Hub navigation water flow condition optimization method based on intelligent regulation and control

Abstract

The invention discloses a hub navigation water flow condition optimization method based on intelligent regulation and control. The method comprises the steps that multi-source heterogeneous water area monitoring data are collected in real time, time synchronization alignment and correction are carried out, abnormal values are corrected through principal component analysis dimension reduction, Kalman filtering denoising and mean value replacement, and a high-integrity feature set is generated. Secondly, on the basis of the pre-trained hydrodynamic model, parameters are finely adjusted through transfer learning, an online calibration algorithm is combined, and a multi-scale water level and flow velocity prediction result is generated. Then ship and water flow correlation characteristics are extracted through a space-time attention mechanism, parameters such as the navigational speed and the water gate opening degree are optimized in combination with mixed integer programming, and a preliminary regulation and control scheme is generated; and finally, screening a high-score scheme by using reinforcement learning and a genetic algorithm, and adjusting equipment parameters in real time by using a distributed control algorithm after digital twinborn simulation verification. According to the invention, safe, efficient and low-energy-consumption navigation of the ship can be realized in a complex and changeable water area environment, and the intelligent level of water area navigation management is improved.

Description

Technical Field

[0001] The present invention belongs to the field of intelligent water conservancy and transportation technology, and in particular relates to a method for optimizing hub navigation flow conditions based on intelligent regulation. Background Art

[0002] Waterway navigation management is a key area for ensuring the safety and efficiency of water transportation. It involves complex systems such as ship scheduling, water flow regulation, and environmental monitoring, and has a significant impact on the national economy and public safety. Currently, waterway management relies on traditional manual scheduling and single-sensor monitoring, which are unable to cope with complex and changing hydrological and meteorological conditions and ship dynamics. This leads to low navigation efficiency, high energy consumption, and difficulty in real-time control of safety risks. Existing methods have the following main limitations: single data collection methods and lack of multi-source collaboration; insufficient model prediction accuracy and difficulty adapting to multi-scale dynamic changes; insufficient optimization of control schemes and difficulty balancing multi-objective constraints such as efficiency, energy consumption, and safety; slow emergency response and lack of intelligent support for equipment health management. These shortcomings make waterway management unable to meet the needs of modern high-density navigation. Summary of the Invention

[0003] In order to solve the above technical problems, the present invention proposes a method for optimizing hub navigation flow conditions based on intelligent regulation to solve the problems existing in the above-mentioned prior art.

[0004] To achieve the above objectives, the present invention provides a method for optimizing hub navigation flow conditions based on intelligent regulation, comprising:

[0005] Obtain multi-source heterogeneous water navigation detection data and pre-process it to obtain a standard fusion data set;

[0006] constructing a time series model based on the standard fusion data set, and obtaining a complete water environment feature set based on the time series model and the standard fusion data set;

[0007] Based on the pre-trained hydrodynamic model, transfer learning is used to perform multi-scale dynamic prediction of the complete water environment feature set. If the prediction deviation exceeds the threshold, the model parameters are optimized through the gradient descent method to generate the hydrodynamic prediction results;

[0008] Based on the hydrodynamic prediction results, a spatiotemporal correlation feature between the ship trajectory and the water flow is obtained; and based on the spatiotemporal correlation feature between the ship trajectory and the water flow and a feature clustering algorithm, a multidimensional spatiotemporal feature vector is obtained;

[0009] A mixed integer programming algorithm is used to perform multi-objective optimization on the multi-dimensional spatiotemporal feature vector to obtain an initial set of control schemes. The initial set of control schemes is scored, and the optimal control scheme is obtained based on the scores to optimize the navigation flow conditions of the hub.

[0010] Optionally, the process of preprocessing multi-source heterogeneous water navigation detection data includes:

[0011] A time synchronization protocol is used to perform real-time collaborative collection of the multi-source heterogeneous water navigation detection data to obtain a time-aligned multi-source data set; if the timestamp deviation of the data packet in the multi-source data set exceeds a preset threshold, the timestamp is corrected by a linear interpolation method to obtain a time-synchronized data set; the time-synchronized data set is subjected to dimensionality reduction processing to extract the main features to obtain a reduced-dimensionality feature set; the reduced-dimensionality feature set is smoothed by a Kalman filter algorithm to obtain a smoothed feature set; if the outliers in the smoothed feature set exceed a preset threshold, the outliers are corrected by a mean replacement method to obtain a corrected feature set; the corrected feature set is preprocessed by a standardization method to obtain a standardized feature set; if the data distribution in the standardized feature set deviates from the preset threshold, it is adjusted by a mean normalization method to obtain a standard fused data set.

[0012] Optionally, the multi-source heterogeneous water navigation detection data includes data collected through underwater sonar, drone aerial survey and fiber optic sensing.

[0013] Optionally, the process of constructing a time series model based on the standard fusion dataset and obtaining a complete water environment feature set based on the time series model and the standard fusion dataset includes:

[0014] The standard fusion data set is modeled into time series by using sequence analysis methods to extract dynamic change patterns and obtain a time series model. Based on the time series model, the data is segmented using a sliding window method to generate a segmented feature set. If there are missing values in the segmented feature set, the missing values are filled using a linear interpolation method to obtain a complete feature set. The complete feature set is grouped using a density clustering algorithm. Based on the grouping results, the time series characteristics of the water environment are generated to obtain a complete water environment feature set.

[0015] Optionally, the process of generating hydrodynamic prediction results includes:

[0016] Based on the pre-trained hydrodynamic model, transfer learning is used to adjust the parameters of the water environment feature set to obtain an initial dynamic prediction model; if the model output deviation exceeds the preset threshold, the model parameters are adjusted in real time through an online calibration algorithm, and iterative optimization is combined with the gradient descent method until the deviation converges to within the threshold, and the hydrodynamic prediction results are obtained through the trained model.

[0017] Optionally, the process of obtaining a multi-dimensional spatiotemporal feature vector includes:

[0018] A spatiotemporal attention mechanism is adopted to extract the spatiotemporal characteristics of ship dynamics and water flow changes from the hydrodynamic prediction results as a multidimensional initial spatiotemporal feature vector; a feature clustering algorithm is used to group the multidimensional initial spatiotemporal feature vector to obtain a feature cluster set; if the distribution density of the feature cluster set is lower than a preset threshold, an anomaly detection module is used to identify isolated feature clusters and generate an abnormal feature subset; based on the abnormal feature subset, dynamic change data of the corresponding time window is extracted from the hydrodynamic prediction results to generate a dynamic change sequence; a time series analysis method is used to perform trend decomposition on the dynamic change sequence to obtain multi-scale trend features and perform segmentation processing to obtain a segmented trend vector; based on the segmented trend vector, the weight parameters of the spatiotemporal attention mechanism are updated to obtain a final multidimensional spatiotemporal feature vector.

[0019] Optionally, the process of obtaining the optimal control solution includes:

[0020] Score the initial control scheme set and obtain the optimal control scheme based on the score

[0021] The feature weights of each scheme are extracted from the preliminary control scheme set, and the comprehensive score of the scheme is calculated using the feature weighting method to obtain a score sequence; it is determined whether there is a scheme below the preset first threshold in the score sequence. If so, the low-scoring scheme is eliminated to generate a selected scheme set; cluster analysis is used to extract the central features of the scheme cluster, and the dynamic parameters are optimized through time series smoothing to generate the final control scheme; if the score is lower than the preset second threshold, the load distribution strategy is dynamically updated through reinforcement learning, and the feature vector is cross-mutated in combination with the genetic algorithm to obtain the optimal control scheme that meets multiple objective constraints.

[0022] Optionally, if the score is lower than a preset second threshold, the characteristic vector of the final control plan is used as the state space, and the comprehensive score is used as the reward function; a genetic algorithm is used to perform cross-mutation on the characteristic vector to generate an updated set of high-scoring plans; and the optimal control plan that meets the multi-objective constraints is screened through the weighted sum evaluation index.

[0023] Optionally, it also includes using digital twin technology to conduct virtual simulation of the optimal control plan, combining real-time feedback to iteratively adjust parameters, and generate a dynamic control parameter set; based on the dynamic control parameter set, real-time adjustment of ship scheduling and water flow control equipment is carried out through distributed control algorithms until the navigation status meets the safety threshold.

[0024] The present invention also provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0025] Compared with the prior art, the present invention has the following advantages and technical effects:

[0026] The present invention discloses a method for optimizing hub navigation flow conditions based on intelligent regulation. By collecting and integrating multi-source heterogeneous data, a high-integrity water environment feature set is constructed, and transfer learning is performed using a pre-trained hydrodynamic model to achieve dynamic prediction of complex hydrological and meteorological conditions. Based on the hydrodynamic prediction results, the present invention extracts the key spatiotemporal characteristics of ship dynamics and water flow changes, and optimizes the regulation scheme through mixed integer programming and reinforcement learning algorithms. Finally, digital twin technology is used for virtual simulation to obtain dynamic regulation parameters, thereby achieving real-time adjustment of ship scheduling and water flow regulation equipment. The present invention can achieve safe, efficient, and low-energy navigation of ships in complex and changeable water environments, and improve the level of intelligence in water navigation management. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] The accompanying drawings, which constitute part of this application, are intended to provide a further understanding of this application. The exemplary embodiments and descriptions of this application are intended to explain this application and do not constitute an improper limitation on this application. In the accompanying drawings:

[0028] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0029] It should be noted that, in the absence of conflict, the embodiments and features of the embodiments in this application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0030] It should be noted that the steps shown in the flowcharts of the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and that, although a logical order is shown in the flowcharts, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0031] Example 1

[0032] like Figure 1 As shown, this embodiment provides a method for optimizing hub navigation flow conditions based on intelligent regulation, including:

[0033] Obtain multi-source heterogeneous water navigation detection data and pre-process it to obtain a standard fusion data set;

[0034] As a specific implementation method, the process of preprocessing multi-source heterogeneous water navigation detection data includes:

[0035] A time synchronization protocol is used to collect multi-source heterogeneous water navigation detection data in real time to obtain a time-aligned multi-source data set; if the timestamp deviation of the data packet in the multi-source data set exceeds the preset threshold, the timestamp is corrected by the linear interpolation method to obtain a time-synchronized data set; the time-synchronized data set is subjected to dimensionality reduction processing to extract the main features and obtain a reduced-dimensionality feature set; the reduced-dimensionality feature set is smoothed by the Kalman filter algorithm to obtain a smoothed feature set; if the outlier in the smoothed feature set exceeds the preset threshold, the mean replacement method is used to correct the outlier to obtain a corrected feature set; the normalization method is used to preprocess the corrected feature set to obtain a standardized feature set; if the data distribution in the standardized feature set deviates from the preset threshold, it is adjusted by the mean normalization method to obtain a standard fusion data set.

[0036] Specifically, the acquisition of heterogeneous data from multiple sources is fundamental to data fusion, involving technologies such as underwater sonar, drone-based aerial surveys, and fiber optic sensing. For example, underwater sonar can be used to detect seabed topography and generate depth maps, with a sampling rate of 10 times per second; drone-based aerial surveys capture sea surface images with a high-resolution camera at a frame rate of 30 frames per second; and fiber optic sensors monitor underwater vibration signals at a sampling rate of 100 times per second. These data sources come in various formats and require unified processing to enable collaborative analysis.

[0037] The time synchronization protocol uses the Network Time Protocol to ensure consistent timestamps across multiple data sources. For example, the sonar data timestamp is 2025-04-22 10:00:00.123, the drone imagery timestamp is 2025-04-22 10:00:00.150, and the fiber optic sensor timestamp is 2025-04-22 10:00:00.100. If the timestamp deviation exceeds a preset threshold of 50 milliseconds, linear interpolation can be used to correct it. Specifically, if the drone imagery timestamp is ahead, interpolation can be performed using the preceding and following frames to generate an image frame aligned with the sonar data. This correction ensures temporal consistency across the time-synchronized datasets, improving the accuracy of subsequent analysis.

[0038] The principal component analysis algorithm is used for dimensionality reduction and key feature extraction. For example, sonar data contains multidimensional features such as depth and reflection intensity, drone imagery includes information such as color and texture, and fiber optic sensors provide information such as vibration frequency and amplitude. Principal component analysis can compress this high-dimensional data into principal feature vectors, retaining 90% of the information variance. This generates a reduced-dimensionality feature set, reduces computational complexity, and highlights key patterns.

[0039] The Kalman filter algorithm is used to smooth out the reduced-dimensionality feature set and eliminate noise. For example, sonar data may contain random noise due to water disturbances. Kalman filtering generates a smooth depth curve through prediction and update steps. This method effectively improves data stability and ensures the reliability of subsequent analysis.

[0040] If an outlier in a smoothed feature set exceeds a preset threshold, for example, if the vibration amplitude of a fiber optic sensor suddenly increases to three times its normal value, a mean replacement method can be used to correct the outlier using the average of the preceding and following data. This correction prevents the outlier from interfering with subsequent groupings and enhances the robustness of the feature set.

[0041] Normalization preprocesses the fused feature set, eliminating dimensional differences between different data sources and producing a standardized feature set. For example, underwater sonar data may be measured in decibels, drone aerial survey data in meters, and fiber optic sensor data in microstrain, all of which have significant dimensional differences. Using the z-score normalization method, the feature values of each data source are converted to dimensionless values with a mean of 0 and a standard deviation of 1. Specifically, for the sonar data sequence [120, 130, 125] decibels, a mean of 125 and a standard deviation of 5 are calculated, and then normalized to [-1, 1, 0], thus unifying the dimensions and facilitating subsequent analysis.

[0042] If the data distribution in the standardized feature set deviates from the preset threshold, it is adjusted by the mean normalization method. Assuming that the preset threshold is that the absolute value of the data skewness is less than 0.5, if the skewness of the sonar data is 0.7, which exceeds the threshold, mean normalization is used to map the data to the [0,1] interval.

[0043] A time series model is constructed based on the standard fusion dataset, and a complete water environment feature set is obtained based on the time series model and the standard fusion dataset;

[0044] As a specific implementation method, a time series model is constructed based on a standard fusion dataset. The process of obtaining a complete water environment feature set based on the time series model and the standard fusion dataset includes:

[0045] The standard fusion data set is modeled into time series through sequence analysis method, and the dynamic change rules are extracted to obtain the time series model. According to the time series model, the data is segmented by the sliding window method to generate a segmented feature set. If there are missing values in the segmented feature set, the missing values are filled by linear interpolation method to obtain a complete feature set. The complete feature set is grouped by the density clustering algorithm. Based on the grouping results, the time series characteristics of the water environment are generated to obtain the complete water environment feature set.

[0046] Specifically, a time series model is constructed on the standard fusion dataset using sequence analysis methods to extract dynamic patterns of change and generate a time series model. Sequence analysis employs autoregressive models to capture temporal variations in characteristics such as temperature and pressure in aquatic environments. For example, analyzing the trends and periodicity of temperature data collected by fiber optic sensors at [20, 21, 20.5] degrees Celsius and building a model to predict future changes can help understand the dynamics of aquatic environments. Based on the time series model, a sliding window approach is used to segment the data and generate segmented feature sets.

[0047] Specifically, we set the window size to 3 seconds and the step size to 1 second. We extract a set of features every 3 seconds of the sonar data sequence, generating multi-segment feature vectors, such as [120, 125, 130] and [125, 130, 128], to facilitate analysis of local trends. If there are missing values in the segmented feature set, they are filled in using linear interpolation to obtain a complete feature set. If the drone aerial survey data is missing a certain second of altitude due to signal interruption, such as in the sequence [10, missing, 12] meters, we can interpolate the values 10 and 12 to fill the gap to 11 meters to ensure data continuity.

[0048] Based on the complete feature set, a density clustering algorithm is used to group the data and determine the resulting water environment feature groups. Density clustering algorithms, such as the DBSCAN algorithm, classify sonar, aerial survey, and fiber optic data into feature groups for different water regions, such as deep and shallow waters, based on the spatial density of the feature set, facilitating the differentiation of environmental characteristics. Based on these grouping results, time series features of the water environment are generated, resulting in a highly complete water environment feature set. For example, the time series features of deep-water sonar data are combined with fiber optic temperature features to generate a comprehensive feature vector that fully characterizes changes in the water environment, providing a reliable data foundation for subsequent analysis.

[0049] Based on the pre-trained hydrodynamic model, transfer learning is used to perform multi-scale dynamic prediction of the complete water environment feature set. If the prediction deviation exceeds the threshold, the model parameters are optimized through the gradient descent method to generate the hydrodynamic prediction results;

[0050] As a specific implementation method, the process of generating hydrodynamic prediction results includes:

[0051] Based on the pre-trained hydrodynamic model, transfer learning is used to adjust the parameters of the water environment feature set to obtain an initial dynamic prediction model; if the model output deviation exceeds the preset threshold, the model parameters are adjusted in real time through an online calibration algorithm, and iterative optimization is combined with the gradient descent method until the deviation converges to within the threshold, and the hydrodynamic prediction results are obtained through the trained model.

[0052] Specifically, pre-trained hydrodynamic models can be used to process multi-scale data sets representing aquatic environmental characteristics. These models are typically constructed based on extensive historical hydrological data, covering characteristics such as flow velocity, water depth, and temperature. For example, a model for a particular river basin might incorporate observational data from multiple hydrological stations over a period of years, enabling a preliminary understanding of the dynamics of the aquatic environment. By adjusting parameters through transfer learning, the model can be adapted to the characteristics of a specific watershed.

[0053] For example, for a small or medium-sized lake, transfer learning can leverage nearly a year's worth of water level and flow rate data to fine-tune model parameters to better reflect the lake's seasonal fluctuations. Specifically, if the initial prediction model is insufficiently adaptable to hydrological and meteorological conditions, such as when the predicted water level deviates from the actual observed value by more than 10%, the parameters can be optimized using a gradient descent algorithm. Preferably, a learning rate of 0.01 is set, and the model weights are iteratively updated until the prediction deviation converges to within a preset threshold.

[0054] For example, in a reservoir forecasting scenario, the optimized model more accurately reflects the rapid rise in water levels after rainfall. Optimized forecasting models can be used to extract dynamic characteristics, such as periodic fluctuations in water flow velocity. Time series analysis methods can generate trend sequences based on historical data.

[0055] For example, analyzing 30 consecutive days of flow velocity data for a particular river section can generate a trend series reflecting the difference between daytime and nighttime flow velocity. If an outlier appears in the series, such as a sudden increase in flow velocity to twice the normal value on a single day, it can be smoothed using a median filter. Specifically, the outlier is replaced by the median flow velocity of the three preceding and following days to ensure continuity of the trend series.

[0056] Exemplarily, the sliding window method can segment the smooth trend sequence to generate a multi-scale segmented feature set. For example, the window size is set to 7 days and the step size is 1 day, and the flow velocity and water temperature characteristics of each week are extracted to form a feature set that reflects short-term and medium-term changes. This method can capture the dynamic characteristics of the water environment, such as a short surge in flow velocity after rainfall or seasonal water temperature changes. For example, the density clustering algorithm can group the multi-scale segmented feature set to identify different dynamic patterns of the water environment. Assuming that the feature set of a lake contains water temperature and flow velocity data, the clustering result may divide the data into two groups: high temperature and low flow velocity and low temperature and high flow velocity, corresponding to the typical states of summer and winter respectively. The grouping result can further generate dynamic prediction features, such as predicting the flow velocity change trend in the next week.

[0057] The resulting prediction model is able to adapt to multi-scale changes and generate dynamic prediction features of the aquatic environment. For example, the model can predict changes in water level and flow velocity in a specific river section within 48 hours after heavy rainfall, providing a reference for flood control and scheduling. Ideally, the model can also incorporate meteorological forecast data to further improve prediction accuracy. This multi-scale dynamic prediction capability helps to more comprehensively understand the changing patterns of the aquatic environment.

[0058] Furthermore, if the output deviation of the dynamic prediction model exceeds a preset threshold, the deviation value is obtained through the deviation detection module to determine the conditions that trigger calibration. The model parameters are adjusted in real time using an online calibration algorithm to obtain calibrated model parameters. Based on the calibrated model parameters, the dynamic prediction model is updated to generate new hydrodynamic prediction results. If the deviation of the new prediction results still exceeds the preset threshold, the model parameters are iteratively optimized using a gradient descent algorithm to obtain optimized model parameters. Based on the optimized model parameters, the hydrodynamic prediction results are regenerated to determine the stability of the prediction results.

[0059] Based on the hydrodynamic prediction results, the spatiotemporal correlation characteristics of ship trajectories and water currents are obtained. Based on the spatiotemporal correlation characteristics of ship trajectories and water currents and the feature clustering algorithm, a multi-dimensional spatiotemporal feature vector is obtained.

[0060] As a specific implementation method, the process of obtaining a multi-dimensional spatiotemporal feature vector includes:

[0061] A spatiotemporal attention mechanism is adopted to extract the spatiotemporal characteristics of ship dynamics and water flow changes from the hydrodynamic prediction results as the multidimensional initial spatiotemporal feature vector; the multidimensional initial spatiotemporal feature vector is grouped by the feature clustering algorithm to obtain a feature cluster set; if the distribution density of the feature cluster set is lower than the preset threshold, the anomaly detection module is used to identify isolated feature clusters and generate an abnormal feature subset; based on the abnormal feature subset, the dynamic change data of the corresponding time window is extracted from the hydrodynamic prediction results to generate a dynamic change sequence; the time series analysis method is used to perform trend decomposition on the dynamic change sequence to obtain multi-scale trend features and perform segmentation processing to obtain a segmented trend vector; based on the segmented trend vector, the weight parameters of the spatiotemporal attention mechanism are updated to obtain the final multidimensional spatiotemporal feature vector.

[0062] Specifically, in the field of hydrodynamic prediction, the spatiotemporal attention mechanism can efficiently extract key information by analyzing the spatiotemporal characteristics of ship dynamics and water flow changes.

[0063] For example, during a ship's voyage, the speed and direction of currents vary over time and space. The spatiotemporal attention mechanism can identify the current regions that have the greatest impact on the ship's motion. For example, if the current speed in a waterway changes from 2 m / s to 5 m / s over a period of time, the mechanism assigns a higher weight to the region with the most dramatic speed change, generating a multidimensional spatiotemporal feature vector containing speed, direction, and timestamps. This vector accurately describes the interaction between the current and the ship.

[0064] Specifically, the feature clustering algorithm groups multidimensional spatiotemporal feature vectors to generate a set of feature clusters. Ship trajectories in different sections exhibit different patterns due to the influence of water currents, and the clustering algorithm can group similar trajectories.

[0065] For example, the trajectory feature vectors of ships in the rapids region of a waterway are grouped into one cluster, and those in the flat flow region into another. If the density of the feature clusters falls below a threshold, such as if a cluster contains only a small number of vectors, the anomaly detection module identifies it as an isolated feature cluster. For example, an abnormal cluster in the rapids region may correspond to abnormal ship oscillations caused by eddies, generating an abnormal feature subset.

[0066] Anomaly feature subsets are used to extract dynamic change data within corresponding time windows, generating a dynamic change sequence. For example, an anomaly subset might indicate a 15-degree ship heading deviation within a 10-minute window. The sequence records the deviation angle and the change in water velocity. Time series analysis methods further decompose the sequence trend, such as decomposing the deviation into short-term fluctuations and long-term trends, to obtain multi-scale trend features. Short-term fluctuations may reflect current pulses, while long-term trends indicate overall flow direction changes.

[0067] The sliding window method segments multi-scale trend features into segments, generating segmented trend vectors. For example, using a 5-minute window to analyze ship heading trends, a vector is generated for each segment, reflecting the impact of local currents. These segmented trend vectors are used to update the weight parameters of the spatiotemporal attention mechanism.

[0068] For example, if a segment of the vector shows a significant change in water velocity, the mechanism increases the attention weight for that segment, generating an optimized feature vector. This optimized vector more accurately represents the dynamic relationship between the water flow and the ship, providing a reliable basis for subsequent predictions.

[0069] The optimized eigenvectors can be used to improve ship navigation path planning, avoid eddy currents, and enhance navigation stability.

[0070] The multi-dimensional spatiotemporal feature vectors are optimized with multiple objectives through a mixed integer programming algorithm to obtain an initial set of control schemes. The initial set of control schemes is scored, and the optimal control scheme is obtained based on the scores to optimize the navigation flow conditions of the hub.

[0071] As a specific implementation method, the process of obtaining the optimal control solution includes:

[0072] Score the initial control scheme set and obtain the optimal control scheme based on the score

[0073] The feature weights of each scheme are extracted from the preliminary control scheme set, and the comprehensive score of the scheme is calculated using the feature weighting method to obtain a score sequence; it is determined whether there is a scheme below the preset first threshold in the score sequence. If so, the low-scoring scheme is eliminated to generate a selected scheme set; cluster analysis is used to extract the central features of the scheme cluster, and the dynamic parameters are optimized through time series smoothing to generate the final control scheme; if the score is lower than the preset second threshold, the load distribution strategy is dynamically updated through reinforcement learning, and the feature vector is cross-mutated in combination with the genetic algorithm to obtain the optimal control scheme that meets multiple objective constraints.

[0074] Furthermore, if the score is lower than the preset second threshold, the characteristic vector of the final control plan is used as the state space, and the comprehensive score is used as the reward function; a genetic algorithm is used to perform cross-mutation on the characteristic vector to generate an updated set of high-scoring plans; and the optimal control plan that meets multiple objective constraints is screened through weighted sum evaluation indicators.

[0075] Specifically, a mixed integer programming algorithm is used to optimize multidimensional spatiotemporal feature vectors, incorporating efficiency, energy consumption, and safety constraints. For example, in a ship navigation control scenario, assuming that a ship needs to navigate complex waters, the goal is to optimize the route to reduce energy consumption and ensure safety. The algorithm generates a set of control plans by defining variables (such as speed and heading) and constraints (such as a maximum energy consumption of 1000 kWh and a minimum safety distance of 500 meters). Each plan contains parameters such as a speed of 15 knots and a heading adjustment of 10 degrees to meet the constraints. This method ensures the feasibility of the plan through global optimization.

[0076] When extracting feature weights from the initial set of control options, a weighted approach can be used to calculate a comprehensive score. For example, consider setting the efficiency weight to 0.4, the energy consumption weight to 0.3, and the safety weight to 0.3. Assuming Option A has an efficiency score of 80, an energy consumption score of 70, and a safety score of 90, the overall score is 80 × 0.4 + 70 × 0.3 + 90 × 0.3 = 80. The scoring sequence reflects the merits of the options, and options below a threshold of 75 are eliminated. This scoring mechanism clearly quantifies the overall performance of the options, facilitating subsequent screening.

[0077] After the filtering module eliminates low-scoring solutions, a set of selected solutions is generated. For example, if there are 10 solutions initially, three with scores below 75 are eliminated, leaving the remaining seven in the selected set. The selected set of solutions is grouped using cluster analysis, hypothetically into three clusters: high-speed, low-energy consumption, safety-first, and balanced. Clustering is based on feature similarity, such as the Euclidean distance of parameters like speed and energy consumption. This grouping method facilitates identification of commonalities and differences among solutions.

[0078] Specifically, the central features are extracted from the set of solution clusters, and the solution cluster that is most similar to the space-time vector is determined through feature matching. For example, the central features of the preferred solution cluster include a speed of 12 knots and an energy consumption of 800 kWh, which has the highest matching degree. The matching process is based on the time window (such as 10 minutes) and spatial range (such as 1 km) of the space-time vector. This method ensures that the solution is highly correlated with the actual water flow and ship dynamics. For example, for the preferred solution cluster, the time series smoothing method processes the dynamic features and generates a smoothed feature sequence. Suppose the speed sequence of a certain solution is [12, 13, 11, 14] knots, and after smoothing it is [12.2, 12.5, 12.3, 12.8] knots. Smoothing reduces noise interference, makes the features more stable, and facilitates subsequent analysis. This processing method is particularly important in dynamic waters.

[0079] The control plan's weight parameters are updated based on the smoothed feature sequence to generate the final control plan. For example, if the smoothed feature sequence shows a stable speed of around 12 knots, the weight parameters are adjusted to favor efficiency, resulting in a final plan: 12 knots and a 5-degree optimized course. This plan balances efficiency and safety in complex waters. This dynamic adjustment mechanism ensures that control more closely matches actual needs.

[0080] Specifically, if the comprehensive score is lower than the preset second threshold, the multidimensional feature vector of each control scheme is obtained from the scheme set through the feature extraction module to obtain a feature vector set. According to the feature vector set, the feature vector is iteratively optimized using the reinforcement learning algorithm, the scheme parameters are adjusted, and an optimized feature vector set is generated. If the comprehensive score of the optimized feature vector set is still lower than the preset threshold, the optimized feature vector set is cross-mutated using the genetic algorithm to obtain an updated feature vector set. According to the updated feature vector set, the efficiency, energy consumption, and safety indicators corresponding to each feature vector are calculated to generate an indicator evaluation set. Through the indicator evaluation set, the efficiency, energy consumption, and safety indicators are comprehensively scored using the weighted summation method to obtain an updated score sequence. If there is a score below the preset threshold in the updated score sequence, the corresponding feature vector is eliminated to generate a selected feature vector set. Based on the selected feature vector set, an optimized control scheme that meets the constraints is generated.

[0081] A method for optimizing hub navigation flow conditions based on intelligent control also includes virtually simulating the optimal control scheme through digital twin technology, iteratively adjusting parameters in combination with real-time feedback, and generating a dynamic control parameter set; based on the dynamic control parameter set, real-time adjustment of ship scheduling and water flow control equipment is performed through a distributed control algorithm until the navigation status meets the safety threshold.

[0082] Specifically, a water environment model is constructed through digital twin technology, a virtual simulation scene is obtained, and initial navigation status data is obtained. Based on the initial navigation status data, a data acquisition module is used to obtain real-time dynamic information in the water environment to obtain an environmental status data set. If the navigation status in the environmental status data set deviates from the preset threshold, a virtual test is run through the simulation scene to generate real-time feedback data. Based on the real-time feedback data, a dynamic control algorithm is used to adjust the parameters in the control strategy to obtain an optimized parameter set. The virtual simulation scene is updated through the optimized parameter set, and the updated navigation status data is obtained to generate a control effect data set. If the navigation status in the control effect data set still does not meet the preset threshold, the feedback mechanism is used to iteratively optimize the control strategy to obtain a selected control parameter set. Based on the selected control parameter set, a dynamic control scheme that meets the water navigation constraints is generated.

[0083] Furthermore, water area operating environment data is collected through environmental sensors to generate a real-time environmental data set. If the flow rate or flow in the real-time environmental data set deviates from the preset threshold, a distributed control algorithm is used to adjust the water flow equipment parameters to generate an optimized equipment parameter set. The water flow equipment operating status is updated according to the optimized equipment parameter set, and the adjusted water area flow state data is obtained. Real-time ship position and speed data are obtained through the ship positioning system to generate a ship operation data set. A collaborative adjustment mechanism is adopted to combine the water area flow state data and the ship operation data set to generate a ship scheduling optimization plan. If the navigation status in the ship scheduling optimization plan does not meet the safety status threshold, feedback data is generated through the status monitoring module, and the control parameters are iteratively updated. Based on the iteratively updated control parameters, a distributed control algorithm is used to generate the final ship scheduling and water flow equipment control plan to obtain a safe and efficient navigation state.

[0084] This embodiment further provides a computer device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above method.

[0085] The above are merely preferred embodiments of the present application, but the scope of protection of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by a person skilled in the art within the technical scope disclosed in this application should be included in the scope of protection of the present application. Therefore, the scope of protection of the present application should be based on the scope of protection of the claims.

Claims

1. A method for optimizing hub navigation flow conditions based on intelligent regulation, characterized in that: The following steps are involved: Obtain multi-source heterogeneous water navigation detection data and pre-process it to obtain a standard fusion data set; constructing a time series model based on the standard fusion data set, and obtaining a complete water environment feature set based on the time series model and the standard fusion data set; Based on the pre-trained hydrodynamic model, transfer learning is used to perform multi-scale dynamic prediction of the complete water environment feature set. If the prediction deviation exceeds the threshold, the model parameters are optimized through the gradient descent method to generate the hydrodynamic prediction results; Based on the hydrodynamic prediction results, a spatiotemporal correlation feature between the ship trajectory and the water flow is obtained; and based on the spatiotemporal correlation feature between the ship trajectory and the water flow and a feature clustering algorithm, a multidimensional spatiotemporal feature vector is obtained; A mixed integer programming algorithm is used to perform multi-objective optimization on the multi-dimensional spatiotemporal feature vector to obtain an initial set of control schemes. The initial set of control schemes is scored, and the optimal control scheme is obtained based on the scores to optimize the navigation flow conditions of the hub.

2. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: The process of preprocessing multi-source heterogeneous water navigation detection data includes: A time synchronization protocol is used to perform real-time collaborative collection of the multi-source heterogeneous water navigation detection data to obtain a time-aligned multi-source data set; if the timestamp deviation of the data packet in the multi-source data set exceeds a preset threshold, the timestamp is corrected by a linear interpolation method to obtain a time-synchronized data set; the time-synchronized data set is subjected to dimensionality reduction processing to extract the main features to obtain a reduced-dimensionality feature set; the reduced-dimensionality feature set is smoothed by a Kalman filter algorithm to obtain a smoothed feature set; if the outliers in the smoothed feature set exceed a preset threshold, the outliers are corrected by a mean replacement method to obtain a corrected feature set; the corrected feature set is preprocessed by a standardization method to obtain a standardized feature set; if the data distribution in the standardized feature set deviates from the preset threshold, it is adjusted by a mean normalization method to obtain a standard fused data set.

3. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: The multi-source heterogeneous water navigation detection data includes data collected through underwater sonar, drone aerial survey and fiber optic sensing.

4. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: The process of constructing a time series model based on the standard fusion data set and obtaining a complete water environment feature set based on the time series model and the standard fusion data set includes: The standard fusion data set is modeled into time series by using sequence analysis methods to extract dynamic change patterns and obtain a time series model. Based on the time series model, the data is segmented using a sliding window method to generate a segmented feature set. If there are missing values in the segmented feature set, the missing values are filled using a linear interpolation method to obtain a complete feature set. The complete feature set is grouped using a density clustering algorithm. Based on the grouping results, the time series characteristics of the water environment are generated to obtain a complete water environment feature set.

5. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: The process of generating hydrodynamic predictions involves: Based on the pre-trained hydrodynamic model, transfer learning is used to adjust the parameters of the water environment feature set to obtain an initial dynamic prediction model; if the model output deviation exceeds the preset threshold, the model parameters are adjusted in real time through an online calibration algorithm, and iterative optimization is combined with the gradient descent method until the deviation converges to within the threshold, and the hydrodynamic prediction results are obtained through the trained model.

6. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: The process of obtaining a multidimensional space-time feature vector includes: A spatiotemporal attention mechanism is adopted to extract the spatiotemporal characteristics of ship dynamics and water flow changes from the hydrodynamic prediction results as a multidimensional initial spatiotemporal feature vector; a feature clustering algorithm is used to group the multidimensional initial spatiotemporal feature vector to obtain a feature cluster set; if the distribution density of the feature cluster set is lower than a preset threshold, an anomaly detection module is used to identify isolated feature clusters and generate an abnormal feature subset; based on the abnormal feature subset, dynamic change data of the corresponding time window is extracted from the hydrodynamic prediction results to generate a dynamic change sequence; a time series analysis method is used to perform trend decomposition on the dynamic change sequence to obtain multi-scale trend features and perform segmentation processing to obtain a segmented trend vector; based on the segmented trend vector, the weight parameters of the spatiotemporal attention mechanism are updated to obtain a final multidimensional spatiotemporal feature vector.

7. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: The process of obtaining the optimal control solution includes: Score the initial control scheme set and obtain the optimal control scheme based on the score The feature weights of each scheme are extracted from the preliminary control scheme set, and the comprehensive score of the scheme is calculated using the feature weighting method to obtain a score sequence; it is determined whether there is a scheme below the preset first threshold in the score sequence. If so, the low-scoring scheme is eliminated to generate a selected scheme set; cluster analysis is used to extract the central features of the scheme cluster, and the dynamic parameters are optimized through time series smoothing to generate the final control scheme; if the score is lower than the preset second threshold, the load distribution strategy is dynamically updated through reinforcement learning, and the feature vector is cross-mutated in combination with the genetic algorithm to obtain the optimal control scheme that meets multiple objective constraints.

8. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: If the score is lower than the preset second threshold, the characteristic vector of the final control plan is used as the state space, and the comprehensive score is used as the reward function; a genetic algorithm is used to perform cross-mutation on the characteristic vector to generate an updated set of high-scoring plans; and the optimal control plan that meets multiple objective constraints is screened through weighted sum evaluation indicators.

9. The method for optimizing hub navigation flow conditions based on intelligent regulation according to claim 1 is characterized in that: It also includes virtual simulation of the optimal control scheme through digital twin technology, iteratively adjusting parameters based on real-time feedback, and generating a dynamic control parameter set; Based on the dynamic control parameter set, the ship scheduling and water flow control equipment are adjusted in real time through the distributed control algorithm until the navigation status meets the safety threshold.

10. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method according to any one of claims 1 to 9.

Images