An energy consumption optimization and automatic charging control system for intelligent pet robots

Abstract

The present invention relates to the field of robot control technology, and in particular to an energy consumption optimization and automatic charging control system for an intelligent pet robot. It comprises a robot data acquisition module, an energy consumption status evaluation module, an energy efficiency optimization module, an automatic charging scheduling module and a battery health management module which are sequentially connected in communication. The system is driven by multi-dimensional sensor data, and extracts energy consumption characteristics by combining potential field mapping and sparse modeling to construct an energy consumption impact map; utilizes quantum behavior particle swarm algorithm to achieve joint optimization of paths and servo strategies; schedules charging paths through graph attention networks; and dynamically predicts battery health status based on incremental learning algorithms to improve robot operation efficiency, endurance and system intelligence. The present invention constructs a closed-loop control process from energy consumption evaluation to energy efficiency optimization, automatic charging and battery health management, and has good system integration and intelligent response capabilities, significantly improving the endurance and operational stability of pet robots in complex application scenarios.

Description

Technical Field

[0001] The present invention relates to the field of robot control technology, and in particular to an energy consumption optimization and automatic charging control system for an intelligent pet robot. Background Art

[0002] With the continued development of artificial intelligence, the Internet of Things, and robotic control technologies, smart pet robots are gradually becoming part of family life, assuming roles such as companionship, interaction, and comfort. To achieve natural behavioral responses and continuous user companionship, smart pet robots typically incorporate multimodal perception, path planning, voice interaction, and emotion recognition. These functions require high-frequency sensor data acquisition, complex decision-making calculations, and continuous servo motion control, resulting in high overall energy consumption.

[0003] Currently, in the field of intelligent robotics, some progress has been made in optimizing modules such as path planning, task scheduling, and behavior control. However, in-depth modeling and quantitative evaluation of the relationship between multi-dimensional operational data and energy consumption behaviors are lacking. Some systems have attempted to introduce energy consumption estimation methods based on static models, but these methods struggle to adapt to the dynamic characteristics of multiple scenarios and behaviors. Furthermore, charging management generally relies on a trigger mechanism with preset power thresholds, which makes it difficult to implement intelligent scheduling and proactive control based on task priority, battery health, and environmental factors. This can easily lead to low energy utilization and abnormal battery cell degradation. Summary of the Invention

[0004] To solve the above problems, the present invention provides an intelligent pet robot energy consumption optimization and automatic charging control system, which solves the problem of how to achieve accurate modeling and evaluation of the energy consumption behavior of the intelligent pet robot in a dynamic operating environment with multiple scenarios and multiple behaviors, and combines the task urgency and battery health status to achieve adaptive energy efficiency optimization and intelligent charging scheduling, thereby significantly improving the pet robot's endurance and operating stability in complex application scenarios.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] An intelligent pet robot energy consumption optimization and automatic charging control system includes a robot data acquisition module, an energy consumption status evaluation module, an energy efficiency optimization module, an automatic charging scheduling module, and a battery health control module that are sequentially connected in communication;

[0007] The robot data acquisition module is used to collect multi-dimensional operation data of the pet robot during operation in real time;

[0008] The energy consumption status assessment module is used to perform structured aggregation on the multi-dimensional operation data based on the potential field mapping algorithm and the feature sparse modeling mechanism, extract the space-action correlation features of the energy consumption behavior, and generate an energy consumption influencing factor map;

[0009] The energy efficiency optimization module is used to search for the optimal energy consumption solution between the task execution path and the servo control strategy based on the energy consumption influencing factor map by using the quantum behavior particle swarm algorithm, and generate a low-energy execution instruction set;

[0010] The automatic charging scheduling module is used to construct a heterogeneous graph structure with robots and charging piles as nodes based on the robot's current power level, task urgency, spatial location information, and energy consumption evaluation results. It uses a graph attention network algorithm to adaptively model the state weights of the charging path. Combined with the task execution time window constraint scheduling mechanism, it dynamically generates a charging priority queue and an optimal path plan, and controls the pet robot to autonomously navigate to the charging station.

[0011] The battery health control module is used to perform online modeling based on battery charge and discharge cycle data, internal resistance change rate and temperature fluctuation information using an incremental learning algorithm to evaluate battery degradation trends, generate health control instructions, and adjust maximum charge and discharge current and temperature control parameters.

[0012] Furthermore, the multi-dimensional operation data includes the pet robot's three-dimensional motion inertial navigation data, joint servo current waveform, visual information, sound response, posture information, interactive reaction, current and voltage data, and environmental temperature and humidity gradient distribution.

[0013] Furthermore, the operation process of the energy consumption status assessment module includes the following steps:

[0014] Constructing a robot behavior state tensor for energy consumption modeling, and performing spatiotemporal alignment on the multi-dimensional operation data to form an original feature tensor structure with time, spatial position, and behavior category as index dimensions;

[0015] Based on the potential energy field mapping algorithm, the original feature tensor is mapped to the energy response field, the energy consumption density function of the behavior path is constructed, and the high energy consumption concentration area and energy consumption mutation point are identified through gradient analysis;

[0016] For the identified high-energy consumption areas and action segments, a feature sparse modeling mechanism is used to construct an energy consumption attribution mapping matrix. A channel compression algorithm based on the minimum entropy increase criterion is used to mine the behavioral feature subsets and their activation timing patterns that are most sensitive to unit energy consumption growth.

[0017] Combining task identification and spatial location, a cross-temporal and spatial energy consumption behavior impact map is constructed. Based on the energy consumption behavior impact map, the influence centrality and intervention priority of each characteristic factor are calculated, and an energy consumption impact factor map is output.

[0018] Furthermore, the formula of the potential energy field mapping algorithm is as follows:

[0019] ;

[0020] in, represents the energy consumption potential density at the robot's spatial position (x, y) and time t; N represents the number of historical behavior samples; Indicates the current space coordinates; Represents the spatial coordinates of the i-th historical data; and are the timestamps of the current and i-th data respectively; Indicates the activation intensity of the energy consumption behavior corresponding to the i-th behavior data; represents the environmental disturbance factor; It represents the behavior type modulation factor, which is used to distinguish the impact of different actions on energy consumption; It represents the spatial scale control parameter, which determines the smooth range of potential energy diffusion in the spatial dimension; represents the time diffusion scale, reflecting the time decay rate of the historical behavior’s response to current energy consumption; Represents the angle between the direction vector of the i-th behavior and the current energy consumption gradient direction; Represents the directional coupling adjustment coefficient.

[0021] Furthermore, the energy consumption attribution mapping matrix is specifically a sparse behavior attribution matrix constructed based on the minimum entropy increase criterion to compress the feature channel. The rows represent the key behavioral feature dimensions that are sensitive to energy consumption, and the columns represent the divided task behavior time periods. The matrix elements are used to characterize the attribution intensity of each feature to the unit energy consumption increment. It is generated by channel compression, time series division and attribution modeling of multi-dimensional operation data.

[0022] Furthermore, the operation process of the energy efficiency optimization module includes the following steps:

[0023] Based on the energy consumption influencing factor map, the corresponding task execution sequence, spatial movement path and servo control instructions are extracted to construct a joint optimization variable set;

[0024] Initializing the optimization population, mapping each parameter combination of the joint optimization variable set into a search individual;

[0025] The quantum-behaved particle swarm algorithm is used to jointly optimize the path and servo strategy. The particle fitness function is calculated in each generation, and the particle state is iteratively updated based on the historical optimal solution and the current search direction.

[0026] In each iteration, the search individuals are comprehensively evaluated based on multi-dimensional indicators such as task execution time limit, path continuity, behavior control stability, and unit energy consumption level to generate the optimal energy consumption control plan;

[0027] The optimal energy consumption control scheme is mapped into a low-energy consumption control instruction set, which includes a target task sequence, a servo joint angle scheduling curve, a speed constraint parameter and a state feedback frequency.

[0028] Furthermore, the formula of the quantum-behaved particle swarm algorithm is as follows:

[0029] ;

[0030] in, represents the optimal control parameter solution of the pet robot in the current task cycle; Represents the historical optimal solution position of the particle in the kth generation; Represents the dynamic step coefficient, which is set based on the task urgency, its own power level and energy consumption evaluation value; Indicates the global optimal solution position in the current iteration, obtained by combining the historical behavior graph and energy consumption index evaluation; represents the nonlinear convergence adjustment factor; represents a random disturbance variable.

[0031] Furthermore, the operation process of the automatic charging scheduling module includes the following steps:

[0032] The robot's current battery percentage, remaining mission time, spatial coordinates, and accessibility information to surrounding charging stations are collected to construct a heterogeneous directed graph with the robot node and the charging station node as vertices. The edge weights of the heterogeneous directed graph are initialized and modeled by combining spatial distance, energy consumption estimation, and path risk coefficient.

[0033] The heterogeneous directed graph is trained and modeled based on a graph attention network. The attention mechanism is used to perform weighted learning on the power level, task urgency, and path cost between nodes. The robot's preference for candidate charging pile nodes is adaptively adjusted, and a path decision attention matrix is output.

[0034] Combined with the path decision attention matrix, a path optimization function based on the remaining time of the task and navigation energy consumption is constructed to dynamically generate a set of charging candidate paths, calculate the scheduling priority score of each path, and select the optimal charging scheduling path with the minimum path cost and meeting the time window constraint;

[0035] According to the optimal charging scheduling path, a navigation control instruction set is generated to drive the pet robot to autonomously navigate to the designated charging pile to perform the charging task; the navigation control instruction set includes a path point sequence, steering angle control, obstacle avoidance trigger conditions and remaining power threshold trigger mechanism.

[0036] Furthermore, the formula of the path optimization function is as follows:

[0037] ;

[0038] in, represents the scheduling priority score of the candidate charging path r; represents the energy consumption per unit distance of path r; Indicates the Euclidean path length between the robot's current spatial coordinates and the candidate charging pile; Indicates the maximum navigation distance that the current remaining battery power of the robot can support; represents the estimated risk cost coefficient of path r; Represents the ratio of the path time to the remaining time window of the task; Represents the pet robot's preference weight for each key node in the path; and Represents the cost factor weight coefficient.

[0039] Furthermore, the battery health control module specifically has a graded health warning mechanism, which divides the battery status into four levels: normal, mild degradation, critical degradation and severe failure based on the battery life prediction result and the health threshold range.

[0040] The beneficial effects of the present invention are:

[0041] The present invention uses a robot data acquisition module to collect multidimensional operational data, including motion inertial navigation, servo current, posture, and environmental data. This data is aggregated and processed based on potential field mapping and feature sparse modeling. This allows accurate capture of the correlation between energy consumption behavior and spatial motion, improving the timeliness and accuracy of energy consumption analysis. The energy efficiency optimization module, based on an energy factor map, employs a quantum-behaved particle swarm algorithm to jointly optimize the execution path and control strategy, effectively searching for a globally optimal low-energy solution. This prevents traditional methods from falling into local optimality, improving robot operational efficiency and power utilization. By constructing a heterogeneous graph structure with robots and charging stations as nodes and introducing a graph attention network model to dynamically model charging path status, charging tasks can be flexibly scheduled based on power status, task urgency, and location information, enhancing the intelligence and optimality of charging path selection and achieving truly autonomous navigation charging. The battery health control module uses an incremental learning algorithm for online battery status modeling and trend prediction. It dynamically adjusts the health control strategy based on charge-discharge and thermal characteristics, improving the system's responsiveness to battery anomalies, extending battery life, and ensuring device operational safety. Each module works together through communication connections, building a closed-loop control process from data collection, energy consumption assessment to energy efficiency optimization, automatic charging and battery health management. It has good system integration and intelligent response capabilities, significantly improving the pet robot's endurance and operational stability in complex application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 This is a module schematic diagram of an energy consumption optimization and automatic charging control system for an intelligent pet robot according to the present invention.

[0043] Figure 2 It is a flowchart of the operation process of the energy efficiency optimization module provided by one embodiment of the present invention.

[0044] Figure 3 The figure is a flow chart of the operation process of the automatic charging scheduling module provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0045] See also Figure 1-3 As shown, the present invention relates to an energy consumption optimization and automatic charging control system for an intelligent pet robot.

[0046] Example

[0047] An intelligent pet robot energy consumption optimization and automatic charging control system includes a robot data acquisition module, an energy consumption status evaluation module, an energy efficiency optimization module, an automatic charging scheduling module, and a battery health control module that are sequentially connected in communication;

[0048] The robot data acquisition module is used to collect multi-dimensional operation data of the pet robot in real time during its operation; the multi-dimensional operation data includes the pet robot's three-dimensional motion inertial navigation data, joint servo current waveform, visual information, sound response, posture information, interactive reaction, current and voltage data and environmental temperature and humidity gradient distribution.

[0049] It should be noted that the hardware composition and distribution layout are as follows:

[0050] Inertial Measurement Unit (IMU): This integrates a three-axis accelerometer and a three-axis gyroscope, located in the robot's trunk or center mass. It collects three-dimensional acceleration and angular velocity with a resolution of 0.01g / ° / s and a sampling frequency that can be set from 100 to 1000Hz.

[0051] Current / voltage sensor: connected to the robot servo control board and battery management system, collects the current waveform of each joint servo motor and the power supply voltage fluctuation for evaluating the execution status and power consumption;

[0052] Visual sensor system: Equipped with a wide-angle camera (RGB-D depth camera is optional) to collect image data and depth information within the robot's field of view, supporting tasks such as target recognition and user identification;

[0053] Microphone array: usually a MEMS microphone matrix (such as a 4-microphone ring arrangement), used to capture sound response information, such as voice commands and ambient sound source positioning;

[0054] Joint angle and posture encoder: embedded in the servo drive unit, it records the angle changes of each moving joint in real time and combines it with IMU data for posture fusion solution;

[0055] Tactile sensor and interaction unit: including pressure-sensitive membranes or capacitive touch modules distributed on the case, head, and back, used to detect user interaction behaviors such as touching and tapping;

[0056] Environmental sensor: It is equipped with an integrated temperature and humidity module (such as DHT22) and an infrared thermal array module to construct a temperature and humidity distribution map of the surrounding space and form environmental gradient information.

[0057] To reduce data redundancy and power consumption, the module also has a dynamic sampling and event-driven mechanism: when the robot is stationary or in standby state, non-critical sensors such as vision and audio modules enter low-power mode and wake up only when triggered by external stimuli; the main controller regularly evaluates data correlation and energy consumption, and uses information gain thresholds to control the sampling rate of each sensor; it supports edge-adaptive compression coding (such as T-DPCM + Huffman) mechanisms, and locally compresses continuously sampled data before transmission, reducing communication bandwidth consumption.

[0058] The energy consumption status assessment module is used to perform structured aggregation on the multi-dimensional operation data based on the potential field mapping algorithm and the feature sparse modeling mechanism, extract the space-action correlation features of the energy consumption behavior, and generate an energy consumption influencing factor map;

[0059] The operation process of the energy consumption status assessment module includes the following steps:

[0060] Constructing a robot behavior state tensor for energy consumption modeling, and performing spatiotemporal alignment on the multi-dimensional operation data to form an original feature tensor structure with time, spatial position, and behavior category as index dimensions;

[0061] Based on the potential energy field mapping algorithm, the original feature tensor is mapped to the energy response field, the energy consumption density function of the behavior path is constructed, and the high energy consumption concentration area and energy consumption mutation point are identified through gradient analysis;

[0062] It should be noted that the behavioral state tensor is input into the pre-built potential field mapping module, and the system constructs a three-dimensional energy consumption field model based on the behavioral activation intensity, movement posture change amplitude and power consumption indicators.

[0063] Specifically, the system estimates the energy density of continuous actions based on the robot's path trajectory and the execution instruction sequence, generating a spatial distribution map. Each spatial node in the map is labeled with the energy consumption value per unit action within the corresponding time period, forming an intuitive and visual energy consumption "heat map."

[0064] During this process:

[0065] Perform motion-energy response analysis on each segment of the trajectory; mark spatial segments where posture changes frequently or servo current continues to fluctuate abnormally; automatically detect areas where energy gradients change sharply and mark them as "mutation points"; identify high-energy consumption patterns that recur in specific postures, movements, or interactive behaviors, such as frequent waving, continuous standing, and sharp turns.

[0066] The output of this stage includes: high-energy consumption hotspot areas, energy consumption mutation behavior segments and their corresponding location coordinates and time windows, providing preliminary screening basis for feature screening and attribution analysis.

[0067] For the identified high-energy consumption areas and action segments, a feature sparse modeling mechanism is used to construct an energy consumption attribution mapping matrix. A channel compression algorithm based on the minimum entropy increase criterion is used to mine the behavioral feature subsets and their activation timing patterns that are most sensitive to unit energy consumption growth.

[0068] Specifically, this stage adopts the channel compression and attribution modeling mechanism, which includes the following steps:

[0069] In the identified high-energy consumption sections, the temporal distribution of all original features (such as servo current, acceleration, angular velocity, temperature, visual frame change rate, etc.) is statistically analyzed; the correlation between the changing trend of each feature and the change in energy consumption per unit time is analyzed; the feature channels are sorted according to the importance distribution of the features, and the top several channels with significant influence are retained.

[0070] A complete task behavior is divided into multiple stages according to logic or time periods, such as "startup stage", "interaction stage", "turn stage", "return stage", etc.; each stage is bound to the system log or user command to form a traceable behavior period label.

[0071] During the critical behavioral period, the compressed features are modeled to construct a sparse behavioral attribution matrix; each matrix element represents the degree to which a "specific behavioral feature" drives energy consumption fluctuations in the "specific task stage"; after normalization, the matrix can be used to make horizontal comparisons across multiple task scenarios and identify general and scenario-specific energy consumption factors.

[0072] Combined with the activation timing of sensor data, the changing rhythm of characteristic values before and after the energy consumption peak can be identified to analyze whether there are patterns such as "pre-activation", "co-activation" or "delayed activation". It can further analyze whether certain high energy consumption points are triggered by multiple features in coordination or caused by a single factor, thereby accurately controlling the optimization object.

[0073] The final output is: a set of key feature dimensions, their role windows in task behavior, and corresponding energy consumption attribution weights, providing a source of factors for graph construction.

[0074] Combining task identification and spatial location, a cross-temporal and spatial energy consumption behavior impact map is constructed. Based on the energy consumption behavior impact map, the influence centrality and intervention priority of each characteristic factor are calculated, and an energy consumption impact factor map is output.

[0075] It should be noted that the energy consumption behavior influence graph is a directed weighted graph structure, which includes a number of nodes and edges;

[0076] The nodes represent the key behavior units decomposed from the robot's task flow. Each node contains the corresponding behavior type, execution time period, spatial location, and unit energy consumption index.

[0077] The edge represents the coupling association between behavioral units in the spatial path or energy consumption characteristics, and is used to describe the collaborative relationship or adjacent scheduling relationship between two behaviors during the execution process. Furthermore, an edge weight is set on the edge to quantify the strength of the coupling association. The edge weight is comprehensively modeled based on the co-occurrence frequency of energy consumption characteristic factors and the spatial path overlap, reflecting the propagation effect of characteristic factors among multiple behavioral units and the energy consumption superposition trend.

[0078] Specifically, based on the complete task process of the robot, it is broken down into multiple task behavior units, such as "starting movement", "environmental perception", "task response", "path correction", etc.; each behavior unit is a graph node, with its timestamp, position coordinates, main control instructions and energy consumption label.

[0079] Based on the "multiple behaviors - single feature" or "single behavior - multiple features" relationship found in the sparse behavior attribution matrix, edges are established between behavior nodes. The weight of the edge is calculated based on the degree of sharing of these features among behaviors and the degree of energy consumption contribution. At the same time, spatial path coupling is considered. That is, although some behaviors belong to different task units, their execution paths physically overlap, and path influence labels are attached to the edges.

[0080] Using the graph theory betweenness centrality or propagation centrality algorithm, the structural influence of each characteristic factor (or behavior node) in the entire energy consumption behavior influence graph is calculated; this analysis not only considers the energy consumption contribution of the node itself, but also considers its triggering effect on the energy consumption chain of other nodes.

[0081] Influence is integrated with execution frequency, location concentration, and other information to generate a feature intervention priority list. High-priority features or behavior segments will be marked for key constraints, replacements, or adjustments in subsequent energy efficiency optimization modules.

[0082] The final energy consumption influencing factor map is a multi-dimensional graph structure that not only visualizes the complex relationship between behavior and energy consumption, but also provides a basis for task path reconstruction, strategy switching, and charging scheduling.

[0083] The energy consumption attribution mapping matrix is specifically a sparse behavior attribution matrix constructed by compressing feature channels based on the minimum entropy increase criterion. The rows represent the key behavioral feature dimensions that are sensitive to energy consumption, and the columns represent the divided task behavior time periods. The matrix elements are used to characterize the attribution intensity of each feature to the unit energy consumption increment. It is generated by channel compression, time series division and attribution modeling of multidimensional operation data.

[0084] Furthermore, the formula of the potential energy field mapping algorithm is as follows:

[0085] ;

[0086] in, Represents the energy consumption potential density at the robot's spatial position (x, y) and time t; N represents the number of historical behavior samples, which is extracted from the three-dimensional motion inertial navigation data, joint servo current waveform, and voltage and current data collected by the robot data acquisition module; Indicates the current space coordinates; Represents the spatial coordinates of the i-th historical data; and are the timestamps of the current and i-th data, respectively, and the time tags from the robot operation data; It represents the activation intensity of the energy-consuming behavior corresponding to the i-th behavior data, which is normalized by the instantaneous amplitude, voltage fluctuation and posture stability of the joint servo current, and is used to quantify the direct contribution of the behavior to the overall energy consumption; It represents the environmental disturbance factor, which combines the temperature and humidity gradient change rate and the sound response signal amplitude change to reflect the driving disturbance of the local environment on energy consumption behavior; It represents the behavior type modulation factor, which is used to distinguish the impact of different actions on energy consumption; It represents the spatial scale control parameter, which determines the smooth range of potential energy diffusion in the spatial dimension and is set according to the activity radius range of each component of the robot; represents the time diffusion scale, reflecting the time decay rate of the historical behavior’s response to current energy consumption; Represents the angle between the direction vector of the i-th behavior and the current energy consumption gradient direction, calculated by combining the posture change trend and the energy consumption density gradient direction; It represents the directional coupling adjustment coefficient, which is used to control the coupling strength between the robot's behavior direction and the energy consumption field gradient's response to the final energy consumption density.

[0087] The energy efficiency optimization module is used to search for the optimal energy consumption solution between the task execution path and the servo control strategy based on the energy consumption influencing factor map by using the quantum behavior particle swarm algorithm, and generate a low-energy execution instruction set;

[0088] The operation process of the energy efficiency optimization module includes the following steps:

[0089] Based on the energy consumption influencing factor map, the corresponding task execution sequence, spatial movement path and servo control instructions are extracted to construct a joint optimization variable set;

[0090] Specifically, the system first calls the energy consumption influencing factor map generated by the energy consumption status assessment module to extract the relevant task units, spatial path nodes and key control features in the current task scenario.

[0091] Specifically include:

[0092] Task execution sequence: For example, "move to interaction point → interact with user → return to base station", etc., extracting the logical sequence and execution window of each behavior;

[0093] Spatial movement paths: Identify the set of spatial trajectories that the robot may choose, including the distances, environmental constraints, and historical energy consumption labels of different path segments;

[0094] Servo control parameters: A set of servo parameters for each task phase, such as joint rotation angle, execution speed, response inertia, etc.

[0095] The system combines the information of the above three dimensions into a set of joint optimization variables. Each set of variables contains a complete configuration combination of task path-action control, which is used as the basic unit of the solution space for subsequent search algorithms.

[0096] Initializing the optimization population, mapping each parameter combination of the joint optimization variable set into a search individual;

[0097] The quantum-behaved particle swarm algorithm is used to jointly optimize the path and servo strategy. The particle fitness function is calculated in each generation, and the particle state is iteratively updated based on the historical optimal solution and the current search direction.

[0098] It should be noted that this step mainly consists of the following sub-processes:

[0099] 1. Particle representation structure: Each particle is a candidate solution, and its state vector contains three parts: a set of codes for the order in which tasks are executed (for example, an integer sequence is used to represent the task arrangement); a path selection index (selecting a path from a library of candidate paths); and a set of servo parameter combinations consisting of continuous variables (such as angular velocity limits, joint rotation ranges, acceleration thresholds, etc.). Each particle represents a complete "execution solution."

[0100] 2. Particle behavior update mechanism: In each iteration, the particle state is updated by sampling the quantum position update model rather than simply evolving the velocity and position. The current global optimal particle (gbest) is used to guide the probability distribution of all particles. A "center guidance mechanism" is introduced to ensure that each particle is close not only to the individual historical optimum but also to the group mean, reflecting the intelligent aggregation of the group. In the update of path selection and task sequence variables, probabilistic perturbations and reordering strategies are used to introduce jumps to maintain population diversity.

[0101] 3. Coupled Optimization Mechanism: This module performs a multivariable joint optimization problem, namely, simultaneously adjusting the discrete task sequence, path selection index, and continuous servo parameters. Task sequence changes directly affect time window scheduling and servo load rhythm; path selection changes the overall spatial energy consumption structure.

[0102] Control parameter adjustment affects the energy consumption of single behavior and system stability; there is a high degree of coupling between the three, and the joint evolution of the solution space is achieved through a unified particle state coding structure.

[0103] In each iteration, the search individuals are comprehensively evaluated based on multi-dimensional indicators such as task execution time limit, path continuity, behavior control stability, and unit energy consumption level to generate the optimal energy consumption control plan;

[0104] It's important to note that after each iteration, the system evaluates all particle solutions, selects the best candidate solution, and guides the next round of search. This evaluation is no longer based on a single metric, but rather a weighted, comprehensive assessment based on multiple performance indicators to ensure the final solution is practical and stable in actual operation.

[0105] The main evaluation indicators are as follows:

[0106] 1. Task Completion Indicator: Checks whether the solution represented by the particle can complete all tasks within the set time window, and whether there are any omissions or repeated calls. If there are any logical loopholes, the solution will be judged as low fitness.

[0107] 2. Path rationality index: The movement path is checked for continuity, closure, and accessibility, including avoiding fallback paths, non-physically feasible paths, or paths beyond the range of obstacles.

[0108] 3. Servo control stability index: Analyze the changing trend of each motion control curve to detect whether the servo moves too frequently, accelerates and decelerates too quickly, or moves incoherently. These can lead to energy waste and unstable execution.

[0109] 4. Unit Energy Consumption Index: Based on the weight labels in the energy consumption influencing factor map, combined with the simulation execution path and servo control model, the estimated energy consumption value of each particle per unit time and unit distance is calculated. This value is one of the most important target parameters of the optimization algorithm.

[0110] 5. System Response Redundancy: Analyze whether sufficient buffer time and servo tolerance are reserved for each task node in the particle solution. For example, when executing certain highly dynamic tasks (such as rapid steering), is it possible to allow the servo to warm up or release inertia to avoid extreme loads?

[0111] After integrating these multi-dimensional indicators, the system generates a final fitness score for each particle through normalization and weighted fusion. The score not only determines the current optimal particle but also guides the quantum behavior iteration of the next generation of particles.

[0112] The optimal energy consumption control scheme is mapped into a low-energy consumption control instruction set, which includes a target task sequence, a servo joint angle scheduling curve, a speed constraint parameter and a state feedback frequency.

[0113] It should be noted that after completing all iterations (or converging early to meet the termination conditions), the system will automatically extract the solution of the global optimal particle and convert it into a structured low-energy control instruction set, which will be sent to the robot motion controller and task executor.

[0114] The control instruction set mainly consists of the following four parts:

[0115] 1. Task sequence planning instructions: including task number, execution order, and start trigger conditions; each task is accompanied by an execution time window (earliest-latest start time); support for interrupt recovery and task jump instruction nesting to improve fault tolerance.

[0116] 2. Path navigation and posture transition instructions: Clearly specify the node number and spatial displacement target of the robot's movement path; include posture connection mode (smooth steering / step adjustment); configure specific path weights for different terrain types or interaction areas.

[0117] 3. Servo joint scheduling parameter set: provides the angle change curve of each servo joint within the task segment; specifies the starting speed, peak speed, and terminal deceleration control of each motion segment; and specifically sets servo motor power constraints, response time limits, and safe braking strategies.

[0118] 4. State feedback and monitoring frequency setting: Specify the data collection frequency for key nodes (e.g., 10 samples per second); enable the dynamic feedback adjustment mechanism. When a task segment is overloaded or time-consuming, the instruction can temporarily adjust subsequent parameters to achieve online self-adaptation.

[0119] The final generated control instruction set will be uploaded to the robot's embedded controller as an execution template and can be used iteratively in subsequent tasks, cooperating with the system learning module to achieve long-term energy consumption optimization.

[0120] Furthermore, the formula of the quantum-behaved particle swarm algorithm is as follows:

[0121] ;

[0122] in, represents the optimal control parameter solution of the pet robot in the current task cycle; Represents the historical optimal solution position of the particle in the kth generation; Represents the dynamic step coefficient, which is set based on the task urgency, its own power level and energy consumption evaluation value; Indicates the global optimal solution position in the current iteration, obtained by combining the historical behavior graph and energy consumption index evaluation; It represents the nonlinear convergence adjustment factor, reflecting the expected range of path continuity and servo control smoothness, and is calculated from the servo speed constraint and path smoothness index in the joint optimization variable set; Represents a random perturbation variable in the interval (0,1). It is dynamically generated by combining the quantum search mechanism in the energy efficiency optimization module and the global search diversity requirements to simulate the perturbation information entropy of the search space.

[0123] The automatic charging scheduling module is used to construct a heterogeneous graph structure with robots and charging piles as nodes based on the robot's current power level, task urgency, spatial location information, and energy consumption evaluation results. It uses a graph attention network algorithm to adaptively model the state weights of the charging path. Combined with the task execution time window constraint scheduling mechanism, it dynamically generates a charging priority queue and an optimal path plan, and controls the pet robot to autonomously navigate to the charging station.

[0124] The operation process of the automatic charging scheduling module includes the following steps:

[0125] The robot's current battery percentage, remaining mission time, spatial coordinates, and accessibility information to surrounding charging stations are collected to construct a heterogeneous directed graph with the robot node and the charging station node as vertices. The edge weights of the heterogeneous directed graph are initialized and modeled by combining spatial distance, energy consumption estimation, and path risk coefficient.

[0126] Specifically, the system first collects core status data related to scheduling decisions in real time in the scheduling server or local edge computing unit, including: the current remaining battery percentage of the pet robot; the remaining time window of the task being executed or to be executed; the spatial coordinates of the current location (two-dimensional or three-dimensional map coordinates); the spatial location, occupancy status, and queue status of all nearby charging piles; the path connectivity, distance, obstacle information and risk assessment (such as narrow passages, electromagnetic interference areas, etc.) between the robot's current location and each charging pile.

[0127] Based on this, the system constructs a heterogeneous directed graph structure, and the node types of the heterogeneous directed graph structure include: robot nodes and charging pile nodes;

[0128] An edge represents the path that a robot can take to reach a charging station. The initialization of edge weights includes: spatial path distance; estimated energy consumption of the current path (combined with slope, turning frequency, etc.); and path safety or stability level (i.e., path risk coefficient, such as aisle congestion, interference area, etc.).

[0129] The heterogeneous directed graph is trained and modeled based on a graph attention network. The attention mechanism is used to perform weighted learning on the power level, task urgency, and path cost between nodes. The robot's preference for candidate charging pile nodes is adaptively adjusted, and a path decision attention matrix is output.

[0130] Specifically, after establishing the heterogeneous graph structure, the system calls the graph attention network model (GAT) deployed locally or on the server to perform preference learning and state modeling on the charging path.

[0131] The process includes the following:

[0132] 1. Node feature encoding: The robot node encodes its current battery level and task urgency (remaining time and remaining tasks). The charging station node encodes whether it is occupied, the path length to the robot's location, and the current waiting queue length. All node features are converted into a unified vector format through an embedding mechanism and used as graph network input.

[0133] 2. Attention mechanism training: The graph attention network is centered around each robot and performs edge-weighted attention modeling on all candidate charging pile nodes to which it is connected. The training process dynamically adjusts the robot's preference weights for different charging piles by weighted learning path costs, power urgency, and task execution pressure. The training output is a path decision attention matrix, which represents the robot's willingness or priority in choosing different charging paths.

[0134] Combined with the path decision attention matrix, a path optimization function based on the remaining time of the task and navigation energy consumption is constructed to dynamically generate a set of charging candidate paths, calculate the scheduling priority score of each path, and select the optimal charging scheduling path with the minimum path cost and meeting the time window constraint;

[0135] Specifically, after obtaining the path decision attention matrix, the system further performs refined path optimization, constructs a comprehensive scoring function based on task time limit, path energy consumption, and behavior continuity, and selects the final charging path.

[0136] This step includes the following:

[0137] 1. Build a candidate charging path set: Select several charging pile nodes with high attention scores. For each target node, call the map module to plan the robot's actual path to that node, including the sequence of waypoints and estimated time. Eliminate currently unavailable charging piles (e.g., those that are occupied or temporarily blocked).

[0138] 2. Calculate the scheduling priority score for each path, which mainly includes: the remaining task time minus the navigation time (i.e., the task urgency minus the navigation time margin); the matching degree of the path's cumulative energy consumption with the robot's current battery level; the path preference score output by the attention network; and the impact of the current path's obstacle avoidance complexity, turning frequency, and other factors on navigation execution stability.

[0139] 3. Screening the optimal path solution: Standardize all scoring results; eliminate paths that do not meet the time window constraints or exceed energy consumption limits; and select the path with the highest overall score as the optimal scheduling path for this round.

[0140] According to the optimal charging scheduling path, a navigation control instruction set is generated to drive the pet robot to autonomously navigate to the designated charging pile to perform the charging task; the navigation control instruction set includes a path point sequence, steering angle control, obstacle avoidance trigger conditions and remaining power threshold trigger mechanism.

[0141] Specifically, the instruction set includes the following:

[0142] 1. Path point sequence: Identify the continuous path nodes that the robot needs to pass through during movement, and mark the spatial coordinates and expected arrival time of each node.

[0143] 2. Steering angle control information: This includes the steering angle, rotation direction, and angle change rate required by the robot at each node to ensure smooth and energy-efficient movement.

[0144] 3. Obstacle avoidance trigger conditions: Configure local obstacle avoidance trigger rules for each path segment, such as switching to local path planning mode or pause mode when the distance to the obstacle is less than a certain threshold.

[0145] 4. Remaining battery threshold control mechanism: Set an "emergency return threshold". If the system detects that the battery level is below the set threshold while en route to the charging station, it will immediately switch to an alternative charging route or the nearest charging station. Configure a status reassessment mechanism to reassess the route when the route is interrupted or the mission time window changes.

[0146] Furthermore, the formula of the path optimization function is as follows:

[0147]

[0148] in, Represents the scheduling priority score of the candidate charging path r, which is used to sort and filter in the path set. The smaller the value, the higher the priority. represents the energy consumption per unit distance of path r, which is calculated based on the current servo behavior control parameters of the task and the map terrain information; Indicates the Euclidean path length between the robot's current spatial coordinates and the candidate charging pile; Indicates the maximum navigation distance supported by the current remaining battery power of the robot, which is calculated by inversely comparing the remaining battery capacity evaluated in the robot's battery health control module and the unit navigation energy consumption; represents the estimated risk cost coefficient of path r; It represents the ratio of the path time to the remaining time window of the task, which is used to characterize whether the current path can complete the charging task within the task time limit; Represents the pet robot's preference weight for each key node in the path; and It represents the weight coefficient of the cost factor, dynamically adjusting the relative importance of the two dimensions of "energy consumption-distance" and "risk-credibility", and is usually allocated by the strategy layer based on the importance of the current task.

[0149] The battery health control module is used to perform online modeling based on battery charge and discharge cycle data, internal resistance change rate and temperature fluctuation information using an incremental learning algorithm to evaluate battery degradation trends, generate health control instructions, and adjust maximum charge and discharge current and temperature control parameters. The battery health control module specifically has a graded health warning mechanism that divides the battery status into four levels: normal, mild degradation, critical degradation, and severe failure based on the battery life prediction results and health threshold range.

[0150] It should be noted that the battery multivariable degradation prediction model is established based on the incremental learning algorithm, which specifically includes:

[0151] The incremental extreme learning machine (IELM) is used to input the recent multi-dimensional time series characteristics of the battery, dynamically adjust the hidden layer weights, and continuously fine-tune the model's fitting accuracy for the battery SOH;

[0152] Integrate recursive linear regression models (such as RLS) to track the change trajectory of internal resistance and derive the maximum attenuation rate within a certain period in the future;

[0153] A Bayesian confidence interval prediction mechanism is introduced to generate credible intervals for prediction results, evaluate prediction reliability, and provide feedback to the scheduling module for task adaptation and adjustment.

[0154] Special event identification: Combining the increase in battery temperature with the nonlinear increase in internal resistance, the autoencoder anomaly detection network is used to identify thermal runaway or early failure risks.

[0155] Specifically, after obtaining the predicted health status results, this submodule generates the following control parameter instructions based on the standardized battery behavior model:

[0156] Limiting the maximum charge rate (C-rate): When the battery is in a mild or moderate degradation stage, reduce the C-rate from 1.0C to 0.5C to slow down the formation of lithium dendrites;

[0157] Adjust the charge termination voltage threshold: Dynamically set the maximum charge voltage based on temperature conditions (e.g., from 4.2V to 4.1V);

[0158] Control the discharge cut-off threshold: increase the minimum operating voltage (e.g., from 3.0V to 3.2V) to avoid over-discharge;

[0159] Triggering active cooling: If the temperature exceeds the limit but has not yet entered a fault state, the PWM fan or Peltier component is controlled to quickly cool down the system.

[0160] Dynamically adjust task scheduling priorities: If the battery is approaching a critical health threshold, the task will be transferred to other robot nodes to improve system robustness.

[0161] It should be noted that the battery health control module has a set of graded health warning mechanisms, which are used to classify batteries according to the predicted battery life parameters and current health status, and trigger corresponding response strategies. Specifically:

[0162] When the system assesses that the battery is in good health, that is, the state of health value (SOH) is greater than or equal to 90%, and the internal resistance change and temperature rise trend are within the normal fluctuation range, the system determines that the battery is in normal condition and no special intervention is required, allowing the robot to continue to operate according to the current task execution plan;

[0163] If the system detects that the battery SOH is between 70% and 90%, or there is a slight increase in internal resistance and abnormal temperature fluctuation, but it does not affect the task execution time and current output stability, the system will classify the battery as a mild degradation state. In this state, the system will automatically reduce the charge rate and prompt the user to perform regular maintenance to delay performance degradation;

[0164] When the battery's SOH drops to between 50% and 70%, or the model predicts that its usable life is less than one typical mission cycle, the system determines that the battery is in a critical degradation state and immediately triggers the operation restriction mechanism, lowering the servo drive upper limit power and prioritizing the transfer of remaining tasks to other robot nodes;

[0165] If the system detects that the battery's SOH is lower than 50%, or finds nonlinear failure signs such as a sharp temperature rise, a sudden voltage drop, or a sudden change in internal resistance, the battery is considered to have entered a serious failure state. The system will terminate all current task processes, force power-off protection, and notify the user through voice broadcast or mobile application to replace the battery as soon as possible to ensure safe use.

[0166] The above-mentioned health level classification and dynamic response can form an autonomous closed-loop mechanism, enabling the smart pet robot to have battery safety prediction, risk avoidance and adaptive control capabilities during long-term operation.

[0167] In summary, this invention integrates multiple sensors, including IMU, vision, audio, current, and voltage, and combines them with potential field mapping and sparse modeling to not only achieve spatial-behavioral aggregated assessment of energy consumption but also visually generate a map of energy consumption influencing factors, providing structural support for subsequent optimization. By employing the minimum entropy increase criterion and sparse channel compression strategy, it extracts highly sensitive behavioral features and their activation patterns, identifies the sources of energy consumption anomalies during different task phases, and enables interpretable analysis and intervention priority ranking of high-energy-consuming behaviors.

[0168] This paper introduces a quantum behavior update model and a swarm center guidance mechanism in the joint optimization of task paths and servo strategies, enabling coupled searches across task sequences, spatial trajectories, and servo parameters. This significantly improves the global optimization capability and search stability of low-energy instruction set generation. A heterogeneous graph scheduling framework based on the GAT network integrates power consumption, task urgency, and path cost information in real time, enabling the coordinated optimization of charging priority sorting and path selection, improving the accuracy and robustness of scheduling responses.

[0169] This invention uses incremental extreme learning and recursive regression models to track battery state of health (SOH) in real time. Combined with Bayesian prediction and anomaly detection mechanisms, it enables multi-level health state recognition and adaptive adjustment of corresponding charge and discharge parameters, effectively extending battery life and ensuring operational safety. By deeply coupling energy consumption maps with scheduling algorithms, the system can adjust routing, servo response, and charging timing in real time based on task changes, forming a closed-loop optimization control mechanism centered on energy consumption. This significantly reduces system operating power consumption, improves endurance, and enhances task execution stability.

[0170] The above embodiments are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering technicians in this field should fall within the scope of protection determined by the claims of the present invention.

Claims

1. An intelligent pet robot energy consumption optimization and automatic charging control system, characterized in that: It includes a robot data acquisition module, an energy consumption status evaluation module, an energy efficiency optimization module, an automatic charging scheduling module and a battery health control module which are sequentially connected in communication; The robot data acquisition module is used to collect multi-dimensional operation data of the pet robot during operation in real time; The energy consumption status assessment module is used to perform structured aggregation on the multi-dimensional operation data based on the potential field mapping algorithm and the feature sparse modeling mechanism, extract the space-action correlation features of the energy consumption behavior, and generate an energy consumption influencing factor map; The operation process of the energy consumption status assessment module includes the following steps: Constructing a robot behavior state tensor for energy consumption modeling, and performing spatiotemporal alignment on the multi-dimensional operation data to form an original feature tensor structure with time, spatial position, and behavior category as index dimensions; Based on the potential energy field mapping algorithm, the original feature tensor is mapped to the energy response field, the energy consumption density function of the behavior path is constructed, and the high energy consumption concentration area and energy consumption mutation point are identified through gradient analysis; For the identified high-energy consumption areas and action segments, a feature sparse modeling mechanism is used to construct an energy consumption attribution mapping matrix. A channel compression algorithm based on the minimum entropy increase criterion is used to mine the behavioral feature subsets and their activation timing patterns that are most sensitive to unit energy consumption growth. Combining task identification and spatial location, a cross-temporal and spatial energy consumption behavior impact map is constructed. Based on the energy consumption behavior impact map, the influence centrality and intervention priority of each characteristic factor are calculated, and an energy consumption impact factor map is output; The energy efficiency optimization module is used to search for the optimal energy consumption solution between the task execution path and the servo control strategy based on the energy consumption influencing factor map by using the quantum behavior particle swarm algorithm, and generate a low-energy execution instruction set; The automatic charging scheduling module is used to construct a heterogeneous graph structure with robots and charging piles as nodes based on the robot's current power level, task urgency, spatial location information, and energy consumption evaluation results. It uses a graph attention network algorithm to adaptively model the state weights of the charging path. Combined with the task execution time window constraint scheduling mechanism, it dynamically generates a charging priority queue and an optimal path plan, and controls the pet robot to autonomously navigate to the charging station. The battery health control module is used to perform online modeling based on battery charge and discharge cycle data, internal resistance change rate and temperature fluctuation information using an incremental learning algorithm to evaluate battery degradation trends, generate health control instructions, and adjust maximum charge and discharge current and temperature control parameters.

2. The intelligent pet robot energy consumption optimization and automatic charging control system according to claim 1, characterized in that: The multi-dimensional operation data includes the pet robot's three-dimensional motion inertial navigation data, joint servo current waveform, visual information, sound response, posture information, interactive reaction, current and voltage data, and environmental temperature and humidity gradient distribution.

3. The intelligent pet robot energy consumption optimization and automatic charging control system according to claim 1, characterized in that: The formula of the potential energy field mapping algorithm is as follows: ; in, represents the energy consumption potential density at the robot's spatial position (x, y) and time t; N represents the number of historical behavior samples; Indicates the current space coordinates; Represents the spatial coordinates of the i-th historical data; and are the timestamps of the current and i-th data respectively; Indicates the activation intensity of the energy consumption behavior corresponding to the i-th behavior data; represents the environmental disturbance factor; It represents the behavior type modulation factor, which is used to distinguish the impact of different actions on energy consumption; It represents the spatial scale control parameter, which determines the smooth range of potential energy diffusion in the spatial dimension; represents the time diffusion scale, reflecting the time decay rate of the historical behavior’s response to current energy consumption; Represents the angle between the direction vector of the i-th behavior and the current energy consumption gradient direction; Represents the directional coupling adjustment coefficient.

4. The intelligent pet robot energy consumption optimization and automatic charging control system according to claim 1, characterized in that: The energy consumption attribution mapping matrix is specifically a sparse behavior attribution matrix constructed by compressing feature channels based on the minimum entropy increase criterion. The rows represent the key behavioral feature dimensions that are sensitive to energy consumption, and the columns represent the divided task behavior time periods. The matrix elements are used to characterize the attribution intensity of each feature to the unit energy consumption increment. It is generated by channel compression, time series division and attribution modeling of multidimensional operation data.

5. The energy consumption optimization and automatic charging control system of an intelligent pet robot according to claim 1, characterized in that: The operation process of the energy efficiency optimization module includes the following steps: Based on the energy consumption influencing factor map, the corresponding task execution sequence, spatial movement path and servo control instructions are extracted to construct a joint optimization variable set; Initializing the optimization population, mapping each parameter combination of the joint optimization variable set into a search individual; The quantum-behaved particle swarm algorithm is used to jointly optimize the path and servo strategy. The particle fitness function is calculated in each generation, and the particle state is iteratively updated based on the historical optimal solution and the current search direction. In each iteration, the search individuals are comprehensively evaluated based on the task execution time limit, path continuity, behavior control stability and unit energy consumption level to generate the optimal energy consumption control plan; The optimal energy consumption control scheme is mapped into a low-energy consumption control instruction set, which includes a target task sequence, a servo joint angle scheduling curve, a speed constraint parameter and a state feedback frequency.

6. The intelligent pet robot energy consumption optimization and automatic charging control system according to claim 5, characterized in that: The formula of the quantum-behaved particle swarm algorithm is as follows: ； in, represents the optimal control parameter solution of the pet robot in the current task cycle; Represents the historical optimal solution position of the particle in the kth generation; Represents the dynamic step coefficient, which is set based on the task urgency, its own power level and energy consumption evaluation value; Indicates the global optimal solution position in the current iteration, obtained by combining the historical behavior graph and energy consumption index evaluation; represents the nonlinear convergence adjustment factor; represents a random disturbance variable.

7. The intelligent pet robot energy consumption optimization and automatic charging control system according to claim 1, characterized in that: The operation process of the automatic charging scheduling module includes the following steps: The robot's current battery percentage, remaining mission time, spatial coordinates, and accessibility information to surrounding charging stations are collected to construct a heterogeneous directed graph with the robot node and the charging station node as vertices. The edge weights of the heterogeneous directed graph are initialized and modeled by combining spatial distance, energy consumption estimation, and path risk coefficient. The heterogeneous directed graph is trained and modeled based on a graph attention network. The attention mechanism is used to perform weighted learning on the power level, task urgency, and path cost between nodes. The robot's preference for candidate charging pile nodes is adaptively adjusted, and a path decision attention matrix is output. Combined with the path decision attention matrix, a path optimization function based on the remaining time of the task and navigation energy consumption is constructed to dynamically generate a set of charging candidate paths, calculate the scheduling priority score of each path, and select the optimal charging scheduling path with the minimum path cost and meeting the time window constraint; According to the optimal charging scheduling path, a navigation control instruction set is generated to drive the pet robot to autonomously navigate to the designated charging pile to perform the charging task; the navigation control instruction set includes a path point sequence, steering angle control, obstacle avoidance trigger conditions and remaining power threshold trigger mechanism.

8. The intelligent pet robot energy consumption optimization and automatic charging control system according to claim 7, characterized in that: The formula of the path optimization function is as follows: ； in, represents the scheduling priority score of the candidate charging path r; represents the energy consumption per unit distance of path r; Indicates the Euclidean path length between the robot's current spatial coordinates and the candidate charging pile; Indicates the maximum navigation distance that the current remaining battery power of the robot can support; represents the estimated risk cost coefficient of path r; Represents the ratio of the path time to the remaining time window of the task; Represents the pet robot's preference weight for each key node in the path; and Represents the cost factor weight coefficient.

9. The intelligent pet robot energy consumption optimization and automatic charging control system according to claim 1, characterized in that: The battery health control module specifically has a graded health warning mechanism, which divides the battery status into four levels: normal, mild degradation, critical degradation and severe failure based on the battery life prediction results and health threshold range.

Images