Harbor district power distribution network fault recovery control method

Abstract

The invention discloses a harbor district power distribution network fault recovery control method, which comprises the steps of preprocessing operation real-time data of an electric ship, an electric vehicle, an energy storage battery and a power grid and topological data of the power grid by adopting a sliding window adaptive standardization method, and calculating coupling degrees among different power supply areas; extracting fault features in combination with wavelet transform and singular value decomposition; flexible resource characteristics are identified, resource classification is carried out, and the resource schedulability is evaluated; searching a reconstruction path; and generating a reconstruction scheme, and selecting an optimal execution sequence. According to the invention, rapid and accurate recovery of harbor district power distribution network system faults is realized, and system operation reliability is improved.

Description

Technical Field

[0001] The present invention belongs to the field of power system fault recovery and control, and particularly relates to a fault recovery control method for a port distribution network. Background Art

[0002] With the deep promotion of the intelligent and green transformation of ports, ship shore power has become an important technical means for ports to reduce emissions and increase efficiency. As a typical three-layer coupled system, the fault recovery control of the port ship-shore power grid system is of great significance for ensuring the power supply reliability and operation efficiency of the port. During the use of ship shore power, there are significant differences between the load characteristics of ships and traditional shore-based loads, and there are complex coupling relationships among ship electrical equipment, shore power facilities, and the port power grid, which poses higher requirements for the fault recovery control of the system. Especially in large container terminals, the power consumption load of a single quay crane can reach 2-3 MW, and the load fluctuations and frequent starts and stops during the simultaneous operation of multiple quay cranes will have a significant impact on the power supply system. Therefore, how to achieve rapid recovery in case of faults is particularly crucial.

[0003] Currently, the research on fault recovery of port power supply systems mainly focuses on single-layer network reconstruction and simple hierarchical control. Traditional methods usually use static network models to describe the system topology, use heuristic algorithms for fault location and path search, or adopt simple hierarchical control strategies to coordinate the actions of devices at different levels. Some research has begun to focus on the impact of ship load characteristics on system reliability and proposed recovery strategies based on load classification; some researchers have also tried to introduce flexible resources such as energy storage systems into the fault recovery process and determine the resource scheduling scheme through simple capacity matching and power balance calculations. These studies have provided useful explorations for improving the power supply reliability of ports.

[0004] However, there are still many limitations in the existing technologies when dealing with the fault recovery problems of port distribution network systems: First, in terms of data processing, existing methods are difficult to effectively process multi-source heterogeneous data generated by ships, shore power, and grid equipment. Especially when the data quality is unstable and the sampling frequencies are inconsistent, it is impossible to accurately extract fault characteristics and inter-layer coupling relationships; second, in terms of flexible resource management, existing static classification methods cannot adapt to the dynamic changes of resource characteristics, resulting in low resource scheduling efficiency; third, in terms of fault recovery path search, traditional algorithms do not fully consider inter-layer conversion constraints and are difficult to find the optimal reconstruction path under multi-objective constraints; finally, in terms of control sequence generation, existing methods often ignore the time-sequence dependence relationship of device actions and cannot ensure the executability of control instructions. These technical problems seriously restrict the efficiency and reliability of the fault recovery of port power supply systems. Summary of the Invention

[0005] Objective of the Invention: To provide a fault recovery control method for the port area distribution network, aiming to solve at least one technical problem existing in the prior art.

[0006] Technical Solution: The fault recovery control method for the port area distribution network includes the following steps:

[0007] S1. Receive the real-time data of the ship side, shore power facilities (including electric ships, electric vehicles, energy storage) and the grid operation, as well as the grid topology data, and perform preprocessing, calculate the coupling degree between different power supply areas, extract the fault characteristics, and generate the grid system state data;

[0008] S2. Obtain the operation data of flexible resources, extract the characteristics of flexible resources and classify the resources, evaluate the schedulable ability of the resources, and calculate the resource response characteristic data;

[0009] S3. Reconstruct the path based on the grid system state data, calculate the weights of each reconstructed path, and generate the path weight data;

[0010] S4. Based on the path weight data, generate a reconstruction plan and evaluate the feasibility of the plan in combination with the resource response characteristic data to obtain an evaluation result; based on the evaluation result and the path weight data, select the optimal execution plan to obtain the reconstruction control sequence data.

[0011] Beneficial Effects: The present invention can accurately identify the system fault characteristics and the inter-layer coupling relationship, realize the optimal allocation of resources; quickly generate the optimal reconstruction path, and formulate a detailed execution plan, ensuring the reliability and efficiency of fault recovery; at the same time, it not only improves the success rate and efficiency of fault recovery, but also can adapt to the complexity and dynamics of the port area ship-shore power grid system, providing a strong guarantee for the safe and stable operation of the port area power system. Description of the Drawings

[0012] Figure 1 It is a flowchart of the present invention.

[0013] Figure 2 It is a flowchart of step S1 of the present invention.

[0014] Figure 3 It is a flowchart of step S2 of the present invention.

[0015] Figure 4 It is a flowchart of step S3 of the present invention.

[0016] Figure 5 It is a flowchart of step S4 of the present invention.

[0017] Figure 6 It is a flowchart of step S5 of the present invention. Detailed Embodiment

[0018] The following describes the present application in more detail with reference to specific embodiments. As Figure 1 shown, the present application proposes a fault recovery control method for the port area distribution network, including the following steps:

[0019] S1. Receive real-time data of the ship side, real-time data of shore power facilities, real-time data of power grid operation, and power grid topology data, and perform preprocessing using the sliding window adaptive standardization method to generate a standardized data set; based on the standardized data set, use the Pearson correlation coefficient calculation method to generate coupling degree data between different power supply areas; combine the standardized data set and the coupling degree data, and use a combined method of wavelet transform and singular value decomposition to extract fault features to obtain fault feature data; based on the coupling degree data and the fault feature data, generate power grid system state data through a three-dimensional tensor construction method; where the real-time data of the ship side and the real-time data of shore power facilities include the battery status of electric ships, electric vehicles, and energy storage devices accessing the port area; the power grid topology data includes power supply area division;

[0020] S2. Obtain the operation data of flexible resources, use the recursive least squares method to extract the characteristic parameters of flexible resources to generate resource characteristic data; based on the resource characteristic data, classify the flexible resources through an improved K-means clustering method to obtain resource classification data; combine the resource classification data and the power grid system state data, and use a fuzzy evaluation method to evaluate the schedulable ability of flexible resources to generate resource schedulable data; based on the resource characteristic data and the resource schedulable data, calculate the response characteristics of various resources through a dynamic response time prediction method to obtain resource response characteristic data;

[0021] S3. Based on the power grid system state data, use a multi-layer network modeling method to construct a system network model to generate multi-layer network data; according to the multi-layer network data and the fault feature data, search for a feasible reconstruction path through an improved A* algorithm to obtain reconstruction path data; combine the reconstruction path data and the resource schedulable data, and use a multi-objective weight allocation method to calculate the weights of each reconstruction path to generate path weight data;

[0022] S4. Obtain the reconstruction path data and the path weight data, use the Pareto optimization method to generate a predetermined number of reconstruction schemes to obtain reconstruction scheme data; according to the reconstruction scheme data and the resource response characteristic data, evaluate the feasibility of the scheme through a timing constraint test method to generate scheme feasibility data; combine the scheme feasibility data and the path weight data, and use an improved dynamic programming method to select the optimal execution scheme to obtain reconstruction control sequence data.

[0023] As Figure 2 shown, according to one aspect of the present application, step S1 is further:

[0024] S11. Obtain the real-time ship-side data including voltage, current, and power from the ship monitoring system, the real-time shore power facility data including voltage, current, and power from the shore power management system, and the real-time grid operation data including voltage, current, and power from the grid monitoring system, and obtain the grid topology data; preprocess the ship-side real-time data, shore power facility real-time data, grid operation real-time data, and grid topology data by using the sliding window adaptive standardization method to generate a standardized data set;

[0025] S12. Based on the standardized data set, calculate the coupling relationship between different power supply areas respectively through the Pearson correlation coefficient calculation method to generate the coupling degree data representing the degree of association between systems;

[0026] S13. Based on the standardized data set and the coupling degree data, perform multi-scale decomposition by using the wavelet transform method to obtain the feature decomposition data; perform singular value decomposition on the feature decomposition data, extract the fault features, and generate the fault feature data including the fault type, location, and degree;

[0027] S14. Based on the coupling degree data and the fault feature data, integrate the node states of each layer, the inter-layer relationships, and the time series information through the three-dimensional tensor construction method to generate the grid system state data representing the overall operating state of the system.

[0028] In an embodiment of the present application, the Pearson correlation coefficient calculation method: R(t) = [ω(t)·(X(t) - μx(t))·(Y(t) - μy(t))] / [σx(t)·σy(t)]; where ω(t) = exp(-λ·|t - t0|) is the time decay weight; X(t) and Y(t) are the time series data of two layers of systems for which the correlation is to be calculated respectively; μx(t) and μy(t) are the local means of X(t) and Y(t) at time t respectively; σx(t) and σy(t) are the local standard deviations of X(t) and Y(t) at time t respectively; λ is the time decay coefficient used to control the influence degree of historical data; t0 is the current time; t is the historical time; R(t) is the improved Pearson correlation coefficient, and its value range is [-1, 1]. The larger the value, the higher the positive correlation degree, and the smaller the value, the higher the negative correlation degree.

[0029] Wavelet-singular value decomposition combination method: F(t) = ∑[ψj,k(t)·cj,k] is the wavelet decomposition, where ψj,k(t) is the wavelet basis function, cj,k is the wavelet coefficient, j is the scale, and k is the translation parameter; D = U·Σ·V Tis the singular value decomposition, where D is the wavelet coefficient matrix, U is the left singular vector matrix, Σ is the singular value diagonal matrix, and V is the right singular vector matrix; the eigenvector E = U(:, 1:r)·Σ(1:r, 1:r), where r is the truncation order, and r = min{i|∑(σi 2 ) / ∑(σj 2 )≥η}, η is the energy threshold, and σi is the singular value.

[0030] This embodiment not only improves the real-time performance and accuracy of data processing, but also can accurately identify the fault type, location and degree, providing reliable data support for subsequent fault recovery control. At the same time, the three-dimensional tensor construction method is used to integrate the information of each layer, enabling the system to comprehensively grasp the dynamic operation state of the coupled system and providing a complete decision-making basis for the formulation of the fault recovery strategy.

[0031] According to one aspect of the present application, step S11 is further:

[0032] S111. Obtain the real-time ship-side data including voltage, current and power from the ship monitoring system, divide the real-time ship-side data into data segments of fixed length through the time window segmentation method to generate segmented ship data; calculate the baseline level of the data using the moving average method based on the pre-stored historical ship data to generate baseline reference data; calculate the data deviation degree through the adaptive threshold algorithm according to the segmented ship data and the baseline reference data to generate ship data deviation; obtain the segmented ship data and the ship data deviation, and identify the out-of-limit data points using the confidence interval analysis method to generate ship data marks;

[0033] S112. Obtain the real-time shore power facility data including voltage, current and power from the shore power management system, decompose the real-time shore power facility data into different frequency components through the wavelet decomposition method to generate shore power frequency data; calculate the energy distribution of each frequency band using the energy density analysis method according to the shore power frequency data to generate shore power energy data; extract the effective signal components through the adaptive filtering algorithm according to the shore power frequency data and the shore power energy data to generate shore power effective data; combine the shore power effective data and the pre-stored shore power reference data, and identify the position of the missing points using the data integrity inspection method to generate shore power missing marks;

[0034] S113. Obtain the real-time grid operation data including voltage, current, and power from the grid monitoring system, convert the real-time grid operation data to the frequency domain space through the Fourier transform method to generate grid frequency domain data; calculate the frequency component characteristics using the spectral analysis method based on the grid frequency domain data to generate grid spectral characteristic data; obtain the grid frequency domain data and the grid spectral characteristic data, reconstruct the signal through the inverse transform filtering algorithm to generate grid reconstruction data; combine the grid reconstruction data and the pre-stored grid standard data, and use the data normalization inspection method to identify abnormal fluctuations to generate grid anomaly marks.

[0035] S114. Obtain the grid topology data, clean the grid topology data to obtain the cleaned grid topology data; parse the cleaned grid topology data, extract the information of nodes and edges, and construct a grid topology structure diagram; extract the key features from the grid topology structure diagram to generate topology feature data; perform normalization processing on the topology feature data to generate standardized topology data; generate grid topology marks based on the standardized topology data.

[0036] S115. Obtain the ship data mark, shore power missing mark, grid anomaly mark, and grid topology mark, construct a data quality assessment matrix through the multi-source data fusion method to generate a data quality matrix; calculate the data credibility index using the fuzzy inference algorithm based on the data quality matrix to generate data credibility; construct a data correction model through the weighted combination method based on the segmented ship data, shore power valid data, grid reconstruction data, standardized topology data, and data credibility to generate fusion correction data.

[0037] S116. Based on the fusion correction data, construct the corresponding relationship of multi-source data through the timestamp alignment method to generate time series alignment data; supplement the data points at the missing moments using the interpolation algorithm based on the time series alignment data to generate complete time series data; obtain the complete time series data, eliminate the sudden interference through the median filtering method to generate a smooth data set; combine the smooth data set and the data credibility, and perform normalization processing using the sliding window adaptive normalization method to generate a standardized data set.

[0038] In an embodiment of the present application, the sliding mean method: y(t) = ∑[w(i)·x(t - i)] / ∑w(i) is the weighted sliding mean; where w(i) = exp(-i 2 / 2h 2 ) is the Gaussian weight function; exp is the exponential function, h is the bandwidth parameter, h = median{|x(t) - median(x)|}·0.9; i is the time delay, i ranges from 0 to M - 1; M is the sliding window size, M = ceil(fs / f0); fs is the sampling frequency; f0 is the fundamental frequency, and ceil( ) represents the ceiling function.

[0039] Adaptive threshold algorithm: Th(t) = μ(t) + k(t)·σ(t) is the dynamic threshold; where μ(t) = α·μ(t - 1) + (1 - α)·x(t) is the exponentially smoothed mean; σ(t) = sqrt[β·σ 2 (t - 1) + (1 - β)·(x(t) - μ(t)) 2 is the dynamic standard deviation; k(t) = k0·exp(-λ·|dx(t) / dt|) is the adaptive coefficient; α, β are smoothing factors; λ is the sensitivity parameter, k0 is the initial adaptive coefficient, and x(t) is the observed value at time t.

[0040] Confidence interval analysis method: CI(t) = [μ(t) ± Z(α / 2)·σ(t) / sqrt(n)] is the dynamic confidence interval; where Z(α / 2) is the α / 2 quantile of the standard normal distribution; n is the number of samples; Anomaly determination: O(t) = |x(t) - μ(t)| / σ(t) > F -1 (p); where F -1 (p) is the p quantile of the F distribution; p is the confidence level.

[0041] Adaptive filtering algorithm: y(t) = x(t) - ∑[h(k, t)·x(t - k)] is the filtered output; where x(t) is the input signal at time t, h(k, t) is the adaptive filter coefficient: h(k, t) = h(k, t - 1) + μ·e(t)·x(t - k); e(t) is the error signal: e(t) = d(t) - y(t); d(t) is the desired signal at time t, and μ is the step size parameter: μ = α / (Δ + x T (t)·x(t)); α is the convergence coefficient; Δ is the regularization parameter; k is the filter order, T denotes the transpose.

[0042] Inverse transform filtering algorithm: y(t) = IFFT[H(f)·X(f)] is the filtering result; where H(f) is the adaptive filter: H(f) = G(f) / (G(f) + η·N(f)); G(f) is the signal power spectrum; N(f) is the noise power spectrum; η is the smoothing factor: η = ρ·exp(-|dG(f) / df|); ρ is the basic smoothing coefficient.

[0043] This embodiment not only overcomes technical problems such as complex power supply environment, diverse data sources, and severe signal interference in the port area, but also provides a high-quality data basis for fault recovery control, improving the reliability of fault recovery control.

[0044] According to one aspect of the present application, step S12 is further as follows:

[0045] S121. Obtain a standardized data set, extract the data of the ship layer, shore power layer, and power grid layer respectively through a data classification method to generate a stratified data set; according to the stratified data set, use the principal component analysis method to extract the main feature dimensions of each layer of data to generate feature dimension data; obtain the feature dimension data, calculate the weight coefficients of each dimension through a feature importance evaluation algorithm to generate dimension weight data; combine the stratified data set and the dimension weight data, and use a weighted combination method to construct an inter-layer feature vector to generate feature vector data;

[0046] S122. Obtain the data of the ship layer and shore power layer in the feature vector data, calculate the short-term correlation index through a sliding time window method to generate ship-shore short-term association data; according to the data of the ship layer and shore power layer in the feature vector data, use a long-term trend analysis method to calculate the steady-state correlation index to generate ship-shore long-term association data; obtain the ship-shore short-term association data and the ship-shore long-term association data, calculate the comprehensive association degree through a multi-scale fusion algorithm to generate ship-shore coupling data;

[0047] S123. Obtain the data of the shore power layer and power grid layer in the feature vector data, calculate the dynamic correlation index through a time-frequency analysis method to generate shore-grid dynamic association data; according to the data of the shore power layer and power grid layer in the feature vector data, use a state space reconstruction method to calculate the non-linear correlation index to generate shore-grid non-linear association data; obtain the shore-grid dynamic association data and the shore-grid non-linear association data, calculate the comprehensive association degree through an adaptive weighting algorithm to generate shore-grid coupling data;

[0048] S124. Obtain the data of the ship layer and power grid layer in the feature vector data, evaluate the information transfer characteristics through a mutual information calculation method to generate ship-grid information association data; according to the data of the ship layer and power grid layer in the feature vector data, use a conditional entropy analysis method to calculate the information dependence degree to generate ship-grid dependence association data; obtain the ship-grid information association data and the ship-grid dependence association data, calculate the comprehensive association degree through an information fusion algorithm to generate ship-grid coupling data;

[0049] S125. Obtain the ship-shore coupling data, shore-grid coupling data, and ship-grid coupling data, unify the coupling degrees of different types to the same scale through a data standardization method to generate standard coupling data; according to the standard coupling data, use a coupling stability evaluation algorithm to calculate the stability degree of the coupling relationship between each layer to generate coupling stability data; obtain the standard coupling data and the coupling stability data, and construct an inter-layer comprehensive association matrix through a Pearson correlation coefficient calculation method to generate coupling degree data.

[0050] In the coupling system of the ship-shore power grid in the port area of this embodiment, through the combination of methods such as principal component analysis, multi-scale correlation analysis, and information fusion, the accurate quantification of the inter-layer coupling relationship is realized. The main characteristic dimensions of the ship layer, shore power layer, and power grid layer are extracted; the dynamic influence of the ship's electrical load on the shore power system is accurately reflected; the dynamic coupling characteristics between the shore power facilities and the power grid are accurately evaluated; the degree of influence of the change in ship load on the power grid stability is quantified.

[0051] According to one aspect of the present application, step S13 is further as follows:

[0052] S131. Obtain a standardized data set, decompose the data in the standardized data set by using the multi-scale wavelet decomposition method to generate feature decomposition data containing different frequency components; for the high-frequency components in the feature decomposition data, identify the mutation points through the adaptive threshold detection method to obtain high-frequency mutation data;

[0053] S132. Based on the feature decomposition data and the high-frequency mutation data, use the time-frequency feature extraction method to analyze the energy distribution and phase relationship of each frequency component; combine the energy distribution, phase relationship, and coupling degree data to calculate the system state change characteristics and generate time-frequency feature data;

[0054] S133. Based on the time-frequency feature data, use the singular value decomposition method to construct a feature matrix, calculate the main singular values and the corresponding eigenvectors, extract the main feature components, and generate feature component data;

[0055] S134. Based on the feature component data, use the feature mapping method based on pattern matching to match the main feature components with the preset fault feature templates, calculate the similarity of various faults, and generate fault feature data including fault types, positions, and degrees.

[0056] In an embodiment of the present application, the multi-scale wavelet decomposition method: Wj(k) = ∑[h(m - 2k)·Wj-1(m)] is wavelet decomposition; where m is the time point or sample point of the signal, k is the position index in the wavelet decomposition, Wj-1(m) is the wavelet coefficient in the decomposition of the (j - 1)th layer, h(n) is the wavelet filter coefficient, and j is the decomposition scale; the reconstructed signal value is: Rj(t)= ∑[g(m - 2k)·Wj(k)]; g(n) is the reconstruction filter coefficient; the energy distribution: Ej = ∑|Wj(k)| 2 ; scale selection: J = argmax{Ej / E0>ε}; ε is the energy threshold, and E0 is the initial energy of the signal.

[0057] Adaptive Threshold Detection Method: Th(j) = σj·sqrt(2·log(N)) is the scale adaptive threshold; where log(N) is the logarithmic function; σj is the standard deviation of the noise at scale j: σj = median(|Wj|) / 0.6745; Mutation Point Detection: D(k)= {1, if |Wj(k)| > Th(j)}; Connected Region: R = {k | D(k)=1 & |k-k'|≤Δ}; Δ is the time tolerance, and k' is another position index.

[0058] Time-Frequency Feature Extraction Method: C(t, f) = ∑[x(τ)·K(τ-t)·exp(-j2πfτ)] is the Cohen class time-frequency distribution; where K(t) is the kernel function: K(t) = exp(-πt 2 / σ 2 ); Instantaneous Frequency: fi(t) = (1 / 2π)·d[arg(C(t, f))] / dt; Frequency Center: fc(t) = ∑[f·|C(t, f)| 2 / ∑|C(t, f)| 2 ; Bandwidth: B(t) =sqrt{∑[(f-fc(t)) 2 ·|C(t, f)| 2 / ∑|C(t, f)| 2}.

[0059] Singular Value Decomposition Method: W = U·Σ·V T is the standard SVD decomposition; Feature Matrix Construction: M(t) = [W(t), W(t+τ),..., W(t+(d-1)τ)]; where τ is the time delay parameter and d is the embedding dimension: d = ceiling(2·log(N)); Reconstruction Dimension Selection: r = min{k | ∑(i=1 to k)[λi] / ∑(i=1 to n)[λi]≥η*}; where λi are the singular values and η* is the energy threshold; Reconstruction Matrix: R = U(:,1:r)·Σ(1:r,1:r)·V(:,1:r) T .

[0060] Pattern Matching Feature Mapping Method: S(x, y) = exp(-||Fx - Fy|| 2 / 2σ 2) is the feature similarity; where Fx and Fy are feature vectors; Pattern library matching: M(x) = max{S(x, yi) | yi∈Y}; where Y is the preset pattern library; Adaptive threshold: Th(x) = μM + k·σM; where μM and σM are the mean and standard deviation of the matching degree; Mapping function: φ(x) = ∑[wi·yi·S(x, yi)] / ∑[wi·S(x, yi)]; wi is the pattern weight.

[0061] In this embodiment, in the port ship-shore power grid system, through the organic combination of the multi-scale wavelet decomposition and the time-frequency feature extraction method, the accurate identification of the system fault characteristics is realized. The fault characteristics and the normal operation fluctuations are effectively separated; the limitations of the traditional fixed threshold method in the complex power supply environment are overcome; the accurate determination of the fault type, location and degree is realized. It can effectively distinguish the fault characteristics from the normal fluctuations, improve the accuracy of fault diagnosis, and provide a reliable basis for the formulation of the fault recovery control strategy.

[0062] According to one aspect of the present application, step S14 is further as follows:

[0063] S141. Obtain the coupling degree data, split the coupling relationship into two parts of direct coupling and indirect coupling through the data decomposition method, and generate the coupling decomposition data; according to the coupling decomposition data, adopt the correlation path analysis method to identify the key coupling links and generate the coupling path data; obtain the coupling path data, calculate the influence weights of each path through the path importance evaluation algorithm, and generate the path weight data; combine the coupling decomposition data and the path weight data, and adopt the weighted mapping method to construct the inter-layer relationship matrix and generate the relationship matrix data;

[0064] S142. Obtain the fault feature data, divide the fault information into different feature dimensions through the feature decomposition method, and generate the feature decomposition data; according to the feature decomposition data, adopt the dimension importance analysis method to evaluate the contribution degree of each feature dimension and generate the dimension contribution data; obtain the feature decomposition data and the dimension contribution data, and generate the compact representation of the fault state through the feature reconstruction algorithm and generate the state representation data;

[0065] S143. Obtain the standardized data set, divide the data into multiple time scales through the time series segmentation method, and generate the multi-scale data; according to the multi-scale data, adopt the time series feature extraction method to calculate the state features at each scale and generate the time series feature data; obtain the multi-scale data and the time series feature data, and integrate the information of different time scales through the scale fusion algorithm and generate the time series fusion data;

[0066] S144. Obtain relationship matrix data and state representation data, establish the inter-layer state correspondence relationship through the state mapping method, and generate state mapping data; according to the state mapping data and the time-series fusion data, use the spatio-temporal feature fusion algorithm to construct a comprehensive feature tensor and generate feature tensor data; obtain the feature tensor data, and extract the main feature patterns through the tensor decomposition method to generate pattern decomposition data.

[0067] S145. Obtain the pattern decomposition data, calculate the contribution degree of each feature pattern through the pattern importance evaluation method, and generate pattern weight data; according to the pattern decomposition data and the pattern weight data, use the pattern reconstruction algorithm to restore the system state information and generate reconstructed state data; obtain the reconstructed state data, and integrate the node state, inter-layer relationship and time-series information through the three-dimensional tensor construction method to generate the power grid system state data.

[0068] In an embodiment of the present application, the data decomposition method: X(t) = T(t) + S(t) + R(t) is a three-component decomposition; where T(t) is the trend term: T(t) = ∑[αi·pi(t)], pi(t) is a polynomial basis function; S(t) is the periodic term: S(t) = ∑[βi·sin(2πfit + φi)]; R(t) is the random term, which is decomposed by EMD: R(t) = ∑[cj(t)] + rn(t); where cj(t) is the intrinsic mode function and rn(t) is the residual term.

[0069] The associated path analysis method: P(i, j) = {pk | pk = (v0, v1,..., vn), v0 = i, vn = j} is the path set; the path weight: w(p) = ∏[w(vi, vi+1)] is the product of edge weights, and ∏ is the product operator; the criticality: C(p) = w(p)·exp(-λ·l(p)); where l(p) is the path length; λ is the attenuation coefficient; the path selection: P* = {p | C(p) > η·max(C)}; η is the selection threshold.

[0070] The feature decomposition method: F = Q·Λ·Q T is the feature decomposition expression; where Q is the eigenvector matrix; Λ is the eigenvalue diagonal matrix; the feature importance: I(λi) = λi / ∑λj; the dimension selection: d = min{k | ∑(i = 1 to k)[I(λi)]≥ρ}; where ρ is the cumulative contribution rate threshold; the feature reconstruction: F' = Q(:, 1:d)·Λ(1:d, 1:d)·Q(:, 1:d) T ; the orthogonality constraint: Q T ·Q = I, where I is the identity matrix.

[0071] Time series segmentation method: CP(t) = {ti | L(ti-1:ti) + L(ti:ti+1) - L(ti-1:ti+1)>γ} is the set of change points; where L(a:b) is the likelihood function of the interval [a, b]: L(a:b) = -n / 2·log(2π) - n / 2·log(σ 2 ) - ∑(x(t)-μ) 2 / (2σ 2 ); γ is the decision threshold; minimum interval constraint: |ti – t(i-1)|≥Δt; where Δt is the minimum segmentation length.

[0072] Time series feature extraction method: Feature vector F(t) = [μ(t), σ(t), s(t), k(t), h(t)]; where μ(t) is the local mean; σ(t) is the local standard deviation; s(t) is the local skewness: s(t) = E[(x-μ) 3 / σ 3 ; k(t) is the local kurtosis: k(t) = E[(x-μ) 4 / σ 4 ; h(t) is the local entropy: h(t) = -∑p(x)·log(p(x)); local window length: w(t) =α·σ(t) / |dμ(t) / dt|.

[0073] State mapping method: Φ(s) = W·tanh(U·s + b) is the non-linear mapping function; where W, U are weight matrices; b is the bias vector; parameter update: W(t+1) = W(t) - η·ΞL / ΞW; loss function: L = ||Φ(s1) - Φ(s2)|| 2 + λ·||W|| 2 ; state consistency constraint: ||Φ(s1) - Φ(s2)|| ≤ ε·d(s1, s2); where d(s1, s2) is the state distance, Ξ is the partial derivative, η is the learning rate, λ is the regularization parameter, ε is the threshold in the state consistency constraint.

[0074] Spatio-temporal feature fusion algorithm: Z = f(Xs, Xt) is the fusion function; where Xs is the spatial feature: Xs = Conv(X); Xt is the temporal feature: Xt = LSTM(X); attention weight: A = softmax(V·tanh(WX + b)); fusion expression: Z = A·Xs + (1-A)·Xt; feature normalization: Z' = (Z - μz) / σz, μz is the mean of feature Z, σz is the standard deviation of feature Z.

[0075] Tensor decomposition method: X≈∑[a1 r Θa 2 r Θ... Θa N r is CP decomposition; where Θ is the tensor outer product; r is the decomposition rank; a N r is the Nth element in the rth decomposition factor; Tucker decomposition: X ≈ G ×1 U1 ×2 U2... × n U n ; where G is the core tensor; U i is the factor matrix; Rank selection: R = argmin{||X - X r || F / ||X|| F < ε}; where X r is the approximation of rank r, || || F is the Frobenius norm, and ε is the energy threshold.

[0076] This embodiment not only overcomes the limitations of traditional single-layer system modeling methods, but also can comprehensively reflect the complex dynamic characteristics of the ship-shore power grid system in the port area, provides a complete decision-making basis for fault recovery control, and improves the accuracy and reliability of fault recovery.

[0077] As Figure 3 shown, according to one aspect of the present application, step S2 is further:

[0078] S21. Obtain flexible resource operation data including capacity, charge and discharge power, and response speed from the flexible resource management system, and use the recursive least squares method to identify the dynamic response characteristics, adjustable range, and adjustment rate of various resources in the flexible resource operation data, and generate resource characteristic data representing the dynamic characteristics of the resources;

[0079] S22. Based on the resource characteristic data, dynamically classify the flexible resources by an improved K-means clustering method, classify the resources with similar response characteristics, and generate resource classification data representing the characteristics of different types of resources;

[0080] S23. Based on the resource classification data and the power grid system state data, use the fuzzy evaluation method to construct a resource schedulability evaluation model, calculate the schedulability of various resources at different times, and generate resource schedulability data;

[0081] S24. Based on the resource characteristic data and the resource schedulability data, calculate the response time of different types of resources by the dynamic response time prediction method, and generate resource response characteristic data representing the response characteristics of various resources.

[0082] In an embodiment of the present application, the recursive least squares identification method: θ(t) = θ(t - 1) + K(t)·[y(t) - φ T (t)·θ(t - 1)]; where K(t) = P(t - 1)·φ(t)·[λ + φ T (t)·P(t - 1)·φ(t)] -1 is the gain matrix; P(t) = [I - K(t)·φ T (t)]·P(t - 1) / λ is the covariance matrix; φ(t) is the regression vector; y(t) is the measurement output; θ(t) is the parameter estimate value; λ is the forgetting factor, which is used to balance the weights of historical data and new data.

[0083] The improved K-means clustering method: d(x, c) = β1·||x - c|| 2 + β2·|v(x) - v(c)| + β3·|r(x) - r(c)| is the improved distance metric; where x is the resource feature vector, c is the clustering center; v(x), v(c) are the response speed features; r(x), r(c) are the adjustment range features; βi is the adaptive weight coefficient, βi = exp(-γi·σi 2 ), σi is the standard deviation of the corresponding feature; update of the clustering center: c(t + 1) = c(t) + η·∑[ω(x)·(x - c(t))], where η is the learning rate, ω(x) = exp(-d(x, c) / T) is the sample weight, and T is the temperature parameter.

[0084] The fuzzy evaluation method: μij(t) = exp[-((xij(t) - vj) / σj) 2 is the fuzzy membership function; where xij(t) is the value of the i-th type of resource on the j-th index; vj is the index standard value; σj is the fuzzy half-width; comprehensive calculation of the evaluation index: E(t) = ∑[wj(t)·μij(t)]; where wj(t) is the dynamic weight, wj(t) = sj(t) / ∑sk(t), sj(t)= -∑[μij(t)·ln(μij(t))] is the entropy weight; E(t) is the resource schedulability evaluation value.

[0085] This embodiment not only improves the accuracy of flexible resource classification and the efficiency of resource scheduling, but also can adjust the resource classification results and the evaluation of schedulability in real time according to the dynamic changes of the system state, providing a scientific basis for the optimal allocation of resources during the fault recovery process.

[0086] According to one aspect of the present application, step S21 is further:

[0087] S211. Obtain the operation data of flexible resources including capacity, charge-discharge power, and response speed. Use the data segmentation method to divide the operation data according to different working conditions, extract the typical operation segments under each working condition, and generate segmented operation data.

[0088] S212. Obtain the segmented operation data, use the recursive least squares method to construct a parameter estimation model, update the parameter estimation values through iterative calculation to obtain parameter estimation data; according to the parameter estimation data, calculate the error and convergence characteristics of the parameter estimation to generate estimation accuracy data.

[0089] S213. Obtain the parameter estimation data and estimation accuracy data, use the recursive least squares method to identify and optimize the dynamic response characteristics, adjustable range, and adjustment rate, and generate resource characteristic data representing the dynamic characteristics of the resources.

[0090] In an embodiment of the present application, the data segmentation method: WC(t) = argmax{P(ci|x(t))} is the result of working condition division; where P(ci|x(t)) = N(x(t)|μi, Σi) / ∑N(x(t)|μk, Σk) is the working condition probability; N(x|μ, Σ) is the multivariate Gaussian distribution; working condition parameter update: μi = ∑[x(t)·I(WC(t)=i)] / ni; Σi = ∑[(x(t)-μi)(x(t)-μi) T ·I(WC(t)=i)] / ni; where ni is the number of samples in working condition i; I(·) is the indicator function.

[0091] Typical segment extraction method: D(xi, xj) = ∑[wk·d(xi(k), xj(k))] is the sequence similarity; where d(·, ·) is the dynamic time warping distance; wk is the feature weight; representative score: R(x) = ∑[exp(-D(x, xi) / σ 2 )]; typical segment selection: S* = argmax{∑R(x) | x∈S, |S|=m}; where m is the number of required segments.

[0092] This embodiment not only overcomes the problem that the traditional fixed-parameter model cannot adapt to the dynamic changes of resource characteristics, but also can track the performance changes of flexible resources in real time, and accurately evaluate the availability and response ability of the resources. For the scenario in the port area where there are various types of energy storage devices and adjustable loads and complex usage conditions, this embodiment improves the accuracy of flexible resource modeling and provides a reliable parameter basis for subsequent resource optimization scheduling.

[0093] According to one aspect of the present application, step S22 is further:

[0094] S221. Obtain resource characteristic data, map high-dimensional characteristic parameters to a low-dimensional feature space through a data dimensionality reduction method to generate dimensionality-reduced feature data; according to the dimensionality-reduced feature data, use a feature importance evaluation algorithm to calculate the discrimination ability of each feature dimension to generate feature weight data; obtain the dimensionality-reduced feature data and the feature weight data, and construct a resource feature vector through a weighted feature extraction method to generate feature vector data; combine the feature vector data and use a normalization processing method to unify the feature scale to generate normalized feature data.

[0095] S222. Obtain normalized feature data, calculate the sample distribution characteristics through a density estimation method to generate density distribution data; according to the density distribution data, use a local neighborhood analysis method to identify the distribution structure of data clusters to generate cluster structure data; obtain the density distribution data and the cluster structure data, and determine the initial position of the clustering center through an adaptive threshold algorithm to generate initial center data; combine the normalized feature data and the initial center data, and use a distance calculation method to evaluate the similarity between the sample and the clustering center to generate similarity data.

[0096] S223. Obtain similarity data, evaluate the belonging degree of the sample to each category through a membership degree calculation method to generate membership degree data; according to the membership degree data, use a boundary sample recognition algorithm to determine the boundary features between categories to generate boundary feature data; obtain the membership degree data and the boundary feature data, and adjust the category division standard through a category optimization method to generate optimized category data; combine the similarity data and the optimized category data, and use a sample allocation method to determine the category belonging of the sample to generate initial classification data.

[0097] S224. Obtain initial classification data, evaluate the tightness of the classification through an intra-class difference degree calculation method to generate intra-class difference data; according to the initial classification data, use an inter-class distance calculation method to evaluate the separation degree of the classification to generate inter-class distance data; obtain the intra-class difference data and the inter-class distance data, and calculate the classification effect index through a classification quality evaluation algorithm to generate classification quality data; combine the initial classification data and the classification quality data, and use an iterative optimization method to improve the category division result to generate optimized classification data.

[0098] S225. Obtain optimized classification data, evaluate the reliability of the classification result through a stability analysis method to generate classification stability data; according to the optimized classification data and the classification stability data, use a classification correction algorithm to optimize the abnormal classification result to generate corrected classification data; obtain the corrected classification data, and associate the classification result with the resource characteristics through a feature mapping method to generate feature mapping data; combine the feature mapping data and the original resource characteristic data, and use an improved K-means clustering method to determine the final resource category to generate resource classification data.

[0099] In one embodiment of the present application, the data dimensionality reduction method: Y = W T ·X is linear dimensionality reduction; where W = argmax{tr(W T ·Sb·W) / tr(W T ·Sw·W)}; Sb is the between-class scatter matrix; Sw is the within-class scatter matrix; local preservation constraint: ||yi - yj|| 2 ≤ε·||xi - xj|| 2 , when (i, j) ∈ N; where N is the neighbor set.

[0100] Feature importance evaluation: I(f) = H(Y) - H(Y|f) is the information gain; where H(Y) is the class entropy; H(Y|f) is the conditional entropy; stability index: S(f) = 1 - σ 2 (I(f)) / μ 2 (I(f)); comprehensive importance: W(f) = α·I(f) + (1 - α)·S(f); where α is the balance factor.

[0101] Normalization processing method: z(x) = (x - μ(t)) / [k(t)·σ(t)] is dynamic normalization; where μ(t) = β·μ(t - 1) + (1 - β)·x(t) is the sliding mean; σ(t) = sqrt[γ·σ 2 (t - 1) + (1 - γ)·(x(t) - μ(t)) 2 is the dynamic standard deviation; k(t) = log(1 + |dx / dt|) is the adaptive coefficient.

[0102] This embodiment can not only accurately identify the characteristic differences of different types of flexible resources, but also adjust the classification results in a timely manner according to the dynamic changes of resource performance, improving the accuracy and adaptability of the flexible resource management in the port area, and providing a scientific basis for the optimal allocation of resources during the fault recovery process.

[0103] According to one aspect of the present application, step S23 is further as follows:

[0104] S231. Obtain resource classification data, construct an evaluation index system using the fuzzy membership degree calculation method, calculate the membership degree values of various resources in three dimensions of capacity availability, response timeliness, and adjustment accuracy respectively, and generate resource membership degree data;

[0105] S232. Based on the resource membership degree data and the power grid system state data, use the fuzzy comprehensive evaluation method to calculate the weight coefficients of each evaluation index; based on the weight coefficients, determine the relative importance between the indexes through the analytic hierarchy process, and generate index weight data;

[0106] S233. Based on the resource membership data and the index weight data, use the fuzzy matrix synthesis method to perform weighted calculation on the evaluation indexes of various resources to obtain comprehensive evaluation data;

[0107] S234. Based on the comprehensive evaluation data, use the fuzzy decision method to determine the resource schedulability level; based on the resource schedulability level and the time series characteristics of the comprehensive evaluation data, calculate the schedulability degree at different times, and generate resource schedulability data representing the schedulability of various resources.

[0108] In an embodiment of the present application, the fuzzy membership calculation method: μ(x, t) = ∑[wi(t)·exp(-(x - vi) T ·Ai(t)·(x - vi))] is an adaptive membership function; where wi(t) is the dynamic weight: wi(t) = exp(-λ·||x - vi|| 2 ) / ∑exp(-λ·||x - vk|| 2 );Ai(t) is the adaptive covariance matrix: Ai(t) = [Mi(t) + εI] -1 ; Mi(t) is the local scatter matrix; ε is the regularization parameter.

[0109] The analytic hierarchy process: A = [aij] is the judgment matrix; the consistency index: CI = (λmax - n) / (n - 1); where λmax is the maximum eigenvalue; the weight calculation: w = ∑[∏aij 1 / n / ∑∑[∏aij 1 / n ; the consistency test: CR = CI / RI < 0.1; the dynamic weight update: w(t) = α·w(t - 1) + (1 - α)·w'(t); where w'(t) is the newly calculated weight, RI represents the random consistency index, and α is the smoothing factor in the weight update.

[0110] This embodiment not only overcomes the limitations of the traditional single-index evaluation method, but also can accurately reflect the actual availability of the flexible resources in the port area under different operating conditions, providing a reliable resource scheduling basis for the fault recovery control.

[0111] According to one aspect of the present application, step S24 is further as follows:

[0112] S241. Obtain the resource characteristic data, analyze the change law of the response characteristics of the resources through the dynamic feature extraction method to generate dynamic feature data; according to the dynamic feature data, use the time series pattern recognition algorithm to identify the typical response patterns to generate response pattern data; obtain the response pattern data, and construct the dynamic change model of the response characteristics through the pattern evolution analysis method to generate characteristic evolution data.

[0113] S242. Obtain resource schedulable data, decompose the resource schedulability into multiple influencing factors by a state decomposition method to generate influencing factor data; according to the influencing factor data, construct an influence mechanism model of the response time by a factor correlation analysis method to generate mechanism model data; obtain the mechanism model data and characteristic evolution data, and establish a comprehensive prediction model by a model fusion algorithm to generate prediction model data.

[0114] S243. Obtain prediction model data, construct multiple possible response scenarios by a scenario generation method to generate response scenario data; according to the response scenario data, train a response time prediction model by an adaptive learning algorithm to generate model parameter data; obtain the model parameter data, and improve the model prediction accuracy by a parameter optimization method to generate optimized parameter data.

[0115] S244. Obtain the optimized parameter data and real-time resource schedulable data, calculate the response time under different scenarios by a dynamic prediction method to generate predicted time data; according to the predicted time data, evaluate the reliability of the prediction result by a confidence interval estimation method to generate prediction credibility data; obtain the predicted time data and prediction credibility data, and optimize the abnormal prediction result by a prediction correction algorithm to generate corrected prediction data.

[0116] S245. Obtain the corrected prediction data, evaluate the temporal consistency of the prediction result by a temporal correlation analysis method to generate temporal consistency data; according to the corrected prediction data and temporal consistency data, verify the rationality of the prediction result by a comprehensive evaluation method to generate prediction evaluation data; obtain the prediction evaluation data, and determine the final response characteristics of various resources by a dynamic response time prediction method to generate resource response characteristic data.

[0117] In an embodiment of the present application, the dynamic feature extraction method: F(t) = [m(t), v(t), s(t), r(t)] is a feature vector; where m(t) is the autoregressive moving average model (ARMA) parameter: m(t) = ∑[ai·m(t - i)] + ∑[bj·e(t - j)]; v(t) is the rate of change feature: v(t) = ∑[wi·|dm(t - i) / dt|]; s(t) is the spectral feature: s(t) =∑[pk·|X(fk, t)| 2 ; r(t) is the recursive feature: r(t) = ∑[qi·RR(i, t)]; where RR is the recurrence plot feature, ai is the coefficient of the autoregressive model, bj is the coefficient of the moving average model, e(t - j) is the error term at time t - j, wi is the weight coefficient in the rate of change feature, pk is the frequency component in the spectral feature, X(fk, t) is the signal value at frequency fk and time t, and qi is the weight coefficient in the recursive feature.

[0118] Sequential pattern recognition algorithm: P(x|λ) = ∑[αt(i)·βt(i)] is the pattern probability; where α and β are the forward and backward probabilities; P(x|λ) is the probability of the observed data x given the model parameter λ; State transition: aij(t) = η·aij(t - 1) + (1 - η)·P(qt = j|qt - 1 = i, x); aij(t) is the transition probability from state i to state j at time t, and η is the smoothing factor of the state transition probability; P(qt = j|qt - 1 = i, x) represents the conditional probability that the state is i at time t - 1, the state is j at time t, and the observed data is x; Output probability: bj(x, t) = ∑[wk·N(x|μk, Σk)], where wk is the weight coefficient in the output probability; N(x|μk, Σk) is a Gaussian distribution with mean μk and covariance matrix Σk; Pattern adaptation: λ(t) = argmax{P(x|λ)}, where λ(t) is the model parameter at time t, and argmax{P(x|λ)} represents the model parameter λ that maximizes the probability P(x|λ) of the observed data x.

[0119] Pattern evolution analysis method: dM / dt = F(M, t) + G(M, t)·ξ(t) is the stochastic dynamics equation; where F(M, t) is the deterministic term; G(M, t) is the stochastic perturbation term; ξ(t) is Gaussian white noise; Stability analysis: λmax(J) < 0; where J is the Jacobian matrix; Evolution prediction: M(t + τ) = M(t) + ∫[F(M, s) + G(M, s)·ξ(s)]ds.

[0120] This embodiment can not only accurately estimate the response time of different types of flexible resources, but also dynamically adjust the prediction model according to the real-time changes of the system state, improving the timeliness and reliability of the flexible resource scheduling in the port area.

[0121] S25. Obtain resource response characteristic data and resource schedulable data, establish a differentiated regulation strategy for electric ships, electric vehicles, and energy storage batteries through the resource control strategy method, and generate resource control strategy data; According to the resource control strategy data, use the parameter configuration method to determine key control parameters such as power, voltage, and frequency, and generate control parameter data;

[0122] S251. Obtain resource response characteristic data, design dedicated control strategies for different types of resources through the resource classification control method, and generate classification control data; based on the classification control data, design a power regulation model for the electric ship's power consumption load using the ship load control algorithm, and generate ship load control data; obtain the classification control data, construct an electric vehicle charging and discharging power regulation model through the V2G control algorithm, and generate V2G control data; based on the classification control data, design a charge and discharge strategy for the energy storage battery using the energy storage control algorithm, and generate energy storage control data;

[0123] S252. Obtain ship load control data, V2G control data, and energy storage control data, evaluate the optimal working modes of various resources through the control mode selection method, and generate control mode data; based on the control mode data, calculate the set values of active power and reactive power of various resources using the power distribution algorithm, and generate power distribution data; obtain the control mode data and power distribution data, determine the control response rate through the response rate matching method, and generate rate control data; based on the power distribution data and rate control data, tune the controller parameters using the parameter optimization method, and generate control parameter data;

[0124] S26. Obtain resource control strategy data and control parameter data, construct a multi-resource collaborative working mechanism through the resource coordination control method, and generate coordination control data; according to the coordination control data, design a real-time regulation strategy for coping with load fluctuations using the dynamic adjustment method, and generate dynamic adjustment data;

[0125] S261. Obtain resource control strategy data, determine the dominant and subordinate relationships of resource control through the master-slave coordination method, and generate control hierarchy data; based on the control hierarchy data, establish an information interaction mechanism between resources using the communication protocol design method, and generate communication mechanism data; obtain the control hierarchy data and communication mechanism data, process multi-resource control conflicts through the conflict resolution algorithm, and generate conflict handling data; based on the control hierarchy data and conflict handling data, determine the resource coordination control strategy using the collaborative decision-making method, and generate collaborative strategy data;

[0126] S262. Obtain collaborative strategy data, estimate the short-term load change trend through the load prediction method, and generate load prediction data; based on the load prediction data, dynamically adjust the control parameters using the adaptive control algorithm, and generate adaptive control data; obtain the load prediction data and adaptive control data, construct a control strategy library for typical load scenarios through the multi-scenario adaptation method, and generate scenario strategy data; based on the adaptive control data and scenario strategy data, implement a smooth transition of the control strategy using the strategy switching method, and generate dynamic adjustment data;

[0127] S27. Obtain coordinated control data and dynamic adjustment data, construct a linkage mechanism for distribution network reconstruction and resource regulation through a network resource coordination method, and generate network resource coordination data; according to the network resource coordination data, design resource regulation strategies for different fault types by using a fault type adaptation method, and generate fault adaptation data; obtain the network resource coordination data and the fault adaptation data, construct a resource regulation case library under typical fault scenarios through a regulation case generation method, and generate regulation case data;

[0128] S271. Obtain network resource coordination data, coordinate the execution timings of network reconstruction and resource regulation through a timing matching method, and generate timing matching data; based on the timing matching data, use a key node identification algorithm to determine the network nodes that need to be supported with priority, and generate key node data; obtain the timing matching data and the key node data, construct a support mechanism of resources for network key nodes through a support strategy design method, and generate support strategy data; based on the key node data and the support strategy data, design a maintenance strategy for node voltage and frequency by using a node response method, and generate node response data;

[0129] S272. Obtain the node response data and the fault adaptation data, establish a fault type identification mechanism through a fault classification method, and generate fault classification data; based on the fault classification data, use a resource matching algorithm to determine the resource combinations suitable for different fault types, and generate resource matching data; obtain the fault classification data and the resource matching data, determine the resource call order through a priority allocation method, and generate priority data; based on the resource matching data and the priority data, generate resource regulation cases for different fault types through a typical case construction method, and generate regulation case data;

[0130] In an embodiment of the present application, the resource classification control method: C(r) = {cs(r), cv(r), cb(r)} is a resource classification control function; where cs(r) is a ship load control function: cs(r) = min{Ps,max, max{Ps,min, αs·P* + βs·ΔV + γs·Δf}}; cv(r) is a V2G control function: cv(r) = k·SOC·Pv,rated·sigmoid(λ·ΔV)·(1 - exp(-t / τv)); cb(r) is a energy storage control function: cb(r) = kb·Pb,rated·tanh(αb·ΔV + βb·Δf)·fs(SOC); where P* is the target power, ΔV is the voltage deviation, Δf is the frequency deviation, SOC is the state of charge, Px,rated is the rated power, and fs(SOC) is the SOC constraint function.

[0131] Resource coordination control method: H(C) = ∑[wi(t)·Ci(r)] is the coordination control function; where wi(t) is the dynamic weight: wi(t) = exp(vi(t)) / ∑exp(vj(t)); Hierarchical coordination: M(r) = {rm, rs}, rm is the main control resource, and rs is the subordinate resource; Main control rule: Cm(rm) = G(P*, V*, f*); Subordinate rule: Cs(rs) = G(P* - Pm, ΔV, Δf); Conflict resolution: D(ri, rj) = argmini,j{Ci(ri) + Cj(rj) | ∀k, gk(ri, rj) ≤0}; where gk(ri, rj) is the constraint function.

[0132] Network resource collaboration method: J(N, R) = ∑∑[aij·ni·rj] is the collaboration mapping function; where N={ni} is the network node set, R={rj} is the resource set, and aij is the association strength; Temporal matching: T(ni, rj) = max{0, (tN(ni)- tR(rj)) / τij}; where tN(ni) is the control time of network node ni, tR(rj) is the response time of resource rj, and τij is the characteristic time constant; Support mechanism: S(n*, r) = β·Pn*·(1 + α·ΔVn*)·(1 + γ·Δfn*); where n is the key node, Pn is the node power demand, ΔVn is the voltage deviation, and Δfn is the frequency deviation.

[0133] As Figure 4 shown, according to one aspect of the present application, step S3 is further:

[0134] S31. Obtain the power grid system status data, use the multi-layer network modeling method to construct a system network model including a node set, an edge set, and weights, map the topological relationship and operating status of the ship layer, shore power layer, and power grid layer into the network structure, and generate multi-layer network data;

[0135] S32. Based on the multi-layer network data and fault feature data, search for the reconstruction path from the fault point to the available power supply through the improved A* algorithm, and combine the pre-stored node on-off status and edge connection relationship to generate reconstruction path data including multiple feasible reconstruction paths;

[0136] S33. Based on the reconstruction path data and resource schedulable data, use the multi-objective weight allocation method to calculate the conversion time, energy loss, and reliability weight indicators of each reconstruction path, and generate path weight data representing the quality of the reconstruction path.

[0137] In one embodiment of the present application, for the multi-layer network modeling method: G = {V, E, W} is a multi-layer network model; where V = {V1∪V2∪V3} is the node set, and Vi represents the nodes of the i-th layer; E = {Eintra∪Einter} is the edge set, Eintra is the intra-layer edge, and Einter is the inter-layer edge; edge weight calculation: w(e) = α·f(e) + β·c(e) + γ·r(e); where f(e) is the traffic weight, c(e) is the capacity weight, and r(e) is the reliability weight; α, β, and γ are weight coefficients and satisfy α + β + γ = 1.

[0138] Improved A* algorithm: f(n) = g(n) + h(n) is the evaluation function; where g(n) = ∑[wi·ti + pi·ci] is the actual cost, wi is the intra-layer transfer weight, ti is the intra-layer transfer time, pi is the inter-layer switching penalty factor, and ci is the number of inter-layer switches; heuristic function: h(n) = min{d(n, g) / v + q·l(n, g)}; where d(n, g) is the network distance from node n to the target node g, v is the energy transmission speed, l(n, g) is the minimum number of required inter-layer switches, and q is the switching cost coefficient.

[0139] Multi-objective weight allocation method: W(p) = [ρ1·T(p)·exp(-λ1·t) + ρ2·L(p)·exp(-λ2·e) + ρ3·R(p)·exp(-λ3·r)] is the path weight function; where T(p) is the path time evaluation value, L(p) is the energy loss evaluation value, and R(p) is the reliability evaluation value; t is the normalized time, e is the normalized energy consumption, and r is the normalized risk degree; ρi is the basic weight coefficient, and λi is the attenuation coefficient; and ∑ρi = 1.

[0140] This embodiment not only improves the efficiency and accuracy of the fault recovery path search, but also can dynamically adjust the search strategy and evaluation weights according to the real-time state of the system, providing an optimal execution path for the fault recovery control.

[0141] In another embodiment of the present application, S34. Obtain the reconstruction path data and path weight data, establish the correspondence between the path and the actual switching device through the topology mapping method, and generate the switching operation mapping data; according to the switching operation mapping data, use the switching sequence optimization algorithm to sort the key switching actions and generate the switching operation sequence data;

[0142] S341. Obtain the reconstructed path data, decompose the path into node-level connection relationships by the path segmentation method to generate path segmentation data; according to the path segmentation data, use the device mapping algorithm to identify the physical switch devices corresponding to each path segment to generate device correspondence data; obtain the path segmentation data and the device correspondence data, determine the target state of each switch device by the state conversion method to generate switch state data; combine the switch state data and the power grid topology data, and use the conflict detection method to identify the conflicting operations in the state conversion to generate conflict marker data;

[0143] S342. Obtain the switch state data and the conflict marker data, adjust the switch operation order by the conflict resolution method to generate initial operation order data; according to the initial operation order data, use the closed-loop detection method to evaluate the circulating current risk during the operation to generate circulating current risk data; obtain the initial operation order data and the circulating current risk data, construct the safety operation rules by the safety constraint method to generate safety constraint data; combine the initial operation order data and the safety constraint data, and use the sequential optimization method to adjust the switch action timing to generate optimized order data;

[0144] S343. Obtain the optimized order data, divide the switch actions into opening operations and closing operations by the operation classification method to generate classified operation data; according to the classified operation data, use the timing arrangement method to assign an execution time window to each switch action to generate timing arrangement data; obtain the classified operation data and the timing arrangement data, identify the pre-and post-relationship between actions by the dependency analysis method to generate dependency relationship data; combine the optimized order data and the dependency relationship data, and use the parallel evaluation method to determine the switch operations that can be executed in parallel to generate parallel grouping data;

[0145] S344. Obtain the parallel grouping data and the timing arrangement data, evaluate the executability of the switch operations by the execution verification method to generate execution verification data; according to the execution verification data, use the fault response method to construct a backup strategy for switch operation failure to generate fault response data; obtain the execution verification data and the fault response data, and determine the execution order of the switch operations by the final sorting method to generate switch operation sequence data;

[0146] S35. Obtain the switch operation sequence data and the resource schedulable data, construct the cooperation strategy between the switch operation and the resource scheduling by the collaborative optimization method to generate collaborative control data; according to the collaborative control data, use the timing coordination method to ensure the time matching between the switch operation and the resource scheduling to generate timing coordination data; obtain the collaborative control data and the timing coordination data, ensure the power balance during the reconstruction process by the load balancing method to generate load balancing data; combine the timing coordination data and the load balancing data, and integrate the switch operation sequence and the resource scheduling strategy by the solution integration method to generate integrated solution data;

[0147] S351. Obtain collaborative control data, plan the voltage control strategy during the reconstruction process through the voltage management method to generate voltage control data; according to the voltage control data, design the frequency maintenance strategy through the frequency regulation method to generate frequency regulation data; obtain the voltage control data and the frequency regulation data, evaluate the system stability margin through the stability analysis method to generate stability margin data; combine the load balance data and the stability margin data, and predict the system transient response characteristics through the transient evaluation method to generate transient characteristic data;

[0148] S352. Obtain the integrated solution data and the transient characteristic data, evaluate the feasibility of the integrated solution through the solution verification method to generate solution verification data; according to the solution verification data, test the solution effect under multiple typical fault scenarios through the scenario simulation method to generate scenario test data; obtain the solution verification data and the scenario test data, optimize the control parameters through the solution tuning method to generate tuned solution data; combine the tuned solution data and the switch operation sequence data, and construct a switch operation strategy library for multiple fault types through the scenario adaptation method to generate switch strategy library data;

[0149] In an embodiment of the present application, the topology mapping method: M(p) = {(ei, si) | ei ∈ p, si ∈ S} is the mapping from the path to the switch; where ei is the edge in the path p, si is the corresponding physical switch, and S is the set of switches; the bidirectional correspondence: R(ei, si) = {0, 1, 2}, representing {open, closed, maintain the current state}; the device operation timing: T(si)= min{tj | f(M(pj)) = si}, where tj is the weighted time of the path pj; the device state conflict: C(si) = |{R(ei, si) | ei ∈ E, ei ≠ ej}|>1; the operation sequence generation: O = sortj,k{(sj, R(ej, sj), T(sj)) | C(sj) = 0}.

[0150] The switch sequence optimization algorithm: L(O) = ∑[wi · Ci(O)] is the sequence cost; where Ci(O) is various cost indicators; the timing constraint: T(si+1) - T(si) ≥ Δtmin; the safety constraint: ∀t, P(t) ≤ Pmax, V(t) ∈[Vmin, Vmax]; the minimum number of operations constraint: |O| = min{|O'| | O' meets the reconstruction requirements}; the optimal sequence: O* =argminO{L(O) | meets all constraints}.

[0151] According to one aspect of the present application, step S31 is further as follows:

[0152] S311. Based on the power grid system state data, respectively extract the state information of the ship layer, shore power layer, and power grid layer through a hierarchical extraction method to generate hierarchical state data; according to the hierarchical state data, use a topological feature extraction method to identify the network connection relationships of each layer to generate topological feature data; based on the topological feature data, determine the roles of key nodes through a node importance evaluation method to generate node characteristic data;

[0153] S312. Based on the node characteristic data, evaluate the connection strength between nodes through an edge weight calculation method to generate edge weight data; according to the edge weight data, use a path analysis method to identify the key paths in the network to generate path feature data; obtain the edge weight data and path feature data, and construct the in-layer core network structure through a network backbone extraction algorithm to generate backbone network data;

[0154] S313. Based on the hierarchical state data and backbone network data, map the operating state information to network nodes through a state mapping method to generate node state data; according to the node state data, use a dynamic characteristic analysis method to evaluate the change characteristics of the node state to generate dynamic characteristic data; based on the node state data and dynamic characteristic data, construct the state association relationship between nodes through a pre-configured state propagation model to generate state association data;

[0155] S314. Based on the backbone network data and state association data, construct the connection relationships between different levels through an inter-layer mapping method to generate inter-layer connection data; according to the inter-layer connection data, use a coupling strength evaluation method to calculate the tightness of the inter-layer connection to generate coupling strength data; based on the inter-layer connection data and coupling strength data, integrate the network structures of each layer through a multi-layer structure construction algorithm to generate multi-layer structure data;

[0156] S315. Based on the multi-layer structure data, evaluate the reliability of the network model through a structural stability analysis method to generate structural stability data; according to the multi-layer structure data and structural stability data, use a model optimization method to adjust the network structure parameters to generate optimized structure data; based on the optimized structure data, construct a complete system network model through a multi-layer network modeling method to generate multi-layer network data.

[0157] In an embodiment of the present application, the hierarchical extraction method: L(G) = {G1, G2, G3} is the hierarchical decomposition; the in-layer connection strength: w(i, j) = exp(-d(i, j) / σ 2 )·cos(θij); where d(i, j) is the node distance; θij is the included angle of the state vectors; the inter-layer mapping matrix: M(k, l) = [mij]; mij = ρ·exp(-||xi k - xj l || 2 / h2 ); where ρ is the coupling strength; h is the bandwidth parameter; Hierarchical feature: f(k) = [λ1 k , λ2 k ,..., cn k ; where λi k is the eigenvalue of the k-th layer; ci k is the centrality.

[0158] Topological feature extraction method: C(i) = 2|E(Ni)| / (ki(k(i - 1))) is the clustering coefficient; where E(Ni) is the edge set within the neighborhood of node i; ki is the node degree; Betweenness centrality: B(v) = ∑[σst(v) / σst]; where σst is the number of shortest paths; Eigenvector centrality: x = α·A·x; where A is the adjacency matrix; α is the attenuation factor; Hierarchical feature: T(G) = [C, B, x, H]; where H is the hierarchical entropy.

[0159] Node importance evaluation method: I(v) = ∑[wi·fi(v)] is the comprehensive importance; where fi(v) is the node feature; Relative importance: R(v) = I(v) / max{I(u)}; Dynamic weight: wi(t) = si(t) / ∑sk(t); where si(t)= -∑[pk·log(pk)] is the feature entropy; Stability index: S(v) = 1 - var(R(v)) / mean 2 (R(v)).

[0160] This embodiment not only overcomes the limitation that the traditional single - layer network model cannot describe the inter - layer coupling relationship, but also can accurately reflect the complex interaction relationship among the ship's electrical equipment, shore power facilities and grid equipment in the port - area ship - shore power grid system, provides a complete network structure basis for the search of fault recovery paths, and improves the feasibility of the fault recovery plan.

[0161] According to one aspect of the present application, step S32 is further as follows:

[0162] S321. Obtain multi - layer network data and fault feature data, use the node analysis method to identify the location and influence range of the fault node, establish the connection relationship between the available power nodes and the fault nodes, and generate node connection data.

[0163] S322. Obtain the node connection data, use the improved A* algorithm to construct a heuristic function, quantify the inter - layer conversion cost and intra - layer transfer cost in the network structure respectively, comprehensively consider the available state, on - off state of the nodes and the edge connection state, generate path cost data; according to the path cost data, use the dynamic evaluation method to calculate the priority of path search, and generate search priority data.

[0164] S323. Obtain path cost data and search priority data, establish an open list and a closed list based on the improved A* algorithm, adopt an adaptive step size search strategy to iteratively expand the path, calculate the actual cost and the estimated cost for each candidate node, and generate path expansion data; according to the path expansion data, adopt a pruning optimization method to remove the paths that do not meet the constraint conditions and generate optimized path data.

[0165] S324. Obtain the optimized path data, adopt a multi-path parallel evaluation method to verify the feasibility of the searched path, conduct a comprehensive evaluation based on indicators such as path length, number of conversions, and reliability, and generate reconstruction path data including multiple feasible reconstruction paths.

[0166] In an embodiment of the present application, for the node analysis method: P(F|O) = P(O|F)·P(F) / P(O) is the posterior probability of the fault; where O is the observation set; conditional probability: P(O|F) = ∏N(oi|μf, Σf); fault location: F* = argmax{P(F|O)}; influence range: IR(F) = {v | d(v, F) ≤ r & P(v|F) ≥ η}; where r is the influence radius; η is the probability threshold, P(F|O) is the posterior probability of the fault F under the given observation set O; P(O|F) is the conditional probability of the observation set O under the given fault F; P(F) is the prior probability of the fault F; P(O) is the probability of the observation set O; N(oi|μf, Σf) is the probability of the observed data oi under the Gaussian distribution with the mean of μf and the covariance matrix of Σf; d(v, F) is the distance from the node v to the fault F, and P(v|F) is the probability of the node v under the given fault F.

[0167] Heuristic function construction method: h(n) = w1·ht(n) + w2·he(n) + w3·hr(n) is the combined heuristic function; where ht(n) is the time heuristic term: ht(n) = d(n, g) / v + ∑[τi·si(n)]; he(n) is the energy heuristic term: he(n) = ∑[pi·ei(n, g)]; hr(n) is the reliability heuristic term: hr(n) = -log(∏ri(n)); dynamic weight update: wi(t + 1) = wi(t)·exp(-α·Δi). Where w1, w2, and w3 represent the weight coefficients of each heuristic term in the combined heuristic function; d(n, g) is the distance from node n to the target node g; v is a time-related parameter; τi is the weight coefficient in the time heuristic term; si(n) is the eigenvalue in the time heuristic term; pi is the weight coefficient in the energy heuristic term; ei(n, g) is the energy consumption from node n to the target node g; ri(n) is the reliability of node n; wi(t) is the weight coefficient at time t; α is the smoothing factor in the dynamic weight update, and Δi is the change amount in the weight update.

[0168] This embodiment can not only quickly find the optimal reconstruction path from the fault point to the available power supply, but also fully consider the special constraints of the interlayer conversion in the port area power supply system, improving the efficiency and accuracy of the fault recovery path search.

[0169] According to one aspect of the present application, step S33 is further as follows:

[0170] S331. Obtain the reconstruction path data, divide each reconstruction path into key operation sequences through the path decomposition method to generate operation sequence data; according to the operation sequence data, use the time series dependency analysis method to construct the dependency relationship between operations to generate dependency relationship data; obtain the operation sequence data and the dependency relationship data, and determine the core link of the reconstruction process through the critical path recognition algorithm to generate critical path data.

[0171] S332. Obtain the critical path data, establish a path energy transmission model through the energy flow analysis method to generate energy flow data; according to the energy flow data, use the loss evaluation method to calculate the loss characteristics in the energy conversion process to generate energy loss data; obtain the energy flow data and the energy loss data, and construct a path energy efficiency model through the efficiency evaluation algorithm to generate efficiency evaluation data.

[0172] S333. Obtain the critical path data and the resource schedulable data, evaluate the risk degree of path execution through the reliability analysis method to generate reliability data; according to the reliability data, use the risk propagation model to calculate the cumulative effect of risks to generate risk cumulative data; obtain the reliability data and the risk cumulative data, and construct a path reliability model through the reliability evaluation algorithm to generate reliability evaluation data.

[0173] S334. Obtain efficiency evaluation data and reliability evaluation data, establish a multi-objective evaluation system through the target decomposition method, and generate evaluation index data; according to the evaluation index data, use the analytic hierarchy process method to calculate the relative importance of each index, and generate index importance data; obtain the evaluation index data and index importance data, and calculate the comprehensive score of the path through the comprehensive evaluation algorithm to generate path scoring data.

[0174] S335. Obtain path scoring data, adjust the weight allocation strategy through the dynamic weight calculation method, and generate dynamic weight data; according to the path scoring data and dynamic weight data, use the weight optimization method to balance the relationship between multiple objectives, and generate optimized weight data; obtain the optimized weight data, and determine the final weight of each reconstructed path through the multi-objective weight allocation method to generate path weight data.

[0175] In an embodiment of the present application, the path decomposition method: P = {p1, p2,..., pk} is the path set; the operation sequence extraction: S(p) = {(ai, ti)} is the time-sequence operation sequence; where ai is the operation type; ti is the execution time; the time-sequence constraint: C(i, j) = {(ti - tj) ∈ [lij, uij]}; the basic operation cost: c(ai) = wt·t(ai) + we·e(ai)+ wr·r(ai); the total path cost: C(p) = ∑c(ai) + ∑λij·vio(Cij); where vio(Cij) is the constraint violation degree.

[0176] The energy flow analysis method: E(t) = P(t) + jQ(t) is the complex power flow; the energy transmission efficiency: η(p) = Eout(p) / Ein(p); the loss calculation: L(p) = ∑[ri·|Ii| 2+ Ki·|Si|]; where ri is the impedance; Ki is the switching loss coefficient; Power constraint: |S(v)| ≤ Smax(v); Smax(v) is the maximum allowable power value; Voltage constraint: Vmin ≤ |V(v)| ≤ Vmax; Vmin is the minimum allowable voltage value and Vmax is the maximum allowable voltage value; Power flow equation: Pi = ∑|Vi|·|Vj|·|Yij|·cos(θij + Δj - Δi), where P(t) is the active power, Q(t) is the reactive power, j is the imaginary unit, Eout(p) is the output energy, Ein(p) is the input energy, Pi is the active power of node i, |Vi| and |Vj| represent the voltage amplitudes of nodes i and j, |Yij| is the admittance between nodes i and j, θij is the phase angle difference, Δj and Δi are the phase angles of different nodes, Ii is the current at node i, and Si is the apparent power at node i.

[0177] This embodiment not only considers multiple key factors such as energy efficiency, reliability, and risk in the port area power supply system, but also can flexibly adjust the evaluation weights according to the dynamic changes of the system state, improving the scientificity and adaptability of the reconfiguration path selection.

[0178] As Figure 5 shown, according to one aspect of the present application, step S4 is further as follows:

[0179] S41. Obtain the reconfiguration path data and path weight data, and use the Pareto optimization method to weigh a predetermined number of reconfiguration objectives to generate reconfiguration scheme data including a predetermined number of optional execution schemes;

[0180] S42. Based on the reconfiguration scheme data and resource response characteristic data, evaluate the temporal feasibility of each scheme through the temporal constraint checking method, and combine the pre-stored device action timings and response delays to generate scheme feasibility data representing the implementation conditions of the scheme;

[0181] S43. Based on the scheme feasibility data and path weight data, use the improved dynamic programming method to determine the optimal execution order of control actions and generate reconfiguration control sequence data including detailed control steps.

[0182] In an embodiment of the present application, the dynamic programming method is improved: V(s, t) = min{C(s, a) + γ·E[V(s', t+1)]} is the optimal value function; where s is the system state, a is the control action, C(s, a) is the immediate cost, γ is the discount factor, and s' is the new system state; state transition constraint: Pr(s'|s, a) = ω·φ(s, a)·ψ(t); where φ(s, a) is the action feasibility function, ψ(t) is the timing constraint function, and ω is the normalization factor; action selection: π(s) = argmin{Q(s, a) + λ·H(s, a)}; where Q(s, a) is the action value function, H(s, a) is the timing entropy, and λ is the balance coefficient.

[0183] This embodiment not only realizes the multi-objective optimization of the fault recovery scheme, but also can ensure the timing feasibility and execution efficiency of the control sequence, providing a reliable execution scheme for fault recovery control.

[0184] According to one aspect of the present application, step S41 is further as follows:

[0185] S411. Obtain the reconstruction path data and path weight data, decompose the reconstruction objective into multiple independent optimization objectives through the target decoupling method to generate optimization objective data; according to the optimization objective data, use the constraint extraction method to identify the constraint conditions of each objective to generate constraint condition data; obtain the optimization objective data and constraint condition data, and determine the solution space of the optimization problem through the feasible region construction algorithm to generate feasible region data.

[0186] S412. Obtain the feasible region data, divide the feasible region into multiple subspaces through the solution space division method to generate subspace data; according to the subspace data, use the search strategy generation method to construct a local search model to generate search strategy data; obtain the subspace data and search strategy data, and search for candidate solutions in each subspace through the parallel search algorithm to generate candidate solution data.

[0187] S413. Obtain the candidate solution data, evaluate the non-dominated degree of the solutions through the dominance relationship analysis method to generate dominance relationship data; according to the dominance relationship data, use the hierarchical sorting method to divide the candidate solutions into different levels to generate solution level data; obtain the dominance relationship data and solution level data, and determine the non-dominated solution set through the Pareto front construction algorithm to generate non-dominated solution data.

[0188] S414. Obtain the non-dominated solution data, calculate the distribution characteristics of the solution set through the diversity evaluation method to generate diversity data; according to the diversity data, use the solution set optimization method to adjust the distribution of the solutions to generate distribution optimization data; obtain the non-dominated solution data and distribution optimization data, and convert the optimized solutions into specific reconstruction schemes through the scheme mapping algorithm to generate initial scheme data.

[0189] S415. Obtain the initial solution data, analyze the characteristics of each solution through a solution evaluation method to generate solution characteristic data; according to the solution characteristic data, adopt a solution screening method to remove duplicate and inferior solutions to generate screened solution data; obtain the screened solution data, and determine the final set of reconstruction solutions through a Pareto optimization method to generate reconstruction solution data.

[0190] In an embodiment of the present application, the target decoupling method: F(x) = [f1(x), f2(x),..., fm(x)] is the target vector; the target correlation: ρij = cov(fi, fj) / (σi·σj); the principal component transformation: F' = W T ·F; where W is the eigenvector matrix; the independence constraint: I(fi', fj') ≤ ε; the target reconstruction: F = V·F' + b; where V is the reconstruction matrix; b is the bias vector.

[0191] The constraint extraction method: g(x) ≤ 0 is the inequality constraint; h(x) = 0 is the equality constraint; the feasibility test: μ(x)= max{0, g(x)} + |h(x)|; the constraint violation degree: v(x) = ∑[wi·max{0, gi(x)}] + ∑[λj·|hj(x)|]; the weight adaptation: wi(t + 1) = wi(t)·(1 + Δ·vi(t)); where vi(t) is the violation degree of the i-th constraint.

[0192] The feasible region construction algorithm: Ω(t) = {x | gi(x) ≤ εi(t), hj(x) = 0} is the dynamic feasible region; where εi(t) = ε0·exp(-γ·t) is the adaptive tolerance; the boundary feature extraction: B(t) = {x | ∃i, |gi(x)| ≤Δ*}; the feasible region partition: P(Ω) = {Ωk | ∪Ωk = Ω, ∩Ωk = ∅}; the sub-region volume: V(Ωk) = ∫I(x∈Ωk)dx; the partition criterion: max{min{V(Ωk)}} subject to max{diam(Ωk)}≤d, where gi(x) represents the constraint function for defining the boundary of the feasible region; ε0 is the initial tolerance; γ is the decay factor of the adaptive tolerance; B(t) is the set of boundary features; Δ* is the threshold in the boundary feature extraction; P(Ω) is the set of feasible region partitions; Ωk is the k-th sub-region in the feasible region partition; d is the maximum allowable value of the sub-region diameter, and ∃i means there exists an i.

[0193] This embodiment can not only simultaneously meet multiple optimization objectives such as power supply reliability, energy efficiency, and conversion time in the port area power supply system, but also ensure the uniform distribution of the disaggregation, providing multiple optional high-quality solutions for decision-makers.

[0194] According to one aspect of the present application, step S43 is further as follows:

[0195] S431. Obtain the scheme feasibility data and path weight data, construct a state transition matrix using an improved dynamic programming method, regard each control action in the reconstruction process as a state node, and generate state transition data based on the transition cost and timing constraint relationship between states.

[0196] S432. Obtain the state transition data, construct a multi-stage decision-making model through an adaptive optimization algorithm, set a dynamic reward function for each decision-making stage, comprehensively consider the operation timing, device response characteristics, and system stability, evaluate and score different decision-making paths to generate decision-making score data; based on the decision-making score data, use the backtracking update method to optimize the decision-making parameters and generate parameter optimization data.

[0197] S433. Obtain the decision-making score data and parameter optimization data, use the state space search method to determine the optimal state transition sequence, calculate the cumulative cost and execution risk for each transition sequence to generate transition sequence data; based on the transition sequence data, verify the timing feasibility of the sequence through a parallel evaluation method to generate sequence verification data.

[0198] S434. Obtain the transition sequence data and sequence verification data, use the sequence optimization method to perform timing configuration and parameter tuning on the control actions, and generate reconstruction control sequence data including detailed control steps.

[0199] In one embodiment of the present application, the method for constructing the state transition matrix: T(s'|s, a) = ω(s, a)·φ(s, s')·ψ(t) is the state transition probability; where ω(s, a) is the action feasibility: ω(s, a) = exp(-v(s, a) / τ); v(s, a) is the cost of the action, τ is the temperature parameter; φ(s, s') is the state compatibility: φ(s, s') = exp(-d(s, s') / σ 2 ); d(s, s') is the distance between states, σ is the standard deviation; ψ(t) is the timing constraint: ψ(t) = I(tmin ≤ t ≤ tmax); I( ) is the indicator function, and tmin and tmax are the minimum and maximum times; the transition cost: C(s, s') = wt·Δt + we·ΔE + wr·(1 - R); wt is the time difference weight, Δt is the time difference, we is the energy difference weight, ΔE is the energy difference, wr is the reliability difference weight, and 1 - R is the reliability difference.

[0200] Dynamic reward function design: R(s, a) = α(t)·Rt(s, a) + β(t)·Re(s, a) + γ(t)·Rs(s, a) is the combined reward; where α(t), β(t), and γ(t) are the weight coefficients of the time reward, energy reward, and state reward respectively; Rt is the time reward: Rt = -log(t / tmax); Re is the energy reward: Re = (Emax - E) / Emax; Rs is the state reward: Rs = exp(-d(s, g) / r*), d(s, g) is the distance from state s to the target state g, and r* is the distance parameter; Weight update: α(t) = softmax(vα / T); where vα is the value estimate and T is the temperature parameter.

[0201] Backtracking update method: Q(s, a) = (1 - α)·Q(s, a) + α·[r + γ·maxQ(s', a')] is the Q-value update; where r is the immediate reward, γ is the discount factor, maxQ(s', a') is the maximum Q-value in the new state s', α is the learning rate, Parameter adaptation: α(t) = α0 / (1 + βt), α0 is the initial learning rate, and β is the learning rate decay factor; Experience replay: E = {(si, ai, ri, s'i)}; Priority sampling: p(i) = (|Δi| + ε) η ; where Δi is the TD error and ε is a small constant in priority sampling, η is the exponential parameter in priority sampling; Importance weight: w(i) = (N·p(i)) -β , N is the size of the experience replay set; Gradient update: θ = θ - ρ·w(i)·Δi·▽θQ, ρ is the step size parameter, and ▽θQ is the gradient of the Q-value function with respect to the parameter θ.

[0202] This embodiment not only ensures the coordinated cooperation of various devices in the port area power supply system, but also can minimize the time and energy losses during the reconstruction process and improve the execution efficiency of the fault recovery control.

[0203] As Figure 6 shown, according to one aspect of the present application, it further includes:

[0204] S5. Obtain the reconstructed control sequence data, convert it into specific device control instructions by using an event-driven instruction generation method to obtain control instruction data; according to the control instruction data and the power grid system status data, track the instruction execution situation through a real-time status monitoring method to generate execution status data; combine the execution status data and the power grid system status data, and calculate the system recovery effect by using a multi-dimensional evaluation method to obtain recovery evaluation data. S51. Adopt an event-driven instruction generation method to convert the reconstructed control sequence data into specific control parameters and action instructions for each device, and generate control instruction data including execution time, control object, and specific instructions;

[0205] S52. Obtain the control instruction data and the power grid system status data, track the execution progress and effect of each control instruction through a real-time status monitoring method, and generate execution status data representing the instruction execution situation;

[0206] S53. Obtain the execution status data and the power grid system status data, calculate the system recovery time, recovery quality, and energy loss by using a multi-dimensional evaluation method, and generate recovery evaluation data representing the recovery effect.

[0207] In this embodiment, through the combination of an event-driven instruction generation and a multi-dimensional evaluation method, the precise execution and effect evaluation of fault recovery control are realized. This embodiment not only ensures the accurate execution of control instructions, but also can monitor the execution effect in real time and conduct a comprehensive evaluation, providing a reliable feedback basis for the continuous optimization of fault recovery control.

[0208] According to one aspect of the present application, step S53 is further as follows:

[0209] S531. Obtain the execution status data and the power grid system status data, calculate the time index, quality index, and energy consumption index of system recovery respectively by using a multi-dimensional feature extraction method, establish the correlation relationship between the indexes, and generate multi-dimensional index data.

[0210] S532. Obtain the multi-dimensional index data, construct an evaluation model through a dynamic weight allocation algorithm, adaptively adjust the weight coefficients of each dimension index for different operation scenarios, calculate the comprehensive performance of system recovery in real time, and generate comprehensive performance data; based on the comprehensive performance data, adopt a trend analysis method to evaluate the dynamic characteristics of the recovery process and generate dynamic characteristic data.

[0211] S533. Obtain the comprehensive performance data and the dynamic characteristic data, adopt a multi-level evaluation method to quantitatively analyze the recovery effects of the ship layer, shore power layer, and power grid layer respectively, establish the mapping relationship of the recovery effects between layers, and generate recovery evaluation data representing the recovery effect.

[0212] This embodiment can not only accurately reflect the comprehensive effect of the power supply system fault recovery in the port area, but also timely discover problems during the recovery process, providing a reliable feedback basis for the continuous optimization of the system.

[0213] The present invention realizes the full-process intelligent control of the fault recovery of the port distribution network system by organically combining technologies such as data processing, resource evaluation, path optimization, scheme generation, and execution control. First, through efficient data processing and feature extraction technologies, the system fault characteristics and inter-layer coupling relationships are accurately identified; then, combined with the flexible resource characteristic identification and classification technologies, the optimal allocation of resources is realized; through multi-layer network modeling and improved path search algorithms, the optimal reconstruction path is quickly generated; the multi-objective optimization and dynamic programming methods are used to formulate a detailed execution plan; finally, through precise instruction execution and effect evaluation, the reliability and efficiency of the fault recovery are ensured. The present invention not only improves the success rate and efficiency of fault recovery, but also can adapt to the complexity and dynamics of the ship-shore power grid system in the port area, providing a strong guarantee for the safe and stable operation of the port power system.

[0214] The preferred embodiments of the present invention have been described in detail above. However, the present invention is not limited to the specific details in the above embodiments. Within the scope of the technical concept of the present invention, various equivalent transformations can be made to the technical solutions of the present invention, and these equivalent transformations all fall within the protection scope of the present invention.

Claims

1. A fault recovery control method for a port distribution network, characterized in that: The steps include: S1, receiving real-time data of electric ships, electric vehicles, energy storage batteries and power grid operation and power grid topology data and pre-processing them, calculating the coupling degree between different power supply areas, extracting fault characteristics, and generating power grid system status data; S2. Obtain flexible resource operation data, extract flexible resource characteristics and classify resources, evaluate resource dispatchability, and calculate resource response characteristic data; S3, reconstructing the path based on the power grid system status data, calculating the weight of each reconstructed path, and generating path weight data; S4. Based on the path weight data, a reconstruction plan is generated and the feasibility of the plan is evaluated in combination with the resource response characteristic data to obtain an evaluation result; based on the evaluation result and the path weight data, the optimal execution plan is selected to obtain the reconstruction control sequence data.

2. The port area distribution network fault recovery control method according to claim 1 is characterized in that: Step S1 is further as follows: S11. Acquire real-time data on the ship side from the ship monitoring system, acquire real-time data on shore power facilities from the shore power management system, acquire real-time data on power grid operation from the power grid monitoring system, and acquire power grid topology data; Preprocess them separately to generate standardized data sets; S12. Based on the standardized data set, the coupling relationship between different power supply areas is calculated by the Pearson correlation coefficient calculation method to generate coupling degree data; S13. Based on the standardized data set and coupling degree data, multi-scale decomposition and singular value decomposition are performed to extract fault features and generate fault feature data including fault type, location and degree; S14. Based on the coupling degree data and fault characteristic data, the node status of each layer, the relationship between layers and the timing information are integrated to generate the power grid system status data.

3. The port area distribution network fault recovery control method according to claim 2 is characterized in that: Step S2 is further as follows: S21, obtaining flexible resource operation data from the flexible resource management system, identifying the dynamic response characteristics, adjustable range, and adjustment rate of various types of resources therein, and generating resource characteristic data; S22. Based on the resource characteristic data, the flexible resources are dynamically classified, resources with similar response characteristics are classified, and resource classification data representing the characteristics of different types of resources is generated; S23. Based on resource classification data and power grid system status data, a resource dispatchability evaluation model is constructed to calculate the dispatchability of various resources at different times and generate resource dispatchability data; S24. Based on the resource characteristic data and the resource schedulability data, the response time of different types of resources is calculated to generate resource response characteristic data.

4. The port area distribution network fault recovery control method according to claim 3 is characterized in that: Step S3 is further as follows: S31, obtaining power grid system status data, building a system network model, mapping the topological relationship and operating status of the ship layer, shore power layer and power grid layer to the network structure of the system network model, and generating multi-layer network data; S32, based on the multi-layer network data and the fault characteristic data, searching for a reconstruction path from the fault point to the available power source, and generating reconstruction path data in combination with the pre-stored node on / off states and edge connection relationships; S33. Based on the reconstructed path data and resource schedulable data, the conversion time, energy loss and reliability weight index of each reconstructed path are calculated to generate path weight data.

5. The port area distribution network fault recovery control method according to claim 4 is characterized in that: Step S4 is further as follows: S41, obtaining reconstruction path data and path weight data, weighing predetermined reconstruction targets, and generating reconstruction solution data; S42, based on the reconstruction scheme data and resource response characteristic data, evaluating the feasibility of each scheme in terms of timing, and generating scheme feasibility data in combination with the pre-stored device action timing and response delay; S43. Based on the scheme feasibility data and the path weight data, determine the optimal control action execution sequence and generate reconstructed control sequence data.

6. The port area distribution network fault recovery control method according to claim 5 is characterized in that: Also includes: S5. Obtain the reconstructed control sequence data, convert it into specific device control instructions, and obtain control instruction data; track the instruction execution status according to the control instruction data and the power grid system status data, and generate execution status data; combine the execution status data and the power grid system status data to calculate the system recovery effect and obtain recovery evaluation data.

7. The port area distribution network fault recovery control method according to claim 6 is characterized in that: Step S11 is further as follows: S111, acquiring real-time data from the ship side and dividing it into data segments of fixed length to generate segmented ship data; combining with pre-stored historical ship data, calculating the degree of data deviation, identifying over-limit data points, and generating ship data tags; S112, obtaining real-time data of shore power facilities and decomposing it into different frequency components, calculating the energy distribution of each frequency band, and obtaining shore power energy data; combining with pre-stored shore power reference data, extracting effective signal components, generating shore power effective data and identifying the missing point position, and obtaining a shore power missing mark; S113, acquiring real-time data of power grid operation and converting it into frequency domain space, calculating frequency component characteristics, and obtaining power grid spectrum characteristic data; combining with pre-stored power grid standard data, reconstructing signals, generating power grid reconstruction data and identifying abnormal fluctuations, and obtaining power grid abnormality marks; S114, acquiring power grid topology data, cleaning and parsing it, constructing a power grid topology structure diagram, extracting key features and normalizing them, and generating standardized topology data and power grid topology tags; S115, obtaining a ship data mark, a shore power loss mark, a power grid abnormality mark, and a power grid topology mark, calculating a data credibility index, and generating data credibility; Generate fused correction data based on segmented ship data, shore power effective data, grid reconstruction data, standardized topology data and data credibility; S116. Based on the fused corrected data, the corresponding relationship of multi-source data is constructed, the data points at the missing moments are supplemented, the interference of mutations is eliminated, and a smooth data set is generated; the smooth data set and the data credibility are combined for normalization to generate a standardized data set.

8. The port area distribution network fault recovery control method according to claim 6, characterized in that: Step S13 is further as follows: S131, obtaining a standardized data set and decomposing it to generate feature decomposition data; for the high-frequency components therein, identifying mutation points to obtain high-frequency mutation data; S132, based on the characteristic decomposition data and high-frequency mutation data, analyzing the energy distribution and phase relationship of each frequency component; combining the coupling degree data, calculating the system state change characteristics, and generating time-frequency characteristic data; S133, constructing a feature matrix based on the time-frequency feature data, calculating the main singular values ​​and the corresponding eigenvectors, extracting the main feature components, and generating feature component data; S134. Based on the characteristic component data, the main characteristic components are matched with the preset fault characteristic templates, the similarity of various faults is calculated, and the fault characteristic data is generated.

9. The port area distribution network fault recovery control method according to claim 6, characterized in that: Step S23 is further as follows: S231, obtaining resource classification data, calculating the membership values ​​of various types of resources in three dimensions of capacity availability, response timeliness, and adjustment accuracy, and generating resource membership data; S232, based on the resource affiliation data and the power grid system status data, calculating the weight coefficient of each evaluation index, determining the relative importance between the indicators, and generating indicator weight data; S233, based on the resource affiliation data and the indicator weight data, weighted calculation is performed on the evaluation indicators of various types of resources to obtain comprehensive evaluation data; S234. Determine the resource dispatchability level based on the comprehensive evaluation data; Combined with the time series characteristics of the comprehensive evaluation data, the schedulability at different times is calculated to generate resource schedulability data.

10. The port area distribution network fault recovery control method according to claim 6, characterized in that: Step S31 is further as follows: S311, based on the power grid system status data, respectively extract the status information of the ship layer, shore power layer and power grid layer, generate hierarchical status data, identify the network connection relationship of each layer and determine the role of key nodes, and obtain node characteristic data; S312, based on the node characteristic data, evaluate the connection strength between nodes, identify the key paths in the network, build the core network structure within the layer, and generate backbone network data; S313, based on the hierarchical state data and backbone network data, mapping the operation state information to the network nodes, evaluating the change characteristics of the node states, building the state association relationship between the nodes, and generating state association data; S314, based on the backbone network data and state association data, construct the connection relationship between different layers, the evaluation method calculates the tightness of the connection between the layers, integrates the network structure of each layer, and generates multi-layer structure data; S315. Based on the multi-layer structure data, evaluate the reliability of the network model, adjust the network structure parameters, build a complete system network model, and generate multi-layer network data.

Images