Harbor district flexible resource distributed cooperative scheduling method based on swarm intelligence

Abstract

The invention discloses a harbor flexible resource distributed cooperative scheduling method based on swarm intelligence, and the method comprises the steps: obtaining harbor flexible resource basic data and real-time operation data, and carrying out the data preprocessing and fusion; a hierarchical prediction architecture is adopted to predict ship arrival, load demands and renewable energy output, and prediction confidence is calculated; a multi-objective optimization model is constructed based on the prediction result, and a scheduling strategy is solved and generated; and decomposing the scheduling strategy into edge execution instructions and carrying out real-time monitoring and evaluation. Through the multi-layer data processing, prediction and optimization framework, the utilization efficiency of flexible resources in the harbor district and the operation reliability of the system are improved, and the operation cost is reduced.

Description

Technical Field

[0001] The present invention belongs to the field of distributed resource scheduling, and in particular, to a flexible resource distributed collaborative scheduling method for port areas based on swarm intelligence. Background Art

[0002] With the deepening of the electrification and intelligent transformation of ports, the scale of flexible resources in port areas is continuously expanding, and the collaborative scheduling problem of multiple types of resources such as shore power systems, charging piles, energy storage devices, and distributed photovoltaics has become increasingly prominent. Especially in large container ports, the charging demand of ships arriving at the port has the characteristics of high power, short duration, and randomness. At the same time, the volatility and intermittency of photovoltaic power generation bring impacts to the port area distribution network. Therefore, how to achieve efficient collaboration and precise scheduling of multiple types of flexible resources is of great significance for ensuring the reliability of power supply in port areas, improving the consumption capacity of renewable energy, and reducing operating costs.

[0003] Currently, the scheduling of flexible resources in port areas mainly adopts centralized optimization methods, and unified scheduling is carried out by constructing deterministic models. For example, the energy storage-photovoltaic collaborative scheduling method based on mixed integer programming formulates a scheduling strategy by minimizing the operating cost and network loss; the charging load demand response method based on time-of-use electricity price realizes load smoothing by shifting peaks and filling valleys; the multi-time scale scheduling strategy based on model predictive control improves the real-time performance of scheduling through rolling optimization. However, these methods mainly focus on the optimization of single or local resources, lack of collaborative consideration of the overall system, and the optimization model is too dependent on deterministic information and is difficult to cope with the uncertainties in actual operation.

[0004] The existing technical solutions have the following specific problems: First, in data preprocessing, traditional standardization methods are difficult to process multi-source heterogeneous data, and information distortion is easily caused especially when dealing with extreme values and outliers; the construction of resource topological relationships only considers physical distances and ignores electrical characteristics and operating constraints. Second, in prediction modeling, the existing single-layer prediction structure cannot effectively characterize the coupling relationship between the arrival of ships at the port, load demand, and renewable energy output; the prediction confidence evaluation lacks consideration of the uncertainty of the model structure. Third, in optimal scheduling, existing methods often adopt multi-objective optimization with fixed weights and are difficult to dynamically balance multiple objectives such as operating cost, renewable energy consumption, and load fluctuation; in the distributed execution process, the consideration of communication delay and bandwidth constraints is lacking, which affects the real-time performance and reliability of instruction issuance. Finally, in strategy evaluation, the existing evaluation indicators are relatively single and are difficult to comprehensively reflect the economy, reliability, and stability of scheduling strategies; the model update mechanism lacks in-depth analysis of the sources of prediction errors, which affects the pertinence and effectiveness of model optimization. Summary of the Invention

[0005] Objective of the invention: To provide a distributed collaborative scheduling method for flexible resources in a port area based on swarm intelligence, aiming to solve at least one technical problem existing in the prior art.

[0006] Technical solution: A distributed collaborative scheduling method for flexible resources in a port area based on swarm intelligence includes the following steps:

[0007] S1. Obtain the basic data set and real-time operation data set of flexible resources in the port area, perform data preprocessing and fusion to generate a standardized basic data set and a fusion data set;

[0008] S2. Build a hierarchical prediction architecture based on the fusion data set, predict the arrival of ships, load demand, and renewable energy output, and generate a prediction result data set and a prediction confidence data set;

[0009] S3. Based on the standardized basic data set, prediction result data set, and prediction confidence data set, build a multi-objective optimization model including minimizing system operation cost, maximizing renewable energy consumption, and minimizing load fluctuation, and solve it to generate a collaborative scheduling strategy data set, where the collaborative scheduling strategy data set includes energy storage charge and discharge power strategy data and charging pile charging power strategy data;

[0010] S4. Decompose the collaborative scheduling strategy data set into edge execution instructions and send them to each control unit, collect the actual execution data set; compare it with the prediction result data set, calculate the execution deviation data, update the hierarchical prediction model, and generate an optimization evaluation data set.

[0011] Beneficial effects: The present invention ensures data quality through multi-level data processing and fusion; realizes accurate prediction of the future state of the system based on a hierarchical prediction architecture; adopts multi-objective optimization and fuzzy decision-making methods to generate an optimal scheduling strategy, and ensures the effective execution of the strategy through a real-time monitoring and evaluation mechanism; not only realizes the deep integration of data, model, and control, but also ensures the continuous improvement of the system through a closed-loop optimization mechanism, improves the utilization efficiency of flexible resources in the port area and the reliability of system operation, and provides a reliable technical support for the intelligent transformation of the port area. 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 5This is the flowchart of step S4 of the present invention. Detailed implementation manners

[0017] The following describes the present application in more detail in combination with specific embodiments. As Figure 1 shown, the present application proposes a distributed collaborative scheduling method for flexible resources in a port area based on swarm intelligence, including the following steps:

[0018] S1. Obtain the basic data set and real-time operation data set of flexible resources in the port area, extract the static information and real-time operation status information of charging piles, energy storage devices, and photovoltaic devices from the basic data set and real-time operation data set of flexible resources in the port area, and respectively perform data preprocessing by using the adaptive dynamic quantization standardization method and the self-organizing multi-scale anomaly detection method to obtain a standardized basic data set and a cleaned operation data set; based on the standardized basic data set and the cleaned operation data set, perform data fusion by using the weighted evidence theory method to generate a fusion data set;

[0019] S2. Divide the fusion data set according to a preset time window, extract time series features, spatial features, and association features to construct prediction samples; based on the prediction samples, perform model training by using a hierarchical prediction architecture to generate a hierarchical prediction model; input the fusion data set into the hierarchical prediction model, calculate the prediction results and prediction confidence levels for future time periods, and respectively generate a prediction result data set and a prediction confidence level data set;

[0020] S3. Based on the standardized basic data set, generate a resource topology relationship data set and construct network constraints; combine the prediction result data set and the prediction confidence level data set to construct dynamic constraints; combine the network constraints and the dynamic constraints to generate a constraint condition data set; based on the constraint condition data set, construct a multi-objective optimization model and solve it by using a hybrid adaptive swarm collaborative optimization algorithm to generate an optimization result data set; based on the optimization result data set, generate sub-period scheduling instructions to obtain a collaborative scheduling strategy data set; wherein the collaborative scheduling strategy data set includes strategy data for regulating the charging and discharging power of energy storage and strategy data for regulating the charging power of charging piles;

[0021] S4. Decompose the collaborative scheduling strategy data set into edge execution instructions, send them to each control unit, and collect the actual execution data set; compare the actual execution data set with the prediction result data set, calculate the execution deviation to obtain deviation data; based on the deviation data, update the hierarchical prediction model to generate an optimized evaluation data set.

[0022] As Figure 2 shown, according to one aspect of the present application, step S1 is further as follows:

[0023] S11. Read the basic data set of flexible resources in the port area, extract the static information of the number of charging piles, the capacity of charging piles, the types of energy storage devices, the capacity of energy storage devices, and the installed capacity of photovoltaic devices; process the static information using the adaptive dynamic quantization and standardization method to obtain the standardized basic data set; based on the physical location information of each flexible resource in the standardized basic data set, construct the physical connection relationship between each flexible resource to generate the resource topology relationship data set.

[0024] S12. Collect the real-time operation data set of the operating status, operating power, operating voltage, and operating current of various devices; use the self-organizing multi-scale anomaly detection method to identify and correct outliers in the real-time operation data set to obtain the cleaned operation data set; obtain the fluctuation characteristics of the historical data of each measurement point and calculate the data credibility data set of each measurement point.

[0025] S13. Pair the standardized basic data set with the cleaned operation data set according to the unique resource identifier to obtain the paired data set; based on the paired data set, the data credibility data set, and the resource topology relationship data set, use the weighted evidence theory method for data fusion to generate the fusion data set containing static information and dynamic information.

[0026] In an embodiment of the present application, read the basic data set B of flexible resources in the port area, including static information such as the number and capacity of charging piles, the types and capacity of energy storage devices, and the photovoltaic installation; process it using the adaptive dynamic quantization and standardization method to obtain the standardized basic data set B'; construct the resource topology relationship matrix T to represent the physical connection relationship between each flexible resource. Obtain the real-time measurement data set R(t), including the operating status, power, voltage, etc. of various devices; use the self-organizing multi-scale anomaly detection method to remove outliers to obtain the cleaned data set R'(t), and calculate the credibility weight vector W of each measurement point based on historical data. Pair the standardized basic data set B' with the cleaned data set R'(t) according to the resource ID, and use the weighted Dempster-Shafer evidence theory, combined with the credibility weight vector W, to output the fusion data set F(t).

[0027] The specific method of using the adaptive dynamic quantization and standardization method is as follows: construct the similarity matrix Φ(i, j) between samples to represent the similarity between the i-th and j-th samples; calculate the local density factor ρ(i) of each sample = Σexp(-||x(i)-x(j)|| 2 / σ 2 );Based on the density factor, design the adaptive quantization function: Q(x) = [1 + exp(-αρ(x))] -β* x; where x is the original data, α is the density-sensitive coefficient, β is the non-linear adjustment factor, and ρ(x) is the local density factor at the data point x; the output is the standardized data x' = Q(x).

[0028] The self-organizing multi-scale anomaly detection method is specifically as follows: construct a multi-scale time series decomposition: decompose the original sequence into subsequences of different time scales; for each sequence at each scale: calculate the local trend matrix L(t, s), where t is the time point and s is the scale; extract the dynamic fluctuation feature: V(t, s) = ||L(t, s) - L(t-1, s)||; construct a self-organizing feature map: F(t) = Σw(s)·V(t, s), where w(s) is the adaptive weight at each scale; identify and correct outliers based on the feature map F(t).

[0029] In this embodiment, by introducing an adaptive dynamic quantization and standardization method, a self-organizing multi-scale anomaly detection method, and a weighted evidence theory fusion method, a multi-level data preprocessing and fusion framework is constructed. It can effectively process the heterogeneous data of the flexible resources in the port area, improve the data quality and reliability, and provide a high-quality data basis for subsequent prediction and optimization.

[0030] According to one aspect of the present application, step S11 is further as follows:

[0031] S111. Read the basic dataset of the flexible resources in the port area, extract the charging pile identifier, charging pile location coordinates, rated capacity, operating voltage level, charging interface type, maximum output power per gun, energy storage device identifier, energy storage device type, total energy storage capacity, rated power of the power conversion system, number of energy storage units, battery chemistry type, photovoltaic device identifier, photovoltaic array layout location, string composition method, rated power of the inverter, and grid connection point location, and generate an original basic data table; adopt the adaptive quantization and standardization method to map the numerical data in the original basic data table to the interval [-1, 1], and at the same time convert the charging interface type, energy storage device type, and battery chemistry type into numerical representations to generate a standardized basic dataset;

[0032] S112. Obtain the location coordinate information in the standardized basic dataset, calculate the Euclidean distance matrix between devices to generate a device distance dataset; read the pre-stored distance threshold configuration table, and based on the device distance dataset and the distance threshold configuration table, adopt a threshold determination method to identify the spatially adjacent device groups to generate a physical connection relationship dataset; based on the physical connection relationship dataset, extract the grid connection point information of each device, and adopt a depth-first search method to identify the upstream and downstream relationships of the power supply loop to generate an electrical connection relationship dataset; weighted superposition of the physical connection relationship dataset and the electrical connection relationship dataset according to a preset weight to generate a preliminary topological relationship dataset;

[0033] S113. Read the pre-stored dataset of the operating constraints of the port area power grid, and extract the constraint parameters including the rated capacity of the power supply circuit, voltage level, and power factor; map the constraint parameters one by one with the connection relationships in the preliminary topology relationship dataset to generate a constraint mapping dataset; based on the constraint mapping dataset, use the shortest path analysis method to identify the key connection nodes in the network and generate a key node dataset; based on the key node dataset, use the minimum spanning tree algorithm to identify the redundant connection paths in the network and generate a redundant path dataset;

[0034] S114. Merge the preliminary topology relationship dataset, the constraint mapping dataset, the key node dataset, and the redundant path dataset to construct a multi-layer network topology graph and generate a topology graph dataset; based on the topology graph dataset, calculate the degree centrality, betweenness centrality, and closeness centrality of each node to generate a node importance dataset; combine the topology graph dataset with the node importance dataset to generate the final resource topology relationship dataset.

[0035] In an embodiment of the present application, an adaptive quantization normalization method is used to process the numerical data in the original basic data table. The specific process is as follows: for physical quantity data (such as power, voltage, current), perform normalization based on its rated value: x_normalized = (x - nominal_value) / nominal_value; for statistical data, use the improved Z-score method: x_normalized = (x - median) / MAD, where MAD is the median absolute deviation; use the Sigmoid function for the final mapping: x_final = 2 / (1 + exp(-k*x_normalized))-1, where k is an adaptive coefficient that is dynamically adjusted according to the data distribution, x_normalized is the normalized data value, x is the original data value, nominal_value is the rated value, median is the median, and x_final is the data value after the final mapping.

[0036] In another embodiment of the present application, the adaptive dynamic quantization normalization method: X_norm(i, t) = α(t)·[x(i, t) - μ(t)] / [σ(t) + ε] + β(t)·tanh[γ(t)·x(i, t)]; where μ(t) = Σx(i, t) / N(t) is the time-varying mean; σ(t) = sqrt[Σ(x(i, t) - μ(t)) 2 / N(t)] is the time-varying standard deviation; α(t) = exp[-λ·|σ(t) / σ_ref - 1|] is the adaptive scale factor; β(t) = 1 - α(t) is the complementary weight; γ(t) = k·log[1 + var(t)] is the non-linear mapping factor; ε is a small constant to prevent division by zero; x(i, t) is the original data; i is the data index; t is the timestamp; N(t) is the number of samples at time t; λ is the attenuation coefficient; k is the gain coefficient; σ_ref is the reference standard deviation; var(t) is the variance within the time window.

[0037] Key node identification method: Node_importance(i) = α·BC(i) + β·CC(i) + γ·DC(i)·exp[-λ·Path_cost(i)]; where BC(i) = Σ[g_st(i) / g_st] is the betweenness centrality of node i, g_st is the total number of shortest paths, and g_st(i) is the number of shortest paths passing through node i; CC(i) = (n - 1) / Σd(i, j) is the closeness centrality of node i, n is the total number of nodes, and d(i, j) is the distance between nodes; DC(i) = k_i / k_max is the degree centrality of node i, k_i is the degree of node i, and k_max is the maximum degree; Path_cost(i) = Σ[w_ij·Z_ij] is the path impedance cost, w_ij is the line weight, and Z_ij is the line impedance; α, β, and γ are importance weight coefficients; λ is the attenuation coefficient, and Node_importance(i) is the importance score of node i.

[0038] In another embodiment of the present application, step S111 further includes outlier identification and processing, specifically: AD_score(i) = |x(i) - μ| / (σ + ε)·exp[-κ·(x(i) - μ) 2 / σ 2 ; where x(i) is the original data point; μ is the local mean; σ is the local standard deviation; κ is the scale parameter; ε is the stability factor; when AD_score(i) exceeds the threshold τ, it is marked as an outlier.

[0039] It also includes extreme value smoothing processing, specifically: x_smooth(i) = α·x(i) + (1 - α)·x_med(i); where x_med(i) is the local median; α = exp[-β·AD_score(i)] is the adaptive smoothing coefficient; β is the control parameter.

[0040] Step S112 further includes electrical topology constraints, including calculating the comprehensive distance: D_comp(i, j) = w1·D_euc(i, j) + w2·D_elec(i, j) + w3·D_imp(i, j); where D_euc(i, j) is the Euclidean distance; D_elec(i, j) is the electrical connection distance; D_imp(i, j) is the impedance distance; w1, w2, and w3 are weight coefficients;

[0041] Perform electrical topology constraint verification: TC_score(i, j) = min{1, P_max(i, j) / P_rated}·exp[-λ·Z_path(i, j)]; where P_max(i, j) is the maximum transmission power of the path; P_rated is the rated power; Z_path(i, j) is the path impedance; λ is the attenuation coefficient.

[0042] In this embodiment, by adopting the adaptive dynamic quantization normalization method and the resource topology relationship construction method, the in-depth processing and relationship modeling of flexible resource basic data are realized. It not only solves the problem of heterogeneous data standardization, but also provides basic constraint conditions for subsequent collaborative scheduling by establishing accurate resource topology relationships, effectively improving the accuracy and reliability of data processing.

[0043] In another embodiment of the present application, step S11 can also be:

[0044] S11a. Read the flexible resource basic data set of the port area, extract the rated capacity, geographical location, and technical parameters of charging piles, energy storage devices, and photovoltaic devices, and generate an equipment parameter data set;

[0045] S11b. Normalize the parameters in the equipment parameter data set to eliminate the influence of dimensions and generate a standardized basic data set;

[0046] S11c. Based on the geographical location information in the standardized basic data set, construct the physical connection relationship between each flexible resource to generate a physical topology data set; based on the pre-stored power system wiring situation, construct the electrical connection relationship to generate an electrical topology data set;

[0047] S11d. Integrate the physical topology data set and the electrical topology data set to construct a multi-layer network topology graph; calculate the degree centrality, betweenness centrality, and closeness centrality of each node to generate a node importance data set; based on the node importance data set and the multi-layer network topology graph, generate a resource topology relationship data set.

[0048] According to one aspect of the present application, step S12 is further:

[0049] S121. Read the original operation dataset collected in real time, extract the monitoring data of the output voltage, output current, output power, charging interface temperature, charging gun connection status of the charging pile, the charge and discharge power of the energy storage device, the battery pack voltage, battery pack temperature, state of charge, and state of health, the DC bus voltage, DC current, AC output voltage, AC current, and inverter temperature of the photovoltaic device, establish a unified timestamp index, and generate a time-series measurement point dataset; classify the time-series measurement point dataset according to the device type and physical quantity type to generate a classified measurement point dataset.

[0050] S122. Read the classified measurement point dataset, construct the feature space of the measurement point data using the self-organizing mapping method to generate a feature mapping dataset; perform multi-scale decomposition on the feature mapping dataset based on wavelet transform to extract the data features at different time scales to generate a multi-scale feature dataset; input the multi-scale feature dataset into the self-organizing neural network to train the network model parameters to generate a self-organizing model dataset.

[0051] S123. Read the classified measurement point dataset, construct a self-organizing multi-scale anomaly detection model using the parameters in the self-organizing model dataset, identify the anomaly points in the time-series data to generate an anomaly detection dataset; perform data correction on the anomaly data identified in the anomaly detection dataset in combination with the local linear regression method to generate a data correction dataset; perform time-series reconstruction and data completion on the data correction dataset to generate a cleaned operation dataset.

[0052] S124. Read the multi-scale feature dataset, calculate the data fluctuation characteristics of each measurement point at different time scales to generate a fluctuation characteristic dataset; read the pre-stored historical data fluctuation characteristic set, compare and analyze it with the fluctuation characteristic dataset to evaluate the credibility of the current data to generate a credibility evaluation dataset; update the historical credibility index of each measurement point based on the exponential smoothing method to generate a data credibility dataset.

[0053] In an embodiment of the present application, the self-organizing multi-scale anomaly detection method: AD_score(i, t) = ω1(t)·WD(i, t) + ω2(t)·CD(i, t) + ω3(t)·TD(i, t); where WD(i, t) = Σ|<x(t), ψ_j>| / ||ψ_j|| is the wavelet decomposition anomaly degree, and ψ_j is the wavelet basis function of the j scale; CD(i, t) = exp[-||x(i, t) - c_k(t)|| 2 / 2σ_k 2is the clustering anomaly degree, c_k(t) is the k-th clustering center; TD(i, t) = |x(i, t) - MA(t)| / σ_MA(t) is the trend anomaly degree, MA(t) is the moving average; ω1(t), ω2(t), ω3(t) are adaptive weights, dynamically updated through the LSTM network; x(i, t) is the data point to be detected; i is the data index; t is the timestamp; σ_k is the standard deviation of the k-th class; σ_MA(t) is the moving average standard deviation, x(t) is the data point at time t, and < > represents the inner product.

[0054] Data credibility evaluation method: Trust_score(i, t) = θ1·CS(i, t) + θ2·TS(i, t) + θ3·RS(i, t)·exp[-λ·ΔE(i, t)]; where CS(i, t) = exp[-||x(i, t) - x_pred(i, t)|| 2 / σ_pred 2 is the consistency score; TS(i, t) = 1 - |dH(x(i, t)) / dt| / H_max is the temporal stability score, H(·) is the sample entropy; RS(i, t) = min{1, N_valid(t) / N_required} is the data integrity score; ΔE(i, t) = |E(i, t) - E_ref| / E_ref is the energy deviation; θ1, θ2, θ3 are weight coefficients; λ is the decay factor; x_pred(i, t) is the predicted value; σ_pred is the predicted standard deviation; N_valid(t) is the number of valid samples; N_required is the number of required samples; E(i, t) is the measured energy; E_ref is the reference energy.

[0055] Timestamp unified indexing method: Time_align(x, t) = ω1(t)·x_prev(t) + ω2(t)·x_next(t) + ω3(t)·x_interp(t)·exp[-Δ·ΔT(t)]; where x_prev(t) is the nearest historical data point; x_next(t) is the nearest future data point; x_interp(t) is the interpolation estimate value; ΔT(t) is the time interval; ω1(t)= exp[-α·t_prev] is the historical data weight; ω2(t) = exp[-β·t_next] is the future data weight; ω3(t)= 1 - ω1(t) - ω2(t) is the interpolation weight; Δ is the time decay coefficient; Time_align(x, t) is the aligned data value; t_prev is the time difference from the historical point; t_next is the time difference from the future point; α, β are the time weight coefficients.

[0056] Local linear regression method: LLR(x_i) = θ0(i) + θ1(i)·[x(i) - x_c(i)]·exp[-γ·d(i, c)]; where θ0(i) and θ1(i) are local regression coefficients; x_c(i) is the local center point; d(i, c) is the distance to the center point; γ is the localization parameter; LLR(x_i) is the regression estimated value; x(i) is the input data point; exp[-γ·d(i, c)] is the distance weight function.

[0057] In another embodiment of the present application, step S122 further includes: defining the abnormal determination criterion, specifically: performing multi-dimensional abnormal feature extraction: AF_score(i, t) = {statistical feature: ST(i, t) = |x(i, t) - μ(t)| / σ(t); time series feature: TF(i, t) = |dx(i, t) / dt| / v_max; energy feature: EF(i, t) = |E(i, t) - E_ref| / E_ref}. The data credibility calculation mechanism is: Trust(i, t) = Σ[w_k(t)·Trust_k(i, t)]; where the weight update: w_k(t) = w_k(t-1) + η·[Acc_k(t) - w_k(t-1)], Acc_k(t) is the historical accuracy; η is the learning rate.

[0058] In this embodiment, by introducing the self-organizing multi-scale anomaly detection method and the credibility evaluation mechanism based on historical data, high-quality cleaning and credibility quantification of real-time operation data are achieved. It can not only effectively identify and process various abnormal data, but also provide a reliable weight basis for data fusion through the credibility evaluation mechanism, improving the data quality and usability.

[0059] In another embodiment of the present application, step S12 can also be:

[0060] S12a. Collect the real-time operation data of various devices, including power, voltage, current, and temperature parameters, to form the original real-time operation data set;

[0061] S12b. Perform wavelet transform on the original real-time operation data set, decompose the data into different frequency components, and generate a multi-scale data set;

[0062] S12c. Based on the multi-scale data set, construct a self-organizing neural network, perform clustering analysis on the data of each scale, identify outliers, and generate an outlier marked data set;

[0063] S12d. Based on the outlier marked data set, use the local linear regression method to correct the data points marked as abnormal, and generate the cleaned operation data set;

[0064] S12e. Based on the cleaned operation dataset, combined with the statistical characteristics of historical data in the same period, calculate the credibility scores of the data at each measurement point, and generate a data credibility dataset.

[0065] According to one aspect of the present application, step S13 is further as follows:

[0066] S131. Read the standardized basic dataset and the cleaned operation dataset, construct a mapping relationship table of static information and dynamic information according to the device unique identification code, and generate a data mapping table; based on the data mapping table, pair and combine the corresponding static information and dynamic information to generate an initial fusion dataset; use the time window method to segment the dynamic data in the initial fusion dataset to generate a segmented fusion dataset;

[0067] S132. Read the data credibility dataset, extract the credibility scores of each data measurement point, construct a basic probability assignment function, and generate a probability assignment dataset; read the resource topology relationship dataset, extract the connection relationship strength between devices, construct an evidence weight matrix, and generate a weight matrix dataset; combine the probability assignment dataset and the weight matrix dataset to construct an evidence body, and generate an evidence body dataset;

[0068] S133. Based on the segmented fusion dataset and the evidence body dataset, use the Dempster combination rule to calculate the trust degree of the fusion data, and generate a trust degree dataset; based on the trust degree dataset, use the trust degree threshold screening method to identify the credible data segments, and generate a credible dataset; perform time series reorganization on the data in the credible dataset to generate a reorganized dataset;

[0069] S134. Based on the reorganized dataset, use the weighted evidence theory method to deeply fuse the static information and dynamic information to generate a deep fusion dataset; combine the topology relationship information in the resource topology relationship dataset with the deep fusion dataset to construct the association relationship between data, and generate the final fusion dataset.

[0070] In an embodiment of the present application, the weighted evidence theory fusion method: m(A)=K -1 ·Σ[w(i)·m_i(A)·exp(-D_i)]; where K = 1 - Σ[w(i)·w(j)·m_i(B)·m_j(C)] is the normalization factor, B∩C=∅; D_i = -Σ[m_i(A)·log(m_i(A))] is the evidence conflict degree; w(i) = exp[-β·(S_i - S_min) / (S_max - S_min)] is the evidence weight; m_i(A) is the basic trust degree of the i-th evidence source for proposition A; S_i is the support degree of evidence source i; β is the weight adjustment parameter; A, B, C are subsets of the proposition set.

[0071] Dempster combination rule calculation method: m_combined(A) = k -1 ·Σ[m1(B)·m2(C)·exp(-H_conf)]; where k = 1 - Σ[m1(B)·m2(C)] is the normalization factor, B∩C = ∅; H_conf = -Σ[m(A)·log(m(A))] is the evidence conflict entropy; m_combined(A) is the basic belief degree after combination; m1(B), m2(C) are the basic belief degrees of two evidence sources; A, B, C are subsets of the focal element set, and ∅ is the empty set.

[0072] In this embodiment, by constructing a mapping relationship table between static information and dynamic information, adopting an adaptive dynamic quantization normalization method and a self-organizing multi-scale anomaly detection method for data preprocessing, the standardization level and cleaning quality of the data are effectively improved, thus ensuring the accuracy of the data; through strict data preprocessing and fusion methods, the security and integrity in the data processing process are ensured, effectively avoiding problems such as data loss and error transmission, and improving the overall reliability of the system.

[0073] As Figure 3 shown, according to one aspect of the present application, step S2 is further:

[0074] S21. Based on the fusion data set, construct a sliding time window along the time dimension, extract time series change features, spatial distribution features, and resource association features, and generate a prediction feature data set; match the pre-stored historical operation data set with the prediction feature data set to obtain the actual operation data for the corresponding period; use the actual operation data as the prediction target, and combine it with the prediction feature data set to construct a prediction sample data set;

[0075] S22. Based on the prediction sample data set, construct a three-level cascade prediction architecture for a ship arrival prediction model, a load demand prediction model, and a renewable energy output prediction model; based on the three-level cascade prediction architecture, use a hybrid adaptive group collaborative optimization algorithm for model training to generate a hierarchical prediction model;

[0076] S23. Input the fusion data set into the hierarchical prediction model to generate a prediction result data set for the future period; based on the numerical fluctuation characteristics in the prediction result data set, calculate the uncertainty interval for each prediction point to generate a prediction confidence data set.

[0077] In one embodiment of the present application, based on the fusion dataset F(t), a sliding time window is constructed to extract temporal features, spatial features, and association features, generating a prediction sample set P. A three-level cascaded prediction architecture is constructed: the first layer: ship arrival prediction based on a hybrid adaptive swarm collaborative optimization algorithm; the second layer: load demand prediction based on interval type fuzzy inference; the third layer: renewable energy output prediction based on swarm intelligence optimization; outputting a hierarchical prediction model group M. The fusion dataset F(t) is input into the hierarchical prediction model group M to generate a future period prediction result set Y(t + k), calculating the prediction uncertainty interval to obtain a prediction confidence matrix C.

[0078] The hybrid adaptive swarm collaborative optimization algorithm is specifically as follows: construct a multi-agent swarm G = {g 1 , g 2 ,..., g n}, assign a dynamic learning strategy to each agent: S(gi) = {local exploration, global exploration, swarm learning}, define a collaborative potential energy function: E(G) = Σλ 1 ||x(i)-x*(i)|| 2 + λ 2 Σφ(x(i), x(j)); where x(i) is the position of agent i, x*(i) is the historical optimal position of agent i, φ(x(i), x(j)) is the interaction potential energy between agents, λ 1 , λ 2 are adaptive weight coefficients; drive the swarm evolution based on the collaborative potential energy: dx(i) / dt = -▽E(G) + η(t), where η(t) is an adaptive perturbation term and ▽ is the gradient.

[0079] In this embodiment, by constructing a hierarchical prediction architecture and using a hybrid adaptive swarm collaborative optimization algorithm for model training, accurate prediction of the future state of flexible resources in the port area is achieved. At the same time, not only the characteristics and mutual influences of different types of loads are considered, but also the uncertainty quantification of the prediction results is provided through a confidence evaluation mechanism, providing a reliable decision-making basis for subsequent optimal scheduling.

[0080] According to one aspect of the present application, step S21 is further as follows:

[0081] S211. Read the fusion dataset, segment the data using the sliding time window method, set the window length to twice the prediction duration, and set the sliding step to the sampling period to generate a time window dataset; perform data integrity checks on each window in the time window dataset, and eliminate windows with a data missing rate exceeding a preset threshold to generate a valid window dataset.

[0082] S212. Read the effective window dataset, extract the first-order change rate and second-order change rate of the data using differential operations, calculate statistical features including mean, standard deviation, kurtosis, and skewness, and generate a time-series change feature dataset; calculate the spatial autocorrelation coefficient for the device location information in the effective window dataset, extract the spatial distribution pattern, and generate a spatial distribution feature dataset; extract the topological connection relationship and operation status correlation degree between devices from the effective window dataset, and generate a resource association feature dataset.

[0083] S213. Concatenate the time-series change feature dataset, spatial distribution feature dataset, and resource association feature dataset to generate a prediction feature dataset.

[0084] S214. Read the pre-stored historical operation dataset, split it according to the same time span as the effective window dataset to generate a historical window dataset; extract the actual operation data corresponding to each time window from the historical window dataset as the prediction target, and generate a prediction sample dataset.

[0085] In an embodiment of the present application, the method for calculating the spatial autocorrelation coefficient: Moran_I(t) = [n·Σ(w_ij·(x_i - μ)·(x_j - μ))] / [S0·Σ(x_i - μ) 2 ·exp[-ρ·d_ij]; where n is the number of samples; w_ij is the element of the spatial weight matrix; x_i, x_j are the observed values at spatial positions i, j; μ is the mean of the observed values; S0 is the sum of the spatial weights; d_ij is the distance between positions i, j; ρ is the distance decay coefficient; Moran_I(t) is the spatial autocorrelation coefficient.

[0086] In this embodiment, by constructing a feature extraction framework based on a sliding time window, a comprehensive feature representation of the prediction samples is achieved. It not only comprehensively depicts the time-series, spatial, and association characteristics of the data but also provides rich prediction input information through the combination of features, providing high-quality training samples for the subsequent prediction model.

[0087] According to an aspect of the present application, step S22 is further as follows:

[0088] S221. Read the prediction sample dataset, divide the data into a ship arrival data subset, a load demand data subset, and a renewable energy data subset according to the prediction target type to generate a classification sample dataset; divide each subset in the classification sample dataset into a training set and a validation set in a ratio of 8:2 to generate a training dataset and a validation dataset.

[0089] S222. Read the training data set related to the ship's arrival at the port, construct a ship arrival prediction model including a time series feature extraction layer, an attention mechanism layer, and a prediction output layer, optimize the model parameters using a hybrid adaptive group collaborative optimization algorithm to generate a ship prediction model; input the verification data set related to the ship's arrival at the port into the ship prediction model, calculate the prediction accuracy, and generate a ship model evaluation data set.

[0090] S223. Combine the prediction results of the ship prediction model with the training data set related to the load demand, construct a load prediction model considering the ship charging demand, optimize the model parameters using a hybrid adaptive group collaborative optimization algorithm to generate a load prediction model; input the verification data set related to the load demand into the load prediction model, calculate the prediction accuracy, and generate a load model evaluation data set.

[0091] S224. Combine the prediction results of the load prediction model with the training data set related to renewable energy, construct a renewable energy output prediction model considering the load characteristics, optimize the model parameters using a hybrid adaptive group collaborative optimization algorithm to generate a renewable prediction model; input the verification data set related to renewable energy into the renewable prediction model, calculate the prediction accuracy, and generate a renewable model evaluation data set.

[0092] S225. Construct a three - level cascade prediction architecture for the ship prediction model, the load prediction model, and the renewable prediction model according to the prediction dependency relationship to generate a cascade architecture data set; combine the ship model evaluation data set, the load model evaluation data set, and the renewable model evaluation data set to generate a model evaluation data set; combine the cascade architecture data set with the model evaluation data set to generate a final hierarchical prediction model.

[0093] In an embodiment of the present application, the hierarchical prediction architecture optimization method: y_l(t + h)=f_l[Φ_l(X_t)·W_l(t)+Ψ_l(Y_l - 1)·V_l(t)]·α_l(t); where Φ_l(X_t)=LSTM[x(t - τ:t)] is the feature extraction function of the l - th layer; Ψ_l(Y_l - 1)=CNN[y_l - 1(t - k:t)] is the inter - layer association function; W_l(t), V_l(t) are adaptive weight matrices; α_l(t)=sigmoid[β_l·SNR_l(t)] is the hierarchical confidence; x(t) is the input feature vector; y_l(t) is the output of the l - th layer; h is the prediction step; τ is the time window length; k is the historical data length; SNR_l(t) is the signal - to - noise ratio of the l - th layer; β_l is the sensitivity coefficient.

[0094] Attention mechanism method: Attn_score(q, k)=Σ[softmax(q·k T[(q·k / sqrt(d_k))·v]·exp[-λ·T_gap]; where q is the query vector; k is the key vector; v is the value vector; d_k is the dimension of the key vector; T_gap is the time interval; softmax(x) = exp(x) / Σexp(x) is the normalization function; λ is the time decay coefficient; Attn_score(q, k) is the attention score; sqrt(d_k) is the scaling factor; q·k T is the inner product of the query and the key.

[0095] In another embodiment of the present application, step S221 further includes defining a coupling metric: Coupling_score(i, j) = MI(X_i, X_j) / sqrt[H(X_i)·H(X_j)]; where MI is the mutual information; H is the information entropy; X_i, X_j are the inputs of different prediction models.

[0096] This embodiment not only considers the dependence relationship between different prediction objects, but also improves the prediction accuracy through information transmission between models, providing a reliable decision basis for scheduling optimization.

[0097] According to one aspect of the present application, step S23 is further as follows:

[0098] S231. Based on the fusion dataset, perform data resampling according to a preset prediction time scale to generate a resampled dataset; convert the resampled dataset into the input format required by the hierarchical prediction model to generate a model input dataset;

[0099] S232. Input the model input dataset into the ship arrival prediction model in the hierarchical prediction model to generate a ship prediction dataset; based on the ship prediction dataset, obtain the probability distribution output by the model during the prediction process, calculate the prediction intervals of the ship arrival time and the charging demand, and generate a ship prediction interval dataset;

[0100] S233. Combine the ship prediction dataset and the model input dataset and input them into the load demand prediction model in the hierarchical prediction model to generate a load prediction dataset; based on the load prediction dataset, use the bootstrap resampling method to calculate the confidence interval of the load prediction value and generate a load prediction interval dataset;

[0101] S234. Combine the load prediction dataset and the model input dataset and input them into the renewable energy output prediction model in the hierarchical prediction model to generate a renewable prediction dataset; based on the renewable prediction dataset, use the ensemble learning method to calculate the prediction error distribution of the renewable energy output and generate a renewable prediction interval dataset;

[0102] S235. Align and merge the ship prediction dataset, load prediction dataset, and renewable prediction dataset according to timestamps to generate a prediction result dataset; perform confidence mapping and normalization processing on the ship prediction interval dataset, load prediction interval dataset, and renewable prediction interval dataset to generate a prediction confidence dataset.

[0103] In an embodiment of the present application, the prediction confidence calculation method: Conf_score(t + h) = [1 - λ·E_norm(t)]·exp[-Δ·VAR(t)]·R(t); where E_norm(t) = ||y_pred(t) - y_true(t)|| / ||y_true(t)|| is the normalized prediction error; VAR(t) = Σ(y_i(t) - y_mean(t)) 2 / N is the integrated prediction variance; R(t) = exp[-γ·|dx(t) / dt|] is the data stability index; y_pred(t) is the predicted value; y_true(t) is the true value; y_i(t) is the predicted value of the i-th base learner; y_mean(t) is the integrated average; N is the number of base learners; λ, Δ, γ are adjustment parameters.

[0104] The bootstrap resampling method: CI(x, t) = μ_boot(x, t) ± z_α·σ_boot(x, t)·exp[-γ·N_eff(t)]; where μ_boot(x, t) is the resampling mean; σ_boot(x, t) is the resampling standard deviation; z_α is the confidence level coefficient; N_eff(t) is the number of effective samples; γ is the sample number decay coefficient; CI(x, t) is the confidence interval; x is the prediction variable; t is the time point.

[0105] In another embodiment of the present application, step S231 further includes: calculating the modified prediction confidence: Conf_total(t) = Conf_model(t)·Conf_data(t)·Conf_struct(t); where Conf_model(t) = exp[-λ1·σ_ensemble 2(t) is the model uncertainty, Conf_data(t) = exp[-λ2·VAR_input(t)] is the data uncertainty, Conf_struct(t) = exp[-λ3·D_coupling(t)] is the structural uncertainty, λ1 is the weight adjustment parameter for model uncertainty, σ_ensemble(t) is the standard deviation of the model ensemble prediction results, λ2 is the weight adjustment parameter for data uncertainty, VAR_input(t) is the variance of the input data, λ3 is the weight adjustment parameter for structural uncertainty, and D_coupling(t) is the coupling degree between different components or subsystems in the system.

[0106] In this embodiment, by establishing a confidence evaluation mechanism for multi-layer prediction results, accurate quantification of prediction uncertainty is achieved. Also, by considering the prediction characteristics of different types of loads, a risk control basis is provided for subsequent optimal scheduling, improving the reliability of decision-making.

[0107] As Figure 4 shown, according to one aspect of the present application, step S3 is further as follows:

[0108] S31. Based on the resource topology relationship dataset, extract network topology information and device capacity information to construct network static constraints; based on the prediction result dataset and the prediction confidence dataset, construct dynamic operation constraints; combine the network static constraints and the dynamic operation constraints to generate a constraint condition dataset;

[0109] S32. Based on the constraint condition dataset, construct a multi-objective optimization model that includes minimizing the system operation cost, maximizing the consumption of renewable energy, and minimizing the load fluctuation; based on the multi-objective optimization model, use a hybrid adaptive swarm collaborative optimization algorithm to solve it and generate an optimization result dataset;

[0110] S33. Based on the optimization result dataset, use a fuzzy decision method to select the optimal solution, generate a time-segmented scheduling instruction sequence, and output a collaborative scheduling strategy dataset.

[0111] In an embodiment of the present application, network constraints are constructed based on the resource topology relationship matrix T, dynamic constraints are constructed by combining the prediction result set Y(t + k) and the prediction confidence matrix C, and the constraint condition set H is output. Define the objective functions: f1 is to minimize the system operation cost; f2 is to maximize the consumption of renewable energy; f3 is to minimize the load fluctuation; use an improved multi-objective grey wolf algorithm to solve it and output the Pareto optimal solution set S. Select the optimal solution from the Pareto optimal solution set S based on the fuzzy decision theory, generate time-segmented scheduling instructions, and output the collaborative scheduling strategy D.

[0112] In another embodiment of the present application, the fuzzy decision-making method is specifically as follows: Fuzzy decision score FD_score(i) = Σ[μ_k(x_i)·w_k(t)·exp(-ρ·σ_k 2 )]; where μ_k(x_i) = 1 / [1 + ((x_i - c_k) / a_k) 2 b_k] is the membership function of the k-th fuzzy rule; w_k(t) = s_k(t) / Σs_k(t) is the dynamic rule weight; s_k(t) = exp[η·(r_k(t) - r_min) / (r_max - r_min)] is the rule strength; σ_k 2 is the variance of rule k; x_i is the decision variable; c_k is the fuzzy center; a_k is the fuzzy scale parameter; b_k is the shape parameter; r_k(t) is the responsiveness of rule k at time t; ρ is the penalty coefficient; η is the gain coefficient, r_max is the maximum responsiveness, and r_min is the minimum responsiveness.

[0113] In this embodiment, by comprehensively considering the resource topology relationship, prediction results, and prediction confidence, a multi-objective optimization framework is constructed. It can not only ensure the physical feasibility and economy of the scheduling strategy but also achieve the dynamic balance between multiple objectives through the fuzzy decision-making method, improving the practicability and reliability of the scheduling strategy.

[0114] According to one aspect of the present application, step S31 is further as follows:

[0115] S311. Read the resource topology relationship dataset, extract the physical connection relationship between network nodes, and generate a connection relationship dataset; based on the connection relationship dataset, construct a network power flow constraint equation, including node voltage constraints and line power constraints, and generate a power flow constraint dataset.

[0116] S312. Read the resource topology relationship dataset, extract the rated power of charging piles, the capacity of energy storage devices, and the capacity of photovoltaic devices, and generate a device capacity dataset; based on the device capacity dataset, construct device operation constraints, including upper and lower limits of charging power, the range of energy storage state of charge, and photovoltaic output limits, and generate a device constraint dataset.

[0117] S313. Read the prediction result dataset, extract the ship arrival time, load demand curve, and predicted values of renewable energy output, and generate a prediction constraint dataset; read the prediction confidence dataset, calculate the fluctuation interval of the predicted values, and generate a fluctuation constraint dataset.

[0118] S314. Combine the power flow constraint dataset and the device constraint dataset to generate a static constraint dataset; combine the prediction constraint dataset and the fluctuation constraint dataset to generate a dynamic constraint dataset; perform constraint merging and conflict resolution on the static constraint dataset and the dynamic constraint dataset to generate the final constraint condition dataset.

[0119] In this embodiment, by constructing a comprehensive constraint condition system, comprehensive constraints on the physical limitations and operating requirements of the system are realized. It not only comprehensively considers the physical limitations of the system and the device characteristics, but also improves the adaptability and reliability of the scheduling strategy by introducing dynamic constraints, providing a rigorous constraint framework for the optimization solution.

[0120] According to one aspect of the present application, step S32 is further as follows:

[0121] S321. Based on the constraint condition dataset, extract static constraint conditions and dynamic constraint conditions to generate a constraint classification dataset; based on the constraint classification dataset, construct a system operation cost objective function, including charging cost, energy storage operation cost, and grid power purchase cost, to generate a cost objective dataset; based on the constraint classification dataset, construct a renewable energy consumption objective function, calculate the photovoltaic power generation utilization rate, to generate a consumption objective dataset; based on the constraint classification dataset, construct a load fluctuation objective function, calculate the load curve smoothness, to generate a fluctuation objective dataset.

[0122] S322. Based on the cost objective dataset, the consumption objective dataset, and the fluctuation objective dataset, construct a normalization processing function to generate an objective normalization dataset; based on the objective normalization dataset, use the analytic hierarchy process to calculate the weight coefficients of each objective to generate an objective weight dataset; combine the objective normalization dataset and the objective weight dataset to construct a comprehensive objective function to generate a multi-objective function dataset.

[0123] S323. Based on the constraint classification dataset and the multi-objective function dataset, initialize the parameter configuration of the hybrid adaptive population collaborative optimization algorithm, including population size, number of iterations, and adaptive factor, to generate an algorithm parameter dataset; based on the algorithm parameter dataset, construct an individual coding scheme, map the decision variables to optimized individuals, to generate a coding scheme dataset.

[0124] S324. Based on the algorithm parameter dataset and the coding scheme dataset, initialize the optimization population to generate an initial population dataset; evaluate the fitness of the individuals in the initial population dataset to generate a fitness dataset; based on the fitness dataset, use the non-dominated sorting method to stratify the population to generate a stratified sorting dataset.

[0125] S325. According to the hierarchical sorted data set, use the hybrid adaptive swarm collaborative optimization algorithm for iterative optimization, update the population position and velocity, and generate an iterative optimization data set; screen out the non-dominated solution set from the iterative optimization data set to generate a Pareto solution set data set; decode and map the Pareto solution set data set into actual control variables to generate a final optimized result data set.

[0126] In an embodiment of the present application, the hybrid adaptive swarm collaborative optimization algorithm: v_i(t + 1) = w(t)·v_i(t) + c1·r1·[p_i(t) - x_i(t)] + c2·r2·[g(t) - x_i(t)] + c3·r3·[l_i(t) - x_i(t)]·exp[-d(i,j) / σ 2 ; where w(t) = w_max - (w_max - w_min)·(t / T_max) α is the adaptive inertia weight; c1, c2, and c3 are learning factors; r1, r2, and r3 are random numbers; p_i(t) is the individual optimal solution; g(t) is the global optimal solution; l_i(t) is the local optimal solution; d(i,j) is the distance between individual i and j; σ is the neighborhood range parameter; x_i(t) is the individual position; v_i(t) is the individual velocity; t is the number of iterations; T_max is the maximum number of iterations; α is the adaptive exponent, w_max is the maximum inertia weight, and w_min is the minimum inertia weight.

[0127] The weight calculation method of the analytic hierarchy process: w_i = [Π(a_ij) 1 / n / (Σ[Π(a_ij) 1 / n )·exp[-β·CI_ratio]; where a_ij is the judgment matrix element; n is the number of criteria; CI_ratio is the consistency ratio; β is the consistency penalty coefficient; w_i is the weight of the i-th criterion; Π is the product symbol; a_ij is the importance ratio of criterion i to criterion j.

[0128] In this embodiment, by designing a multi-objective optimization framework and using the hybrid adaptive swarm collaborative optimization algorithm to solve, the global optimization of the scheduling strategy is realized. It can not only take into account multiple objectives such as economy, environmental protection, and stability at the same time, but also improve the solution efficiency through the hybrid optimization algorithm and obtain a high-quality Pareto optimal solution set.

[0129] According to one aspect of the present application, step S33 is further:

[0130] S331. Read the optimized result data set, extract the objective function values of each Pareto optimal solution, including the operating cost index, the renewable energy consumption index, and the load fluctuation index, and generate an objective index data set; construct a fuzzy membership function, calculate the membership values of each index, and generate a membership data set.

[0131] S332. Read the objective index data set, construct a fuzzy rule base based on expert experience, including the cost priority rule, the consumption priority rule, and the stable operation rule, and generate a decision rule data set; according to the real-time operation state of the system, dynamically adjust the activation intensity of the rules, and generate a rule weight data set.

[0132] S333. Read the membership data set and the rule weight data set, use the fuzzy inference method to calculate the comprehensive evaluation values of each optimization solution, and generate an evaluation result data set; select the optimal solution from the evaluation result data set based on the maximum membership principle, and generate an optimal solution data set.

[0133] S334. Read the optimal solution data set, map the optimal solution to the charging pile power command, the energy storage charge and discharge command, and the photovoltaic output command, and generate a device command data set; sort the device command data set according to the time series, and generate a time series command data set.

[0134] S335. Read the time series command data set, correct the command execution time based on the device response characteristics, and generate a corrected command data set; perform a safety check on the corrected command data set to ensure that the physical constraints are met, and generate the final coordinated scheduling strategy data set.

[0135] In this embodiment, by establishing a scheduling strategy generation mechanism based on fuzzy decision-making, the intelligent selection of the optimal solution and the reasonable decomposition of the execution instructions are realized. It not only realizes the reasonable trade-off between multiple objectives, but also improves the reliability of strategy execution through the reasonable decomposition and verification of instructions, providing a reliable control basis for the actual operation of the system.

[0136] In another embodiment of the present application, step S3 is further as follows:

[0137] S3a. Based on the standardized basic data set, extract the network topology information and the device capacity information, and construct the network static constraints; based on the prediction result data set and the prediction confidence data set, construct the dynamic operation constraints; combine the network static constraints and the dynamic operation constraints to generate a constraint condition data set;

[0138] S3b. Based on the constraint condition data set, construct a system operation cost objective function, a renewable energy consumption objective function, and a load fluctuation objective function, and generate a multi-objective function data set; initialize the parameters of the hybrid adaptive population collaborative optimization algorithm, construct an individual coding scheme, and generate a coding scheme data set.

[0139] S3c. Based on the multi-objective function dataset and the coding scheme dataset, use the hybrid adaptive swarm collaborative optimization algorithm for iterative solution to generate the Pareto solution set dataset; use the fuzzy decision-making method to select the optimal solution from the Pareto solution set dataset to generate the collaborative scheduling strategy dataset.

[0140] According to one aspect of the present application, step S3c is further as follows:

[0141] S3c1. Use the hybrid adaptive swarm collaborative optimization algorithm to perform iterative calculation on the multi-objective function, where the individual position update is based on the combination of the adaptive inertia weight and multiple learning strategies, and generate the optimization iteration process dataset;

[0142] S3c2. Based on the optimization iteration process dataset, judge whether the optimization algorithm reaches the termination condition. If it reaches, output the final non-dominated solution set to generate the Pareto solution set dataset; if not, return to S331 to continue iterative optimization;

[0143] S3c3. Construct a fuzzy rule base based on expert experience, construct the membership functions of the system operation cost, the renewable energy consumption rate, and the load volatility, and generate the fuzzy rule dataset;

[0144] S3c4. According to the real-time state of the system, dynamically adjust the activation intensity of the fuzzy rule dataset, and perform fuzzy comprehensive evaluation on each solution in the Pareto solution set dataset to generate the evaluation result dataset;

[0145] S3c5. Select the solution with the highest comprehensive score from the evaluation result dataset as the final scheduling plan to generate the collaborative scheduling strategy dataset, including the energy storage charge and discharge power strategy data and the charging pile charging power strategy data.

[0146] As Figure 5 shown, according to one aspect of the present application, step S4 is further as follows:

[0147] S41. According to the pre-stored resource distribution situation, decompose the collaborative scheduling strategy dataset into edge execution instructions, send them to each control unit, and collect and record the actual execution dataset;

[0148] S42. Compare the actual execution dataset with the prediction result dataset, calculate the strategy execution deviation to obtain the execution deviation data; based on the execution deviation data, update the parameters of the hierarchical prediction model to generate the optimization evaluation dataset.

[0149] In an embodiment of the present application, decompose the collaborative scheduling strategy D into edge execution instructions, send them to each control unit, and record the actual execution data E(t). Compare the actual execution data E(t) with the prediction result set Y(t + k), calculate the strategy execution deviation, update the prediction model group M and the optimization parameters, and output the optimization evaluation report V.

[0150] In this embodiment, by implementing a real-time monitoring and model update mechanism, the closed-loop optimization of the scheduling strategy is achieved. This not only improves the real-time response ability of the system but also enhances the operating efficiency and reliability of the entire system through continuous optimization of the model.

[0151] According to one aspect of the present application, step S41 is further as follows:

[0152] S411. Read the collaborative scheduling strategy data set, divide the control instructions by region according to the physical distribution of the devices to generate a regional instruction data set; based on the communication delay and bandwidth constraints of each region, sort the instructions by priority to generate an instruction priority data set.

[0153] S412. Read the regional instruction data set, parse the charging pile control instructions into device-level power instructions, voltage instructions, start / stop instructions to generate a charging pile instruction data set; parse the energy storage device control instructions into charge / discharge power instructions and operating mode instructions to generate an energy storage instruction data set; parse the photovoltaic device control instructions into active power instructions, reactive power instructions, and grid connection point voltage instructions to generate a photovoltaic instruction data set.

[0154] S413. Read the charging pile instruction data set, energy storage instruction data set, and photovoltaic instruction data set, perform instruction format conversion according to the communication protocol requirements of each control unit to generate a protocol instruction data set; construct a timing queue for instruction issuance to generate an instruction queue data set.

[0155] S414. Read the protocol instruction data set, instruction queue data set, and instruction priority data set, issue the instructions to the corresponding edge control units in the order of priority, record the instruction issuance status to generate an instruction execution status data set; collect the instruction reception confirmation information of the edge control units to generate an instruction confirmation data set.

[0156] S415. Read the instruction execution status data set and instruction confirmation data set, collect the operating data of each device in real time, including the output power of the charging pile, the charge / discharge power of the energy storage, and the output power of the photovoltaic to generate an operating collection data set; perform timestamp alignment and data completion on the operating collection data set to generate a final actual execution data set.

[0157] In an embodiment of the present application, the instruction parsing method: Cmd_parse(x) = α·Base_cmd(x) +β·Delta_cmd(x) + γ·Correct_cmd(x)·exp[-Δ·T_delay]; where Base_cmd(x) is the basic instruction value; Delta_cmd(x) is the incremental correction value; Correct_cmd(x) is the correction value; T_delay is the communication delay; α, β, and γ are the weights of the instruction components; Δ is the delay attenuation coefficient; Cmd_parse(x) is the parsed instruction value; and x is the original instruction.

[0158] In this embodiment, by designing a distributed instruction execution and monitoring mechanism, the precise execution and real-time monitoring of the scheduling strategy are achieved. It not only solves the instruction coordination problem in the distributed system but also provides feedback information on the strategy execution through real-time monitoring, providing actual operation data support for system optimization.

[0159] According to one aspect of the present application, step S42 is further as follows:

[0160] S421. Align the actual execution data set and the prediction result data set according to the time stamp to generate a data comparison data set; based on the data comparison data set, calculate the vessel arrival time deviation, load demand deviation, and renewable energy output deviation to generate an execution deviation data set; based on the execution deviation data set, use the sliding time window method to statistically calculate the mean and standard deviation of various deviations to generate a statistical feature data set;

[0161] S422. Based on the execution deviation data set, use the adaptive weighting method to calculate the prediction error contribution degree of each prediction model to generate an error analysis data set; based on the statistical feature data set, identify the main sources and change trends of the prediction errors to generate an error attribution data set; combine the error analysis data set and the error attribution data set to generate a comprehensive evaluation data set;

[0162] S423. Based on the comprehensive evaluation data set, extract the performance indicators of each prediction model to generate a model performance data set; based on the model performance data set, calculate the update amount of the model parameters through the least mean square error criterion to generate a parameter update data set; apply the parameter update data set to the hierarchical prediction model to generate an updated model data set;

[0163] S424. Based on the updated model data set, conduct a prediction performance evaluation on the pre-stored verification data set to generate a performance evaluation data set; compare the performance evaluation data set with the preset indicators to generate an indicator achievement data set; evaluate the model update effect based on the indicator achievement data set to generate an update evaluation data set;

[0164] S425. Calculate the economic index, reliability index, and stability index of the policy execution based on the execution deviation dataset, error attribution dataset, performance evaluation dataset, and update evaluation dataset, and generate a comprehensive index dataset; compare and analyze the comprehensive index dataset with the pre-stored historical evaluation results to generate a final optimized evaluation dataset.

[0165] In an embodiment of the present application, the optimized evaluation index calculation method is: Perf_index(t) = θ_e·EI(t) + θ_r·RI(t) + θ_s·SI(t)·exp[-ε·ΔT(t)]; where EI(t) = (C_ref - C(t)) / C_ref is the economic index, C(t) is the operating cost, and C_ref is the reference operating cost; RI(t) = exp[-μ·MTTF(t) / MTTF_ref] is the reliability index, MTTF is the mean time to failure, and MTTF_ref is the reference mean time to failure; SI(t) = 1 - |P_var(t)| / P_max is the stability index, P_var is the power fluctuation, and P_max is the maximum power; ΔT(t) is the evaluation period; θ_e, θ_r, and θ_s are weight coefficients; ε is the time decay coefficient; and μ is the reliability sensitivity coefficient.

[0166] The adaptive weighting method is: w_adaptive(i, t) = η·w_hist(i, t) + (1 - η)·w_curr(i, t)·exp[-λ·E_rate(t)]; where w_hist(i, t) is the historical weight, w_curr(i, t) is the current weight, E_rate(t) is the error change rate, η is the historical information weight coefficient, λ is the error penalty coefficient, w_adaptive(i, t) is the adaptive weight, i is the feature index, and t is the time point.

[0167] In this embodiment, by constructing a comprehensive evaluation and model update mechanism, continuous optimization of the prediction model and dynamic evaluation of the system performance are realized. At the same time, not only the adaptive optimization of the prediction model is achieved, but also a quantitative basis for the system optimization effect is provided through the comprehensive index evaluation, providing reliable decision support for the continuous improvement of the system.

[0168] In another embodiment of the present application, step S4 can also be:

[0169] S4a. According to the pre-stored resource distribution, decompose the cooperative scheduling policy dataset into energy storage control instructions, charging pile control instructions, and photovoltaic control instructions, perform format conversion according to the communication protocol, and generate a protocol instruction dataset; issue it to each edge control unit in the order of priority to collect the actual execution dataset.

[0170] S4b. Align the actual execution dataset with the prediction result dataset, calculate the deviation of the ship's arrival time, the deviation of the load demand, and the deviation of the renewable energy output, and generate an execution deviation dataset; based on the execution deviation dataset, calculate the contribution degree of the prediction error and generate an error analysis dataset.

[0171] S4c. Based on the error analysis dataset, calculate the update amount of the hierarchical prediction model parameters and generate an updated model dataset; calculate the economic index, reliability index, and stability index of the strategy execution and generate an optimization evaluation dataset.

[0172] According to one aspect of the present application, step S4a is further as follows:

[0173] S4a1. Extract the energy storage charge and discharge power strategy data, the charging pile charging power strategy data, and the photovoltaic power control strategy data from the collaborative scheduling strategy dataset to generate a device control strategy dataset.

[0174] S4a2. Based on the pre-stored resource distribution, divide the device control strategy dataset according to the control area to generate a regional control dataset.

[0175] S4a3. According to the communication protocol requirements of each edge control unit, convert the regional control dataset into the corresponding protocol format to generate a protocol instruction dataset.

[0176] S4a4. Based on the pre-stored communication delay and bandwidth constraints, sort the instructions in the protocol instruction dataset by priority to generate an instruction priority dataset; according to the instruction priority dataset, send the control instructions to each edge control unit in sequence.

[0177] S4a5. Collect the execution status and device operation parameters of each edge control unit to generate an actual execution dataset.

[0178] The present invention discloses a distributed collaborative scheduling method for flexible resources in a port area based on swarm intelligence, including: acquiring the basic data and real-time operation data of the flexible resources in the port area, and performing data preprocessing and fusion; using a hierarchical prediction architecture to predict the ship arrival, load demand, and renewable energy output, and calculating the prediction confidence; constructing a multi-objective optimization model based on the prediction results, and using a hybrid adaptive swarm collaborative optimization algorithm to solve and generate a scheduling strategy; decomposing the scheduling strategy into edge execution instructions and performing real-time monitoring and evaluation. The present invention realizes the intelligent collaborative scheduling of the flexible resources in the port area by constructing a complete data processing, prediction modeling, optimization scheduling, and execution evaluation framework. First, the data quality is ensured through multi-level data processing and fusion, then the accurate prediction of the future state of the system is realized based on the hierarchical prediction architecture, then the optimal scheduling strategy is generated by using multi-objective optimization and fuzzy decision-making methods, and finally the effective execution of the strategy is ensured through the real-time monitoring and evaluation mechanism. The present invention not only realizes the deep integration of data, model, and control, but also ensures the continuous improvement of the system through the closed-loop optimization mechanism, improves the utilization efficiency of the flexible resources in the port area and the reliability of the system operation, and provides a reliable technical support for the intelligent transformation of the port area.

[0179] 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 belong to the protection scope of the present invention.

Claims

1. A distributed collaborative scheduling method for flexible resources in port areas based on swarm intelligence, characterized in that: The steps include: S1. Obtain the basic data set of flexible resources and the real-time operation data set of the port area, perform data preprocessing and fusion, and generate standardized basic data set and fusion data set; S2. Build a hierarchical prediction architecture based on the fused data set to predict ship arrival, load demand and renewable energy output, and generate a prediction result data set and a prediction confidence data set; S3. Based on the standardized basic data set, the prediction result data set and the prediction confidence data set, a multi-objective optimization model including minimization of system operation cost, maximization of renewable energy consumption and minimization of load fluctuation is constructed and solved to generate a collaborative scheduling strategy data set, where the collaborative scheduling strategy data set includes energy storage charging and discharging power strategy data and charging pile charging power strategy data; S4, decomposing the collaborative scheduling strategy data set into edge execution instructions and sending them to each control unit to collect the actual execution data set; Compare it with the prediction result dataset, calculate the execution deviation data, update the hierarchical prediction model, and generate an optimized evaluation dataset.

2. The distributed collaborative scheduling method for flexible resources in port areas based on swarm intelligence according to claim 1 is characterized in that: The step S2 comprises: S21. Based on the fused data set, a sliding time window is constructed to extract time series change characteristics, spatial distribution characteristics and resource association characteristics to generate a prediction feature data set; combined with the pre-stored historical operation data set, the actual operation data of the corresponding period is obtained to construct a prediction sample data set; S22. Based on the prediction sample data set, a three-layer cascade prediction architecture of ship arrival prediction model, load demand prediction model and renewable energy output prediction model is constructed; a hybrid adaptive swarm collaborative optimization algorithm is used for model training to generate a hierarchical prediction model; S23, input the fused data set into the hierarchical prediction model to generate a prediction result data set for the future period; calculate the uncertainty interval of each prediction point based on the prediction result data set to generate a prediction confidence data set.

3. The distributed collaborative scheduling method for flexible port resources based on swarm intelligence according to claim 1 is characterized in that: The step S1 comprises: S11. Read the basic data set of flexible resources in the port area, extract the static information of charging piles, energy storage equipment and photovoltaic equipment, perform standardization processing, and generate a standardized basic data set; construct the physical connection relationship and electrical connection relationship between each flexible resource, and generate a resource topology relationship data set; S12. Collect real-time operation data sets of various equipment, use self-organizing multi-scale anomaly detection methods to identify and correct outliers, and generate cleaned operation data sets; based on the fluctuation characteristics of historical data, calculate the data credibility data sets of each measuring point; S13. Pair the standardized basic data set with the cleaned running data set to generate a paired data set; based on the paired data set, the data credibility data set and the resource topology relationship data set, use the weighted evidence theory method to fuse the data to generate a fused data set.

4. The distributed collaborative scheduling method for flexible port resources based on swarm intelligence according to claim 1 is characterized in that: The step S3 comprises: S31. Based on the standardized basic data set, extract network topology information and equipment capacity information to construct network static constraints; based on the prediction result data set and the prediction confidence data set, construct dynamic operation constraints; combine the network static constraints and the dynamic operation constraints to generate a constraint condition data set; S32. Based on the constraint condition data set, construct the system operation cost objective function, renewable energy consumption objective function and load fluctuation objective function, and generate a multi-objective function data set; initialize the hybrid adaptive group collaborative optimization algorithm parameters, construct an individual coding scheme, and generate a coding scheme data set; S33. Based on the multi-objective function data set and the coding scheme data set, a hybrid adaptive group collaborative optimization algorithm is used for iterative solution to generate a Pareto solution set data set; a fuzzy decision method is used to select the optimal solution from the Pareto solution set data set to generate a collaborative scheduling strategy data set.

5. The distributed collaborative scheduling method for flexible resources in port areas based on swarm intelligence according to claim 1 is characterized in that: The step S4 comprises: S41. According to the pre-stored resource distribution, the collaborative scheduling strategy data set is decomposed into energy storage control instructions, charging pile control instructions and photovoltaic control instructions, and the format is converted according to the communication protocol to generate a protocol instruction data set; it is sent to each edge control unit in order of priority to collect the actual execution data set; S42, aligning the actual execution data set with the prediction result data set, calculating the deviation of the ship arrival time, the load demand deviation and the renewable energy output deviation, and generating an execution deviation data set; based on the execution deviation data set, calculating the prediction error contribution, and generating an error analysis data set; S43. Based on the error analysis data set, calculate the update amount of the hierarchical prediction model parameters to generate an updated model data set; calculate the economic index, reliability index and stability index of the strategy execution to generate an optimized evaluation data set.

6. The distributed collaborative scheduling method for flexible port resources based on swarm intelligence according to claim 2 is characterized in that: The step S23 comprises: S231, inputting the fused data set into the first-layer ship arrival prediction model in chronological order to generate a ship arrival prediction data set; calculating the ship arrival prediction interval to generate a ship prediction confidence data set; S232, integrating the ship arrival prediction data set and the fusion data set, and inputting them into the second-layer load demand prediction model to generate a load demand prediction data set; calculating the load demand prediction interval, and generating a load prediction confidence data set; S233, integrating the load demand prediction data set, the ship arrival prediction data set and the fusion data set, and inputting them into the third-layer renewable energy output prediction model to generate a renewable energy output prediction data set; calculating the renewable energy output prediction interval, and generating a renewable energy prediction confidence data set; S234. Combine the ship arrival prediction data set, the load demand prediction data set and the renewable energy output prediction data set to generate a prediction result data set; integrate the ship prediction confidence data set, the load prediction confidence data set and the renewable energy prediction confidence data set to generate a prediction confidence data set.

7. The distributed collaborative scheduling method for flexible resources in port areas based on swarm intelligence according to claim 3 is characterized in that: The step S11 comprises: S111. Read the basic data set of flexible resources in the port area, extract the rated capacity, geographical location and technical parameters of charging piles, energy storage equipment and photovoltaic equipment, and generate an equipment parameter data set; S112, normalizing various parameters in the equipment parameter data set to eliminate the dimension effect and generate a standardized basic data set; S113, based on the geographical location information in the standardized basic data set, construct the physical connection relationship between the flexible resources to generate a physical topology data set; based on the pre-stored power system wiring conditions, construct the electrical connection relationship to generate an electrical topology data set; S114. Integrate the physical topology dataset and the electrical topology dataset to construct a multi-layer network topology map; calculate the degree centrality, betweenness centrality, and closeness centrality of each node to generate a node importance dataset; and generate a resource topology relationship dataset based on the node importance dataset and the multi-layer network topology map.

8. The distributed collaborative scheduling method for flexible port resources based on swarm intelligence according to claim 3 is characterized in that: The step S12 comprises: S121. Collect real-time operation data of various devices, including power, voltage, current and temperature parameters, to form an original real-time operation data set; S122, performing wavelet transform on the original real-time running data set, decomposing the data into different frequency components, and generating a multi-scale data set; S123. Based on the multi-scale data set, a self-organizing neural network is constructed to perform cluster analysis on the data of each scale, identify outliers, and generate an outlier labeled data set; S124, based on the outlier marked data set, using a local linear regression method to correct the data points marked as abnormal, and generate a cleaned running data set; S125. Based on the cleaned running data set and in combination with the statistical characteristics of the historical data of the same period, the credibility score of the data at each measuring point is calculated to generate a data credibility data set.

9. The distributed collaborative scheduling method for flexible port resources based on swarm intelligence according to claim 4 is characterized in that: The step S33 comprises: S331, using a hybrid adaptive swarm collaborative optimization algorithm to iteratively calculate the multi-objective function, where individual position updates are based on adaptive inertia weights and a combination of multiple learning strategies to generate an optimization iterative process data set; S332, judging whether the optimization algorithm has reached the termination condition based on the optimization iterative process data set, and if so, outputting the final non-dominated solution set to generate a Pareto solution set data set; if not, returning to S331 to continue iterative optimization; S333, construct a fuzzy rule base based on expert experience, construct membership functions of system operation cost, renewable energy consumption rate and load fluctuation rate, and generate a fuzzy rule data set; S334, dynamically adjusting the activation strength of the fuzzy rule data set according to the real-time state of the system, performing fuzzy comprehensive evaluation on each solution in the Pareto solution set data set, and generating an evaluation result data set; S335. Select the solution with the highest comprehensive score from the evaluation result data set as the final scheduling solution, and generate a collaborative scheduling strategy data set, including energy storage charging and discharging power strategy data and charging pile charging power strategy data.

10. The distributed collaborative scheduling method for flexible port resources based on swarm intelligence according to claim 5 is characterized in that: The step S41 comprises: S411, extracting energy storage charging and discharging power strategy data, charging pile charging power strategy data and photovoltaic power control strategy data from the coordinated scheduling strategy data set to generate a device control strategy data set; S412, based on the pre-stored resource distribution, divide the device control strategy data set according to the control area to generate a regional control data set; S413, according to the communication protocol requirements of each edge control unit, convert the regional control data set into a corresponding protocol format to generate a protocol instruction data set; S414, based on the pre-stored communication delay and bandwidth constraints, prioritize the instructions in the protocol instruction data set to generate an instruction priority data set; and issue control instructions to each edge control unit in sequence according to the instruction priority data set; S415: Collect the execution status and device operation parameters of each edge control unit to generate an actual execution data set.

Images