A path planning system for a levee breach rescue equipment

Abstract

The application discloses a kind of path planning systems of dike breach rescue equipment, comprising: environment perception module, for obtaining environmental data and position information, constructs rescue environment digital model;Path planning model is used to determine path planning based on rescue environment digital model;Path planning model is strategy network and safety prediction network coupling reward function, fusion reliability safety index component intelligent agent decision model, is trained based on PPO algorithm and is constructed generation;Wherein, strategy network is used to obtain shared features, to safety prediction network output action mean, according to total reward determines the path trajectory of equipment target;Interaction module is used to output real-time path planning result, risk probability and safety classification.According to the above technical scheme, the dynamic coupling of geological bearing capacity, water flow power, boat bridge stability and equipment dynamics is comprehensively considered, and global safety constraint is formed;Ensure the transportation safety of equipment in complex environment, greatly improve the efficiency of rescue.

Description

Technical Field

[0001] This invention relates to the interdisciplinary field of emergency rescue technology, artificial intelligence and structural engineering, and more specifically, to a path planning system for dam breach rescue equipment. Background Technology

[0002] In emergency rescue operations following dam breaches, the planning of transportation routes for heavy equipment faces three core challenges: the uneven spatial distribution of geological bearing capacity of land-based embankments and the difficulty in accurately measuring it; the complex and variable hydrodynamic conditions that seriously affect the stability of floating platforms; and the diverse assembly modes of pontoon bridges that lead to dynamic changes in load-bearing characteristics.

[0003] Traditional path planning methods simplify safety issues to static constraints, but cannot quantify dynamic risks in complex environments. While the centroid method of existing reliability theory can quantitatively assess structural safety, it is computationally complex and difficult to integrate into real-time path planning. Reinforcement learning methods can solve path optimization problems, but lack engineering physical constraints, and the planning results may not meet engineering safety requirements.

[0004] Therefore, there is an urgent need for a technical solution to address the contradiction between safety and efficiency in equipment transportation during dam breach rescue operations, so as to achieve transportation safety and improve rescue efficiency in complex environments. Summary of the Invention

[0005] To achieve the above objectives, this application provides a path planning system for dam breach rescue equipment, comprising:

[0006] The environmental perception module is used to acquire environmental data and location information to construct a digital model of the rescue environment. The environmental data includes: geological bearing capacity, water flow velocity and flow field, pontoon bridge information, and equipment information. The location information includes the precise positioning of the equipment and the precise positioning of the pontoon bridge. The digital model of the rescue environment includes a gridded geological bearing capacity field and a time-varying water flow velocity vector field.

[0007] A path planning model is used to determine path planning based on a digital model of the rescue environment. The path planning model is an intelligent agent decision-making model composed of a policy network and a safety prediction network coupled with a reward function and fused with reliability and safety indicators. It is trained and generated based on the PPO algorithm. The policy network is used to acquire shared features, output the average action value to the safety prediction network, and determine the path trajectory of the equipment target based on the total reward.

[0008] The interaction module is used to input environmental data into the environmental perception module, display the load-bearing stability of the equipment and pontoon bridge in real time, and output real-time path planning results, risk probability, and safety classification.

[0009] The shared features include: environmental data, task state vectors, historical state sequences, and security state information;

[0010] The mission state vector contains the relative position of the equipment target point. Relative position of the pontoon bridge target point Relative azimuth of equipment target Relative bearing of ships and targets Distance from equipment to target point Distance from boat to target point Current time Path progress ;

[0011] The historical state sequence contains the position sequence of the past T steps. Past T-step action sequence , m For the action dimension;

[0012] Security status information includes the fusion reliability security metrics from the previous time period. .

[0013] Furthermore, the policy network includes a multimodal feature extraction layer, a feature fusion layer, and a policy output layer;

[0014] Among them, the multimodal feature extraction layer extracts multimodal features from shared features;

[0015] The feature fusion layer concatenates multimodal features and outputs them through the ReLU activation function;

[0016] The policy output layer calculates and outputs the average action value through an activation function.

[0017] Furthermore, the security prediction network includes a shared feature extraction layer, a value assessment layer, and a security prediction layer;

[0018] The shared feature extraction layer is used to extract common features that need to be used from shared features;

[0019] The value assessment layer calculates and outputs a value function based on the fusion reliability and security index;

[0020] Specifically, the security prediction layer consists of a reliability analysis module, a risk probability module, and a security classification module, which outputs a fusion of reliability security indicators, risk probability, and security classification.

[0021] The value function is expressed as follows: ,

[0022] in, The value function for time t The time series difference error at time t, For reward decay coefficient, The attenuation coefficient in generalized dominance estimation The timing difference error is first determined to determine the trajectory length;

[0023] Timing Differential Error The calculation method is as follows: ,in, The reward value for time t, In security status information S t+1 The expected cumulative reward following the current strategy.

[0024] Furthermore, the reliability analysis module is used to calculate the fusion reliability and safety index using environmental data and the precise positioning of equipment and pontoon bridges. ;

[0025] Computational fusion reliability and security indicators First, define the function of the equipment and the pontoon bridge, expressed as:

[0026] ,

[0027] in, The load-bearing capacity function characterizes the structural safety margin. For the maximum bearing capacity of the foundation, For equipment grounding voltage, The stability function characterizes the stability safety margin. To provide the restoring moment for the pontoon bridge platform to resist overturning, The overturning moment caused by external forces;

[0028] Among them, equipment grounding voltage The calculation method is as follows:

[0029] ,

[0030] in, For the total weight of the equipment, For the number of grounding units, For unit grounding area, For equipment tilt angle, For equipment grounding voltage standard deviation, For equipment weight standard deviation, The standard deviation of the grounding area. This represents the total grounding area.

[0031] Furthermore, reliability and safety indicators The calculation method is as follows:

[0032] ,

[0033] in, For stability and safety factor, To bear the safety factor, For penalty coefficient, For the target reliability index, For bearing capacity reliability index, As a stability and reliability indicator; among which, The value is generally between 3.2 and 4.2.

[0034] Among them, stability and reliability indicators and stability safety factor The calculation method is as follows:

[0035] , ,in, The average overturning moment, To restore the mean value of the torque, The standard deviation of the overturning moment, The standard deviation of the restoring torque;

[0036] Bearing capacity reliability index and load-bearing safety factor The calculation method is as follows:

[0037] ,and: ;

[0038] The mean overturning moment and the standard deviation of the overturning moment are calculated as follows:

[0039] ,

[0040] in, The average overturning moment, For the density of water, This is the drag force coefficient. For the area facing the flow, The average water flow velocity. The distance from the point of application of the resultant force of the water flow to the overturning axis. The standard deviation of the overturning moment. For water flow velocity, The standard deviation of the water flow velocity. The standard deviation of the overturning moment;

[0041] The mean and standard deviation of the restoring torque are calculated using the following formulas:

[0042]

[0043] To restore the mean value of the torque, For the drainage volume of the pontoon bridge platform, Let be the stability lever arm function that varies with the tilt angle of the pontoon bridge. For the standard deviation of drainage volume, For the standard deviation of the tilt angle, The standard deviation of the restoring torque.

[0044] Furthermore, the probability of risk is calculated using the following formula:

[0045] ,in, For risk probability, It is the standard normal distribution function;

[0046] Safety is categorized into three classes: safe, warning, and danger, represented as follows:

[0047] ,

[0048] in, For safety classification, These are safety classification indicators.

[0049] Furthermore, the reward function is defined as:

[0050]

[0051]

[0052] in, For the total reward, As a time reward, As a safety reward, To encourage rewards, To smooth out rewards, Weighted by time reward For time step, This represents the Euclidean distance from the current location to the target point. This is the total distance from the starting point to the target point. For security reward weighting, For the target safety threshold, To guide the weighting of rewards, The distance of the previous step. The current step distance, To smooth out reward weights, This is the current control action vector. The control action vector of the previous time, It is the square of the Euclidean norm.

[0053] This invention addresses the conflict between safety and efficiency in equipment transportation during dam breach emergency repairs by deeply embedding structural reliability theory into a reinforcement learning framework. By constructing a multimodal perception network integrating geological bearing capacity, hydrodynamics, and pontoon bridge stability, it fundamentally transforms safety constraints from external limitations to internal optimization goals. This method not only ensures the safety of equipment transportation in complex environments but also significantly improves rescue efficiency through intelligent decision-making. It provides crucial technical support for the autonomous and intelligent development of emergency rescue equipment, possessing significant engineering application value and social significance. Attached Figure Description

[0054] Figure 1 This is a schematic diagram of the path planning system for dam breach rescue equipment provided according to an embodiment of the present invention.

[0055] Figure 2 This is a schematic diagram of a strategy network structure provided in an embodiment of the present invention;

[0056] Figure 3 This is a schematic diagram of a security prediction network structure provided according to an embodiment of the present invention;

[0057] Figure 4 This is a schematic diagram of the path planning for the pontoon bridge and equipment provided according to an embodiment of the present invention;

[0058] Figure 5 This is a graph showing the training performance of the agent decision-making model provided in an embodiment of the present invention. Detailed Implementation

[0059] The specific implementation of the present invention will now be described in detail with reference to the accompanying drawings.

[0060] The path planning system structure of the dam breach rescue equipment provided by this invention is as follows: Figure 1 As shown, it includes:

[0061] The P110 environmental perception module is used to acquire environmental data and build a digital model of the rescue environment.

[0062] Specifically, environmental data includes the detection of geological bearing capacity, water flow velocity and flow field, pontoon bridge information, and equipment information. Environmental data can be manually entered through the interactive module, extracted from the database, or obtained from ground-penetrating radar sensors, water flow velocity meters, and attitude sensors through the interactive module.

[0063] Meanwhile, the environmental perception module also acquires location information, including the precise positioning of the equipment. Precise positioning of pontoon bridges Location information can be obtained from the BeiDou Navigation Satellite System through the interactive module or entered manually.

[0064] The digital model of the rescue environment constructed using environmental data includes a gridded geological bearing capacity field and a time-varying water flow velocity vector field:

[0065] Each grid cell in the gridded geological bearing capacity field contains the standard deviation of bearing capacity. This reflects the geological bearing capacity;

[0066] Each grid cell in the time-varying water flow velocity vector field contains For water density, For drag force coefficient and flow velocity vector Mean water flow velocity and standard deviation of water flow velocity ;

[0067] Equipment information in the digital model of the rescue environment includes: the equipment's heading angle. θ Equipment forward speed Equipment acceleration a Equipment tilt angle Total weight of equipment Number of grounding units Unit grounding area Equipment grounding voltage standard deviation , Equipment weight standard deviation and Standard deviation of grounding area;

[0068] Pontoon bridge information includes: assembly mode, unique hot-swappable code. Pontoon bridge platform displacement Stability lever arm function as the tilt angle of the pontoon bridge changes Restoring moment of pontoon bridge platform against overturning Overturning moment caused by external forces Pontoon bridge facing area Distance from the point of application of the resultant force of the water flow to the overturning axis Standard deviation of overturning moment Standard deviation of drainage volume and the standard deviation of the tilt angle .

[0069] The P120 path planning model is an agent decision-making model composed of a policy network and a safety prediction network coupled with a reward function and a fused reliability and safety index. It is generated by training based on the PPO algorithm and can determine path planning based on a digital model of the rescue environment.

[0070] The training data for training the agent decision-making model includes initial geological bearing capacity, water flow velocity and flow field, equipment information and pontoon bridge information randomly generated by the digital model of the rescue environment, as well as the positioning of equipment and pontoon bridges; the digital model of the rescue environment is constructed using data from historical real-world dam breach scenarios.

[0071] The P121 policy network is used to acquire shared features, output the action mean to the security prediction network, and determine the path trajectory of the equipment target based on the total reward.

[0072] The shared features include: environmental data, task state vectors, historical state sequences, and security state information;

[0073] The mission state vector contains the relative position of the equipment target point. Relative position of the pontoon bridge target point Relative azimuth of equipment target Relative bearing of ships and targets Distance from equipment to target point Distance from boat to target point Current time Path progress ;

[0074] The historical state sequence contains the position sequence of the past T steps. Past T-step action sequence , m For the action dimension;

[0075] Security status information includes the fusion reliability security metrics from the previous time period. .

[0076] The network structure of a policy network is as follows: Figure 2 As shown, it includes a multimodal feature extraction layer, a feature fusion layer, and a policy output layer;

[0077] 1) The multimodal feature extraction layer includes a one-dimensional convolutional layer, a two-dimensional convolutional layer, and an embedding layer, which are used to extract multimodal features from shared features such as equipment information, pontoon bridge information, environmental information, location sequence, and action sequence;

[0078] Specifically, the one-dimensional convolutional layer extracts features from equipment and pontoon bridge information, including: the relative azimuth angle of the equipment target. Relative bearing of ships and targets Distance from equipment to target point Distance from boat to target point Current time and path progress The one-dimensional convolutional layer extracts features from the above information into one-dimensional convolutional features. ;

[0079] Two-dimensional convolutional layers affect geological bearing capacity, water flow velocity and flow field, and the relative position of equipment target points. Relative position of the pontoon bridge target point Environmental information, the position sequence of the past T steps Past T-step action sequence Feature extraction is performed as two-dimensional convolutional features. ;

[0080] Embedded layer performs one-hot encoding of the assembly mode Processed into embedded convolutional features ;

[0081] 2) The feature fusion layer concatenates the extracted multimodal features and outputs them through the ReLU activation function, which is calculated as follows:

[0082]

[0083] in For vector concatenation, As a feature of fusion, For the weights of the feature fusion layer, For the feature fusion layer bias, The reliability and safety indicators for the previous time period;

[0084] 3) Average value of output actions from the strategy output layer Its calculation is as follows:

[0085]

[0086] in For activation function, Output weights for the action mean. This is used to offset the mean of the action output.

[0087] Finally, the policy network outputs the average action at time t. .

[0088] P122 security prediction network, such as Figure 3 As shown, it includes a shared feature extraction layer, a value assessment layer, and a security prediction layer;

[0089] 1) The shared feature extraction layer shares input with the policy network to extract common features; common features include task state vector, historical state sequence and security state information;

[0090] 2) The value assessment layer calculates and outputs a value function based on security status information (integrating reliability and security indicators);

[0091] value function Calculate as follows:

[0092] ,

[0093] in, The time series difference error at time t, For reward decay coefficient, The attenuation coefficient in generalized dominance estimation To determine the trajectory length, first determine the timing difference error, where t+1 represents the next time step and T+1 represents the next step step.

[0094] Timing Differential Error The calculation method is as follows:

[0095] ,

[0096] in, The reward value for time t, The fusion reliability and security index S at the next time point t+1 The expected cumulative reward following the current strategy.

[0097] After the value function value is calculated, it is output by the value assessment layer.

[0098] 3) The security prediction layer consists of a reliability analysis module, a risk probability module, and a security classification module, and outputs a fused reliability and security index. Risk probability Safety Classification SC ;

[0099] The reliability analysis module uses the center point method of the first second moment method in reliability theory to quantitatively assess structural safety. It calculates safety status information through environmental data and the precise positioning of equipment and pontoon bridges.

[0100] First, define the function of the equipment and pontoon bridge. The function is used to reflect the load-bearing stability of the equipment and pontoon bridge in real time, and is defined as follows:

[0101] ,

[0102] in, The load-bearing capacity function characterizes the structural safety margin. For the maximum bearing capacity of the foundation, For equipment grounding voltage, The stability function characterizes the stability safety margin. To provide the restoring moment for the pontoon bridge platform to resist overturning, The overturning moment caused by external forces;

[0103] Among them, the maximum bearing capacity of the foundation The standard deviation of each grid cell in the gridded geological bearing capacity field is used. get;

[0104] Equipment grounding pressure The calculation method is as follows:

[0105] ,

[0106] in, For the total weight of the equipment, For the number of grounding units, For unit grounding area, For equipment tilt angle, For equipment grounding voltage standard deviation, For equipment weight standard deviation, The standard deviation of the grounding area. The total ground area; these parameters were obtained through a digital model of the rescue environment.

[0107] Reliability and safety indicators The calculation method is as follows:

[0108] ,

[0109] in, For stability and safety factor, To bear the safety factor, For penalty coefficient, For the target reliability index, For bearing capacity reliability index, As a stability and reliability indicator; among which, The value is typically between 3.2 and 4.2.

[0110] Furthermore, stability and reliability indicators and stability safety factor The calculation method is as follows:

[0111] ,

[0112] ,

[0113] in, The average overturning moment, To restore the mean value of the torque, The standard deviation of the overturning moment, The standard deviation of the restoring torque;

[0114] Bearing capacity reliability index and load-bearing safety factor The calculation method is as follows:

[0115] ,and: ,

[0116] Statistical parameters of the overturning moment of a pontoon bridge: The mean overturning moment and the overturning moment are calculated as follows:

[0117] ,

[0118] in, The average overturning moment, For the density of water, This is the drag force coefficient. For the area facing the flow, The average water flow velocity. The distance from the point of application of the resultant force of the water flow to the overturning axis. The standard deviation of the overturning moment. For water flow velocity, The standard deviation of the water flow velocity. The standard deviation of the overturning moment;

[0119] Statistical parameters of the restoring moment of a pontoon bridge: The mean and standard deviation of the restoring moment are calculated as follows:

[0120]

[0121] To restore the mean value of the torque, For the drainage volume of the pontoon bridge platform, Let be the stability lever arm function that varies with the tilt angle of the pontoon bridge. For the standard deviation of drainage volume, For the standard deviation of the tilt angle, The standard deviation of the restoring torque.

[0122] Among them, the probability of risk The calculation method is as follows:

[0123]

[0124] It is the standard normal distribution function.

[0125] Safety Classification SC It uses a three-category coding system, representing safety, warning, and danger, as follows:

[0126] ,

[0127] in, These are safety classification indicators.

[0128] During the training process of the agent decision-making model, the policy network acquires environmental data, task state vectors, historical state sequences, and the fusion reliability and security index from the previous time step. The system calculates the average action value output; based on the average action value from the policy network, the security prediction network calculates the fusion reliability security index for the current time. Output the total reward to the policy network.

[0129] Specifically, the reward function is defined as:

[0130]

[0131]

[0132] in, For the total reward, As a time reward, As a safety reward, To encourage rewards, To smooth out rewards, Weighted by time reward For time step, This represents the Euclidean distance from the current location to the target point. This is the total distance from the starting point to the target point. For security reward weighting, For the target safety threshold, To guide the weighting of rewards, The distance of the previous step. The current step distance, To smooth out reward weights, This is the current control action vector. The control action vector of the previous time, It is the square of the Euclidean norm.

[0133] After training, the path planning model determines the path plan based on the total reward.

[0134] The P130 interactive module is used to output real-time path planning results, risk probabilities, and safety classifications.

[0135] Specifically, the interaction module inputs environmental data into the environment perception module through third-party interfaces or management interfaces; based on the functional functions of the equipment and pontoon bridge, it displays the load-bearing stability of the equipment and pontoon bridge in real time; it obtains real-time path planning results, risk probabilities, and safety classifications from the path planning model, generates risk warning information, and displays it through the human-computer interaction interface.

[0136] This invention provides an embodiment: Taking a lake as an example, data on the lake's breach water level reaching 11 meters and flow velocity ranging from approximately 0.5 to 1.8 m / s are collected. Rescue personnel integrate ground-penetrating radar sensors, water flow velocity meters, and attitude sensors into their electronic devices (such as laptops). This allows for real-time detection of geological bearing capacity, water flow velocity and flow field, equipment information, and pontoon bridge information. A digital model of the rescue environment is constructed on-site in real time. The data from this digital model, including geological bearing capacity, water flow velocity and flow field, equipment information, pontoon bridge information, and equipment and pontoon bridge positioning, is transmitted into a path planning model. This enables rapid decision-making for path planning of equipment and pontoon bridges. Figure 4 As shown; the path planning model is generated by pre-training the agent decision-making model with a reward function, and its training metrics are as follows. Figure 5 As shown.

[0137] This invention constructs a physical-data fusion intelligent architecture, embedding reliability theory into intelligent agents to guide and constrain data-driven processes based on physical laws, ensuring the engineering reliability of decision-making. In this invention, a multi-physics field coupled constraint system is established, comprehensively considering the dynamic coupling of geological bearing capacity, hydrodynamics, pontoon bridge stability, and equipment dynamics to form a global safety constraint. An adaptive closed-loop system is constructed, continuously updating during the rescue process through real-time sensing to adapt to dynamically changing environments. This achieves efficient and high-precision simulations, significantly improving the speed of decision-making and safety assessment while ensuring physical consistency, greatly enhancing rescue efficiency. It provides key technical support for the autonomous and intelligent development of emergency rescue equipment, possessing significant engineering application value and social significance.

[0138] The above-disclosed embodiments are merely a few specific examples of the present invention. However, the present invention is not limited thereto, and any variations that can be conceived by those skilled in the art should fall within the protection scope of the present invention.

Claims

1. A path planning system for levee breach rescue equipment, characterized by, The method comprises the following steps: An environment perception module is used to acquire environment data and position information, and to construct a rescue environment digital model; wherein the environment data comprises detected geological bearing capacity, water flow velocity flow field, boat bridge information and equipment information; the position information comprises accurate positioning of the equipment and accurate positioning of the boat bridge; the rescue environment digital model comprises a grid-based geological bearing capacity field and a time-varying water flow velocity vector field; A path planning model is used to determine path planning based on the rescue environment digital model; the path planning model is an intelligent agent decision model composed of a coupling reward function of a strategy network and a safety prediction network, and fusion of reliability safety indexes, and is generated by training based on a PPO algorithm; wherein the strategy network is used to acquire shared features, to output an action mean value to the safety prediction network, and to determine a path trajectory of the equipment target according to a total reward; An interaction module is used to input environment data to the environment perception module, to display the bearing stability of the equipment and the boat bridge in real time, and to output real-time path planning results, risk probability and safety classification; The calculation method of the fusion reliability safety indexes is as follows: ， wherein, is a fusion reliability safety index, is a stability safety factor, is a bearing safety factor, is a penalty factor, is a target reliability index, is a bearing capacity reliability index, is a stability reliability index; wherein, is generally 3.2-4.

2. Wherein, the stability reliable index and the stability safety factor The calculation method is: ， ， wherein is the mean of the overturning moments, is the mean of the restoring moments, is the standard deviation of the overturning moments, is the standard deviation of the restoring moments; The bearing capacity reliability index and the bearing safety factor The calculation method is: and , The calculation of the overturning moment mean value and the standard deviation of the overturning moment is as follows, ， wherein, is the mean of the overturning moment, is the water density, is the drag force coefficient, is the frontal area, is the mean of the water flow velocity, is the distance of the point of action of the water flow resultant from the overturning axis, is the standard deviation of the overturning moment, is the water flow velocity, is the standard deviation of the water flow velocity, is the standard deviation of the overturning moment; The calculation of the mean value of the restoring moment and the standard deviation of the restoring moment is as follows, Mm is the mean of the restoring moment, V is the displacement of the pontoon platform, Mm is the mean of the restoring moment, σV is the standard deviation of the displacement, σθ is the standard deviation of the inclination angle, σMm is the standard deviation of the restoring moment.

2. The system of claim 1, wherein, The shared features comprise environment data, a task state vector, a historical state sequence and safety state information; The task state vector includes equipment target point relative position , boat bridge target point relative position , equipment target relative azimuth , boat target relative azimuth , equipment to target point distance , boat to target point distance , current time , path progress ; The historical state sequence includes a position sequence of past T steps , an action sequence of past T steps , m is an action dimension; The security state information includes a fusion reliability security indicator of a last time .

3. The system of claim 1, wherein, The strategy network comprises a multi-modal feature extraction layer, a feature fusion layer and a strategy output layer; The multi-modal feature extraction layer extracts multi-modal features from the shared features; The feature fusion layer outputs after splicing the multi-modal features through an activation function ReLU; The strategy output layer calculates and outputs an action mean value output through an activation function.

4. The system of claim 1, wherein, The safety prediction network comprises a shared feature extraction layer, a value evaluation layer and a safety prediction layer; The shared feature extraction layer is used to extract common features needed to be used from the shared features; The value evaluation layer calculates and outputs a value function based on the fusion reliability safety indexes; The safety prediction layer is composed of a reliability analysis module, a risk probability module and a safety classification module, and outputs fusion reliability safety indexes, risk probability and safety classification.

5. The system of claim 4, wherein, The value function is expressed as, , wherein, is a value function for time t, is a temporal difference error for time t, is a reward decay coefficient, is a decay coefficient in generalized advantage estimation, is a trajectory length; wherein the timing difference error is calculated as: wherein is the reward value at time t, is the fusion reliability safety index S at time t+1 t+1 follows the current policy.

6. The system of claim 4, wherein, The reliability analysis module is used to calculate the fusion reliability safety index through the environment data and the accurate positioning of the equipment and the boat bridge ; Computing a fusion reliability safety index At this point, the function of the equipment and the bridge is first defined, indicated as: ， wherein, is a structure safety margin characterized by a bearing capacity function, is a maximum bearing capacity of the ground, is an equipment contact pressure, is a stability safety margin characterized by a stability function, is a righting moment of the bridge platform against overturning, is an overturning moment caused by external forces; The calculation method of the ground pressure of the equipment is as follows. ​ ， wherein, is the total weight of the equipment, is the number of grounding units, is the unit grounding area, is the equipment tilt angle, is the equipment grounding pressure standard deviation, is the equipment weight standard deviation, is the grounding area standard deviation, is the total grounding area.

7. The system of claim 1, wherein, The calculation method of the risk probability is as follows, wherein, is the risk probability, is the standard normal distribution function; The safety classification is a three-classification coding, which is safe, warning and dangerous, and is represented as ， wherein is a safety classification, is a safety classification index.

8. The system of claim 1, wherein, The reward function is defined as wherein, is a total reward, is a time reward, is a safety reward, is a guidance reward, is a smoothing reward, is a time reward weight, is a time step, is a Euclidean distance from the current position to the goal point, is a total distance from the start point to the goal point, is a safety reward weight, is a goal safety threshold, is a guidance reward weight, is a previous step distance, is a current step distance, is a smoothing reward weight, is a current control action vector, a control action vector of the previous time, is a squared Euclidean norm.

Images