through the improved NSGA-III algorithm, and a convergence compensation mechanism is introduced to effectively solve the convergence problem under high-dimensional constraints. In particular, in the strategy optimization stage, a deep reinforcement learning method is introduced, and continuous optimization and performance improvement of the control strategy are achieved through the experience replay mechanism and soft update algorithm.

[0121] The technical advantages of the present invention are mainly reflected in the following aspects: first, through multi-layer graph feature extraction and probabilistic time series modeling, the ability to capture the dynamic characteristics of the system is improved; second, through adaptive time window division and multi-scale feature fusion, the ability to respond to load changes at different time scales is enhanced; third, through the combination of multi-objective optimization and reinforcement learning, the continuous optimization and performance improvement of the control strategy are achieved. The scheme shows good adaptability and control effect in practical applications, especially when dealing with complex scenarios such as sudden changes in ship power demand, it shows significant performance advantages.

[0122] Although 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 the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A multi-level interactive control method for port flexible loads oriented to ship-port network collaboration, characterized in that: The steps include: Step S1, obtaining a pre-stored basic data set, wherein the basic data set includes a ship data set, a port equipment data set, and a power grid topology data set, and constructing adjacency matrices of a ship layer, a port equipment layer, and a power grid layer based on the basic data set, respectively, and using a graph attention mechanism to obtain intra-layer feature weights and inter-layer feature weights, and generating a multi-layer graph feature tensor; Step S2, obtaining and calculating the node feature probability distribution of the multi-layer graph feature tensor and the time series change probability distribution of the historical time series data based on the multi-layer graph feature tensor and the pre-stored historical time series data, constructing a multi-dimensional probability tensor, and using a tensor decomposition method to obtain a time series feature matrix; Step S3, obtaining a time series feature matrix, using a dynamic programming algorithm to determine the time window division points, generating a multi-scale time window sequence, and obtaining a multi-scale dynamic feature set based on a cross-scale attention mechanism; Step S4, obtaining pre-stored physical constraint data and logical constraint data, combining with a multi-scale dynamic feature set, constructing an inter-layer static coupling matrix and a dynamic coupling matrix, and then generating a coupling feature matrix; Step S5, obtaining the coupling characteristic matrix and pre-stored system constraint data, constructing a multi-time scale objective function and constraint condition set, and using a multi-objective optimization algorithm to obtain a collaborative control strategy set; Step S6: Obtain the collaborative control strategy set and pre-stored evaluation index data, calculate the performance index value and perform strategy optimization to generate the optimal control strategy.

2. The multi-level interactive control method of port flexible load for ship-port network collaboration according to claim 1 is characterized in that: Step S1 is specifically as follows: Step S11, obtaining a pre-stored ship data set, extracting the ship identification, power demand and time window information therein; obtaining a pre-stored port equipment data set, extracting the equipment identification, capacity and operating status information therein; obtaining a pre-stored power grid topology data set, extracting the node information and edge information therein; generating an initial feature data set according to the ship identification, power demand and time window information, the equipment identification, capacity and operating status information, and the node information and edge information; Step S12: Based on the ship power demand data in the initial feature data set, a ship-level adjacency matrix is ​​calculated; Based on the port equipment operation status information in the initial feature data set, the port equipment layer adjacency matrix is ​​calculated; Based on the grid node and edge information in the initial feature data set, the grid layer adjacency matrix is ​​calculated; Generate a ship-port association matrix based on the connection relationship between ships and port equipment, and generate a port-grid association matrix based on the connection relationship between port equipment and power grid nodes; Step S13, the ship layer adjacency matrix, the port equipment layer adjacency matrix and the power grid layer adjacency matrix are processed respectively using the graph attention mechanism to obtain intra-layer feature weight data; The ship-port association matrix and the port-network association matrix are processed using a cross-layer attention mechanism to obtain the inter-layer feature weight data; the intra-layer feature weight data and the inter-layer feature weight data are combined to generate a multi-layer graph feature tensor.

3. The multi-level interactive control method of port flexible load for ship-port network collaboration according to claim 1 is characterized in that: Step S2 is specifically as follows: Step S21, obtaining node feature data in the multi-layer graph feature tensor, and using a probability distribution estimation method to calculate the node feature probability distribution data; obtaining pre-stored historical time series data, and using a probability distribution estimation method to calculate the time series change probability distribution data; combining the node feature probability distribution data and the time series change probability distribution data to generate a multi-dimensional probability tensor; Step S22: Process the multidimensional probability tensor using the Tucker decomposition method to obtain a core tensor and a factor matrix group; extract the main feature pattern data based on the core tensor, combine it with the factor matrix group, and generate a time series feature matrix.

4. The multi-level interactive control method of port flexible load for ship-port network collaboration according to claim 1 is characterized in that: Step S3 is specifically as follows: Step S31, obtaining a time series feature matrix, calculating the time series correlation between features to obtain a correlation matrix; processing the correlation matrix using a dynamic programming algorithm to obtain an optimal window partition point set; generating a multi-scale time window sequence according to the optimal window partition point set; Step S32: segment the temporal feature matrix based on the multi-scale time window sequence to obtain a local feature matrix set; aggregate the local feature matrix set using a cross-scale attention mechanism to generate a multi-scale dynamic feature set.

5. The multi-level interactive control method of port flexible load for ship-port network collaboration according to claim 1 is characterized in that: Step S4 is specifically as follows: Step S41, obtaining pre-stored physical constraint data and constructing a physical constraint matrix; obtaining pre-stored logical constraint data and constructing a logical constraint matrix; combining the physical constraint matrix and the logical constraint matrix to generate a static coupling matrix; Step S42, obtaining a multi-scale dynamic feature set, calculating the inter-layer time-varying coupling strength to obtain a dynamic coupling matrix; The static coupling matrix and the dynamic coupling matrix are fused to generate a coupling feature matrix.

6. The multi-level interactive control method of port flexible load for ship-port network collaboration according to claim 1 is characterized in that: Step S5 is specifically as follows: Step S51, obtaining pre-stored frequency modulation index data, constructing a millisecond-level objective function; obtaining pre-stored voltage modulation index data, constructing a second-level objective function; obtaining pre-stored peak modulation index data, constructing a minute-level objective function; combining and processing these three objective functions to generate a multi-scale objective function set; Step S52: Obtain the equipment operation limit values ​​in the pre-stored system constraint data to generate an equipment constraint set; obtain the network security parameters in the pre-stored system constraint data to generate a network constraint set; obtain the timing correlation data in the coupling characteristic matrix to generate a timing constraint set; combine and process these three constraint sets to generate a comprehensive constraint condition set; Step S53: construct a multi-objective optimization model based on the multi-scale objective function set and the comprehensive constraint condition set; solve the model using the improved NSGA-III algorithm to generate a collaborative control strategy set.

7. The multi-level interactive control method of port flexible load for ship-port network collaboration according to claim 1 is characterized in that: Step S6 is specifically as follows: Step S61: obtaining the accuracy requirements in the collaborative control strategy set and the pre-stored evaluation index data, and calculating the control accuracy index value; obtaining the time requirements in the collaborative control strategy set and the pre-stored evaluation index data, and calculating the response time index value; Obtain the stability requirements in the collaborative control strategy set and the pre-stored evaluation index data, and calculate the stability index value; combine the three types of index values ​​to generate a comprehensive evaluation index set; Step S62: constructing an evaluation function based on the comprehensive evaluation index set; The reinforcement learning method is used to iteratively optimize the collaborative control strategy set to generate the optimal control strategy.

8. The multi-level interactive control method of port flexible load for ship-port-network collaboration according to claim 2 is characterized in that: Step S11 is specifically as follows: Step S111, obtaining a pre-stored ship data set, encoding the ship identification using a data segmentation algorithm to obtain identification coding data, segmenting the power demand data using a sliding time window method to obtain segmented power data, mapping the time window information using a time series encoder to obtain time feature data, and fusing the identification coding data, segmented power data and time feature data to obtain initial ship feature data; Step S112, obtaining a pre-stored data set of port equipment, using an adaptive quantization method to discretize the equipment capacity data to obtain discrete capacity data, encoding the equipment operation status data to obtain status encoding data, and fusing the discrete capacity data and the status encoding data to obtain initial port characteristic data; Step S113, obtaining a pre-stored power grid topology data set, using a depth-first search algorithm to traverse the node information to obtain node sequence data, using a minimum spanning tree algorithm to simplify the edge information to obtain simplified edge set data, and combining the node sequence data and the simplified edge set data to obtain the initial power grid characteristic data; Step S114: input the initial data of ship characteristics, the initial data of port characteristics and the initial data of power grid characteristics into the autoencoder for dimensionality reduction processing to obtain an initial feature data set.

9. The multi-level interactive control method of port flexible load for ship-port network collaboration according to claim 3 is characterized in that: Step S21 is specifically as follows: Step S211, obtaining a multi-layer graph feature tensor, using an adaptive kernel density estimation method to perform probability distribution calculation on the node feature data to obtain node probability initial data, using a Bayesian network to perform conditional probability reasoning on the node probability initial data to obtain node conditional probability data, and using an information entropy criterion to optimize the node conditional probability data to obtain node feature probability distribution data; Step S212, obtaining pre-stored historical time series data, using a dynamic time warping algorithm to align the historical time series data to obtain aligned time series data, using a hidden Markov model to calculate the state transition probability of the aligned time series data to obtain transition probability data, and using a sequence prediction model to extract evolution features of the transition probability data to obtain time series change probability distribution data; Step S213, obtain node feature probability distribution data and time series change probability distribution data, use a tensor network to perform high-dimensional mapping processing on the probability distribution data to obtain mapping probability data, use an adaptive probability fusion algorithm to integrate the mapping probability data to obtain fused probability data, reconstruct the fused probability data based on the tensor construction criterion, and generate a multidimensional probability tensor.

10. The multi-level interactive control method of port flexible load for ship-port-network collaboration according to claim 4 is characterized in that: Step S31 is specifically as follows: Step S311, obtaining a time series feature matrix, using a local binary pattern algorithm to perform feature encoding processing on the time series data to obtain feature encoding data, using a mutual information criterion to perform correlation measurement processing on the feature encoding data to obtain correlation measurement data, and using an adaptive threshold method to screen the correlation measurement data to obtain a correlation matrix; Step S312, obtaining a correlation matrix, using an interval graph construction algorithm to perform graph structure conversion processing on the correlation data to obtain interval graph data, using a shortest path algorithm to perform optimization traversal processing on the interval graph data to obtain path feature data, and using a dynamic programming algorithm to calculate the partition points of the path feature data to obtain an optimal window partition point set; Step S313, obtain the optimal window division point set, use a multi-scale decomposition algorithm to perform hierarchical division processing on the division point data to obtain hierarchical sequence data, use an adaptive window generation algorithm to perform window construction processing on the hierarchical sequence data to obtain window construction data, use a time series alignment algorithm to perform boundary optimization processing on the window construction data, and generate a multi-scale time window sequence.

Images