Marine work platform resource scheduling system and method based on deep learning

Abstract

The invention discloses a maritime work platform resource scheduling system and method based on deep learning, and relates to the technical field of operation resource scheduling of a floating wind power platform, and the method comprises the steps: collecting a displacement track and an operation state signal in the operation process of a cargo boom; space mapping is carried out in the operation radius range of the cargo boom, and a three-dimensional rasterized space occupation graph is constructed; constructing a construction task atlas; the method comprises the following steps: performing dynamic overlapping prediction on a displacement track in a future cargo boom operation process by using a three-dimensional convolutional network, outputting a predicted displacement track, adding a platform disturbance component of a floating wind power platform, performing disturbance modeling, calculating a conflict probability of the predicted displacement track in a path conflict area list based on the disturbance modeling, and outputting the predicted displacement track. When the conflict probability is larger than a preset conflict threshold value, real-time early warning is conducted through an edge computing node, and cargo boom operation is dispatched. According to the method, the safety and the intelligent level of collaborative operation of the multiple cargo booms are effectively improved, and the method has wide engineering application prospects.

Description

Technical Field

[0001] This invention relates to the field of operational resource scheduling technology for floating wind power platforms, specifically a resource scheduling system and method for offshore engineering platforms based on deep learning. Background Technology

[0002] During the installation and maintenance of floating wind power platforms, the lifting operation of the boom is a critical construction task, which usually involves the hoisting and transportation of large wind turbine components. Since floating platforms are often located in complex marine environments, their attitude is significantly affected by natural disturbances such as wind, waves, and currents, which greatly interferes with the boom's operating trajectory and load stability.

[0003] Currently, traditional methods for planning and scheduling crane boom operation paths mainly rely on preset rules, human experience, or two-dimensional drawings for decision-making assistance. These methods are difficult to respond to changes in platform dynamic disturbances in real time and cannot effectively identify potential path conflicts between multiple crane booms in a limited space.

[0004] In recent years, with the development of deep learning technology, three-dimensional convolutional networks (3D-CNN) have been widely used in video analysis, medical imaging, and spatial event prediction, demonstrating significant advantages in extracting spatiotemporal features. Therefore, a crane boom resource scheduling method is needed that combines the operational scenarios of floating wind power platforms, introduces a platform disturbance modeling mechanism, and integrates deep learning path prediction capabilities to achieve early detection and dynamic avoidance of operational path conflicts, thereby improving the safety and efficiency of offshore construction. Summary of the Invention

[0005] The purpose of this invention is to provide a resource scheduling system and method for offshore engineering platforms based on deep learning, so as to solve the problems raised in the prior art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a resource scheduling method for offshore engineering platforms based on deep learning, comprising: Edge computing nodes and multimodal sensors are deployed in the boom working area of ​​the floating wind power platform to collect displacement trajectory and working status signals during boom operation; the collected displacement trajectory and working status signals are integrated and spatially mapped within the working radius of the boom to construct a three-dimensional rasterized spatial occupancy map. A construction task map is constructed using a three-dimensional rasterized spatial occupancy map; Based on the constructed construction task map, a 3D convolutional network is used to dynamically overlap and predict the displacement trajectory of the crane boom during future operations, outputting the predicted displacement trajectory. Based on the prediction results, the DBSCAN algorithm is used to divide the path conflict areas and display the list of path conflict areas. It also includes adding the platform disturbance component of the floating wind power platform, performing disturbance modeling, calculating the conflict probability of the predicted displacement trajectory based on the disturbance modeling path conflict area list, and issuing a real-time warning through the edge computing node when the conflict probability is greater than the preset conflict threshold, and scheduling the crane boom operation.

[0007] Furthermore, the multimodal sensor group is used to collect monitoring data of the working area of ​​the floating wind power platform's boom, and the monitoring data includes real-time working images and working status signals; The real-time operation images are captured by the set video acquisition device, forming a real-time video image sequence of the acquisition operation area; The operation status signal is obtained by the control system of the boom, including the lifting load of the boom; The edge computing nodes are used to preprocess real-time video image sequences in the acquisition area; and provide data fusion analysis and localized decision-making functions.

[0008] Further, in step S201, the real-time video image sequence of the acquired work area is preprocessed and spatiotemporally aligned with the working radius range of the crane boom. The preprocessing includes data cleaning and standardization; Frame extraction, dehazing, noise reduction, and distortion correction are performed on the visual data; the displacement trajectory of the crane boom is obtained from the real-time video image sequence of the acquired work area; The displacement trajectory is unified in coordinate system and transformed to the three-dimensional coordinate system of the floating wind power platform. The spatiotemporal alignment includes spatiotemporal synchronization correction; timestamping the collected data based on the local clock of the edge computing node; and transforming the pixel coordinates in the video image to the three-dimensional coordinate system of the floating wind power platform through a spatial mapping matrix. Step S202: Discretize the three-dimensional coordinate system space of the floating wind power platform into quantifiable spatial units, record the occupancy status of each unit in real time, and form a three-dimensional rasterized spatial occupancy map. Step S203: Based on the three-dimensional rasterized spatial occupancy map, construct the construction task map G=(V,E); where the node V in the construction task map represents the occupied grid; and the directed edge E in the construction task map represents the displacement trajectory of the crane boom. Step S204: The attributes of the node include the set S of the required grid cells and the historical duration of occupation {t1,t2,…,tn}; the attributes of the directed edge E include the planned start and end times of the job [ts,te].

[0009] Furthermore, the three-dimensional coordinate system of the floating wind power platform is established with the center of the bottom platform of the crane boom as the origin, and the x, y and z axes correspond to the horizontal, longitudinal and vertical heights, respectively.

[0010] Furthermore, based on the operating radius of the floating wind turbine platform's boom, the operating area is divided into a three-dimensional grid, with each grid assigned a unique spatial number; dynamically occupied grids are marked according to the boom's displacement trajectory; and a time dimension attribute is added to each dynamically occupied grid, recording the most recent occupation time and occupation duration.

[0011] Further, in step S301, a spatiotemporal feature cube is constructed as the input to a three-dimensional convolutional network; Step S302: Dynamically overlap and predict the displacement trajectory of the crane boom during future operation based on a three-dimensional convolutional network; Step S303: Perform threshold filtering on the output trajectory overlap probability P, mark the grids with P greater than or equal to the preset threshold as potential overlapping grids, output a list of potential overlapping grids including location, time and crane boom, traverse the list of potential overlapping grids by DBSCAN algorithm, and form continuous spatial regions by clustering the potential overlapping grids, which are marked as path conflict regions. Step S304: Visualize the list of path conflict areas.

[0012] Optionally, in step S3011, based on the occupancy grid set S of each node in the construction task map of step S203, the continuous path coordinate sequence of the work area is extracted to generate the displacement trajectory dataset of the crane boom; and discrete displacement vectors are obtained through the displacement trajectory dataset. Step S3012: Calculate the velocity and acceleration characteristics of the displacement trajectory dataset at each time step, and mark the current execution stage of the job in conjunction with the start and end times of the job plan [ts, te]. Step S3013: Encode the attributes of each occupied grid in the construction task map, and convert the occupancy status and historical occupancy duration {t1,t2,…,tn} into spatial feature vectors; using the three-dimensional grid of the floating wind power platform as the spatial dimension and the discrete time step as the time dimension, fuse the discrete displacement vector with the spatial feature vector to construct a multi-channel spatiotemporal feature cube, where each cell represents the comprehensive operation status of a certain occupied grid at a certain time step.

[0013] Optionally, in step S3021, a three-dimensional convolutional network structure is designed, including an input layer, several 3D convolutional layers, a 3D pooling layer, a fully connected layer, and an output layer; the input layer receives a spatiotemporal feature cube with dimensions (X, Y, Z, T, C), where X, Y, and Z are spatial dimensions, T is the number of time steps, and C is the number of channels; the spatial dimensions are obtained based on the three-dimensional coordinate system of the floating wind power platform. Step S3022: The convolutional layer uses a three-dimensional convolutional kernel to extract the spatial neighborhood features and time series dynamic change patterns of the crane boom displacement trajectory, and the pooling layer reduces the feature dimension through average pooling; wherein, the time series dynamic change pattern is obtained through the velocity features and acceleration features at each time step; Step S3023: The fully connected layer aggregates the extracted local features and generates a predicted displacement trajectory through the output layer. The output layer uses the Sigmoid activation function to output a value P∈[0,1] for each grid-time point in the predicted displacement trajectory, which represents the probability that the grid is simultaneously occupied by the crane boom displacement trajectory at the corresponding time step, denoted as the trajectory overlap probability.

[0014] Furthermore, in the process of using a three-dimensional convolutional network to dynamically overlay and predict the displacement trajectory of the crane boom during future operations, a platform disturbance component of the floating wind power platform is added; the platform disturbance component includes the platform's pitch angle, roll angle, yaw angle, and wind speed data. The pitch angle θ, roll angle Φ and yaw angle ψ of the floating wind power platform are obtained by attitude sensors; an anemometer is integrated at the upper end of the boom to obtain wind speed data W(t); Perform disturbance modeling: Δr(t) = Δr att (t)+Δr wind (t); where Δr(t) represents the disturbance displacement value; The offset Δr of the lifting point is calculated using a first-order approximation method. att (t): Δr att (t) = L * (sinθ + sinΦ + sinψ); where L represents the length of the crane boom; t represents the time marker; The disturbance displacement Δr caused by wind speed is calculated using the simple harmonic vibration process. wind (t): Δr wind (t)=β w *W 2 (t); where β w Indicates the wind speed response coefficient; Based on the path conflict region list from step S304, calculate the expected distance d in the direction of maximum disturbance for the predicted displacement trajectories r1(t) and r2(t) within the path conflict regions. eff (t): d eff (t)=||r1(t)-r2(t)||-||Δr1(t)||-||Δr2(t)||; where Δr1(t) represents the disturbance displacement value of the predicted displacement trajectory r1(t); Δr2(t) represents the disturbance displacement value of the predicted displacement trajectory r2(t); The conflict probability f(t) is calculated based on perturbation modeling: f(t) = 1 / [1 + exp((d eff [(t)-ds) / γ)]; where exp represents an exponential function with the natural constant as the base; ds represents the preset safety distance; and γ represents the collision bandwidth parameter; The conflict probability f(t) is judged, and when f(t) is greater than the preset conflict threshold, a real-time warning is issued through the edge computing node.

[0015] A deep learning-based resource scheduling system for offshore engineering platforms includes a data acquisition module, a data processing module, a construction task map construction module, a dynamic overlap prediction module, a disturbance modeling module, and a scheduling module. The data acquisition module is used to collect the displacement trajectory, operation status signal and platform disturbance component during the operation of the crane boom; The data processing module is used for preprocessing and spatiotemporal alignment. The construction task map construction module is used to integrate the collected displacement trajectory and operation status signals, perform spatial mapping within the operation radius of the crane boom, construct a three-dimensional rasterized spatial occupancy map, and construct a construction task map through the three-dimensional rasterized spatial occupancy map. The dynamic overlap prediction module is used to dynamically predict the displacement trajectory of the crane boom during future operations based on the constructed construction task map and a three-dimensional convolutional network, marking path conflict areas and displaying a list of path conflict areas.

[0016] The disturbance modeling module is used to perform disturbance modeling based on the platform disturbance components of the floating wind power platform; The scheduling module is used to calculate the conflict probability of the predicted displacement trajectory based on the path conflict area list based on disturbance modeling. When the conflict probability is greater than the preset conflict threshold, a real-time warning is issued through the edge computing node, and the crane boom operation is scheduled.

[0017] Compared with existing technologies, the beneficial effects of this invention are as follows: This method deploys edge computing nodes and multimodal sensors in the operating area of ​​the floating wind turbine platform's boom to achieve real-time acquisition and intelligent processing of boom displacement trajectory and operating status signals. The acquired data is then mapped to a unified three-dimensional coordinate system to construct a high-precision three-dimensional rasterized spatial occupancy map, effectively depicting the dynamic occupancy status of the operating area. Furthermore, by combining the constructed construction task map with a three-dimensional convolutional network model, spatiotemporal overlap prediction of future operating paths is performed, and path conflict regions are divided using clustering algorithms, improving the accuracy and interpretability of path conflict identification.

[0018] The significant innovation of this invention lies in introducing a disturbance modeling mechanism for floating wind power platforms. Key disturbance parameters such as pitch angle, roll angle, yaw angle, and wind speed are transformed into displacement disturbances, which are then embedded into a path conflict probability calculation model to establish a dynamic mapping relationship between disturbance factors and path overlap risk. By constructing a conflict probability function based on disturbance correction, the risk level of different operational trajectories at future moments can be quantified, enabling dynamic scheduling optimization under disturbance perception conditions.

[0019] Furthermore, this invention provides localized decision-making and response capabilities through edge computing nodes, enabling abnormal early warning and risk area visualization during operations. This reduces the lag in response to complex environmental changes in traditional manual scheduling, enhances the intelligence, adaptability, and safety of marine platform resource scheduling, and is particularly suitable for large-scale hoisting operations in complex sea conditions such as floating wind power. It has broad engineering application prospects and promotional value. Attached Figure Description

[0020] Figure 1 This is a schematic diagram illustrating the steps of a deep learning-based resource scheduling method for offshore engineering platforms according to the present invention. Detailed Implementation

[0021] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] Example: Figure 1 As shown, this invention provides a technical solution: a resource scheduling method for offshore engineering platforms based on deep learning, comprising: Edge computing nodes and multimodal sensors are deployed in the boom working area of ​​the floating wind power platform to collect displacement trajectory and working status signals during boom operation; the collected displacement trajectory and working status signals are integrated and spatially mapped within the working radius of the boom to construct a three-dimensional rasterized spatial occupancy map. Furthermore, the multimodal sensor group is used to collect monitoring data of the working area of ​​the floating wind power platform's boom, and the monitoring data includes real-time working images and working status signals; The real-time operation images are captured by the set video acquisition device, forming a real-time video image sequence of the acquisition operation area; The operation status signal is obtained by the control system of the boom, including the lifting load of the boom; The edge computing nodes are used to preprocess real-time video image sequences in the acquisition area; and provide data fusion analysis and localized decision-making functions.

[0023] A construction task map is constructed using a three-dimensional rasterized spatial occupancy map; Further, in step S201, the real-time video image sequence of the acquired work area is preprocessed and spatiotemporally aligned with the working radius range of the crane boom. The preprocessing includes data cleaning and standardization; Frame extraction, dehazing and noise reduction (the high humidity environment in the ocean causes image blurring, so the dark channel prior algorithm is used to process it) and distortion correction (distortion caused by vibration of industrial cameras is corrected by the camera intrinsic parameter matrix). The displacement trajectory of the crane boom is obtained by collecting real-time video image sequences of the work area. The displacement trajectory is unified in coordinate system and transformed to the three-dimensional coordinate system of the floating wind power platform. The spatiotemporal alignment includes spatiotemporal synchronization correction; timestamping the collected data based on the local clock of the edge computing node; and transforming the pixel coordinates in the video image to the three-dimensional coordinate system of the floating wind power platform through a spatial mapping matrix. Step S202: Discretize the three-dimensional coordinate system space of the floating wind power platform into quantifiable spatial units, record the occupancy status of each unit in real time, and form a three-dimensional rasterized spatial occupancy map. Step S203: Based on the three-dimensional rasterized spatial occupancy map, construct the construction task map G=(V,E); where the node V in the construction task map represents the occupied grid; and the directed edge E in the construction task map represents the displacement trajectory of the crane boom. Step S204: The attributes of the node include the set S of the required grid cells and the historical duration of occupation {t1,t2,…,tn}; the attributes of the directed edge E include the planned start and end times of the job [ts,te].

[0024] Furthermore, the three-dimensional coordinate system of the floating wind power platform is established with the center of the bottom platform of the crane boom as the origin, and the x, y and z axes correspond to the horizontal, longitudinal and vertical heights, respectively.

[0025] Furthermore, based on the operating radius of the floating wind turbine platform's boom, the operating area is divided into a three-dimensional grid, with each grid assigned a unique spatial number; dynamically occupied grids are marked according to the boom's displacement trajectory; and a time dimension attribute is added to each dynamically occupied grid, recording the most recent occupation time and occupation duration.

[0026] Based on the constructed construction task map, a 3D convolutional network is used to dynamically overlap and predict the displacement trajectory of the crane boom during future operations, outputting the predicted displacement trajectory. Based on the prediction results, the DBSCAN algorithm is used to divide the path conflict areas and display the list of path conflict areas. Further, in step S301, a spatiotemporal feature cube is constructed as the input to a three-dimensional convolutional network; Optionally, in step S3011, based on the occupancy grid set S of each node in the construction task map of step S203, the continuous path coordinate sequence of the work area is extracted to generate the displacement trajectory dataset of the crane boom; and discrete displacement vectors are obtained through the displacement trajectory dataset. In this embodiment, a path coordinate sequence is generated: based on the recorded spatial location point set S={(x1,y1,z1),(x2,y2,z2),...,(x n ,y n ,z n )}, arrange them in chronological order to generate continuous operation paths; obtain discrete displacement vectors from the displacement trajectory dataset, such as: [(x2,y2,z2)-(x1,y1,z1)]; Step S3012: Calculate the velocity and acceleration characteristics of the displacement trajectory dataset at each time step, and mark the current execution stage of the job in conjunction with the start and end times of the job plan [ts, te]. Calculating velocity and acceleration characteristics based on coordinates is an existing technique and will not be elaborated here. It should be noted that this part is used to capture the dynamic characteristics of the displacement trajectory of the task and to provide time-series patterns for subsequent prediction. Step S3013: Encode the attributes of each occupied grid in the construction task map, and convert the occupancy status and historical occupancy duration {t1,t2,…,tn} into spatial feature vectors; using the three-dimensional grid of the floating wind power platform as the spatial dimension and the discrete time step as the time dimension, fuse the discrete displacement vector with the spatial feature vector to construct a multi-channel spatiotemporal feature cube, where each cell represents the comprehensive operation status of a certain occupied grid at a certain time step.

[0027] It should be noted that fusing discrete displacement vectors with spatial feature vectors means that at each time step t, the discrete displacement vector at the corresponding time is connected with the spatial features of the currently involved raster in the same structure. In this embodiment, the fused vector represents the overall operation status, including dimensions such as discrete displacement vector, velocity, and acceleration; Step S302: Dynamically overlap and predict the displacement trajectory of the crane boom during future operation based on a three-dimensional convolutional network; Optionally, in step S3021, a three-dimensional convolutional network structure is designed, including an input layer, several 3D convolutional layers, a 3D pooling layer, a fully connected layer, and an output layer; the input layer receives a spatiotemporal feature cube with dimensions (X, Y, Z, T, C), where X, Y, and Z are spatial dimensions, T is the number of time steps, and C is the number of channels; the spatial dimensions are obtained based on the three-dimensional coordinate system of the floating wind power platform. Step S3022: The convolutional layer uses a three-dimensional convolutional kernel to extract the spatial neighborhood features and time series dynamic change patterns of the crane boom displacement trajectory, and the pooling layer reduces the feature dimension through average pooling; wherein, the time series dynamic change pattern is obtained through the velocity features and acceleration features at each time step; Step S3023: The fully connected layer aggregates the extracted local features and generates a predicted displacement trajectory through the output layer. The output layer uses the Sigmoid activation function to output a value P∈[0,1] for each grid-time point in the predicted displacement trajectory, which represents the probability that the grid is simultaneously occupied by the crane boom displacement trajectory at the corresponding time step, denoted as the trajectory overlap probability.

[0028] In this embodiment, the final output of the fully connected layer is obtained through σ(P)=1 / (1+e -P Mapped to the 0~1 interval; it should be noted that the trajectory overlap probability P∈[0,1] is automatically learned by the 3D convolutional neural network based on the input spatiotemporal feature cube, which is a probabilistic space-time state estimation. The data input to the 3D CNN is a five-dimensional tensor of the form (X,Y,Z,T,C), called the spatiotemporal feature cube. Each "cell" of this tensor represents a multidimensional description of the job state at a certain time and spatial location. The 3D CNN uses multiple three-dimensional convolutional kernels (sliding simultaneously in space and time) to extract features from this tensor: Convolutional kernels can identify patterns of continuous trajectories in space; convolutions in the temporal dimension can capture dynamic changes (such as whether they are close to each other or in the same task stage); the deep representation of the network gradually forms a basis for judging the possibility of "simultaneous occupation of trajectories"; The output layer of the network uses the Sigmoid activation function; if highly overlapping features appear in the input cube (such as two trajectories in the same position or with similar velocity directions at the same time step), the neural network will generate a large P after learning, causing P→1; If the trajectories are spaced far apart or in time, or if they still do not intersect after the disturbance, then P→0; Therefore, the P-value is essentially a probability estimate learned by the neural network, representing the probability that "the current grid and the current time are occupied by multiple job trajectories at the same time".

[0029] The preset threshold is 0.6. When the trajectory overlap probability P > 0.6, the grid is considered to be a potentially overlapping grid. Step S303: Perform threshold filtering on the output trajectory overlap probability P, mark the grids with P greater than or equal to the preset threshold as potential overlapping grids, output a list of potential overlapping grids including location, time and crane boom, traverse the list of potential overlapping grids by DBSCAN algorithm, and form continuous spatial regions by clustering the potential overlapping grids, which are marked as path conflict regions. Step S304: Visualize the list of path conflict areas.

[0030] In this embodiment, the network learning objective is defined as "trajectory conflict marker prediction". For the training samples: Label y=1: ​​indicates that the grid is simultaneously covered by multiple task trajectories at time step t; Label y=0: indicates no conflict or only one trajectory is occupied; The cross-entropy loss function is used during training: Where N is the number of training samples, P i Let be the predicted overlap probability of the i-th grid point; It should be noted that the trajectory overlap probability P is a probability value generated by the output layer of a 3D convolutional neural network through a Sigmoid function after layer-by-layer feature extraction and aggregation of the constructed spatiotemporal feature cube. This probability value reflects the likelihood that a spatial grid will be simultaneously occupied by multiple lifting task paths at a future time step, and is an important basis for path conflict prediction.

[0031] In this embodiment, the model convergence is monitored using a validation set, and the training data is divided into a training set and a validation set; the validation loss and accuracy of the validation set are calculated after each epoch. Training should be stopped when the validation set loss stops decreasing after N consecutive iterations to avoid overfitting.

[0032] It should be noted that the number of iterations can also be set based on experience to improve training efficiency.

[0033] It also includes adding the platform disturbance component of the floating wind power platform, performing disturbance modeling, calculating the conflict probability of the predicted displacement trajectory based on the disturbance modeling path conflict area list, and issuing a real-time warning through the edge computing node when the conflict probability is greater than the preset conflict threshold, and scheduling the crane boom operation.

[0034] Furthermore, in the process of using a three-dimensional convolutional network to dynamically overlay and predict the displacement trajectory of the crane boom during future operations, a platform disturbance component of the floating wind power platform is added; the platform disturbance component includes the platform's pitch angle, roll angle, yaw angle, and wind speed data. The pitch angle θ, roll angle Φ and yaw angle ψ of the floating wind power platform are obtained by attitude sensors; an anemometer is integrated at the upper end of the boom to obtain wind speed data W(t); Perform disturbance modeling: Δr(t) = Δr att (t)+Δr wind (t); where Δr(t) represents the disturbance displacement value; The offset Δr of the lifting point is calculated using a first-order approximation method. att (t): Δr att (t) = L * (sinθ + sinΦ + sinψ); where L represents the length of the crane boom; t represents the time marker; The disturbance displacement Δr caused by wind speed is calculated using the simple harmonic vibration process. wind (t): Δr wind (t)=β w *W 2 (t); where β w Indicates the wind speed response coefficient; It should be noted that the unit of disturbance displacement value is meters (m). Wind speed is measured in m / s. The unit of wind speed response coefficient is: m / (m / s) 2 ; Based on the path conflict region list from step S304, calculate the expected distance d in the direction of maximum disturbance for the predicted displacement trajectories r1(t) and r2(t) within the path conflict regions. eff (t): d eff (t)=||r1(t)-r2(t)||-||Δr1(t)||-||Δr2(t)||; where Δr1(t) represents the disturbance displacement value of the predicted displacement trajectory r1(t); Δr2(t) represents the disturbance displacement value of the predicted displacement trajectory r2(t); |||| is the modulus calculation, which is performed based on the coordinates in the three-dimensional coordinate system of the floating wind power platform in this scheme; The conflict probability f(t) is calculated based on perturbation modeling: f(t) = 1 / [1 + exp((d eff [(t)-ds) / γ)]; where exp represents an exponential function with the natural constant as the base; ds represents the preset safety distance; and γ represents the collision bandwidth parameter; In this embodiment, an exponential function with the natural constant as the base is used as the normalization function to represent the probability of conflict in the predicted displacement trajectory, thus eliminating the dimension problem. The safety distance is preset to ds=1.5m; γ is the slope of the control transition zone, and is set to γ=0.5m; It should be noted that the expected distance d in the direction of maximum disturbance eff (t) may be negative, which in its physical sense means that the trajectory regions overlap under disturbance, and there may be a conflict. The conflict probability f(t) is judged, and when f(t) is greater than the preset conflict threshold, a real-time warning is issued through the edge computing node.

[0035] The scheduling process involves staff operating the crane to avoid conflict areas and change the crane boom's operating trajectory.

[0036] A deep learning-based resource scheduling system for offshore engineering platforms includes a data acquisition module, a data processing module, a construction task map construction module, a dynamic overlap prediction module, a disturbance modeling module, and a scheduling module. The data acquisition module is used to collect the displacement trajectory, operation status signals, and platform disturbance components during the operation of the crane boom; The data processing module is used for preprocessing and spatiotemporal alignment; The construction task map construction module is used to integrate the collected displacement trajectory and operation status signals, perform spatial mapping within the operation radius of the crane boom, construct a three-dimensional rasterized spatial occupancy map, and construct the construction task map through the three-dimensional rasterized spatial occupancy map. The dynamic overlap prediction module is used to dynamically predict the displacement trajectory of the crane boom during future operations based on the constructed construction task map and a 3D convolutional network. It marks path conflict areas and displays a list of path conflict areas.

[0037] The disturbance modeling module is used to perform disturbance modeling based on the platform disturbance components of the floating wind power platform; The scheduling module is used to calculate the collision probability of the predicted displacement trajectory based on the path collision region list based on disturbance modeling. When the collision probability is greater than the preset collision threshold, it provides real-time warning through the edge computing node and schedules the crane boom operation.

[0038] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered in all respects as exemplary and non-limiting, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

Claims

1. A resource scheduling method for offshore engineering platforms based on deep learning, characterized in that: include: Edge computing nodes and multimodal sensors are deployed in the boom working area of ​​the floating wind power platform to collect the displacement trajectory and working status signals of the boom during operation. By integrating the collected displacement trajectory and operation status signals, spatial mapping is performed within the operating radius of the crane boom to construct a three-dimensional rasterized spatial occupancy map; A construction task map is constructed using a three-dimensional rasterized spatial occupancy map; Based on the constructed construction task map, a 3D convolutional network is used to dynamically overlap and predict the displacement trajectory of the crane boom during future operations, outputting the predicted displacement trajectory. Based on the prediction results, the DBSCAN algorithm is used to divide the path conflict areas and display the list of path conflict areas. It also includes adding the platform disturbance component of the floating wind power platform, performing disturbance modeling, calculating the conflict probability of the predicted displacement trajectory based on the disturbance modeling path conflict area list, and issuing a real-time warning through the edge computing node when the conflict probability is greater than the preset conflict threshold, and scheduling the crane boom operation.

2. The resource scheduling method for offshore engineering platforms based on deep learning according to claim 1, characterized in that: include: The multimodal sensor group is used to collect monitoring data of the working area of ​​the floating wind power platform's boom, and the monitoring data includes real-time working images and working status signals; The real-time operation images are captured by the set video acquisition device, forming a real-time video image sequence of the acquisition operation area; The operation status signal is obtained by the control system of the boom, including the lifting load of the boom; The edge computing nodes are used to preprocess real-time video image sequences in the acquisition area; and provide data fusion analysis and localized decision-making functions.

3. The resource scheduling method for offshore engineering platforms based on deep learning according to claim 2, characterized in that: include: Step S201: Preprocess the real-time video image sequence of the acquired work area and align it with the working radius of the crane boom in time and space. The preprocessing includes data cleaning and standardization; Frame extraction, dehazing, noise reduction, and distortion correction are performed on the visual data; the displacement trajectory of the crane boom is obtained from the real-time video image sequence of the acquired work area; The displacement trajectory is unified in coordinate system and transformed to the three-dimensional coordinate system of the floating wind power platform. The spatiotemporal alignment includes spatiotemporal synchronization correction; timestamping the collected data based on the local clock of the edge computing node; and transforming the pixel coordinates in the video image to the three-dimensional coordinate system of the floating wind power platform through a spatial mapping matrix. Step S202: Discretize the three-dimensional coordinate system space of the floating wind power platform into quantifiable spatial units, record the occupancy status of each unit in real time, and form a three-dimensional rasterized spatial occupancy map. Step S203: Based on the three-dimensional rasterized spatial occupancy map, construct the construction task map G=(V,E); where the node V in the construction task map represents the occupied grid; and the directed edge E in the construction task map represents the displacement trajectory of the crane boom. Step S204: The attributes of the node include the set S of the required grid cells and the historical duration of occupation {t1,t2,…,tn}; the attributes of the directed edge E include the planned start and end times of the job [ts,te].

4. The resource scheduling method for offshore engineering platforms based on deep learning according to claim 3, characterized in that: include: The three-dimensional coordinate system of the floating wind power platform is established with the center of the bottom platform of the crane boom as the origin, and the x, y and z axes corresponding to the horizontal, longitudinal and vertical heights, respectively.

5. The resource scheduling method for offshore engineering platforms based on deep learning according to claim 3, characterized in that: Specifically: Specifically, based on the operating radius of the floating wind turbine platform's boom, the operating area is divided into a three-dimensional grid, with each grid assigned a unique spatial number; dynamically occupied grids are marked according to the boom's displacement trajectory; and a time dimension attribute is added to each dynamically occupied grid, recording the most recent occupation time and the duration of occupation.

6. The resource scheduling method for offshore engineering platforms based on deep learning according to claim 5, characterized in that: include: Step S301: Construct a spatiotemporal feature cube as the input to a 3D convolutional network; Step S302: Dynamically overlap and predict the displacement trajectory of the crane boom during future operation based on a three-dimensional convolutional network; Step S303: Perform threshold filtering on the output trajectory overlap probability P, mark the grids with P greater than or equal to the preset threshold as potential overlapping grids, output a list of potential overlapping grids including location, time and crane boom, traverse the list of potential overlapping grids by DBSCAN algorithm, and form continuous spatial regions by clustering the potential overlapping grids, which are marked as path conflict regions. Step S304: Visualize the list of path conflict areas.

7. The resource scheduling method for offshore engineering platforms based on deep learning according to claim 6, characterized in that: Specifically: Step S3011: Based on the occupancy grid set S of each node in the construction task map of step S203, extract the continuous path coordinate sequence of the work area to generate the displacement trajectory dataset of the crane boom; obtain the discrete displacement vector through the displacement trajectory dataset. Step S3012: Calculate the velocity and acceleration characteristics of the displacement trajectory dataset at each time step, and mark the current execution stage of the job in conjunction with the start and end times of the job plan [ts, te]. Step S3013: Encode the attributes of each occupied grid in the construction task map, and convert the occupancy status and historical occupancy duration {t1,t2,…,tn} into spatial feature vectors; using the three-dimensional grid of the floating wind power platform as the spatial dimension and the discrete time step as the time dimension, fuse the discrete displacement vector with the spatial feature vector to construct a multi-channel spatiotemporal feature cube, where each cell represents the comprehensive operation status of a certain occupied grid at a certain time step.

8. The resource scheduling method for offshore engineering platforms based on deep learning according to claim 6, characterized in that: Specifically: Step S3021: Design a three-dimensional convolutional network structure, including an input layer, several 3D convolutional layers, a 3D pooling layer, a fully connected layer, and an output layer; the input layer receives a spatiotemporal feature cube with dimensions (X, Y, Z, T, C), where X, Y, and Z are spatial dimensions, T is the number of time steps, and C is the number of channels; the spatial dimensions are obtained based on the three-dimensional coordinate system of the floating wind power platform; Step S3022: The convolutional layer uses a three-dimensional convolutional kernel to extract the spatial neighborhood features and time series dynamic change patterns of the crane boom displacement trajectory, and the pooling layer reduces the feature dimension through average pooling; wherein, the time series dynamic change pattern is obtained through the velocity features and acceleration features at each time step; Step S3023: The fully connected layer aggregates the extracted local features and generates a predicted displacement trajectory through the output layer. The output layer uses the Sigmoid activation function to output a value P∈[0,1] for each grid-time point in the predicted displacement trajectory, which represents the probability that the grid is simultaneously occupied by the crane boom displacement trajectory at the corresponding time step, denoted as the trajectory overlap probability.

9. A resource scheduling method for offshore engineering platforms based on deep learning according to claim 8, characterized in that: Also includes: In the process of using a 3D convolutional network to dynamically overlay and predict the displacement trajectory of the crane boom during future operations, the platform disturbance component of the floating wind power platform is added. The platform disturbance components include the platform's pitch angle, roll angle, yaw angle, and wind speed data; The pitch angle θ, roll angle Φ and yaw angle ψ of the floating wind power platform are obtained by attitude sensors; an anemometer is integrated at the upper end of the boom to obtain wind speed data W(t); Perform disturbance modeling: Δr(t) = Δr att (t)+Δr wind (t); where Δr(t) represents the disturbance displacement value; The offset Δr of the lifting point is calculated using a first-order approximation method. att (t): Δr att (t) = L * (sinθ + sinΦ + sinψ); where L represents the length of the crane boom; t represents the time marker; The disturbance displacement Δr caused by wind speed is calculated using the simple harmonic vibration process. wind (t): Δr wind (t)=β w *W 2 (t); where β w Indicates the wind speed response coefficient; Based on the path conflict region list from step S304, calculate the expected distance d in the direction of maximum disturbance for the predicted displacement trajectories r1(t) and r2(t) within the path conflict regions. eff (t): d eff (t)=||r1(t)-r2(t)||-||Δr1(t)||-||Δr2(t)||; where Δr1(t) represents the disturbance displacement value of the predicted displacement trajectory r1(t); Δr2(t) represents the disturbance displacement value of the predicted displacement trajectory r2(t); The conflict probability f(t) is calculated based on perturbation modeling: f(t) = 1 / [1 + exp((d eff [(t)-ds) / γ)]; where exp represents an exponential function with the natural constant as the base; ds represents the preset safety distance; and γ represents the collision bandwidth parameter; The conflict probability f(t) is judged, and when f(t) is greater than the preset conflict threshold, a real-time warning is issued through the edge computing node.

10. A deep learning-based resource scheduling system for offshore engineering platforms, applied to the deep learning-based resource scheduling method for offshore engineering platforms as described in any one of claims 1-9, characterized in that: It includes a data acquisition module, a data processing module, a construction task map construction module, a dynamic overlap prediction module, a disturbance modeling module, and a scheduling module; The data acquisition module is used to collect the displacement trajectory, operation status signal and platform disturbance component during the operation of the crane boom; The data processing module is used for preprocessing and spatiotemporal alignment. The construction task map construction module is used to integrate the collected displacement trajectory and operation status signals, perform spatial mapping within the operation radius of the crane boom, construct a three-dimensional rasterized spatial occupancy map, and construct a construction task map through the three-dimensional rasterized spatial occupancy map. The dynamic overlap prediction module is used to dynamically predict the displacement trajectory of the crane boom during future operations based on the constructed construction task map and a three-dimensional convolutional network, mark path conflict areas, and display a list of path conflict areas. The disturbance modeling module is used to perform disturbance modeling based on the platform disturbance components of the floating wind power platform; The scheduling module is used to calculate the conflict probability of the predicted displacement trajectory based on the path conflict area list based on disturbance modeling. When the conflict probability is greater than the preset conflict threshold, a real-time warning is issued through the edge computing node, and the crane boom operation is scheduled.

Images