Mobile wireless charging dynamic scheduling system and method based on multi-mode perception

Abstract

The invention discloses a mobile wireless charging dynamic scheduling system and method based on multi-modal perception, and aims to solve the technical problem that the whole process of mobile wireless charging cannot be automatically managed due to the fact that the current mobile wireless charging dynamic scheduling system cannot be deeply coordinated in the use process. The multi-mode sensing unit is used for monitoring and acquiring information of environment obstacles, a spatial topological structure, dynamic path information, equipment identity, battery health degree, charging state parameters, and spatial alignment and energy transmission efficiency of a transmitting coil and a receiving coil; and the space-time calibration and data fusion module is used for realizing space-time synchronization of the sensor based on an SST-Calib algorithm. Through deep cooperation of multi-mode perception, intelligent decision and accurate execution, full-process automatic management of mobile wireless charging is realized, and the system is suitable for high-efficiency energy scheduling requirements of scenes such as intelligent manufacturing, logistics storage and intelligent transportation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of wireless charging, in particular to a mobile wireless charging dynamic scheduling system and method based on multi-modal perception. BACKGROUND

[0002] Wireless sensor networks are network systems composed of numerous sensor nodes, which play a key role in environmental monitoring, health monitoring, military reconnaissance and other fields. These nodes are usually powered by limited batteries, facing the challenge of energy limitation. With the wide application of wireless sensor networks in forest monitoring, environmental perception, green energy saving, industrial monitoring, health care and other fields, how to effectively manage the energy consumption of nodes, prolong the network life cycle and ensure the quality of monitoring tasks has become the core problem of research in this field.

[0003] In the process of realizing the application, the inventors found that at least the following problems in the prior art have not been solved. In the use process, the traditional mobile wireless charging dynamic scheduling system cannot be deeply coordinated in the use process, so that the whole process of mobile wireless charging cannot be automatically managed. Therefore, a new technical solution needs to be designed to solve it. SUMMARY

[0004] The purpose of the present application is to provide a mobile wireless charging dynamic scheduling system and method based on multi-modal perception to solve the technical problem that the current mobile wireless charging dynamic scheduling system cannot be deeply coordinated in the use process, so that the whole process of mobile wireless charging cannot be automatically managed.

[0005] To achieve the above purpose, the present application provides the following technical scheme: a mobile wireless charging dynamic scheduling system and method based on multi-modal perception, comprising Multi-modal sensing unit: for monitoring and acquiring information of environmental obstacles, spatial topology structure, dynamic path information, device identity, battery health, charging state parameters, spatial alignment degree of transmitting coil and receiving coil, and energy transmission efficiency; Space-time calibration and data fusion module: based on SST-Calib algorithm to realize space-time synchronization of sensors, and eliminate spatial deviation of laser radar and camera through bidirectional semantic alignment loss function; Dynamic scheduling algorithm engine: for battery adaptive priority evaluation and distributed path planning; Energy management and collaborative control module: adopting model predictive control (MPC) algorithm to dynamically adjust output power according to power grid load and device charging demand; Mobile charging unit: for planning of mobile chassis and adjustment of intelligent coupling mechanism; Charging execution module: for adjustment of high-frequency resonant circuit and battery safety protection detection.

[0006] As a preferred embodiment of the present application, the multi-modal sensor array unit comprises environment perception module: for real-time acquisition of environmental obstacle distribution, spatial topology and dynamic path information; device state perception module: for synchronous acquisition of device identity, battery health and charging state parameters; charging coupling state perception module: for real-time monitoring of spatial alignment and energy transmission efficiency of transmitting coil and receiving coil.

[0007] As a preferred embodiment of the present application, the dynamic scheduling algorithm engine comprises adaptive priority evaluation module: for fusion of device remaining power, power decay rate and spatial clustering density parameters; distributed path planning module: for improved ant colony algorithm combined with real-time traffic flow data to realize multi-charging unit collaborative obstacle avoidance.

[0008] As a preferred embodiment of the present application, the mobile charging unit comprises mobile chassis: using omnidirectional wheels and Mecanum wheels for mixed driving, supporting 360° turning; intelligent coupling mechanism: integrating six-axis mechanical arm and adaptive pose adjustment algorithm to complete space pose adjustment of charging coil.

[0009] As a preferred embodiment of the present application, the charging execution module comprises high-frequency resonant circuit module: using GaN power amplifier for power dynamic adjustment; safety protection system: for overcurrent protection, overheat protection and foreign matter detection.

[0010] As a preferred embodiment of the present application, the sensor space-time synchronization based on SST-Calib algorithm eliminates the spatial deviation of laser radar and camera through bidirectional semantic alignment loss function, the SST-Calib algorithm includes one-way (laser point - pixel point) semantic alignment loss calculation, pixel point - laser point loss calculation, bidirectional semantic alignment loss calculation of the Xth iteration, projection equation considering time delay and modified bidirectional loss function for joint space-time parameter estimation: one-way (laser point - pixel point) semantic alignment loss: defined as L s,k =1 / ∣S k ∣∑ i∈Sk ∥p u,i −q u,i ∥2 where S k is the set of laser points in the camera field of view after projection, pu,i is the laser projection point, q u,i is the nearest laser point belonging to the same class, this loss function is used to measure the distance deviation of the laser point projected to the image plane and the pixel point with the same semantic label; Pixel-laser point loss: defined as L p,k =1 / ∣P k ∣∑ j∈Pk ∥p v,j −q v,j ∥2 where P k is the set of pixels participating in the calculation, p v,j is the point after the pixel is projected to the laser point cloud space, q v,j is the nearest laser point belonging to the same class, this loss function is a supplement to the one-way loss, further measuring the matching degree of the two from the direction of pixel to laser point; Bidirectional semantic alignment loss of the Xth iteration: denoted as L (X) =α (X) L s +β (X) L p +γ (X) ∥ξ^ (X) −ξ init ∥2 where α (X) , β (X) , γ (X) are weight coefficients used to adjust the importance of different loss terms; ξ^ (X) is the estimated extrinsic parameter of the Xth iteration, ξ init is the initial extrinsic parameter, ∥ξ^ (X) −ξ init ∥2 is a regularization term, the purpose is to make the extrinsic parameter estimation closer to the initial value, avoid excessive deviation from the initial guess; Projection equation considering time delay: The translation offset between P k and I k+δ can be expressed as t δ,k =v^ k *δ where v^ k is the velocity between adjacent two images I k+δ and I k+δ−1 estimated by visual odometry, δ is the time delay between P k and I k+δ , this equation is used to compensate for the translation deviation between laser point cloud and camera image caused by time delay, which plays a key role in joint space-time parameter calibration The modified bidirectional loss function is used for joint space-time parameter estimation: adjust to L joint (X) =α (X) L s +β (X) L p +γ (X) ∥ξ^ (X) −ξinit∥2+λ (X) ∥δ^ (X) −δ init ∥2 On the basis of the original bidirectional loss function, increase λ (X) ∥δ^ (X) −δ init ∥2, Where λ (X) is the weight coefficient, δ^ (X) is the time delay estimated by the Xth iteration, δ init is the initial time delay estimation value, which is used to estimate the space and time calibration parameters simultaneously, so that the algorithm can accurately estimate the time delay between sensors while optimizing the external parameters.

[0011] As a preferred embodiment of the application, the model predictive control (MPC) algorithm is used to dynamically adjust the output power according to the power grid load and equipment charging demand, and the formula of the model predictive control (MPC) algorithm is specifically: The core of MPC is to solve the optimal control sequence u(k), u(k+1),..., u(k+N−1) in the future finite time domain [k, k+N−1] based on the current state Y(k) at each control time k, so as to minimize the objective function, which usually includes tracking error and control amount penalty, the formula is as follows: J=∑ i=0 N−1 [∥Y(k+i∣k)−Y ref (k+i)∥ Q 2 +∥u(k+i∣k)−u ref (k+i)∥ R 2 ]+∥Y(k+N∣k)−Y ref (k+N)∥ P 2 Where, N: prediction horizon (number of steps of future optimization); Y(k+i∣k): based on the measurement value at time k, predict the system state in the future i steps; u(k+i∣k): the control amount optimized in the future i steps at time k; Y ref , uref : desired state trajectory and control amount trajectory (reference trajectory) Q, R, P: weight matrices (all positive definite matrices), respectively used for penalizing state tracking error, control amount change and terminal state error.

[0012] As a preferred embodiment of the application, the improved ant colony algorithm is combined with real-time traffic flow data to realize multi-charging unit cooperative obstacle avoidance, and the ant colony algorithm formula is as follows:

[0013] Wherein: τ ij (t): pheromone concentration on path (i,j) at t time; η ij 1 / d ij : heuristic factor (d ij is the distance between i and j, reflecting the inherent advantages and disadvantages of the path); α,β: weight coefficients of pheromone and heuristic factor; allowed k : node set not visited by ant k; Improved direction and formula: Dynamic weight coefficient (adaptive α,β) To balance exploration (pheromone) and utilization (heuristic information), let α,β dynamically adjust with iteration (such as increasing β at the beginning and increasing α at the later stage) Where t is the current iteration number, T max is the maximum iteration number, α0,α max ,β0,β max are the initial and maximum weights.

[0014] As a preferred embodiment of the application, the mobile wireless charging dynamic scheduling step is as follows: S1. First, the environmental perception module scans the charging area at a frequency of 20Hz to generate a three-dimensional map containing obstacle coordinates and personnel density, and then the device state perception module synchronously acquires device ID, battery temperature and charging demand through RFID and infrared thermal imaging; S2. Then, the SST-Calib algorithm is used to perform space-time alignment on the laser radar point cloud and camera image, eliminate the spatial deviation between sensors through a bidirectional semantic loss function, and then automatically adjust the sensor weight according to the ambient light intensity through a confidence dynamic weighting module; S3. According to the adaptive priority algorithm, a priority matrix is generated according to the remaining power of the device, the historical charging frequency and the spatial position, and the Hungarian algorithm is used for task allocation; S4. Through the improved ant colony algorithm, the path is dynamically optimized through pheromone concentration, and when multiple charging units are expected to meet within 5m, a path deviation mechanism is triggered: target point information is exchanged through the communication bus; S5. When the charging unit reaches the target position, the six-axis mechanical arm adjusts the posture of the transmitting coil through visual guidance, the high-frequency resonant circuit dynamically adjusts the output according to the demand power feedback by the battery management system, and the foreign matter detection system monitors the charging area in real time through magnetic field distortion analysis, when the charging unit fails, the mechanical emergency release mechanism automatically disconnects the charging connection, and returns to the maintenance point through the backup power supply; S6. The communication bus based on the OPC UA protocol supports real-time synchronization of device state, the two-way communication delay between the charging unit and the device to be charged is less than 10ms, ensuring accurate execution of scheduling instructions, the power grid load monitoring module feeds back the power grid state in real time, when the load exceeds 80% of the rated capacity, the system automatically adjusts the charging power of non-urgent tasks, the model predictive control algorithm predicts the device charging demand according to historical charging data, and schedules the charging unit to the hot spot area in advance.

[0015] As a preferred embodiment of the present application, the Hungarian algorithm is specifically: Suppose there are n tasks and n executors, and the cost matrix is C=(c ij ) n×n , wherein c ij represents the cost of the i th executor completing the j th task; Objective: find a permutation matrix X=(x ij ) n×n (x ij =1 only when the i th person is assigned to complete the j th task, and the rest is 0), so that the total cost is minimized: Compared with the prior art, the beneficial effects of the present application are as follows: The present application forms a closed-loop cooperation through the deep cooperation of multi-modal perception, intelligent decision-making and precise execution, so as to achieve the full-process automation of mobile wireless charging from sensing demand to completing charging without manual intervention, and because of its efficient and automated energy scheduling capability, it can meet the needs of equipment endurance in intelligent manufacturing, robot power supply in logistics and warehousing, vehicle power supply in intelligent transportation and other scenarios, and solve the dynamic charging problems of various mobile devices. BRIEF DESCRIPTION OF DRAWINGS

[0016] Other features, objects and advantages of the present application will become more apparent through reading the detailed description of the non-limiting embodiments with reference to the following drawings: Figure 1 The flowchart of the present application. DETAILED DESCRIPTION

[0017] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0018] Example 1: Refer to Figure 1 A mobile wireless charging dynamic scheduling system and method based on multimodal sensing, including Multimodal sensing unit: used for monitoring and acquiring information such as environmental obstacles, spatial topology, dynamic path information, device identity, battery health, charging status parameters, spatial alignment of transmitting coil and receiving coil, and energy transfer efficiency; Spatiotemporal calibration and data fusion module: Based on the SST-Calib algorithm, it realizes spatiotemporal synchronization of sensors and eliminates the spatial deviation between lidar and camera through a bidirectional semantic alignment loss function; Dynamic scheduling algorithm engine: used for battery adaptive priority evaluation and distributed path planning; Energy Management and Cooperative Control Module: Employs Model Predictive Control (MPC) algorithm to dynamically adjust output power based on grid load and equipment charging demand; Mobile charging unit: used for the planning of mobile chassis and the adjustment of intelligent coupling mechanism; Charging execution module: used for adjusting the high-frequency resonant circuit and detecting battery safety protection.

[0019] By deeply integrating "multimodal perception" (integrating data from multiple sensors to comprehensively understand the environment and equipment status), "intelligent decision-making" (dynamically planning charging priorities and paths based on algorithms), and "precise execution" (mobile charging units efficiently completing charging operations), a closed-loop collaboration is formed, achieving full automation of the mobile wireless charging process from sensing needs to completing charging, without human intervention. Furthermore, due to its efficient and automated energy scheduling capabilities, it can meet the needs of scenarios such as equipment battery life in intelligent manufacturing, robot power supply in logistics and warehousing, and vehicle recharging in intelligent transportation, solving the dynamic charging problem of various mobile devices.

[0020] Specifically, the multimodal sensor array unit includes Environmental perception module: includes ultrasonic sensor (range accuracy ±2mm), lidar (scanning frequency 30Hz), and high-definition industrial camera (resolution 4K@60fps), used to collect environmental obstacle distribution, spatial topology and dynamic path information in real time; Device status sensing module: integrates RFID tags (identification distance 0-5m), infrared thermal imager (temperature resolution 0.05℃), and current and voltage sensors (accuracy ±0.1% FS), and can simultaneously acquire device identity, battery health (SOH) and charging status parameters; Charging coupling status sensing module: It adopts a triaxial magnetometer (accuracy ±0.1μT) and a phase detection circuit to monitor the spatial alignment (error <1mm) and energy transmission efficiency (dynamic adjustment range 20%-95%) of the transmitting coil and receiving coil in real time.

[0021] Furthermore, the dynamic scheduling algorithm engine includes Adaptive priority evaluation module: It integrates parameters such as remaining battery power (threshold <20% triggers emergency scheduling), battery decay rate (typical value for drones is 5% / min), and spatial cluster density (cluster scheduling is initiated when there are more than 3 devices within a radius of 5m), and uses a reinforcement learning model (Q-learning) to dynamically adjust task priorities with a response time of <200ms; Distributed path planning module: An improved ant colony algorithm combined with real-time traffic flow data (update frequency 1Hz) enables multi-charging unit collaborative obstacle avoidance, with path planning error <0.5m and empty driving rate reduced by 40%.

[0022] Furthermore, the mobile charging unit includes Mobile chassis: It adopts a hybrid drive system of omnidirectional wheels and Mecanum wheels, supports 360° steering (turning radius <0.8m), has a maximum moving speed of 2m / s, and is equipped with a dual-winding motor (peak power 5kW) and a mechanical emergency release mechanism; Intelligent coupling mechanism: It integrates a six-axis robotic arm (repeat positioning accuracy ±0.02mm) and an adaptive attitude adjustment algorithm, which can complete the spatial attitude adjustment of the charging coil (pitch angle ±15°, yaw angle ±30°) within 5s.

[0023] It is worth noting that the charging execution module includes High-frequency resonant circuit module: operating frequency 200kHz, using GaN power amplifier (efficiency >95%), supporting dynamic power adjustment from 10-150kW; Safety protection system: used for overcurrent protection, overheat protection and foreign object detection, including overcurrent protection (response time <5μs), overheat protection (temperature threshold 85℃) and foreign object detection (metal recognition accuracy >φ3mm).

[0024] It is worth noting that the sensor spatiotemporal synchronization based on the SST-Calib algorithm eliminates the spatial deviation between the lidar and the camera through a bidirectional semantic alignment loss function. The SST-Calib algorithm includes unidirectional (laser point to pixel) semantic alignment loss calculation, pixel to laser point loss calculation, bidirectional semantic alignment loss calculation in the Xth iteration, a projection equation considering time delay, and a modified bidirectional loss function for joint spatiotemporal parameter estimation. One-way (laser point - pixel point) semantic alignment loss: Defined as L s,k =1 / ∣S k |∑ i∈Sk ∥p u,i -q u,i ∥2 Where S k p is the set of laser points within the camera's field of view after projection. u,i It is the laser projection point, q u,i It is the nearest pixel belonging to the same category. This loss function is used to measure the distance deviation between the laser point and the pixel with the same semantic label after the laser point is projected onto the image plane. Pixel-Laser Point Loss: Defined as L p,k =1 / ∣P k |∑ j∈Pk ∥p v,j -q v,j ∥2 Where P k It is the set of pixels involved in the calculation, p v,j It is the point after the pixel is projected onto the point cloud space of the LiDAR, q v,j It is the nearest LiDAR point belonging to the same category. This loss function is a supplement to the unidirectional loss, and it further measures the degree of matching between the two from the direction from the pixel point to the LiDAR point. The bidirectional semantic alignment loss in the Xth iteration: Represented as L (X) =α (X) L s +β (X) L p +γ (X) ∥ξ^ (X) −ξ init ∥2 Where α (X) β (X) γ (X) ξ^ is a weighting coefficient used to adjust the importance of different loss terms. (X) Let ξ be the extrinsic parameter estimated in the Xth iteration. init As the initial extrinsic parameter, ∥ξ^ (X) −ξinit ∥2 is a regularization term, the purpose of which is to make the extrinsic parameter estimates closer to the initial values ​​and avoid excessive deviation from the initial guesses; Projection equation considering time delay: P k and I k+δ The translational offset between them can be expressed as t δ,k =v^ k *δ Where v^ k It is the estimation of two adjacent frames of images I through visual odometry. k+δ and I k+δ−1 The velocity between them, δ is P k with I k+δ The equation, which compensates for the translational deviation between the lidar point cloud and the camera image caused by the time delay, plays a crucial role in joint spatiotemporal parameter calibration. The modified bidirectional loss function is used for joint spatiotemporal parameter estimation: adjusted to L joint (X) =α (X) L s +β (X) L p +γ (X) ∥ξ^ (X) −ξinit∥2+λ (X) ∥δ^ (X) −δ init ∥2 Based on the original bidirectional loss function, λ is added. (X) ∥δ^ (X) −δ init ∥2, Where λ (X) It is the weighting coefficient, δ^ (X) δ is the time delay estimated for the Xth iteration. init This initial time delay estimate is used to simultaneously estimate spatial and temporal calibration parameters, enabling the algorithm to accurately estimate the time delay between sensors while optimizing extrinsic parameters. It is worth noting that the proposed algorithm employs Model Predictive Control (MPC) to dynamically adjust the output power based on grid load and equipment charging demand. The specific formula for the MPC algorithm is as follows: The core of MPC is to solve for the optimal control sequence u(k), u(k+1), ..., u(k+N−1) over the finite time domain [k,k+N−1] at each control time k, based on the current state Y(k), to minimize the objective function. The objective function typically includes two parts: tracking error and control quantity penalty, as shown in the following formula: J=∑ i=0 N−1 [∥Y(k+i∣k)−Yref (k+i)∥ Q 2 +∥u(k+i∣k)−u ref (k+i)∥ R 2 ]+∥Y(k+N∣k)−Y ref (k+N)∥ P 2 Where N: prediction time domain (number of future optimization steps); Y(k+i|k): Based on the measurement value at time k, predict the system state for the next i steps; u(k+i|k): The control quantity optimized at time k for the next i steps; Y ref u ref : Expected state trajectory and control quantity trajectory (reference trajectory); Q, R, P: Weight matrices (all positive definite matrices), used to penalize state tracking error, control quantity change, and terminal state error, respectively.

[0025] It is worth emphasizing that the improved ant colony algorithm, combined with real-time traffic flow data, enables multi-charging unit cooperative obstacle avoidance. The specific formula for the ant colony algorithm is as follows:

[0026] in: τ ij (t): The pheromone concentration on the path (i,j) at time t; η ij 1 / d ij Heuristic Factor (d) ij (The distance between i and j reflects the inherent advantages and disadvantages of the path). α,β: Weighting coefficients of pheromones and heuristic factors; allowed k The set of nodes that ant k has not visited; Improvement directions and formulas: Dynamic weighting coefficients (adaptive α, β) To balance exploration (pheromones) and utilization (heuristic information), α and β are dynamically adjusted with iteration (e.g., β increases in the early stage and α increases in the later stage). Where t is the current iteration number, T max Let α0, α be the maximum number of iterations. max ,β0,β max These are the initial and maximum weights.

[0027] The dynamic scheduling method for mobile wireless charging based on multimodal perception includes the following steps: S1. First, the environmental perception module scans the charging area at a frequency of 20Hz to generate a 3D map containing obstacle coordinates (accuracy ±5mm) and personnel density (statistical resolution 0.5m×0.5m). Then, the device status perception module synchronously acquires the device ID (64-bit code length), battery temperature (monitoring range -40℃~125℃), and charging requirements (protocol type supports SAE J2954) through RFID and infrared thermal imaging. S2. The SST-Calib algorithm is then used to perform spatiotemporal alignment between the LiDAR point cloud and the camera image. Spatial deviation between sensors is eliminated through a bidirectional semantic loss function (point-to-pixel matching error < 2 pixels). Then, the confidence dynamic weighting module automatically adjusts the sensor weights according to the ambient light intensity (monitoring range 0-10000 lux): the visual weight is increased to 80% in strong light environment and the infrared weight is increased to 70% in low light environment. S3. Then, based on the adaptive priority algorithm, a priority matrix is ​​generated according to the device's remaining power (weight 40%), historical charging frequency (weight 30%), and spatial location (weight 30%). The response time for emergency tasks (remaining power <15%) is <10s. Furthermore, the Hungarian algorithm is used for task allocation to ensure that the average distance between the charging unit and the target device is shortened by 30%, while avoiding path conflicts (conflict detection accuracy <0.3m). S4. By using an improved ant colony algorithm to dynamically optimize the path based on pheromone concentration (update cycle 200ms), combined with real-time traffic flow data (automatic deceleration to 0.5m / s when pedestrian density > 2 people / m²), the path planning efficiency is improved by 50%. When multiple charging units are expected to meet within 5m, a path offset mechanism is triggered: target point information is exchanged through the communication bus, and the path is replanned using the A* algorithm, with an obstacle avoidance success rate > 99%. S5. When the charging unit reaches the target position, the six-axis robotic arm adjusts the attitude of the transmitting coil through visual guidance (positioning accuracy ±0.1mm) to ensure that the coupling coefficient with the receiving coil is >0.85. The high-frequency resonant circuit dynamically adjusts the output according to the power demand feedback from the battery management system (BMS) (adjustment accuracy ±1%), and the charging efficiency is adaptively optimized within the range of 85%-95%. The foreign object detection system monitors the charging area in real time through magnetic field distortion analysis (detection sensitivity ±0.5μT). When a metal foreign object is detected, the power output is immediately cut off (response time <20μs). When the charging unit malfunctions (such as motor overload), the mechanical emergency release mechanism automatically disconnects the charging connection and returns to the maintenance point through the backup power supply (30 minutes of continuous operation). S6. The communication bus built on the OPC UA protocol supports real-time synchronization of device status. The bidirectional communication latency between the charging unit and the device to be charged is <10ms, ensuring the accurate execution of scheduling instructions. The power grid load monitoring module (accuracy ±0.5kW) provides real-time feedback on the power grid status. When the load exceeds 80% of the rated capacity, the system automatically adjusts the charging power of non-urgent tasks (derating by 20%-50%). The Model Predictive Control (MPC) algorithm predicts the charging demand of the device based on historical charging data (24-hour time window) and schedules the charging unit to hotspot areas (such as logistics and warehousing areas) in advance, reducing the empty running rate to below 15%. The system optimizes the charging sequence through the energy management module, prioritizing the allocation of resources to devices with high energy efficiency ratios (such as electric vehicles), resulting in an overall energy efficiency improvement of 12%.

[0028] The Hungarian algorithm is as follows: Suppose there are n tasks and n executors, and the cost matrix is ​​C = (c ij ) n×n , where c ij This represents the cost for the i-th executor to complete the j-th task; Objective: Find a permutation matrix X=(x ij ) n×n (x only when the i-th person is assigned to complete the j-th task) ij (where 1 is constant and the rest are 0), to minimize the total cost: .

[0029] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. It will be apparent to those skilled in the art that the invention is not limited to the details of the exemplary embodiments described above, and that the invention can be implemented in other specific forms without departing from its spirit or essential characteristics. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the scope of the invention. No reference numerals in the claims should be construed as limiting the scope of the claims.

[0030] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.

Claims

1. A mobile wireless charging dynamic scheduling system based on multimodal sensing, characterized in that: include Multimodal sensing unit: used for monitoring and acquiring information such as environmental obstacles, spatial topology, dynamic path information, device identity, battery health, charging status parameters, spatial alignment of transmitting coil and receiving coil, and energy transfer efficiency; Spatiotemporal calibration and data fusion module: Based on the SST-Calib algorithm, it realizes spatiotemporal synchronization of sensors and eliminates the spatial deviation between lidar and camera through a bidirectional semantic alignment loss function; Dynamic scheduling algorithm engine: used for battery adaptive priority evaluation and distributed path planning; Energy Management and Cooperative Control Module: Employs Model Predictive Control (MPC) algorithm to dynamically adjust output power based on grid load and equipment charging demand; Mobile charging unit: used for the planning of mobile chassis and the adjustment of intelligent coupling mechanism; Charging execution module: used for adjusting the high-frequency resonant circuit and detecting battery safety protection.

2. The mobile wireless charging dynamic scheduling system based on multimodal perception according to claim 1, characterized in that: The multimodal sensor array unit includes Environmental perception module: used to collect real-time information on the distribution of environmental obstacles, spatial topology, and dynamic path. Device status sensing module: used to synchronously acquire device identity, battery health and charging status parameters; Charging coupling status sensing module: used to monitor the spatial alignment and energy transfer efficiency of the transmitting coil and receiving coil in real time.

3. The mobile wireless charging dynamic scheduling system based on multimodal perception according to claim 1, characterized in that: The dynamic scheduling algorithm engine includes Adaptive Priority Evaluation Module: Used to integrate device remaining power, power decay rate, and spatial cluster density parameters; Distributed path planning module: used to combine the improved ant colony algorithm with real-time traffic flow data to achieve collaborative obstacle avoidance by multiple charging units. The mobile wireless charging dynamic scheduling system based on multimodal perception according to claim 1, characterized in that: the mobile charging unit includes... Mobile chassis: It adopts a hybrid drive system of omnidirectional wheels and Mecanum wheels, supporting 360° steering; Intelligent coupling mechanism: integrates a six-axis robotic arm and an adaptive attitude adjustment algorithm to complete the spatial attitude adjustment of the charging coil.

4. The mobile wireless charging dynamic scheduling system based on multimodal perception according to claim 1, characterized in that: The charging execution module includes High-frequency resonant circuit module: uses a GaN power amplifier for dynamic power adjustment; Safety protection system: used for overcurrent protection, overheat protection and foreign object detection.

5. The mobile wireless charging dynamic scheduling system based on multimodal perception according to claim 1, characterized in that: The sensor spatiotemporal synchronization based on the SST-Calib algorithm is achieved by eliminating the spatial deviation between the lidar and the camera through a bidirectional semantic alignment loss function. The SST-Calib algorithm includes unidirectional (library point to pixel) semantic alignment loss calculation, pixel to lidar point loss calculation, bidirectional semantic alignment loss calculation in the Xth iteration, a projection equation considering time delay, and a modified bidirectional loss function for joint spatiotemporal parameter estimation. One-way (laser point - pixel) semantic alignment loss: Defined as L s,k =1 / ∣S k |∑ i∈Sk ∥p u,i -q u,i ∥2 Where S k p is the set of laser points within the camera's field of view after projection. u,i It is the laser projection point, q u,i It is the nearest pixel belonging to the same category. This loss function is used to measure the distance deviation between the laser point and the pixel with the same semantic label after the laser point is projected onto the image plane. Pixel-Laser Point Loss: Defined as L p,k =1 / ∣P k |∑ j∈Pk ∥p v,j -q v,j ∥2 Where P k It is the set of pixels involved in the calculation, p v,j It is the point after the pixel is projected onto the point cloud space of the LiDAR, q v,j It is the nearest LiDAR point belonging to the same category. This loss function is a supplement to the unidirectional loss, and it further measures the degree of matching between the two from the direction from the pixel point to the LiDAR point. The bidirectional semantic alignment loss in the Xth iteration is denoted as L. (X) =α (X) L s +β (X) L p +γ (X) ∥ξ^ (X) −ξ init ∥2 Where α (X) β (X) γ (X) ξ^ is a weighting coefficient used to adjust the importance of different loss terms. (X) Let ξ be the extrinsic parameter estimated in the Xth iteration. init As the initial extrinsic parameter, ∥ξ^ (X) −ξ init ∥2 is a regularization term, the purpose of which is to make the extrinsic parameter estimates closer to the initial values ​​and avoid excessive deviation from the initial guesses; Projection equation considering time delay: P k and I k+δ The translational offset between them can be expressed as t δ,k =v^ k *δ; Where v^ k It is the estimation of two adjacent frames of images I through visual odometry. k+δ and I k+δ−1 The velocity between them, δ is P k with I k+δ The equation, which compensates for the translational deviation between the lidar point cloud and the camera image caused by the time delay, plays a crucial role in joint spatiotemporal parameter calibration. The modified bidirectional loss function is used for joint spatiotemporal parameter estimation: adjusted to L joint (X) =a (X) L s +b (X) L p +g (X) ∥ξ^ (X) −ξinit∥2+λ (X) ∥d^ (X) −d init ∥2 Based on the original bidirectional loss function, λ is added. (X) ∥δ^ (X) −δ init ∥2, Where λ (X) It is the weighting coefficient, δ^ (X) δ is the time delay estimated for the Xth iteration. init This is the initial time delay estimate, used to simultaneously estimate spatial and temporal calibration parameters, enabling the algorithm to accurately estimate the time delay between sensors while optimizing extrinsic parameters.

6. The mobile wireless charging dynamic scheduling system based on multimodal perception according to claim 1, characterized in that: The aforementioned model predictive control (MPC) algorithm dynamically adjusts the output power based on grid load and equipment charging demand. The specific formula for the model predictive control (MPC) algorithm is as follows: The core of MPC is to solve for the optimal control sequence u(k), u(k+1), ..., u(k+N−1) over the finite time domain [k,k+N−1] at each control time k, based on the current state Y(k), to minimize the objective function. The objective function typically includes two parts: tracking error and control quantity penalty, as shown in the following formula: J=∑ i=0 N−1 [∥Y(k+i∣k)−Y ref (k+i)∥ Q 2 +∥u(k+i∣k)−u ref (k+i)∥ R 2 ]+∥Y(k+N∣k)−Y ref (k+N)∥ P 2 Where N: prediction time domain (number of future optimization steps); Y(k+i|k): Based on the measurement value at time k, predict the system state for the next i steps; u(k+i|k): The control quantity optimized at time k for the next i steps; Y ref u ref : Expected state trajectory and control quantity trajectory (reference trajectory); Q, R, P: Weight matrices (all positive definite matrices), used to penalize state tracking error, control quantity change, and terminal state error, respectively.

7. The mobile wireless charging dynamic scheduling system based on multimodal perception according to claim 1, characterized in that: The improved ant colony algorithm, combined with real-time traffic flow data, enables multi-charging unit cooperative obstacle avoidance. The specific formula for the ant colony algorithm is as follows:

8. Among them: τ ij (t): The pheromone concentration on the path (i,j) at time t; η ij 1 / d ij Heuristic factor (d) ij (The distance between i and j reflects the inherent advantages and disadvantages of the path). α,β: Weighting coefficients of pheromones and heuristic factors; allowed k The set of nodes that ant k has not visited; Improvement directions and formulas: Dynamic weighting coefficients To balance exploration (pheromones) and utilization (heuristic information), α and β are dynamically adjusted with iteration. Where t is the current iteration number, T max Let α0, α be the maximum number of iterations. max ,β0,β max These are the initial and maximum weights.

9. The mobile wireless charging dynamic scheduling method based on multimodal perception according to claim 1, characterized in that: The dynamic scheduling steps for mobile wireless charging are as follows: S1. First, the environmental perception module scans the charging area at a frequency of 20Hz to generate a 3D map containing obstacle coordinates and personnel density. Then, the device status perception module simultaneously acquires the device ID, battery temperature, and charging requirements through RFID and infrared thermal imaging. S2. Then, the SST-Calib algorithm is used to perform spatiotemporal alignment between the lidar point cloud and the camera image. The spatial deviation between sensors is eliminated through a bidirectional semantic loss function. Finally, the confidence dynamic weighting module automatically adjusts the sensor weights according to the ambient light intensity. S3. Then, based on the adaptive priority algorithm, a priority matrix is ​​generated according to the device's remaining power, historical charging frequency, and spatial location, and the Hungarian algorithm is used for task allocation. S4. By using an improved ant colony algorithm to dynamically optimize the path based on pheromone concentration, when multiple charging units are expected to meet within 5m, a path offset mechanism is triggered, and target point information is exchanged through the communication bus. S5. When the charging unit reaches the target position, the six-axis robotic arm adjusts the attitude of the transmitting coil through visual guidance. The high-frequency resonant circuit dynamically adjusts the output according to the power demand feedback from the battery management system. The foreign object detection system monitors the charging area in real time through magnetic field distortion analysis. When the charging unit fails, the mechanical emergency release mechanism automatically disconnects the charging connection and returns to the maintenance point through the backup power supply. S6. The communication bus built on the OPC UA protocol supports real-time synchronization of device status. The bidirectional communication delay between the charging unit and the device to be charged is <10ms, ensuring the accurate execution of scheduling instructions. The power grid load monitoring module provides real-time feedback on the power grid status. When the load exceeds 80% of the rated capacity, the system automatically adjusts the charging power of non-urgent tasks. The model predictive control algorithm predicts the charging demand of the device based on historical charging data and schedules the charging unit to the hot spot area in advance.

10. The mobile wireless charging dynamic scheduling method based on multimodal perception according to claim 9, characterized in that: The Hungarian algorithm is specifically as follows: Suppose there are n tasks and n executors, and the cost matrix is ​​C = (c ij ) n×n , where c ij This represents the cost for the i-th executor to complete the j-th task; Objective: Find a permutation matrix X=(x ij ) n×n (x only when the i-th person is assigned to complete the j-th task) ij (where 1 is constant and the rest are 0), to minimize the total cost: 。

Images