A deep learning-based offshore platform resource scheduling system and method

Abstract

The application discloses a kind of based on deep learning's sea work platform resource scheduling system and method, it is related to the technical field of floating wind power platform operation resource scheduling, including: collecting displacement trajectory and operation state signal in the process of crane boom operation;Space mapping is carried out in the operation radius range of crane boom, and three-dimensional gridding space occupancy map is constructed;Construction task atlas is constructed;Using three-dimensional convolution network, the displacement trajectory in the future crane boom operation process is dynamically overlapped and predicted, and the predicted displacement trajectory is output, further comprising, adding the platform disturbance component of floating wind power platform, disturbance modeling is carried out, the conflict probability of predicted displacement trajectory in path conflict area list is calculated based on disturbance modeling, when the conflict probability is greater than preset conflict threshold, real-time early warning is carried out through edge computing node, and crane boom operation is scheduled.The method effectively improves the safety and intelligent level of multi-crane boom collaborative operation, and has wide engineering application prospect.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of operation resource scheduling of floating wind power platforms, in particular to a marine platform resource scheduling system and method based on deep learning. BACKGROUND

[0002] In the installation and maintenance process of floating wind power platforms, the crane boom hoisting operation is a key construction task, usually involving the hoisting and transfer of large wind power components. Due to the fact that the floating platform is often in a complex marine environment, the platform posture will be significantly affected by natural disturbances such as wind, waves, and currents, thereby causing great disturbance to the operation trajectory and load stability of the crane boom.

[0003] At present, the traditional crane boom operation path planning and scheduling method mainly relies on preset rules, manual experience or two-dimensional drawings for auxiliary decision-making, which is difficult to respond to changes in platform dynamic disturbances in real time, and also cannot effectively identify potential path conflicts of multiple crane booms in 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, etc., and have significant advantages in extracting spatio-temporal features. Therefore, a crane boom resource scheduling method is needed that combines the floating wind power platform operation scene, introduces platform disturbance modeling mechanism, and integrates the deep learning path prediction capability, to realize the early perception and dynamic avoidance of operation path conflicts, and improve the safety and efficiency of offshore construction. SUMMARY

[0005] The purpose of the present application is to provide a marine platform resource scheduling system and method based on deep learning to solve the problems in the prior art.

[0006] To achieve the above purpose, the present application provides the following technical scheme: a marine platform resource scheduling method based on deep learning, comprising:

[0007] Deploy edge computing nodes and multi-modal sensors in the crane boom operation area of the floating wind power platform, collect displacement trajectories and operation state signals during the crane boom operation process; integrate the collected displacement trajectories and operation state signals, perform spatial mapping within the operation radius of the crane boom, and construct a three-dimensional grid-based space occupancy map;

[0008] Construct a construction task graph through the three-dimensional grid-based space occupancy map;

[0009] Based on the constructed construction task graph, use a three-dimensional convolutional network to dynamically overlap predict the displacement trajectory in the future crane boom operation process, output the predicted displacement trajectory, divide the path conflict area according to the prediction result through the DBSCAN algorithm, and display the path conflict area list;

[0010] Further, the platform disturbance component of the floating wind power platform is added, the disturbance modeling is performed, the conflict probability of the predicted displacement trajectory in the path conflict area list is calculated based on the disturbance modeling, the real-time early warning is performed through the edge computing node when the conflict probability is greater than a preset conflict threshold, and the crane boom operation is scheduled.

[0011] Further, the multi-modal sensor group is used to collect monitoring data of the crane boom operation area of the floating wind power platform, and the monitoring data includes real-time operation images and operation state signals.

[0012] The real-time operation images are collected by the set video collection device to form a real-time video image sequence of the collection operation area.

[0013] The operation state signals are obtained by the control system of the crane boom, and include the hoisting load of the crane boom.

[0014] The edge computing node is used for pre-processing the real-time video image sequence of the collection operation area, and provides data fusion analysis and localized decision-making functions.

[0015] Further, in step S201, the real-time video image sequence of the collection operation area is pre-processed, and is spatio-temporally aligned with the operation radius range of the crane boom.

[0016] The pre-processing includes data cleaning and standardization.

[0017] Frame extraction, fog removal, denoising and distortion correction are performed on the visual data; and the displacement trajectory of the crane boom is obtained through the real-time video image sequence of the collection operation area.

[0018] The displacement trajectory is uniformly converted to the floating wind power platform three-dimensional coordinate system.

[0019] The spatio-temporal alignment includes spatio-temporal synchronization correction; the collected data is time-stamped based on the local clock of the edge computing node; and the pixel coordinates in the video image are converted to the floating wind power platform three-dimensional coordinate system through a space mapping matrix.

[0020] In step S202, the floating wind power platform three-dimensional coordinate system is discretized into quantifiable space units, the occupancy state of each unit is recorded in real time, and a three-dimensional rasterized space occupancy map is formed.

[0021] In step S203, based on the three-dimensional rasterized space occupancy map, a construction task graph G=(V, E) is constructed; wherein, the node V in the construction task graph represents the occupied grid; and the directed edge E in the construction task graph represents the displacement trajectory of the crane boom.

[0022] Step S204, the attributes of the node include a set S of required occupation grids and a historical occupation duration {t1, t2, …, tn}; the attributes of the directed edge E include the planned start and end time [ts, te] of the job.

[0023] Further, 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 transverse, longitudinal and vertical height three-dimensional coordinate system.

[0024] Further, based on the operation radius range of the crane boom of the floating wind power platform, the operation area is divided into three-dimensional grids, each grid is assigned a unique space number; according to the displacement trajectory of the crane boom, mark the dynamic occupation grid; for each dynamically occupied grid, add a time dimension attribute, record the latest occupation time and occupation duration.

[0025] Further, step S301, construct a spatiotemporal feature cube as the input of the three-dimensional convolutional network;

[0026] Step S302, based on the three-dimensional convolutional network, dynamically overlapping prediction is performed on the displacement trajectory in the future crane boom operation process;

[0027] Step S303, threshold filtering is performed on the output trajectory overlapping probability P, and the grid with P greater than or equal to the preset threshold is marked as a potential overlapping grid; output the list of potential overlapping grids including position, time and crane boom involved; through the DBSCAN algorithm, the list of potential overlapping grids is traversed, and through clustering of the potential overlapping grids, a continuous spatial region is formed, which is marked as a path conflict region;

[0028] Step S304, visualize and display the list of path conflict regions.

[0029] Optionally, step S3011, based on the occupation grid set S of each node in the construction task graph of step S203, extract the continuous path coordinate sequence of the operation area, generate the displacement trajectory data set of the crane boom; through the displacement trajectory data set, obtain the discrete displacement vector;

[0030] Step S3012, calculate the speed feature and acceleration feature of the displacement trajectory data set at each time step, and combine the job plan start and end time [ts, te] to mark the execution stage of the current job;

[0031] Step S3013, encode the properties of each occupied grid in the construction task graph, convert the occupancy state and historical occupancy duration {t1, t2, …, tn} into a spatial feature vector; take the three-dimensional grid of the floating wind power platform as the spatial dimension, and take the discrete time step as the time dimension, fuse the discrete displacement vector and the spatial feature vector, and construct a multi-channel spatio-temporal feature cube, each cell representing the comprehensive operation state of a certain occupied grid at a certain time step.

[0032] Optionally, step S3021, design a three-dimensional convolutional network structure, including an input layer, a plurality of 3D convolutional layers, a 3D pooling layer, a fully connected layer, and an output layer; the input layer receives the spatio-temporal 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 channel dimension; the spatial dimension is obtained based on the three-dimensional coordinate system of the floating wind power platform;

[0033] Step S3022, the convolutional layer uses a three-dimensional convolution 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 feature and the acceleration feature at each time step;

[0034] 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 a Sigmoid activation function to output a value P∈[0,1] for each grid-time point in the predicted displacement trajectory, representing 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.

[0035] Further, in the process of using the three-dimensional convolutional network to dynamically predict the overlap of the future crane boom operation process, a platform disturbance component of the floating wind power platform is added; the platform disturbance component includes the pitch angle, roll angle, and yaw angle of the platform, and wind speed data;

[0036] The pitch angle θ, roll angle Φ, and yaw angle ψ of the platform are obtained through the attitude sensor equipped on the floating wind power platform; the wind speed data W(t) is obtained by integrating a wind speed meter at the upper end point of the crane boom;

[0037] Perform disturbance modeling: Δr(t) = Δr att (t) + Δr wind (t); wherein Δr(t) represents the disturbance displacement value;

[0038] The offset Δr att (t) of the hoisting point is calculated by a first-order approximation method:

[0039] Δr att(t) = L * (sinθ + sinΦ + sinψ) ; wherein, L represents the length of the crane boom; t represents the time identifier;

[0040] Calculate the disturbance displacement Δr caused by the wind speed through the simple harmonic vibration process wind (t) :

[0041] Δr wind (t) = β w * W 2 (t) ; wherein, β w represents the wind speed response coefficient;

[0042] Based on the path conflict area list of step S304, the predicted displacement trajectories r1(t) and r2(t) in the path conflict area; calculate the expected distance d in the maximum disturbance direction eff (t) :

[0043] d eff (t) = ||r1(t) - r2(t)|| - ||Δr1(t)|| - ||Δr2(t)|| ; wherein, 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) ;

[0044] Calculate the conflict probability f(t) based on the disturbance modeling:

[0045] f(t) = 1 / [1 + exp ((d eff (t) - ds) / γ) ] ; wherein, exp represents the exponential function with the natural constant as the base; ds represents the preset safety distance; γ represents the conflict bandwidth parameter;

[0046] Judge the conflict probability f(t), and when f(t) is greater than the preset conflict threshold, real-time early warning is performed through the edge computing node.

[0047] A deep learning-based offshore platform resource scheduling system, comprising a data acquisition module, a data processing module, a construction task graph construction module, a dynamic overlap prediction module, a disturbance modeling module and a scheduling module;

[0048] The data acquisition module is used to acquire displacement trajectories, operation state signals and platform disturbance components in the operation process of the crane boom;

[0049] The data processing module is used for pre-processing and space-time alignment;

[0050] The construction task graph construction module is used to integrate the acquired displacement trajectories and operation state signals, perform spatial mapping in the operation radius range of the crane boom, construct a three-dimensional gridded space occupancy graph, and construct a construction task graph through the three-dimensional gridded space occupancy graph;

[0051] The dynamic overlap prediction module is used for using a three-dimensional convolution network to dynamically predict displacement trajectories in future crane boom operation processes based on the constructed construction task map, marking path conflict areas, and displaying a path conflict area list.

[0052] The disturbance modeling module is used for disturbance modeling based on platform disturbance components of the floating wind power platform.

[0053] The scheduling module is used for calculating a conflict probability of the predicted displacement trajectories in the path conflict area list based on disturbance modeling, performing real-time early warning through an edge computing node when the conflict probability is greater than a preset conflict threshold, and scheduling crane boom operation.

[0054] Compared with the prior art, the beneficial effects of the present application are: the method realizes real-time collection and intelligent processing of crane boom displacement trajectories and operation state signals by deploying edge computing nodes and multi-modal sensors in the crane boom operation area of the floating wind power platform, and maps the collected data to a unified three-dimensional coordinate system to construct a high-precision three-dimensional rasterized space occupation map, effectively depicting the dynamic occupation state of the operation area. Further, combined with the constructed construction task map and the three-dimensional convolution network model, the future operation path is spatio-temporally overlapped and predicted, and the path conflict area is divided through a clustering algorithm, improving the accuracy and interpretability of path conflict identification.

[0055] The significant innovation of the present application is to introduce a disturbance modeling mechanism for the floating wind power platform, convert key disturbance parameters such as pitch angle, roll angle, yaw angle and wind speed into displacement disturbance quantities, embed the path conflict probability calculation model, and establish a dynamic mapping relationship between disturbance factors and path overlap risk. By constructing a conflict probability function based on disturbance correction, the risk degree of different operation trajectories at future time can be quantified, and dynamic scheduling optimization under disturbance perception conditions can be realized.

[0056] In addition, the present application provides localized decision response capability through edge computing nodes, realizes abnormal early warning and risk area visualization during operation, reduces the response lag problem of traditional manual scheduling to complex environmental changes, enhances the intelligence, adaptability and safety of offshore platform resource scheduling, and is especially suitable for large-scale hoisting operation scenes in complex sea conditions such as floating wind power, has wide engineering application prospect and popularization value. BRIEF DESCRIPTION OF DRAWINGS

[0057] Figure 1 is a step schematic diagram of the offshore platform resource scheduling method based on deep learning. DETAILED DESCRIPTION

[0058] With reference to the accompanying drawings, the technical solutions in the embodiments of the present application will be clearly and completely described below, obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative work belong to the protection scope of the present application.

[0059] Embodiment: As shown in the figure, the present application provides a technical solution, a deep learning-based offshore platform resource scheduling method, comprising: Figure 1

[0060] Deploying an edge computing node and a multi-modal sensor in the crane boom operation area of the floating wind power platform to collect displacement trajectories and operation state signals in the crane boom operation process; integrating the collected displacement trajectories and operation state signals, performing spatial mapping in the operation radius range of the crane boom, and constructing a three-dimensional gridded space occupation map;

[0061] Further, the multi-modal sensor group is used to collect monitoring data of the crane boom operation area of the floating wind power platform, and the monitoring data includes real-time operation images and operation state signals;

[0062] The real-time operation images are collected by the set video acquisition device to form a real-time video image sequence of the collection operation area;

[0063] The operation state signals are obtained by the control system of the crane boom, including the hoisting load of the crane boom;

[0064] The edge computing node is used to pre-process the real-time video image sequence of the collection operation area; and provide data fusion analysis and localized decision-making functions.

[0065] A construction task map is constructed through the three-dimensional gridded space occupation map;

[0066] Further, step S201, the real-time video image sequence of the collection operation area is pre-processed, and the operation radius range of the crane boom is spatio-temporally aligned;

[0067] The pre-processing includes data cleaning and standardization;

[0068] Frame extraction, defogging and denoising (image blur caused by high humidity environment, dark channel prior algorithm is used for processing), and distortion correction (industrial camera distortion caused by vibration, corrected by camera intrinsic matrix) are performed on the visual data;

[0069] The displacement trajectory of the crane boom is obtained through the collected real-time video image sequence of the collection operation area;

[0070] ​Coordinate unification is performed on the displacement trajectory, and the displacement trajectory is converted to a three-dimensional coordinate system of the floating wind power platform;

[0071] The spatio-temporal alignment includes spatio-temporal synchronization correction; based on a local clock of the edge computing node, the collected data is time-stamped; pixel coordinates in a video image are converted to the three-dimensional coordinate system of the floating wind power platform through a space mapping matrix;

[0072] In step S202, the three-dimensional coordinate system of the floating wind power platform is discretized into quantifiable space units, and the occupancy state of each unit is recorded in real time to form a three-dimensional rasterized space occupancy map;

[0073] In step S203, a construction task graph G=(V, E) is constructed based on the three-dimensional rasterized space occupancy map; wherein, the nodes V in the construction task graph represent the occupied grids; and the directed edges E in the construction task graph represent the displacement trajectory of the crane boom.

[0074] In step S204, the attributes of the nodes include a set S of required occupied grids and a historical occupancy duration {t1, t2, …, tn}; and the attributes of the directed edges E include the planned start and end time [ts, te] of the operation.

[0075] Further, 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 transverse, longitudinal and vertical height, respectively.

[0076] Further, based on the operation radius range of the crane boom of the floating wind power platform, the operation area is divided into three-dimensional grids, each grid is assigned a unique space number; according to the displacement trajectory of the crane boom, the dynamically occupied grids are marked; for each dynamically occupied grid, a time dimension attribute is added to record the latest occupancy time and occupancy duration.

[0077] Based on the constructed construction task graph, a three-dimensional convolutional network is used to dynamically overlap and predict the displacement trajectory in the future operation process of the crane boom, output the predicted displacement trajectory, divide the path conflict area according to the prediction result through the DBSCAN algorithm, and show the path conflict area list.

[0078] Further, in step S301, a spatio-temporal feature cube is constructed as the input of the three-dimensional convolutional network.

[0079] Optionally, in step S3011, based on the occupied grid set S of each node in the construction task graph of step S203, a continuous path coordinate sequence of the operation area is extracted to generate a displacement trajectory data set of the crane boom; and a discrete displacement vector is obtained through the displacement trajectory data set.

[0080] In the embodiment, a path coordinate sequence is generated: based on a recorded spatial position point set S={(x1,y1,z1),(x2,y2,z2),...,(x n ,y n ,z n )}, a continuous operation path is generated in chronological order; a discrete displacement vector is obtained through a displacement trajectory data set, such as [(x2,y2,z2)-(x1,y1,z1)];

[0081] Step S3012, the speed feature and the acceleration feature of the displacement trajectory data set at each time step are calculated, and the execution stage in which the current operation is located is marked by combining the start and end times [ts,te] of the operation plan;

[0082] According to the coordinate calculation of the speed feature and the acceleration feature, the prior art is not repeated here, and it should be noted that this part is used to capture the dynamic characteristics of the operation task displacement trajectory, and provides a time sequence rule for subsequent prediction;

[0083] Step S3013, the properties of each occupied grid in the construction task graph are feature-encoded, the occupancy state and the historical occupancy duration {t1,t2,...,tn} are converted into a spatial feature vector; the three-dimensional grid of the floating wind power platform is taken as the spatial dimension, and the discrete time step is taken as the time dimension, the discrete displacement vector and the spatial feature vector are fused, and a multi-channel spatio-temporal feature cube is constructed, and each cell represents the comprehensive operation state of a certain occupied grid at a certain time step.

[0084] It should be noted that the discrete displacement vector and the spatial feature vector are fused, and the fusion means that at each time step t, the discrete displacement vector corresponding to the time is connected with the spatial feature of the current involved grid in the same structure.

[0085] In the embodiment, the fused vector represents the comprehensive operation state, including the dimensions of the discrete displacement vector, the speed, the acceleration, etc.

[0086] Step S302, based on a three-dimensional convolutional network, the displacement trajectory in the future operation process of the hoist arm is dynamically overlapped and predicted.

[0087] Optionally, step S3021, a three-dimensional convolutional network structure is designed, including an input layer, a plurality of 3D convolutional layers, a 3D pooling layer, a full connection layer, and an output layer; the input layer receives the spatio-temporal feature cube, and the dimension is (X,Y,Z,T,C), wherein X, Y, and Z are spatial dimensions, T is the number of time steps, and C is the channel number dimension; the spatial dimension is obtained based on the three-dimensional coordinate system of the floating wind power platform;

[0088] Step S3022, the convolution layer extracts the spatial neighborhood features and the time sequence dynamic change mode of the crane boom displacement trajectory using a three-dimensional convolution kernel, and the pooling layer reduces the feature dimension through average pooling; wherein the time sequence dynamic change mode is obtained through the speed feature and the acceleration feature at each time step;

[0089] Step S3023, the fully connected layer aggregates the extracted local features, and generates a predicted displacement trajectory through the output layer; the output layer adopts a Sigmoid activation function, outputs a value P∈[0, 1] for each grid-time point in the predicted displacement trajectory, indicating the probability that the grid is simultaneously occupied by the crane boom displacement trajectory at the corresponding time step, which is recorded as a trajectory overlap probability.

[0090] In the embodiment, the final output of the fully connected layer is mapped to the interval of 0~1 through σ(P)=1 / (1+e -P ); It should be noted that the trajectory overlap probability P∈[0, 1] is automatically learned by the three-dimensional convolutional neural network based on the input spatio-temporal feature cube, which is a probabilistic space-time state estimation, and the data input to the 3D CNN is a five-dimensional tensor in the form of (X, Y, Z, T, C), called a spatio-temporal feature cube. Each “cell” of the tensor represents a multi-dimensional description of the job state at a certain time and a certain spatial position, and the 3D CNN extracts features from the tensor using multiple three-dimensional convolution kernels (sliding in space and time):

[0091] The convolution kernel can identify the pattern of the continuous trajectory in space; the convolution in the time dimension can capture the dynamic change (such as whether it is approaching or in the same job phase); the deep layers of the network gradually form the basis for judging the possibility of “trajectory simultaneous occupation”;

[0092] The output layer of the network adopts a Sigmoid activation function; if there are highly overlapping features in the input cube (such as two trajectories in the same position and the same time step speed direction close), the neural network will learn to produce a larger P, so that P→1;

[0093] If the trajectories are far apart in space or time, or have no intersection after disturbance, then P→0;

[0094] Therefore, the P value is essentially a probabilistic estimate learned by the neural network, indicating the probability that the “current grid, current time is simultaneously occupied by multiple job trajectories”.

[0095] The preset threshold is 0.6, and when the trajectory overlap probability P>0.6, the grid is considered as a potential overlapping grid;

[0096] Step S303, threshold screening is performed on the output trajectory overlap probability P, grids with P greater than or equal to a preset threshold are marked as potential overlap grids, a list of potential overlap grids including position, time and involving the crane boom is output, the list of potential overlap grids is traversed through the DBSCAN algorithm, and a continuous spatial region is formed by clustering the potential overlap grids, which is marked as a path conflict region;

[0097] Step S304, visualizing and displaying the list of path conflict regions.

[0098] In this embodiment, the network learning goal is defined as "trajectory conflict label prediction", and for the training sample:

[0099] Label y = 1: indicating that the grid is simultaneously covered by multiple task trajectories at time step t;

[0100] Label y = 0: indicating no conflict or only one trajectory occupying;

[0101] The cross-entropy loss function is used during training: ; Wherein, N is the number of training samples, P i is the predicted overlap probability of the i-th grid point;

[0102] It should be noted that the trajectory overlap probability P is a probability value generated by the output layer through the Sigmoid function after the three-dimensional convolutional neural network performs layer-by-layer feature extraction and aggregation on the constructed spatiotemporal feature cube. The probability value reflects the possibility that a certain spatial grid is simultaneously occupied by multiple crane task paths at a certain time step in the future, and is an important basis for path conflict prediction.

[0103] In this embodiment, the validation set is used to monitor model convergence, and the training data is divided into: training set + validation set; The validation loss and accuracy of the validation set are calculated after each epoch.

[0104] When the validation set loss does not decrease for N consecutive times, stop training; avoid "overfitting".

[0105] It should be noted that the number of iterations can also be set according to experience to improve training efficiency.

[0106] Further comprising adding a platform disturbance component of the floating wind power platform, performing disturbance modeling, calculating the conflict probability of the predicted displacement trajectory in the path conflict region list based on the disturbance modeling, and performing real-time early warning through the edge computing node when the conflict probability is greater than a preset conflict threshold, and scheduling the crane boom operation.

[0107] Further, in the process of using a three-dimensional convolution network to dynamically overlap predict the displacement trajectory in the future crane boom operation process, the platform disturbance component of the floating wind power platform is added; the platform disturbance component includes the pitch angle, roll angle and yaw angle of the platform and wind speed data;

[0108] The pitch angle θ, roll angle Φ and yaw angle ψ of the platform are obtained through the attitude sensor equipped on the floating wind power platform; the wind speed data W(t) is obtained through the wind speed meter integrated at the upper end point of the crane boom;

[0109] Disturbance modeling is performed: Δr(t) = Δr att (t) + Δr wind (t); wherein, Δr(t) represents the disturbance displacement value;

[0110] The offset amount Δr att (t) of the hoisting point is calculated through the first-order approximation method:

[0111] Δr att (t) = L * (sin θ + sin Φ + sin ψ); wherein, L represents the length of the crane boom; t represents the time identifier;

[0112] The disturbance displacement Δr wind (t) caused by the simple harmonic vibration process is calculated:

[0113] Δr wind (t) = β w * W 2 (t); wherein, β w represents the wind speed response coefficient;

[0114] It should be noted that the unit of the disturbance displacement value is: m;

[0115] The unit of the wind speed is: m / s;

[0116] The unit of the wind speed response coefficient is: m / (m / s) 2 ;

[0117] Based on the path conflict area list of step S304, the predicted displacement trajectories r1(t) and r2(t) in the path conflict area are calculated; the expected distance d eff (t) in the maximum disturbance direction:

[0118] d eff (t) = ||r1(t)-r2(t)||-||Δr1(t)||-||Δr2(t)||; wherein, 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 operation, which is calculated based on the coordinates in the three-dimensional coordinate system of the floating wind power platform in the scheme.

[0119] The conflict probability f(t) is calculated based on the disturbance modeling:

[0120] f(t)=1 / [1+exp((d eff (t)-ds) / γ)];wherein exp represents an exponential function with a natural constant as a base; ds represents a preset safety distance; and γ represents a conflict bandwidth parameter;

[0121] In this embodiment, the exponential function with a natural constant as a base is used as a normalization function to represent the conflict probability of the predicted displacement trajectory, and the dimension problem is also eliminated;

[0122] The safety distance is preset as ds=1.5m; and γ is a control transition zone slope, and γ=0.5m is taken;

[0123] It should be noted that the expected distance d eff (t) may be negative, which in its physical sense indicates that the trajectory area under the disturbance overlaps, and there is a conflict possibility;

[0124] The conflict probability f(t) is judged, and when f(t) is greater than a preset conflict threshold, real-time early warning is performed through the edge computing node.

[0125] The scheduling content is that the staff member performs an operation to avoid the conflict area and performs a change of the crane boom operation trajectory.

[0126] A marine platform resource scheduling system based on deep learning includes a data acquisition module, a data processing module, a construction task graph construction module, a dynamic overlap prediction module, a disturbance modeling module, and a scheduling module.

[0127] The data acquisition module is used to acquire displacement trajectories, operation state signals, and platform disturbance components in the crane boom operation process.

[0128] The data processing module is used for preprocessing and spatio-temporal alignment.

[0129] The construction task graph construction module is used to integrate the acquired displacement trajectories and operation state signals, perform spatial mapping in the operation radius range of the crane boom, construct a three-dimensional rasterized space occupation graph, and construct a construction task graph through the three-dimensional rasterized space occupation graph.

[0130] The dynamic overlap prediction module is used to perform dynamic overlap prediction on the displacement trajectories in the future crane boom operation process based on the constructed construction task graph, mark the path conflict area, and display a path conflict area list.

[0131] The disturbance modeling module is used to perform disturbance modeling based on the platform disturbance components of the floating wind power platform.

[0132] The scheduling module is configured to calculate a conflict probability of the predicted displacement trajectory in the path conflict area list based on the disturbance modeling, to perform real-time early warning through the edge computing node when the conflict probability is greater than a preset conflict threshold, and to perform scheduling on the crane boom operation.

[0133] It will be obvious to a person skilled in the art that the application is not limited to the details of the above-described exemplary embodiments, but that the application can be implemented in other concrete forms without departing from the spirit or essential characteristics of the application. The embodiments are therefore to be considered in all respects as illustrative and not restrictive, the scope of the application being indicated by the appended claims rather than by the above description, and it is intended to encompass all changes and modifications that fall within the meaning and scope of equivalents of the claims. Any reference signs in the claims should not be construed as limiting the claims to the figures in which the reference signs are used.

Claims

1. A deep learning-based offshore platform resource scheduling method, characterized in that: The method comprises the following steps: Deploying an edge computing node and a multi-modal sensor in the crane operation area of the floating wind power platform to collect displacement trajectories and operation state signals during the crane operation process; Integrating the collected displacement trajectories and operation state signals, and performing spatial mapping within the operation radius of the crane to construct a three-dimensional rasterized spatial occupancy map; The multi-modal sensor group is used to collect monitoring data of the crane operation area of the floating wind power platform, and the monitoring data includes real-time operation images and operation state signals; the real-time operation images are collected by the video collection device to form a real-time video image sequence of the collection operation area; The specific steps of constructing the three-dimensional rasterized spatial occupancy map are as follows: Step 1. Preprocess the collected real-time video image sequence of the operation area, and perform spatio-temporal alignment with the operation radius of the crane; the preprocessing includes data cleaning and standardization; frame extraction, fog removal, denoising and distortion correction are performed on the visual data; the displacement trajectory of the crane is obtained through the collected real-time video image sequence of the operation area; the displacement trajectory is uniformly converted to the three-dimensional coordinate system of the floating wind power platform; the spatio-temporal alignment includes spatio-temporal synchronization correction; based on the local clock of the edge computing node, the collected data is time-stamped; the pixel coordinates in the video image are converted to the three-dimensional coordinate system of the floating wind power platform through a space mapping matrix; Step 2. Coordinate unification of displacement trajectory, unified conversion to floating wind power platform three-dimensional coordinate system; the spatio-temporal alignment includes spatio-temporal synchronization correction; based on the local clock of the edge computing node, the collected data is time-stamped; the pixel coordinates in the video image are converted to the three-dimensional coordinate system of the floating wind power platform through a space mapping matrix; Step 3. Discretize the three-dimensional coordinate system of the floating wind power platform into quantifiable space units, record the occupancy state of each unit in real time, and form a three-dimensional rasterized spatial occupancy map; Step 4. The attributes of the node include a set of required occupancy grids S and a historical occupancy duration {t1, t2, …, tn}; the attributes of the directed edge E include the planned start and end time [ts, te] of the operation; Construct a construction task graph based on the constructed construction task graph, and use a three-dimensional convolutional network to dynamically overlap and predict the displacement trajectory in the future crane operation process, output the predicted displacement trajectory, divide the path conflict area according to the prediction result through a DBSCAN algorithm, and visually display the path conflict area list, which comprises the following steps: Step S1. Construct a spatio-temporal feature cube as the input of the three-dimensional convolutional network; Step S2. Based on the three-dimensional convolutional network, dynamically overlap and predict the displacement trajectory in the future crane operation process; ​ Step S3. Threshold screening is performed on the output trajectory overlap probability P, and the grid with P greater than or equal to the preset threshold is marked as a potential overlap grid. A list of potential overlap grids including position, time and crane involved is output. The list of potential overlap grids is traversed by a DBSCAN algorithm. The potential overlap grids are clustered to form a continuous spatial region, which is marked as a path conflict region; In step S1, the specific steps of constructing the spatiotemporal feature cube are as follows: Step S1-1. Based on the occupied grid set S of each node in the construction task graph in step 4, the continuous path coordinate sequence of the work area is extracted, and the displacement trajectory data set of the crane is generated. The discrete displacement vector is obtained from the displacement trajectory data set; Step S1-2. The speed feature and acceleration feature of the displacement trajectory data set at each time step are calculated, and the execution stage of the current work is marked by combining the work plan start and end time [ts, te]; Step S1-3. The attributes of each occupied grid in the construction task graph are feature encoded, and the occupancy state and historical occupancy duration {t1, t2, …, tn} are converted into a spatial feature vector. The discrete displacement vector and the spatial feature vector are fused to construct a multi-channel spatiotemporal feature cube, with the three-dimensional grid of the floating wind power platform as the spatial dimension and the discrete time step as the time dimension. Each cell represents the comprehensive work state of a certain occupied grid at a certain time step; The three-dimensional convolutional network includes an input layer, a 3D convolutional layer, a 3D pooling layer, a fully connected layer, and an output layer. The input layer receives the spatiotemporal feature cube; The convolutional layer uses a three-dimensional convolutional kernel to extract the spatial neighborhood features and time sequence dynamic change patterns of the crane displacement trajectory. The pooling layer reduces the feature dimension through average pooling. The time sequence dynamic change pattern is obtained from the speed feature and acceleration feature at each time step; The fully connected layer aggregates the extracted local features, and the output layer generates a predicted displacement trajectory; Further comprising adding a platform disturbance component of the floating wind power platform, performing disturbance modeling, calculating the conflict probability of the predicted displacement trajectory in the path conflict region list based on the disturbance modeling, and performing real-time early warning through the edge computing node when the conflict probability is greater than a preset conflict threshold, and scheduling the crane work.

2. The method of claim 1, wherein: Comprise: The edge computing node and the multi-modal sensor are deployed in the crane work area of the floating wind power platform to collect displacement trajectories and work state signals during the crane work process. Specifically, it includes: The work state signal is obtained by the control system of the crane, including the hoisting load of the crane; The edge computing node is used for pre-processing real-time video image sequences collected in the work area, and provides data fusion analysis and localized decision-making functions.

3. The method of claim 1, wherein the method further comprises: The three-dimensional coordinate system of the floating wind power platform is specifically: The three-dimensional coordinate system of the floating wind power platform is established with the bottom platform center of the crane as the origin, and the x, y and z axes correspond to the horizontal transverse, longitudinal and vertical height, respectively.

4. The method of claim 1, wherein the method is based on deep learning. The spatial mapping in the work radius range of the crane is specifically: Based on the operating radius range of the floating wind platform crane boom, the operating area is divided into three-dimensional grids, each grid is assigned a unique spatial number; according to the displacement trajectory of the crane boom, the dynamic occupied grid is marked; for each dynamic occupied grid, the time dimension attribute is added, and the latest occupation time and occupation duration are recorded.

5. The method of claim 1, wherein: The three-dimensional convolutional network further comprises: 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 channel dimensions; the spatial dimensions are obtained based on the three-dimensional coordinate system of the floating wind platform; The output layer uses a Sigmoid activation function to output a value P∈[0,1] for each grid-time point in the predicted displacement trajectory, representing the probability of the grid being simultaneously occupied by the crane boom displacement trajectory at the corresponding time step, denoted as the trajectory overlap probability.

6. The method of claim 5, wherein: The platform disturbance component of the floating wind platform is added to the disturbance modeling, and the conflict probability of the predicted displacement trajectory in the path conflict area list is calculated based on the disturbance modeling; when the conflict probability is greater than the preset conflict threshold, real-time warning is performed through the edge computing node, and the crane boom operation is scheduled, specifically: In the process of using the three-dimensional convolutional network to dynamically predict the overlap of the displacement trajectory in the future crane boom operation, the platform disturbance component of the floating wind platform is added; the platform disturbance component includes the pitch angle, roll angle, and yaw angle of the platform, and wind speed data; The pitch angle θ, roll angle Φ, and yaw angle ψ of the platform are obtained through the attitude sensor equipped on the floating wind platform; the wind speed data W(t) is obtained by integrating a wind speed meter at the upper end point of the crane boom; Modeling the disturbance: Δr(t) = Δr att (t) + Δr wind (t); where Δr(t) represents the disturbance displacement value; The offset amount Δr of the hoisting point is calculated by a first order approximation method att (t): Δr att (t) = L * (sin θ + sin Φ + sin ψ); where L represents the length of the boom; t represents the time identifier; The perturbation displacement Δr caused by the wind speed is calculated by 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 area list of step S304, the predicted displacement trajectories r1(t) and r2(t) within the path conflict area; calculate the expected distance d in the maximum disturbance direction eff (t): d eff (t) = ||r1(t) - r2(t)|| - ||Dr1(t)|| - ||Dr2(t)||; where Dr1(t) denotes a perturbation displacement value of the predicted displacement trajectory r1(t); Dr2(t) denotes a perturbation displacement value of the predicted displacement trajectory r2(t); The conflict probability f(t) is calculated based on the disturbance modeling: f(t) = 1 / [1 + exp((d eff (t) - ds) / γ)]; where exp denotes an exponential function with a natural constant as a base; ds denotes a preset safety distance; and γ denotes a collision bandwidth parameter; The conflict probability f(t) is judged, and when f(t) is greater than the preset conflict threshold, real-time warning is performed through the edge computing node.

7. A deep learning-based offshore platform resource scheduling system applied to the deep learning-based offshore platform resource scheduling method of any one of claims 1-6. It comprises a data acquisition module, a data processing module, a construction task graph construction module, a dynamic overlap prediction module, a disturbance modeling module, and a scheduling module; The data acquisition module is used to acquire the displacement trajectory, operation state signal, and platform disturbance component in the crane boom operation process; The data processing module is used for preprocessing and spatiotemporal alignment; The construction task graph construction module is used to integrate the acquired displacement trajectory and operation state signal, perform spatial mapping in the operating radius range of the crane boom, construct a three-dimensional gridded spatial occupation graph, and construct a construction task graph through the three-dimensional gridded spatial occupation graph; The dynamic overlap prediction module is used to dynamically predict the overlap of the displacement trajectory in the future crane boom operation process based on the constructed construction task graph, using the three-dimensional convolutional network, mark the path conflict area, and display the path conflict area list; The disturbance modeling module is used to perform disturbance modeling based on the platform disturbance component of the floating wind platform; The scheduling module is used to calculate the conflict probability of the predicted displacement trajectory in the path conflict area list based on the disturbance modeling; when the conflict probability is greater than the preset conflict threshold, real-time warning is performed through the edge computing node, and the crane boom operation is scheduled.

Images