Port mobile equipment operation risk early warning method and system based on risk potential field

Abstract

The port mobile equipment operation risk early warning method and system based on risk potential field belong to the field of port operation safety management and control, are realized based on an early warning system, and have the following steps: firstly, a three-dimensional risk potential field is constructed by taking the three-dimensional dynamic geometric contour of each operation element as a field source, and a scene three-dimensional risk potential field is constructed by taking a boundary obstacle as a field source; secondly, a plurality of sampling points are arranged on the three-dimensional dynamic geometric contour of the main mobile equipment, and then the RAI values of the main mobile equipment in each direction are calculated; thirdly, a prediction model is constructed to train time sequence data sets, and a mobile equipment RAI value prediction model is trained based on a long short-term memory network method; and finally, a section of real-time motion parameters of each operation element is input, and the risk level is judged according to the output main mobile equipment prediction RAI value. The present application can realize the subdivision risk quantification of eight directions in the three-dimensional space around the main mobile equipment, and improve the port mobile equipment operation safety and the wharf risk management and control ability.

Description

Technical Field

[0001] This invention belongs to the field of port operation safety management and control, and relates to a method and system for early warning of port mobile equipment operation risks based on risk potential energy field. Background Technology

[0002] As ports continue to advance high-quality development, improving operational safety has become a major focus, with the application of intelligent detection and monitoring technologies for transportation equipment playing a crucial supporting role. General-purpose terminals handle diverse cargo types and loading / unloading processes, employ complex equipment, and rely heavily on manual labor. Operations involve interactions within three-dimensional space, such as the lifting and lowering of loader booms and the raising and lowering of gantry crane grabs, creating a complex environment for risk perception. Loaders and heavy-duty trucks, common mobile equipment in general-purpose terminals, are less flexible than road vehicles, have larger blind spots, greater braking inertia, and drivers face harsh environmental interference such as dust and noise, posing significant operational safety hazards. Therefore, developing risk perception and early warning methods adapted to the characteristics of mobile equipment in general-purpose terminal operations is crucial for ensuring operational safety. This can help drivers obtain objective and effective risk level information and, in the long term, facilitate the transformation towards unmanned equipment and terminal automation.

[0003] Chinese invention patent CN113096440A addresses the operational safety monitoring needs in complex port environments by proposing an operational safety early warning method based on BeiDou high-precision positioning and ultra-wideband (UWB) ranging technology. This method acquires the real-time BeiDou latitude and longitude coordinates of mobile machinery, fixed equipment, and personnel, as well as the relative distances between these objects obtained through UWB pulse signals. It then dynamically predicts whether trajectories intersect based on movement trends, thereby determining the existence of potential safety hazards. However, this patent relies solely on relative distance to assess risk, lacking consideration for mechanical dimensions and motion states, and is only applicable to assessments within a two-dimensional plane.

[0004] Chinese invention patent CN120599866A proposes a collision avoidance warning method for slow-moving vehicles based on a risk field model. This method comprehensively considers stationary objects, road markings, non-motorized vehicles, and pedestrians to construct a comprehensive risk field model. Based on field force vector analysis, it forms a conflict assessment index, which can be used for risk warning of non-motorized vehicles. This patent incorporates multiple types of subjects into the comprehensive risk assessment and improves the accuracy of risk assessment by considering factors such as steering, braking, and speed. Its approach is suitable for risk assessment in scenarios involving multiple equipment and personnel, such as dock operations.

[0005] Existing methods rely on cameras, sensors, and positioning systems to conduct risk assessments based on obstacle distance. However, they fail to comprehensively incorporate multiple factors such as distance, speed, and motion state into the risk assessment, thus failing to effectively meet the risk monitoring needs of large mobile equipment in the complex operating environment of general-purpose terminals. While the risk field theory in road traffic comprehensively considers scene factors and multiple equipment and personnel factors to assess driving risks, it is rarely applied to terminal operation scenarios. To improve the risk assessment capabilities of port mobile equipment operations and achieve effective risk early warning, this invention proposes a risk early warning method and system for port mobile equipment operations based on a three-dimensional risk potential energy field. This helps improve the safety of loader and heavy-duty truck drivers and promotes the automation transformation of general-purpose terminals. Summary of the Invention

[0006] This invention addresses the interactive characteristics of equipment within the three-dimensional space of port operations by providing a risk early warning method and system for port mobile equipment operations based on a risk potential energy field. It constructs a three-dimensional risk potential energy field model considering the risk impact of operational elements and boundaries in three-dimensional space. Based on the three-dimensional dynamic geometric contour of the mobile equipment, it calculates the operational risk levels in multiple directions around it and achieves short-term prediction of operational risk values ​​based on collected motion parameters. This provides port mobile equipment operators with an objective and effective method and system for risk quantification and early warning, contributing to improved safety in port mobile equipment operations and enhanced port risk management capabilities.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A risk warning method for port mobile equipment operations based on a risk potential energy field, comprising the following steps: The first step is to construct a three-dimensional dynamic geometric profile for each operational element, and then use this profile as the field source to construct the three-dimensional risk potential energy field for each operational element. Then, a three-dimensional risk potential energy field for the scene is constructed using boundary obstacles as the field source. Specifically: Step 1.1: Construct the three-dimensional dynamic geometric contours of each operational element, which includes loaders, heavy-duty trucks, gantry crane grabs, and workers.

[0008] Step 1.1.1: Construct the three-dimensional dynamic geometric profile of the loader.

[0009] Based on the loader's articulation center position, vehicle length, speed, steering angle, lateral safety redundancy, and minimum safety distance, the loader's two-dimensional dynamic geometric profile is determined. It consists of two arc segments centered at the articulation center (front and rear) and four arc segments on the left and right sides. Both the front and rear arc segments are centered at the articulation center. The front and rear arc segments are further divided into left and right arc segments, respectively, based on the orientation of the front and rear sections of the loader's body. The central angles of the four arc segments formed are calculated for each. Taking the left arc segment as an example, its central angle is defined by the following formula: (1) (2) In the formula, and These are the central angles of the left side segments of the front and back arc segments, respectively; It is the velocity profile offset coefficient; and These are the forward and rearward lengths of the loader's body, respectively, from the articulation center. and These represent the radii of the arc segments in the forward and backward directions, respectively. It is a lateral safety redundancy for the loader; This is the minimum safe distance for loader operation; and These are the turning angles of the left-side arc segments in both the forward and backward directions.

[0010] The two arc segments form four endpoints, extending from the loader's hinge center to the left and right respectively along a direction perpendicular to the front of the vehicle. The distance between them forms two other endpoints. From these two endpoints, the endpoints of the front and rear arc segments on each side are connected to form four line segments on the left and right sides of the two-dimensional dynamic geometric contour of the loader.

[0011] A three-dimensional dynamic geometric profile of the loader is constructed by adding additional endpoints to the existing two-dimensional dynamic geometric profile. The two-dimensional dynamic geometric profile is divided into a front profile and a rear profile, each containing one arc segment and two line segments. The front profile is shifted upwards by the maximum height of the loader bucket plus a safety margin above the loader. The rear profile is shifted upwards by the height of the rear section of the loader body plus a safety margin above the loader. This yields the upper front profile, upper rear profile, and upper endpoints of the loader's three-dimensional dynamic geometric profile. Connecting these upper endpoints to the endpoints with the same two-dimensional coordinates in the loader's two-dimensional dynamic geometric profile completes the loader's three-dimensional dynamic geometric profile.

[0012] Step 1.1.2: Construct the three-dimensional dynamic geometric profile of the heavy-duty truck.

[0013] The heavy-duty truck body is simplified into a cuboid, with the connection point between the cab and the cargo box as the center point. First, based on the center point, body length, width, speed, and safety margin, the two-dimensional dynamic geometric profile of the heavy-duty truck is determined. Then, safety margins are added to the base of the simplified cuboid to form a rectangular two-dimensional dynamic geometric profile. The distances from the center point to the front and rear sides of the two-dimensional dynamic geometric profile are calculated in the same way, as are the distances from the center point to the left and right sides. Taking forward movement as an example, the formulas for calculating the distances from the center point to the front and left sides of the two-dimensional dynamic geometric profile are as follows: (3) (4) In the formula, and These are the distances from the center point to the front and left sides of the two-dimensional dynamic geometric contour of the heavy-duty truck, respectively. This refers to the length of the front of the vehicle. This refers to the width of the vehicle body; and These are the basic safety redundancies for the front and left sides of the truck, respectively. and These are the linear velocity profile offset coefficients for the front and left sides, respectively. This represents the speed of the truck.

[0014] The three-dimensional dynamic geometric profile of a heavy-duty truck is a cuboid, with the rectangle of its two-dimensional dynamic geometric profile as the base. The height is the height of the heavy-duty truck body plus the safety redundancy above the heavy-duty truck.

[0015] Step 1.1.3: Construct the three-dimensional dynamic geometric profile of the gantry crane grab bucket.

[0016] The gantry crane grab bucket is connected to the gantry crane body by a suspension rope. A crossbeam is positioned at the connection point, and a connecting rod extending from the crossbeam controls the opening and closing of the two grab bucket segments. The hinge point of the two segments is located directly below the connection point. The grab bucket segments are simplified as triangular prisms with equilateral triangles as their bases. First, the two-dimensional dynamic geometric profile of the gantry crane grab bucket is determined on the longitudinal section perpendicular to the crossbeam at the connection point. Its shape is a curved trapezoid with two line segments on the sides, a line segment on the upper base, and a circular arc on the lower base. The connection point is the midpoint of the upper base of the curved trapezoid. From this point, a longitudinal safety redundancy is extended horizontally to both sides on the longitudinal section perpendicular to the crossbeam, obtaining the two endpoints of the upper base of the curved trapezoid. The circular arc-shaped lower base of the curved trapezoid is formed within the longitudinal section with the connection point as its center. The central angle is positively correlated with the opening angle of the gantry crane grab bucket. The radius is calculated using the following formula: (5) In the formula, The radius of the arc segment; The opening angle of the gantry crane grab; The height difference between the connection point of the gantry crane grab bucket and the hoisting rope and the hinge point of the two bucket segments of the gantry crane grab bucket; The side length of a single canthus; The speed influence coefficient of the gantry crane grab bucket; This refers to the speed of the gantry crane's grab bucket.

[0017] The three-dimensional dynamic geometric profile of the gantry crane grab bucket is a geometric body formed by extending its two-dimensional dynamic geometric profile along the straight line of the crossbeam. The horizontal safety redundancy distance extends from the connection point along the straight line of the crossbeam to both sides.

[0018] Step 1.1.4: Construct the three-dimensional dynamic geometric contours of the staff.

[0019] The 3D dynamic geometric profile of the worker includes cylindrical and elliptical cylinders, with a fixed height of 3 meters. In the stationary worker's 3D dynamic geometric profile, the base is a circle centered on the worker's position, and the radius of the base is set to the spatial influence length along the worker's direction of movement. In the three-dimensional dynamic geometric profile of the moving worker, the base is an ellipse with the worker's position as the focus and the direction of movement pointing towards the center of the ellipse. The length of the minor semi-axis of the base ellipse is the safe distance of the worker. The length of the major semi-axis is the spatial influence length in the direction of the worker's movement. The formula for calculating the spatial influence length in the direction of staff movement is as follows: (6) In the formula, Maintain a safe distance for staff; The speed at which the staff moves; The influence coefficient of staff speed; The spatial influence length in the direction of the worker's movement is taken as the worker's safe distance when the worker is stationary. It serves as the radius of the bottom circle of the three-dimensional dynamic geometric profile and as the length of the major semi-axis of the bottom ellipse of the three-dimensional dynamic geometric profile when the worker is moving.

[0020] Step 1.2: Based on the three-dimensional dynamic geometric contours of each task element, construct the three-dimensional risk potential energy field for each task element. .

[0021] A three-dimensional risk potential field is a physical field that characterizes the degree of influence of the state of various elements and environmental parameters in a scenario on operational risks. Within this physical field, any point in three-dimensional space... The strength of the three-dimensional risk potential energy field directly reflects the risk level experienced by a unit mass object at that point. Considering the three-dimensional dynamic geometric contour of a single operational element as the source of its three-dimensional risk potential energy field, the resulting three-dimensional risk potential energy field of the operational element... Points outside the three-dimensional dynamic geometric contour of the work element in three-dimensional space The formula for calculating the electric field strength is as follows: (7) In the formula, Used to distinguish different types of task elements; Indicates the quality of the task element; Indicates factors affecting road conditions; This is the quality adjustment factor for the work element; Indicates the speed of movement of the task element; This represents the angle between the line connecting the midpoint of the space to the center of the work element and the direction of movement of the work element; Represents job elements Distance point on the three-dimensional dynamic geometric contour The nearest point-to-point The distance vector, the direction of the field strength and same, This represents the magnitude of the distance vector. This is the adjustment factor for the shortest distance from the point to the task element; Adjust the movement direction coefficient for the work element.

[0022] The three-dimensional risk potential energy field of each task element within its three-dimensional dynamic geometric contour. The field strength values ​​are all 1.

[0023] Step 1.3: Construct a three-dimensional risk potential energy field for the scene. ; Boundary obstacles in the dock operation scenario are considered as the source of the scenario's three-dimensional risk potential energy field, and the resulting scenario's three-dimensional risk potential energy field At a point in three-dimensional space The formula for calculating the electric field strength is as follows: (8) In the formula, Used to distinguish different types of boundary obstacles; Indicates the mass of the boundary obstacle; This is the mass adjustment factor for boundary obstacles; Indicates factors affecting road conditions; Indicates boundary obstacles Upper distance point The nearest point-to-point The distance vector, the direction of the field strength and same, This represents the magnitude of the distance vector; For point Adjustment factor for the shortest distance to the boundary obstacle.

[0024] The second step involves setting multiple sampling points on the three-dimensional dynamic geometric contour of the main mobile equipment, calculating the single-point operational risk assessment index (RAI) value for each sampling point, and then calculating the RAI values ​​for each direction of the main mobile equipment. The main mobile equipment includes loaders and heavy-duty trucks; specifically: Step 2.1: Set sampling points on the three-dimensional dynamic geometric profiles of the loader and heavy-duty truck.

[0025] Thirty-two sampling points are set on the three-dimensional dynamic geometric contour of the loader. There are 16 sampling points on the bottom surface of the three-dimensional dynamic geometric contour, including three quarter points on the front and rear arc segments, and the endpoints and midpoints of the four line segments on the left and right sides. The endpoint with the largest height coordinate value that is the same as the two-dimensional coordinate of these 16 sampling points is found on the three-dimensional dynamic geometric contour of the loader, which is the other 16 sampling points on the three-dimensional dynamic geometric contour of the loader.

[0026] Sixteen sampling points were set on the three-dimensional dynamic geometric contour of the heavy-duty truck, including eight vertices of the three-dimensional dynamic geometric contour cuboid and eight trisection points along the four long sides of the truck's length.

[0027] Step 2.2: Based on the combined influence of the three-dimensional risk potential energy fields of each operating element around the main mobile equipment and the three-dimensional risk potential energy field of the scene, calculate the risk assessment indicator (RAI value) for all sampling points confirmed in Step 2.1.

[0028] Similar to the force exerted on an object in a physical field, the main flow equipment also experiences a repulsive force in a three-dimensional risk potential energy field, representing the risk posed by obstacles to the operation of the main flow equipment. Located within the main flow equipment... Points on the three-dimensional dynamic geometric contour Obstacles The formula for calculating the applied field force is: (9) In the formula, Indicates obstacles Points on the three-dimensional dynamic geometric contour of the main flow equipment The field strength of the three-dimensional risk potential energy field generated at the location; Indicates mobile equipment The quality; Indicates the main flow equipment and obstacles The relative velocity relationship factor between them.

[0029] Field forces can cancel each other out in real physical fields. However, for a certain main mobile device in a scenario, the operational risks represented by field forces with different directions generated by obstacles in different directions cannot be canceled out. Therefore, a superimposed scalar is needed to characterize the magnitude of the risk of the main mobile device.

[0030] Three-dimensional safety potential field energy (SPFE) is the energy possessed by an object due to the work done by conservative field forces acting on it in a three-dimensional safety potential field. It includes the three-dimensional safety potential field energy of the working element and the three-dimensional safety potential field energy of the scene. (Main moving equipment) Points on the three-dimensional dynamic geometric contour The formula for calculating the potential energy of the three-dimensional risk field is: (10) (11) In the formula, The number of operational elements that pose a risk to the main mobile equipment; Main mobile equipment Points on the three-dimensional dynamic geometric contour To the The distance between the centers of each task element; Main mobile equipment Points on the three-dimensional dynamic geometric contour Subject to the The field force exerted by each working element; The three-dimensional risk field potential energy of the work elements under the influence of the three-dimensional risk potential energy field of multiple work elements. Main mobile equipment Points on the three-dimensional dynamic geometric contour To the nearest boundary obstacle The distance; Main mobile equipment Points on the three-dimensional dynamic geometric contour Subject to the nearest boundary obstacle The field force acting; The potential energy of the three-dimensional risk field in the scene under the influence of the three-dimensional risk potential energy field.

[0031] The operational risk assessment index (RAI) value of a single sampling point on the three-dimensional dynamic geometric contour of the main mobile equipment is calculated using the following formula: (12) In the formula, and These represent the weights of the three-dimensional risk field potential energy of the task element and the three-dimensional risk field potential energy of the scene, respectively.

[0032] Step 2.3: Group all sampling points on the three-dimensional dynamic geometric contour of the main flow equipment according to direction, and calculate the operational RAI value of the main flow equipment in each direction.

[0033] The RAI values ​​of multiple sampling points in step 2.1 are calculated according to the method in step 2.2. The sampling points are divided into eight groups according to their location: front, left front, right front, front upper, rear, left rear, right rear, and rear upper. The RAI values ​​of all sampling points in each group are averaged to obtain the working RAI values ​​of the main flow equipment in eight directions.

[0034] The third step involves constructing a time-series dataset for predictive model training. A sliding window is used to construct the training sample set from this dataset, and the mobile device RAI value prediction model is trained using the Long Short-Term Memory (LSTM) network method. Specifically: Step 3.1: Construct a time series dataset for training the prediction model.

[0035] Temporal prediction utilizes mathematical models to uncover the temporal evolution patterns of historical data, enabling forward-looking inferences about data values ​​at specific future times. Therefore, it requires constructing long-term temporal sequence datasets. In a dock operation scenario, motion parameters of all operational elements are collected at a frequency of 0.2 seconds. These parameters include the 2D coordinates of the loader's center point, the angle between the boom and the horizontal direction, the angle between the bucket's lower edge and the horizontal direction, the orientation angle, and the magnitude and direction of its velocity; the 2D coordinates of the heavy-duty truck's center point, its height, orientation angle, and the magnitude and direction of its velocity; the 3D coordinates of the gantry crane's grab bucket and the connection point with the hoisting rope, its velocity, and its opening angle; and the 2D coordinates, velocity, and orientation angle of the worker's location. Furthermore, the 3D coordinates of all boundary obstacles are collected and updated every minute. The original temporal sequence dataset consists of the motion parameters of all operational elements and the 3D coordinates of boundary obstacles for each frame.

[0036] Since the operational elements frequently enter and exit the dock operation scene, the number of motion parameters in each frame of the original time series dataset is not fixed. The data of each frame in the original time series dataset is converted into general feature values ​​describing the interaction between the main mobile equipment and the surrounding environment. The general features include the speed of the main equipment, the speed of the nearest obstacle, the speed of the second nearest obstacle, the distance of the nearest obstacle, the relative angle of the nearest obstacle, the relative height of the nearest obstacle, the type of the nearest obstacle, the distance of the second nearest obstacle, and the relative speed of the second nearest obstacle, forming the processed time series dataset.

[0037] Simultaneously, following the method described in the second step above, the eight-directional RAI value of the main mobile device in each frame is calculated. The general feature value and the eight-directional RAI value are then combined to form a single-frame data stamp.

[0038] Finally, all data stamps are arranged in chronological order, forming a structure of length [length missing]. The prediction model is trained on a time series dataset at each time step.

[0039] Step 3.2: Construct a training sample set for the prediction model.

[0040] For the time-series dataset used to train the prediction model formed in step 3.1, a sliding window technique is employed to construct samples. A fixed-length window slides across the time series, segmenting the continuous long sequence into a series of subsequences. In this process, two key parameters need to be set: the prediction window length and the time series data set. and predicted lead time The prediction window length is the same as the historical window length, which is the length of historical information used by the model for prediction. The prediction lead length is the prediction lead amount, which is the number of future time steps the model predicts for the target.

[0041] For a common Each time step includes One general feature and The dataset contains n prediction targets, and the sliding window moves from time step 0 to time step 1. Each time step moves one time step at a time, and each move forms a training sample for the deep learning model. Each sample contains... Frame data stamps. From time step 0 to time step 1 in each sample. The general feature values ​​of each time step constitute Using a dimensional matrix as the input sequence, select the ... The target sequence is formed by the eight-axis RAI values ​​at each time step.

[0042] The final form includes The sample set of a sample is divided into a training set, a validation set, and a test set according to the proportions of 70%, 15%, and 15%, respectively.

[0043] Step 3.3: Train the RAI value prediction model for mobile equipment based on the LSTM method.

[0044] Long Short-Term Memory (LSTM) networks are a common method for temporal prediction in deep learning, often abbreviated as LSTM prediction models. An LSTM prediction model consists of two stacked LSTM layers and an output projection layer; each LSTM layer contains 256 hidden units. From the training set obtained in step 3.2, the LSTM prediction model extracts 64 samples at a time as a batch for parallel processing. The three-dimensional tensor is input into the first LSTM layer, which is traversed... That moment will The general feature mapping of dimension 1 is a 256-dimensional hidden state sequence containing richer information, with an output dimension of 1. The first LSTM layer receives the 3D tensor from the first LSTM layer and reduces its dimension to 1. It includes the past The feature vectors representing key information at each time step are fed into the output projection layer. The output projection layer is a fully connected network that projects the feature vectors of dimension 1. The feature vectors are mapped to the 8-dimensional main body flow device to predict the RAI value, and the output shape is... eigenvectors.

[0045] The LSTM prediction model updates its weight parameters after processing each batch of samples in the training set. One round of training is completed by dividing the training set into several batches and processing each batch separately. After each training round, the LSTM prediction model calculates the error between the predicted and true values ​​on the validation set samples and fine-tunes the weight parameters in the LSTM and fully connected layers. After 80 rounds of training, the predictive performance of the LSTM prediction model is evaluated by calculating the root mean square error, mean absolute error, and coefficient of determination on the test set.

[0046] Once the LSTM prediction model is trained, it will be used as the RAI value prediction model for mobile devices.

[0047] The fourth step is to input a segment of real-time motion parameters of each operational element into the mobile equipment RAI value model, determine the risk level at the predicted time based on the output main mobile equipment predicted RAI value, and execute the corresponding early warning command.

[0048] The mobile equipment RAI value prediction model is deployed to the actual working environment. The real-time motion parameters of the main mobile equipment are processed to obtain general features, which are then input into the mobile equipment RAI value prediction model. The predicted RAI value output by the mobile equipment RAI value prediction model will be used as the basis for risk assessment.

[0049] The mobile equipment is classified into different risk levels based on the magnitude of the predicted RAI value. The risk levels include low risk, medium risk, and high risk. When the predicted RAI value in all eight directions does not exceed 0.4, the mobile equipment is in the low risk level; when the predicted RAI value in any direction is greater than 0.4 but not more than 0.9, the mobile equipment is in the medium risk level; when the predicted RAI value in any direction is greater than 0.9, the mobile equipment is in the high risk level.

[0050] When the mobile equipment is at a low risk level, no risk warning signal is generated, and the mobile equipment alarm device is not triggered.

[0051] When the mobile equipment is at a medium-risk level, the mobile equipment alarm will emit two consecutive short beeps, and the risk information display screen will show the direction with the highest predicted RAI value and flash continuously.

[0052] When the mobile equipment is at a high risk level, the mobile equipment alarm will continuously beep, the risk information display screen will show the direction with the highest predicted RAI value and continuously flash, and the in-vehicle broadcast will issue a voice prompt "Please be aware of the risk in a certain direction".

[0053] A port mobile equipment operation risk early warning system based on a three-dimensional risk potential energy field is used to implement the aforementioned port mobile equipment operation risk early warning method. The port mobile equipment operation risk early warning system includes a data acquisition unit, a central processing unit, and an early warning execution unit. The central processing unit includes a storage module, a calculation module, and a response module. Specifically: The data acquisition units are distributed across all operational elements and boundary obstacles in the port, and are responsible for collecting raw time-series datasets at fixed time intervals.

[0054] The storage unit of the central processing unit is responsible for storing the original time-series dataset and the RAI value time-series dataset; the calculation module performs two functions: first, it reads the original time-series dataset and calculates the multi-directional RAI values ​​of the main flow device; second, it uses the prediction model to train the LSTM prediction model using the time-series dataset; the response module saves the LSTM prediction model trained by the calculation module, reads a segment of real-time acquired original time-series dataset from the storage module, processes it, outputs the predicted RAI value, and determines whether to output risk warning information.

[0055] The aforementioned early warning execution unit is installed in the driver's cab of the main mobile equipment and executes early warning commands based on the risk early warning information provided by the central processing unit response module.

[0056] To realize the various functions of the port mobile equipment operation risk early warning system based on a three-dimensional risk potential energy field, the following devices are required: The data acquisition unit includes: a high-precision positioning sensor and an inertial measurement unit installed on the mobile equipment vehicle; a winch system drum encoder and a hydraulic cylinder built-in displacement sensor installed on the gantry crane; a personnel positioning beacon installed on the worker's safety helmet; and a lidar, millimeter-wave radar, and high-definition camera installed in the scene.

[0057] The central processing unit includes: a database server for the storage module; a multi-core CPU, a graphics processing unit (GPU), a risk value calculation engine based on a three-dimensional risk potential field, and a PyTorch-based deep learning training framework for the computing module; and a real-time inference engine for the response module.

[0058] The aforementioned early warning execution unit includes: an in-vehicle communication device, a signal transmission module, an in-vehicle domain controller, a physical alarm device, and a risk information display screen.

[0059] The beneficial effects of this invention are: First, this invention constructs a three-dimensional risk potential energy field for operational elements and a three-dimensional risk potential energy field for the scene in a three-dimensional space operation scenario at the dock. It comprehensively considers factors such as the attributes of operational elements and boundary obstacles, the relative distance and speed between obstacles and the main mobile equipment, etc., to calculate the RAI value of the main mobile equipment to assess operational risks. It can achieve detailed risk quantification in eight directions in the three-dimensional space around the main mobile equipment.

[0060] Second, this invention transforms a number of variable motion parameters into general features to describe the interaction between the main mobile equipment and the surrounding environment, and further constructs a time-series dataset to train an LSTM prediction model. By outputting the predicted RAI value of the mobile equipment, combined with the risk level setting, it provides early warning of risks to the mobile equipment driver. Attached Figure Description

[0061] Figure 1 This is a schematic diagram illustrating the implementation steps of the method described in this invention; Figure 2 A simplified three-dimensional dynamic geometric contour diagram of the loader's physical frame and under different motion states; Figure 2 (a) and Figure 2 (b) in the figure represents the three-dimensional dynamic geometric profile of the loader when it moves forward at speeds of 1 m / s and 2 m / s, respectively. Figure 2 (c) and Figure 2 In the figure, (d) represents the three-dimensional dynamic geometric profile of the loader when it is moving forward without steering and when it is moving forward and turning right, respectively. Figure 2 (e) and Figure 2 (f) in the figure represents the three-dimensional dynamic geometric profile of the loader when it is stationary with the boom pointing down and when it is stationary with the boom pointing up, respectively; Figure 3 A simplified framework and a schematic diagram of the three-dimensional dynamic geometric contours of a heavy-duty truck under different motion states; Figure 3 (a) and Figure 3 (b) in the figure represents the three-dimensional dynamic geometric profile of the heavy-duty truck when it is stationary and when it is moving forward, respectively; Figure 4 A schematic diagram of the actual gantry crane grab bucket and its three-dimensional dynamic geometric outline; Figure 4 (a) in the image is the actual gantry crane grab bucket. Figure 4 (b) in the figure represents the three-dimensional dynamic geometric profile of the gantry crane grab bucket when the opening angle is 90 degrees; Figure 5 A schematic diagram of the three-dimensional dynamic geometric contour of the staff under different motion states; Figure 5 (a) and Figure 5 (b) in the figure represents the three-dimensional dynamic geometric profile of the worker when he is stationary and when his speed is 1 m / s, respectively. Figure 6 A schematic diagram of the three-dimensional risk potential energy field strength distribution of the loader at different heights when the boom is raised from a stationary position; Figure 6 (a) Figure 6 (b) and Figure 6 (c) in the figure represents the field strength distribution of the three-dimensional risk potential energy field of the loader at heights of 0 meters, 5 meters, and 9 meters, respectively; Figure 7 A schematic diagram of the three-dimensional risk potential energy field strength distribution at different altitudes; Figure 7 (a) Figure 7 (b) and Figure 7 (c) in the figure represents the field strength distribution of the three-dimensional risk potential energy field of the scene at heights of 0 meters, 2 meters, and 4 meters, respectively; Figure 8 A schematic diagram showing the distribution of sampling points on the simplified body and three-dimensional dynamic geometric contour of a loader; Figure 8 (a) and Figure 8 (b) in the figure are a three-dimensional dynamic geometric profile of the loader and a top view, respectively. The red dots and numbers in the figure represent sampling points. Figure 9 A schematic diagram showing the distribution of sampling points on the simplified body and three-dimensional dynamic geometric contour of a heavy-duty truck; Figure 9 (a) and Figure 9 (a) shows the three-dimensional dynamic geometric profile of the heavy-duty truck in a stereo view and a top view, respectively. The red dots and numbers in the figure represent sampling points.

[0062] Figure 4 In the middle: 1. Connection point, 2. Crossbeam, 3. Hinge point, 4. Linkage rod, 5. Bucket petal, 6. Height difference between the connection point of the gantry crane grab bucket and the hoisting rope and the hinge point of the two bucket petals of the gantry crane grab bucket, 7. Side length of a single bucket petal. Detailed Implementation

[0063] The implementation details of this invention will be described in more detail below with reference to specific implementation examples and accompanying drawings.

[0064] Figure 1 The implementation steps of a risk warning method for port mobile equipment operations are demonstrated. The specific steps are explained below: The first step is to construct a three-dimensional dynamic geometric profile for each operational element, and then use this profile as the field source to construct the three-dimensional risk potential energy field for each operational element. Then, a three-dimensional risk potential energy field for the scene is constructed using boundary obstacles as the field source. .

[0065] Step 1.1: Construct the three-dimensional dynamic geometric contours of each operational element, which includes loaders, heavy-duty trucks, gantry crane grabs, and workers.

[0066] Step 1.1.1: Construct the three-dimensional dynamic geometric profile of the loader.

[0067] Based on the loader's articulation center position, vehicle length, speed, steering angle, lateral safety redundancy, and minimum safety distance, the loader's two-dimensional dynamic geometric profile is determined. It consists of two arc segments centered at the articulation center (front and rear) and four line segments on the left and right sides. Both the front and rear arc segments are centered at the articulation center. The front and rear arc segments are further divided into left and right arc segments, respectively, based on the orientation of the front and rear sections of the loader's body. The central angles of the four arc segments formed are calculated for each. Taking the left arc segment as an example, its central angle is defined by the following formula: (1) (2) In the formula and These are the central angles of the left-side arc segments, one forward and one backward; It is the velocity profile offset coefficient; and These are the forward and rearward lengths of the loader's body, respectively, from the articulation center. and These represent the radii of the arc segments in the forward and backward directions, respectively. It is the lateral safety redundancy of the loader, which increases with the increase of speed, and the value ranges from 0.2 to 0.5 meters; This is the minimum safe distance for loader operation; and These are the turning angles of the left-side arc segments in both the forward and backward directions.

[0068] The two arc segments form four endpoints, extending from the loader's hinge center to the left and right respectively along a direction perpendicular to the front of the vehicle. The distance between them forms two more endpoints. From these two endpoints, the endpoints of the front and rear arc segments on each side are connected to form four line segments on the left and right sides of the two-dimensional dynamic geometric contour.

[0069] A three-dimensional dynamic geometric profile of the loader is constructed by adding additional endpoints to the existing two-dimensional dynamic geometric profile. The two-dimensional dynamic geometric profile is divided into a front profile and a rear profile, each containing one arc segment and two line segments. The front profile is shifted upwards by the maximum height of the loader bucket plus a safety redundancy above the loader. The rear profile is shifted upwards by the height of the rear section of the loader body plus a safety redundancy above the loader. This yields the upper front profile, upper rear profile, and upper endpoints of the three-dimensional dynamic geometric profile of the loader. The safety redundancy above the loader is set to 0.3 meters. Connecting the upper endpoints to the endpoints with the same two-dimensional coordinates in the two-dimensional dynamic geometric profile of the loader yields the three-dimensional dynamic geometric profile of the loader.

[0070] Figure 2 The three-dimensional dynamic geometric profile of the loader under different motion states is shown. When the speed is higher, the loader has a greater lateral safety redundancy, and the radius and central angle of the front and rear arc segments are also larger. When turning, both the front and rear profiles rotate around the hinge point. When the boom is raised, the front section of the three-dimensional dynamic geometric profile has a greater height. Figure 2 (a) and Figure 2 (b) in the figure represents the three-dimensional dynamic geometric profile of the loader when it moves forward at speeds of 1 m / s and 2 m / s, respectively. Figure 2 (c) and Figure 2 In the figure, (d) represents the three-dimensional dynamic geometric profile of the loader when it is moving forward without steering and when it is moving forward and turning right, respectively. Figure 2 (e) and Figure 2 In the figure, (f) represents the three-dimensional dynamic geometric profile of the loader when it is stationary with the boom pointing down and when it is stationary with the boom pointing up.

[0071] Step 1.1.2: Construct the three-dimensional dynamic geometric profile of the heavy-duty truck.

[0072] The heavy-duty truck body is simplified into a cuboid, with the connection point between the cab and the cargo box as the center point. First, based on the center point, body length, width, speed, and safety margin, the two-dimensional dynamic geometric profile of the heavy-duty truck is determined. Then, safety margins are added to the base of the simplified cuboid to form a rectangular two-dimensional dynamic geometric profile. The distances from the center point to the front and rear sides of the two-dimensional dynamic geometric profile are calculated in the same way, as are the distances from the center point to the left and right sides. Taking forward movement as an example, the formulas for calculating the distances from the center point to the front and left sides of the two-dimensional dynamic geometric profile are as follows: (3) (4) In the formula and These are the distances from the center point to the front and left sides of the two-dimensional dynamic geometric contour of the heavy-duty truck, respectively. This refers to the length of the front of the vehicle. This refers to the width of the vehicle body; and The basic safety redundancy for the front and left sides of the truck is 0.4 meters for both. and These are the linear velocity profile offset coefficients for the front and left sides, respectively. This represents the speed of the truck.

[0073] The three-dimensional dynamic geometric profile of a heavy-duty truck is a cuboid, with the rectangle of its two-dimensional dynamic geometric profile as the base. The height is the height of the heavy-duty truck body plus the upper safety redundancy of the heavy-duty truck, which is set to 0.3 meters.

[0074] Figure 3 It shows the three-dimensional dynamic geometric profile of a heavy-duty truck when it is stationary and moving forward. When it is moving, the three-dimensional dynamic geometric profile has a larger range in space. Figure 3 (a) and Figure 3 (b) in the figure represents the three-dimensional dynamic geometric profile of the heavy-duty truck when it is stationary and when it is moving forward.

[0075] Step 1.1.3: Construct the three-dimensional dynamic geometric profile of the gantry crane grab bucket.

[0076] The gantry crane grab bucket is connected to the gantry crane body by a suspension rope. A crossbeam is positioned at the connection point, and a connecting rod extending from the crossbeam controls the opening and closing of the two grab bucket segments. The hinge point of the two segments is located directly below the connection point. The grab bucket segments are simplified as triangular prisms with equilateral triangles as their bases. First, the two-dimensional dynamic geometric profile of the gantry crane grab bucket is determined on the longitudinal section perpendicular to the crossbeam at the connection point. Its shape is a curved trapezoid with two line segments on the sides, a line segment on the upper base, and a circular arc on the lower base. The connection point is the midpoint of the upper base of the curved trapezoid. From this point, a longitudinal safety redundancy is extended horizontally to both sides on the longitudinal section perpendicular to the crossbeam, obtaining the two endpoints of the upper base of the curved trapezoid. This longitudinal safety redundancy is related to the size and speed of the grab bucket segments. A circular arc-shaped lower base of the curved trapezoid is formed within the longitudinal section with the connection point as its center. The central angle is positively correlated with the opening angle of the grab bucket. The radius is calculated using the following formula: (5) In the formula, The radius of the arc segment; The opening angle of the gantry crane grab; The height difference between the connection point of the gantry crane grab bucket and the hoisting rope and the hinge point of the two bucket segments of the gantry crane grab bucket; The side length of a single canthus; The speed influence coefficient of the gantry crane grab bucket; This refers to the speed of the gantry crane's grab bucket.

[0077] Figure 4 The exhibit showcases a physical gantry crane grab bucket and its three-dimensional dynamic geometric profile when the bucket is open at a 90-degree angle. Figure 4 (a) in the image is the actual gantry crane grab bucket. Figure 4 (b) shows the three-dimensional dynamic geometric profile of the gantry crane grab when its opening angle is 90 degrees. The three-dimensional dynamic geometric profile of the gantry crane grab is a geometric body formed by extending its two-dimensional dynamic geometric profile along the straight line of the crossbeam. Lateral safety redundancy is extended to both sides from the connection point along the straight line of the crossbeam. The length of the lateral safety redundancy on one side is 0.4 meters more than half the length of the crossbeam of the gantry crane grab, and it increases with the increase of the gantry crane grab's movement speed.

[0078] Step 1.1.4: Construct the three-dimensional dynamic geometric contours of the staff.

[0079] The three-dimensional dynamic geometric profile of the staff includes cylindrical and elliptical cylinders, with the height of the cylinders fixed at 3 meters. Figure 5 (a) and Figure 5(b) represents the three-dimensional dynamic geometric contours of the worker when stationary and when the speed is 1 m / s, respectively. Specifically, the three-dimensional dynamic geometric contours of the worker under different motion states are given. In the three-dimensional dynamic geometric contour of the stationary worker, the base is a circle centered on the worker's position, and the radius of the base is set to the spatial influence length in the direction of the worker's movement. In the three-dimensional dynamic geometric profile of the moving worker, the base is an ellipse with the worker's position as the focus and the direction of movement pointing towards the center of the ellipse. The length of the minor semi-axis of the base ellipse is the safe distance of the worker. The length of the major semi-axis is the spatial influence length in the direction of the worker's movement. The formula for calculating the spatial influence length in the direction of staff movement is as follows: (6) In the formula Maintain a safe distance for staff; The speed at which the staff moves; The influence coefficient of staff speed; The spatial influence length in the direction of the worker's movement is taken as the worker's safe distance when the worker is stationary. It serves as the radius of the bottom circle of the three-dimensional dynamic geometric profile and as the length of the major semi-axis of the bottom ellipse of the three-dimensional dynamic geometric profile when the worker is moving.

[0080] Step 1.2: Based on the three-dimensional dynamic geometric contours of each task element, construct the three-dimensional risk potential energy field for each task element. .

[0081] A three-dimensional risk potential field is a physical field that characterizes the degree of influence of the state of various elements and environmental parameters in a scenario on operational risks. Within this physical field, any point in three-dimensional space... The strength of the three-dimensional risk potential energy field directly reflects the risk level experienced by a unit mass object at that point. Considering the three-dimensional dynamic geometric contour of a single operational element as the source of its three-dimensional risk potential energy field, the resulting three-dimensional risk potential energy field of the operational element... Points outside the three-dimensional dynamic geometric contour of the work element in three-dimensional space The formula for calculating the electric field strength is as follows: (7) In the formula Used to distinguish different types of task elements; Indicates the quality of the task element; Indicates factors affecting road conditions; This is the quality adjustment factor for the work element; Indicates the speed of movement of the task element; This represents the angle between the line connecting the midpoint of the space to the center of the work element and the direction of movement of the work element; Represents job elements Distance point on the three-dimensional dynamic geometric contour The nearest point-to-point The distance vector, the direction of the field strength and same, This represents the magnitude of the distance vector. This is the adjustment factor for the shortest distance from the point to the task element; Adjust the movement direction coefficient for the work element.

[0082] The three-dimensional risk potential energy field of each task element within its three-dimensional dynamic geometric contour. The field strength values ​​are all 1.

[0083] Figure 6 (a) Figure 6 (b) and Figure 6 In the figure, (c) represents the field strength distribution of the loader's three-dimensional risk potential energy field at heights of 0 meters, 5 meters, and 9 meters. As the height increases, the cross-section of the three-dimensional dynamic geometric profile changes from two parts, the front and the rear, to only the front section, and then to no cross-section. At the height where no cross-section is formed, the field strength value of the three-dimensional risk potential energy field decreases overall. When the loader is stationary, the field strength value of its three-dimensional risk potential energy field decreases uniformly from the cross-section of the three-dimensional dynamic geometric profile outwards.

[0084] Step 1.3: Construct a three-dimensional risk potential energy field for the scene. Boundary obstacles in the dock operation scenario are considered as the source of the scenario's three-dimensional risk potential energy field, and the resulting scenario's three-dimensional risk potential energy field At a point in three-dimensional space The formula for calculating the electric field strength is as follows: (8) In the formula Used to distinguish different types of boundary obstacles; Indicates the mass of the boundary obstacle; This is the mass adjustment factor for boundary obstacles; Indicates factors affecting road conditions; Indicates boundary obstacles Upper distance point The nearest point-to-point The distance vector, the direction of the field strength and same, This represents the magnitude of the distance vector; For point Adjustment factor for the shortest distance to the boundary obstacle.

[0085] Figure 7 (a) Figure 7 (b) and Figure 7 (c) in the figure represents the field strength distribution of the three-dimensional risk potential energy field of the scene at heights of 0 meters, 2 meters, and 4 meters, respectively. The boundary obstacles considered include the gantry crane base and the fence. The bases of the two gantry cranes are dispersed on the upper edge of the figure. Due to their high height, they exhibit the same high field strength value area centered on the gantry crane base at different heights. The fence is continuous and located on the lower edge of the figure. It generates a continuous and uniform high field strength value area on the lower side of the scene, but its height is low. There is a high field strength value area on the lower edge of the cross section at a height of 0 meters. As the height increases, the field strength value around it gradually decreases.

[0086] The second step involves setting multiple sampling points on the three-dimensional dynamic geometric contour of the main mobile equipment, calculating the single-point operational risk assessment index (RAI) value for each sampling point, and then calculating the RAI values ​​for each direction of the main mobile equipment. The main mobile equipment includes loaders and heavy-duty trucks, specifically: Step 2.1: Set sampling points on the three-dimensional dynamic geometric profiles of the loader and heavy-duty truck.

[0087] Thirty-two sampling points are set on the three-dimensional dynamic geometric contour of the loader. There are 16 sampling points on the bottom surface of the three-dimensional dynamic geometric contour, including three quarter points on the front and rear arc segments, and the endpoints and midpoints of the four line segments on the left and right sides. The endpoint with the largest height coordinate value that is the same as the two-dimensional coordinate of these 16 sampling points is found on the three-dimensional dynamic geometric contour of the loader, which is the other 16 sampling points on the three-dimensional dynamic geometric contour of the loader.

[0088] Figure 8 The distribution of sampling points on the three-dimensional dynamic geometric contour of the loader is shown. Figure 8 (a) and Figure 8 (b) in the figure shows the three-dimensional dynamic geometric contour of the loader in a perspective view and a top view view, respectively. The red dots and numbers in the figure represent sampling points, which are numbered from 1 to 32.

[0089] Sixteen sampling points were set on the three-dimensional dynamic geometric contour of the heavy-duty truck, including eight vertices of the three-dimensional dynamic geometric contour cuboid and eight trisection points along the four long sides of the truck's length.

[0090] Figure 9 The distribution of sampling points on the three-dimensional dynamic geometric profile of a heavy-duty truck is shown. Figure 9 (a) and Figure 9 (a) shows the three-dimensional dynamic geometric profile of the heavy-duty truck in a stereo view and a top view, respectively. The red dots and numbers in the figure represent sampling points, which are numbered from 0 to 15.

[0091] Step 2.2: Based on the combined influence of the three-dimensional risk potential energy fields of each operating element around the main mobile equipment and the three-dimensional risk potential energy field of the scene, calculate the risk assessment indicator (RAI value) for all sampling points confirmed in Step 2.1.

[0092] Similar to the force exerted on an object in a physical field, the main flow equipment also experiences a repulsive force in a three-dimensional risk potential energy field, representing the risk posed by obstacles to the operation of the main flow equipment. Located within the main flow equipment... Points on the three-dimensional dynamic geometric contour Obstacles The formula for calculating the applied field force is: (9) In the formula Indicates obstacles Points on the three-dimensional dynamic geometric contour of the main flow equipment The field strength of the three-dimensional risk potential energy field generated at the location; Indicates mobile equipment The quality; Indicates the main flow equipment and obstacles The relative velocity relationship factor between them.

[0093] Field forces can cancel each other out in real physical fields. However, for a certain main mobile device in a scenario, the operational risks represented by field forces with different directions generated by obstacles in different directions cannot be canceled out. Therefore, a superimposed scalar is needed to characterize the magnitude of the risk of the main mobile device.

[0094] Three-dimensional safety potential field energy (SPFE) is the energy possessed by an object due to the work done by conservative field forces acting on it in a three-dimensional safety potential field. It includes the three-dimensional safety potential field energy of the working element and the three-dimensional safety potential field energy of the scene. (Main moving equipment) Points on the three-dimensional dynamic geometric contour The formula for calculating the potential energy of the three-dimensional risk field is: (10) (11) In the formula The number of operational elements that pose a risk to the main mobile equipment; Main mobile equipment Points on the three-dimensional dynamic geometric contour To the The distance between the centers of each task element; Main mobile equipment Points on the three-dimensional dynamic geometric contour Subject to the The field force exerted by each working element; The three-dimensional risk field potential energy of the work elements under the influence of the three-dimensional risk potential energy field of multiple work elements. Main mobile equipment Points on the three-dimensional dynamic geometric contour To the nearest boundary obstacle The distance; Main mobile equipment Points on the three-dimensional dynamic geometric contour Subject to the nearest boundary obstacle The field force acting; The potential energy of the three-dimensional risk field in the scene under the influence of the three-dimensional risk potential energy field.

[0095] The operational risk assessment index (RAI) value of a single sampling point on the three-dimensional dynamic geometric contour of the main mobile equipment is calculated using the following formula: (12) In the formula and These represent the weights of the three-dimensional risk field potential energy of the task element and the three-dimensional risk field potential energy of the scene, respectively.

[0096] Step 2.3: Group all sampling points on the three-dimensional dynamic geometric contour of the main flow equipment according to direction, and calculate the operational RAI value of the main flow equipment in each direction.

[0097] The RAI values ​​of multiple sampling points in step 2.1 are calculated according to the method in step 2.2. The sampling points are divided into eight groups according to their location: front, left front, right front, rear, left rear, right rear, front upper, and rear upper. The RAI values ​​of all sampling points in each group are averaged to obtain the working RAI values ​​of the main flow equipment in eight directions.

[0098] For the loader, the 32 sampling points were divided into eight groups according to the grouping method shown in Table 1.

[0099] Table 1. Grouping of sampling points on the three-dimensional dynamic geometric contour of the loader.

[0100] For heavy-duty trucks, the 16 sampling points were divided into eight groups according to the grouping method shown in Table 2.

[0101] Table 2 Grouping of sampling points on the three-dimensional dynamic geometric profile of heavy-duty trucks

[0102] The third step involves constructing a time-series dataset for predictive model training. A sliding window is used to construct the training sample set from this dataset, and the mobile device RAI value prediction model is trained using the Long Short-Term Memory (LSTM) network method. Specifically: Step 3.1: Construct a time series dataset for training the prediction model.

[0103] Temporal prediction utilizes mathematical models to uncover the temporal evolution patterns of historical data, enabling forward-looking inferences about data values ​​at specific future times. Therefore, it requires constructing long-term temporal sequence datasets. In a dock operation scenario, motion parameters of all operational elements are collected at a frequency of 0.2 seconds. These parameters include the 2D coordinates of the loader's center point, the angle between the boom and the horizontal direction, the angle between the bucket's lower edge and the horizontal direction, the orientation angle, and the magnitude and direction of its velocity; the 2D coordinates of the heavy-duty truck's center point, its height, orientation angle, and the magnitude and direction of its velocity; the 3D coordinates of the gantry crane's grab bucket and the connection point with the hoisting rope, its velocity, and its opening angle; and the 2D coordinates, velocity, and orientation angle of the worker's location. Furthermore, the 3D coordinates of all boundary obstacles are collected and updated every minute. The original temporal sequence dataset consists of the motion parameters of all operational elements and the 3D coordinates of boundary obstacles for each frame.

[0104] Since the operational elements frequently enter and exit the dock operation scene, the number of motion parameters in each frame of the original time series dataset is not fixed. The data of each frame in the original time series dataset is converted into general feature values ​​describing the interaction between the main mobile equipment and the surrounding environment. The general features include the speed of the main equipment, the speed of the nearest obstacle, the speed of the second nearest obstacle, the distance of the nearest obstacle, the relative angle of the nearest obstacle, the relative height of the nearest obstacle, the type of the nearest obstacle, the distance of the second nearest obstacle, and the relative speed of the second nearest obstacle, forming the processed time series dataset.

[0105] Simultaneously, following the method described in the second step above, the eight-directional RAI value of the main mobile device in each frame is calculated. The general feature value and the eight-directional RAI value are then combined to form a single-frame data stamp.

[0106] Finally, all data stamps are arranged in chronological order, forming a structure of length [length missing]. The prediction model is trained on a time series dataset at each time step.

[0107] Step 3.2: Construct the sample set for training the prediction model.

[0108] For the time-series dataset used to train the prediction model formed in step 3.1, a sliding window technique is employed to construct samples. A fixed-length window slides across the time series, segmenting the continuous long sequence into a series of subsequences. In this process, two key parameters need to be set: the prediction window length and the time series data set. and predicted lead time The prediction window length is the same as the historical window length, which is the length of historical information used by the model for prediction. The prediction lead length is the prediction lead amount, which is the number of future time steps the model predicts for the target.

[0109] For a common Each time step includes One general feature and The prediction model for predicting n targets is trained on a time series dataset, with the sliding window moving from time step 0 to time step 1. Each time step moves one time step at a time, and each move forms a training sample for the deep learning model. Each sample contains... Frame data stamps. From time step 0 to time step 1 in each sample. The general feature values ​​of each time step constitute Using a dimensional matrix as the input sequence, select the ... The target sequence is formed by the eight-axis RAI values ​​at each time step.

[0110] Through the above processing, a mixture containing The sample set of a sample is divided into a training set, a validation set, and a test set according to the proportions of 70%, 15%, and 15%, respectively.

[0111] Step 3.3: Train the RAI value prediction model for mobile equipment based on the LSTM prediction model.

[0112] Long Short-Term Memory (LSTM) networks are a common method for temporal prediction in deep learning, often abbreviated as LSTM prediction models. An LSTM prediction model consists of two stacked LSTM layers and an output projection layer; each LSTM layer contains 256 hidden units. From the training set obtained in step 3.2, the LSTM prediction model extracts 64 samples at a time as a batch for parallel processing. The three-dimensional tensor is input into the first LSTM layer, which is traversed... That moment will The general feature mapping of dimension 1 is a 256-dimensional hidden state sequence containing richer information, with an output dimension of 1. The first LSTM layer receives the 3D tensor from the first LSTM layer and reduces its dimension to 1. It includes the past The feature vectors representing key information at each time step are fed into the output projection layer. The output projection layer is a fully connected network that projects the feature vectors of dimension 1. The feature vectors are mapped to the 8-dimensional main body flow device to predict the RAI value, and the output shape is... eigenvectors.

[0113] The LSTM prediction model updates its weight parameters after processing each batch of samples in the training set. One round of training is completed by dividing the training set into several batches and processing each batch separately. After each training round, the LSTM prediction model calculates the error between the predicted and true values ​​on the validation set samples and fine-tunes the weight parameters in the LSTM and fully connected layers. After 80 rounds of training, the predictive performance of the LSTM prediction model is evaluated by calculating the root mean square error, mean absolute error, and coefficient of determination on the test set.

[0114] The root mean square error (RMSE) is calculated using the following formula; a smaller value indicates better prediction performance: (13) The formula for calculating the mean absolute error is as follows; the smaller the value, the better the prediction performance: (14) The formula for calculating the coefficient of determination is as follows; the closer the value is to 1, the better the predictive performance: (15) In the formula, Indicates the number of samples in the test set. This represents the eight-way RAI value of the target sequence in each sample of the test set. This represents the 8-dimensional predicted RAI value of the main flow equipment output by the LSTM prediction model after processing each sample. Indicates all The average of the values.

[0115] Once the LSTM prediction model is trained, it will be used as the RAI value prediction model for mobile devices.

[0116] The fourth step is to input a segment of real-time motion parameters of each operational element into the mobile equipment RAI value model, determine the risk level at the predicted time based on the output main mobile equipment predicted RAI value, and execute the corresponding early warning command.

[0117] The mobile equipment RAI value prediction model is deployed to the actual working environment. The real-time motion parameters of the main mobile equipment are processed to obtain general features, which are then input into the mobile equipment RAI value prediction model. The predicted RAI value output by the mobile equipment RAI value prediction model will be used as the basis for risk assessment.

[0118] The mobile equipment is classified into different risk levels based on the magnitude of the predicted RAI value. The risk levels include low risk, medium risk, and high risk. When the predicted RAI value in all eight directions does not exceed 0.4, the mobile equipment is in the low risk level; when the predicted RAI value in any direction is greater than 0.4 but not more than 0.9, the mobile equipment is in the medium risk level; when the predicted RAI value in any direction is greater than 0.9, the mobile equipment is in the high risk level.

[0119] When the mobile equipment is at a low risk level, no risk warning signal is generated, and the mobile equipment alarm device is not triggered.

[0120] When the mobile equipment is at a medium-risk level, the mobile equipment alarm will emit two consecutive short beeps, and the risk information display screen will show the direction with the highest predicted RAI value and flash continuously.

[0121] When the mobile equipment is at a high risk level, the mobile equipment alarm will continuously beep, the risk information display screen will show the direction with the highest predicted RAI value and continuously flash, and the in-vehicle broadcast will issue a voice prompt "Please be aware of the risk in a certain direction".

[0122] The port mobile equipment operation risk early warning system based on a three-dimensional risk potential energy field realizes the above-mentioned port mobile equipment operation risk early warning method. The port mobile equipment operation risk early warning system includes a data acquisition unit, a central processing unit, and an early warning execution unit, wherein the central processing unit includes a storage module, a calculation module, and a response module.

[0123] The data acquisition units are distributed across all operational elements and boundary obstacles in the port, and are responsible for collecting raw time-series datasets at fixed time intervals.

[0124] The storage unit of the central processing unit is responsible for storing the original time-series dataset and the RAI value time-series dataset; the calculation module performs two functions: first, it reads the original time-series dataset and calculates the multi-directional RAI values ​​of the main flow device; second, it uses the prediction model to train the LSTM prediction model using the time-series dataset; the response module saves the LSTM prediction model trained by the calculation module, reads a segment of real-time acquired original time-series dataset from the storage module, processes it, outputs the predicted RAI value, and determines whether to output risk warning information.

[0125] The aforementioned early warning execution unit is installed in the driver's cab of the main mobile equipment and executes early warning commands based on the risk early warning information provided by the central processing unit response module.

[0126] To realize the various functions of the port mobile equipment operation risk early warning system based on a three-dimensional risk potential energy field, the following devices are required: The data acquisition unit includes: a high-precision positioning sensor and an inertial measurement unit installed on the mobile equipment vehicle; a winch system drum encoder and a hydraulic cylinder built-in displacement sensor installed on the gantry crane; a personnel positioning beacon installed on the worker's safety helmet; and a lidar, millimeter-wave radar, and high-definition camera installed in the scene.

[0127] The central processing unit includes: a database server for the storage module; a multi-core CPU, a graphics processing unit (GPU), a risk value calculation engine based on a three-dimensional risk potential field, and a PyTorch-based deep learning training framework for the computing module; and a real-time inference engine for the response module.

[0128] The aforementioned early warning execution unit includes: an in-vehicle communication device, a signal transmission module, an in-vehicle domain controller, a physical alarm device, and a risk information display screen.

[0129] The above-described embodiments are merely illustrative of the implementation methods of the present invention, but should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the protection scope of the present invention.

Claims

1. A method for early warning of operational risks of port mobile equipment based on a risk potential energy field, characterized in that, The aforementioned port mobile equipment operation risk early warning method includes the following steps: The first step is to construct a three-dimensional dynamic geometric contour for each operational element, and then use this contour as the field source to construct a three-dimensional risk potential energy field for each operational element. Finally, using boundary obstacles as the field source, a three-dimensional risk potential energy field for the entire scene is constructed. Specifically: Step 1.1: Construct the three-dimensional dynamic geometric contours of each operational element, which includes loaders, heavy-duty trucks, gantry crane grabs, and workers; Step 1.2: Based on the three-dimensional dynamic geometric contours of each task element, construct the three-dimensional risk potential energy field for each task element. ; Specifically, step 1.2 is as follows: any point in three-dimensional space The three-dimensional risk potential energy field strength reflects the risk level experienced by a unit mass object at a point; the three-dimensional dynamic geometric contour of a single work element is considered as the source of the three-dimensional risk potential energy field of the work element, and the resulting three-dimensional risk potential energy field of the work element... Points outside the three-dimensional dynamic geometric contour of the work element in three-dimensional space The formula for calculating the electric field strength is as follows: In the formula, Used to distinguish different types of task elements; Indicates the quality of the task element; Indicates factors affecting road conditions; This is the quality adjustment factor for the work element; Indicates the speed of movement of the task element; This represents the angle between the line connecting the midpoint of the space to the center of the work element and the direction of movement of the work element; Represents job elements Distance point on the three-dimensional dynamic geometric contour The nearest point-to-point The distance vector, the direction of the field strength and same, Represents the magnitude of the distance vector. This is the adjustment factor for the shortest distance from the point to the task element; Adjustment coefficient for the direction of movement of work elements; The three-dimensional risk potential energy field of each task element within its three-dimensional dynamic geometric contour. The field strength values ​​are all 1; Step 1.3: Construct a three-dimensional risk potential energy field for the scene. ; Boundary obstacles in the dock operation scenario are considered as the source of the scenario's three-dimensional risk potential energy field, and the resulting scenario's three-dimensional risk potential energy field At a point in three-dimensional space The formula for calculating the electric field strength is as follows: In the formula, Used to distinguish different types of boundary obstacles; Indicates the mass of the boundary obstacle; This is the mass adjustment factor for boundary obstacles; Indicates factors affecting road conditions; Indicates boundary obstacles Upper distance point The nearest point-to-point The distance vector, the direction of the field strength and same, Represents the magnitude of the distance vector; For point Adjustment factor for the shortest distance to the boundary obstacle; The second step involves setting multiple sampling points on the three-dimensional dynamic geometric contour of the main mobile equipment, calculating the single-point operational risk assessment index (RAI) value for each sampling point, and then calculating the RAI values ​​for each direction of the main mobile equipment. The main mobile equipment includes loaders and heavy-duty trucks. Specifically: Step 2.1: Set sampling points on the three-dimensional dynamic geometric contours of the loader and heavy-duty truck; Step 2.2: Based on the combined influence of the three-dimensional risk potential energy fields of each operational element around the main mobile equipment and the three-dimensional risk potential energy field of the scene, calculate the risk assessment index of all sampling points confirmed in Step 2.1, referred to as the RAI value; Located in the main flow equipment Points on the three-dimensional dynamic geometric contour Obstacles The formula for calculating the applied field force is: In the formula, Indicates obstacles Points on the three-dimensional dynamic geometric contour of the main flow equipment The field strength of the three-dimensional risk potential energy field generated at the location; Indicates the main flow equipment The quality; Indicates the main flow equipment and obstacles The relative velocity relationship factor between them; The three-dimensional risk field potential energy, abbreviated as SPFE value, includes the three-dimensional risk field potential energy of the operational elements and the three-dimensional risk field potential energy of the scene; the main flow equipment Points on the three-dimensional dynamic geometric contour The formula for calculating the potential energy of the three-dimensional risk field is: In the formula, The number of operational elements that pose a risk to the main mobile equipment; Main mobile equipment Points on the three-dimensional dynamic geometric contour To the The distance between the centers of each task element; Main mobile equipment Points on the three-dimensional dynamic geometric contour Subject to the The field force exerted by each working element; The three-dimensional risk field potential energy of the work elements under the influence of the three-dimensional risk potential energy field of multiple work elements; Main mobile equipment Points on the three-dimensional dynamic geometric contour To the nearest boundary obstacle The distance; Main mobile equipment Points on the three-dimensional dynamic geometric contour Subject to the nearest boundary obstacle The field force acting; The potential energy of the three-dimensional risk field of the scene under the influence of the three-dimensional risk potential energy field of the scene; The operational risk assessment index (RAI) value of a single sampling point on the three-dimensional dynamic geometric contour of the main mobile equipment is calculated using the following formula: In the formula, and These represent the weights of the three-dimensional risk field potential energy of the task element and the three-dimensional risk field potential energy of the scene, respectively. Step 2.3: Group all sampling points on the three-dimensional dynamic geometric contour of the main flow equipment according to direction, and calculate the operational RAI values ​​of the main flow equipment in eight directions; The third step is to construct a time series dataset for predictive model training, use a sliding window to construct a training sample set for predictive model from it, and train a mobile device RAI value prediction model based on the long short-term memory network method. The fourth step involves inputting a segment of real-time motion parameters for each operational element into the mobile equipment RAI value prediction model. Based on the output predicted RAI value of the main mobile equipment, the risk level at the predicted time is determined, and corresponding early warning commands are executed. Specifically: The mobile equipment RAI value prediction model is deployed to the actual working environment. The real-time motion parameters of the main mobile equipment are processed to obtain general features, which are then input into the mobile equipment RAI value prediction model. The predicted RAI value output by the mobile equipment RAI value prediction model will be used as the basis for risk assessment. The mobile equipment is classified into different risk levels based on the magnitude of the predicted RAI value. The risk levels include low risk, medium risk, and high risk. When the predicted RAI value in all eight directions does not exceed 0.4, the mobile equipment is in the low risk level; when the predicted RAI value in any direction is greater than 0.4 but not more than 0.9, the mobile equipment is in the medium risk level; when the predicted RAI value in any direction is greater than 0.9, the mobile equipment is in the high risk level.

2. The port mobile equipment operation risk early warning method based on risk potential energy field according to claim 1, characterized in that, Specifically, step 1.1 is as follows: Step 1.1.1: Construct the three-dimensional dynamic geometric profile of the loader; Based on the loader's articulation center position, vehicle length, speed, steering angle, lateral safety redundancy, and minimum safety distance, the loader's two-dimensional dynamic geometric profile is determined. It consists of two arc segments centered at the articulation center at the front and rear, and four arc segments on the left and right sides, including the front arc segment, rear arc segment, left arc segment, and right arc segment. The central angles of each of the four arc segments are calculated. Taking the left arc segment as an example, its central angle is defined by the following formula: In the formula, and These are the central angles of the left side segments of the front and back arc segments, respectively. It is the velocity profile offset coefficient; and These are the forward and rearward lengths of the loader's body, respectively, from the articulation center. and These represent the radii of the arc segments in the forward and backward directions, respectively. It is a lateral safety redundancy for the loader; This is the minimum safe distance for loader operation; and These are the turning angles of the left-side arc segments in the forward and backward directions, respectively. Based on the two-dimensional dynamic geometric contour of the loader, additional endpoints are added to construct the three-dimensional dynamic geometric contour of the loader. The two-dimensional dynamic geometric contour is divided into a front contour and a rear contour, each containing an arc segment and two line segments. The front contour is shifted upward by the highest height of the loader bucket plus the safety redundancy above the loader. The rear contour is shifted upward by the height of the rear section of the loader body plus the safety redundancy above the loader. This yields the upper front contour, upper rear contour, and upper endpoint of the three-dimensional dynamic geometric contour of the loader. The upper endpoints are connected to the endpoints with the same two-dimensional coordinates in the two-dimensional dynamic geometric contour of the loader to obtain the three-dimensional dynamic geometric contour of the loader. Step 1.1.2: Construct the three-dimensional dynamic geometric profile of the heavy-duty truck; The heavy-duty truck body is simplified to a cuboid, with the connection point between the cab and the cargo box as the center point. First, based on the center point, body length, width, speed, and safety margin, the two-dimensional dynamic geometric profile of the heavy-duty truck is determined. Safety margins are then added to the bottom surface of the cuboid in all directions (front, back, left, right) to form a rectangular two-dimensional dynamic geometric profile. The distances from the center point to the front and rear sides of the two-dimensional dynamic geometric profile are calculated in the same way, as are the distances from the center point to the left and right sides. Taking forward movement as an example, the formulas for calculating the distances from the center point to the front and left sides are as follows: In the formula, and These are the distances from the center point to the front and left sides of the two-dimensional dynamic geometric contour of the heavy-duty truck, respectively. This refers to the length of the front of the vehicle. This refers to the width of the vehicle body; and These are the basic safety redundancies for the front and left sides of the truck, respectively. and These are the linear velocity profile offset coefficients for the front and left sides, respectively. The speed of the truck; The three-dimensional dynamic geometric profile of a heavy-duty truck is a cuboid, with the rectangle of its two-dimensional dynamic geometric profile as the base, and the height is the vehicle height of the heavy-duty truck plus the safety redundancy of the heavy-duty truck. Step 1.1.3: Construct the three-dimensional dynamic geometric profile of the gantry crane grab bucket; The grab bucket of the gantry crane is simplified as a triangular prism with an equilateral triangle as its base. First, the two-dimensional dynamic geometric profile of the grab bucket is determined on the longitudinal section perpendicular to the crossbeam at the connection point. Its shape is a curved trapezoid formed by two line segments on the sides, a line segment on the upper base, and a circular arc on the lower base. The connection point is the midpoint of the upper base of the curved trapezoid. On the longitudinal section perpendicular to the crossbeam, the longitudinal safety redundancy of the grab bucket is extended horizontally from the connection point to both sides, obtaining the two endpoints of the upper base of the curved trapezoid. With the connection point as the center, a circular arc-shaped lower base of the curved trapezoid is formed within the longitudinal section. The central angle is positively correlated with the opening angle of the grab bucket. The formula for calculating the radius is as follows: In the formula, The radius of the arc segment; The opening angle of the gantry crane grab; The height difference between the connection point of the gantry crane grab bucket and the hoisting rope and the hinge point of the two bucket segments of the gantry crane grab bucket; The side length of a single canthus; The speed influence coefficient of the gantry crane grab bucket; The speed of the gantry crane grab bucket; The three-dimensional dynamic geometric profile of the gantry crane grab bucket is a geometric body formed by extending its two-dimensional dynamic geometric profile along the straight line of the crossbeam, and extending the lateral safety redundancy distance from the connection point along the straight line of the crossbeam to both sides respectively. Step 1.1.4: Construct the three-dimensional dynamic geometric contours of the staff; The 3D dynamic geometric profile of the worker includes cylindrical and elliptical cylindrical shapes; in the 3D dynamic geometric profile of a stationary worker, the base is a circle centered on the worker's position, and the radius of the base is set to the spatial influence length in the worker's direction of movement. In the three-dimensional dynamic geometric profile of the moving worker, the base is an ellipse with the worker's position as the focus and the direction of movement pointing towards the center of the ellipse. The length of the minor semi-axis of the base ellipse is the safe distance of the worker. The length of the major semi-axis is the spatial influence length in the direction of the worker's movement. The formula for calculating the spatial influence length in the direction of staff movement is as follows: In the formula, Maintain a safe distance for staff; The speed at which the staff moves; The influence coefficient of staff speed; The spatial influence length in the direction of the worker's movement is taken as the worker's safe distance when the worker is stationary. It serves as the radius of the bottom circle of the three-dimensional dynamic geometric profile and as the length of the major semi-axis of the bottom ellipse of the three-dimensional dynamic geometric profile when the worker is moving.

3. The port mobile equipment operation risk early warning method based on risk potential energy field according to claim 1, characterized in that, The second step is as follows: Step 2.1 specifically involves: 32 sampling points are set on the three-dimensional dynamic geometric contour of the loader, and 16 sampling points are set on the bottom surface of the three-dimensional dynamic geometric contour; 16 sampling points are set on the three-dimensional dynamic geometric contour of the heavy truck, including 8 vertices of its three-dimensional dynamic geometric contour cuboid and 8 trisection points of the 4 long sides in the vehicle length direction. Step 2.3 specifically involves: calculating the RAI values ​​of multiple sampling points from step 2.1 according to the method in step 2.2; dividing the sampling points into eight groups based on their location: front, left front, right front, front upper, rear, left rear, right rear, and rear upper; and averaging the RAI values ​​of all sampling points in each group to obtain the operational RAI values ​​for the eight directions of the main flow equipment.

4. The port mobile equipment operation risk early warning method based on risk potential energy field according to claim 3, characterized in that, The third step is specifically as follows: Step 3.1: Construct a time series dataset for training the prediction model; In the dock operation scenario, motion parameters of all operation elements are collected; in addition, the three-dimensional coordinates of the locations of all boundary obstacles are collected and updated at regular intervals; the original time-series dataset consists of motion parameters of all operation elements and three-dimensional coordinates of boundary obstacles for each frame; Each frame of data in the original time-series dataset is converted into general feature values ​​that describe the interaction between the main mobile device and its surrounding environment, forming the processed time-series dataset. Calculate the eight-directional RAI value of the main mobile device in each frame; merge the general feature value with the eight-directional RAI value to form a single-frame data stamp; Arrange all data stamps in chronological order, forming a structure of length [length missing]. The prediction model is trained on a time series dataset at each time step; Step 3.2: Construct the sample set for training the prediction model; The time series dataset is trained based on the prediction model developed in step 3.

1. A sliding window is used to construct samples, where a fixed-length window slides across the time series to segment the continuous long sequence into a series of subsequences. Two key parameters are set: the prediction window length and the prediction window length. and predicted lead time ; For a common Each time step includes One general feature and The prediction model for predicting n targets is trained on a time series dataset, with the sliding window moving from time step 0 to time step 1. Each time step moves one time step at a time, and each move forms a training sample for the deep learning model. Each sample contains... Frame data stamps; select time step 0 to time step 1 in each sample. The general feature values ​​of each time step constitute Using a dimensional matrix as the input sequence, select the th... The target sequence is composed of eight RAI values ​​at each time step; The final form includes The sample set of each sample is divided into a training set, a validation set, and a test set. Step 3.3: Train the RAI value prediction model for mobile equipment based on the LSTM prediction model; The LSTM prediction model updates its weight parameters after processing each batch of samples in the training set. Dividing the samples in the training set into several batches and processing them separately completes one round of training. After each round of training, the LSTM prediction model calculates the error between the predicted value and the true value on the validation set samples and fine-tunes the weight parameters in the LSTM layer and the fully connected layer. After training is completed, the prediction performance of the LSTM prediction model is evaluated by calculating the root mean square error, mean absolute error, and coefficient of determination on the test set. Once the LSTM prediction model is trained, it will be used as the RAI value prediction model for mobile devices.

5. The port mobile equipment operation risk early warning method based on risk potential energy field according to claim 4, characterized in that, In step 3.1: The motion parameters of all operating elements are collected at a frequency of 0.2 seconds, including the two-dimensional coordinates of the center point of the loader, the angle between the boom and the horizontal direction, the angle between the lower edge of the bucket and the horizontal direction, the orientation angle, the magnitude and direction of the speed, the two-dimensional coordinates of the center point of the heavy-duty truck, the vehicle height, the orientation angle, the magnitude and direction of the speed, the three-dimensional coordinates of the connection point between the gantry crane grab bucket and the hoisting rope, the magnitude of the speed, and the opening angle, and the two-dimensional coordinates, the magnitude of the speed, and the orientation angle of the worker's location. The three-dimensional coordinates of all boundary obstacles are collected and updated every minute. The general features include the speed of the main device, the speed of the nearest obstacle, the speed of the second nearest obstacle, the distance to the nearest obstacle, the relative angle of the nearest obstacle, the relative height of the nearest obstacle, the type of the nearest obstacle, the distance to the second nearest obstacle, and the relative speed of the second nearest obstacle.

6. The port mobile equipment operation risk early warning method based on risk potential energy field according to claim 5, characterized in that, In step 3.3: the LSTM prediction model consists of two stacked LSTM layers and an output projection layer, with each LSTM layer containing 256 hidden units.

7. The port mobile equipment operation risk early warning method based on risk potential energy field according to claim 6, characterized in that, In the fourth step: When the mobile equipment is at a low risk level, no risk warning signal is generated, and the mobile equipment alarm device is not triggered. When the mobile equipment is at a medium risk level, the mobile equipment alarm will emit two consecutive short beeps, and the risk information display screen will show the direction with the highest predicted RAI value and flash continuously. When a mobile device is at a high-risk level, the mobile device alarm will continuously beep, the risk information display screen will show the direction with the highest predicted RAI value and flash continuously, and a voice prompt will be issued.

8. A port mobile equipment operation risk early warning system based on a three-dimensional risk potential energy field, characterized in that, The port mobile equipment operation risk early warning method according to any one of claims 1-7 is implemented through a port mobile equipment operation risk early warning system. The port mobile equipment operation risk early warning system includes a data acquisition unit, a central processing unit, and an early warning execution unit, wherein the central processing unit includes a storage module, a computing module, and a response module; specifically: The data acquisition units are distributed across all operational elements and boundary obstacles in the port, and are used to collect raw time-series datasets at fixed time intervals. The storage unit of the central processing unit is used to store the original time-series sequence dataset and the RAI value time-series dataset; The computing module performs two functions: first, it reads the original time-series dataset and the multi-directional RAI values ​​of the main mobile device; second, it uses the prediction model to train the LSTM prediction model from the time-series dataset. The response module saves the LSTM prediction model trained by the computation module, reads a real-time acquired raw time series dataset from the storage module, processes it, outputs the predicted RAI value, and determines whether to output risk warning information. The aforementioned early warning execution unit is installed in the driver's cab of the main mobile equipment and executes early warning commands based on the risk early warning information provided by the central processing unit response module.

Images