Intelligent robot hybrid energy storage system dynamic power distribution method and device for complex tasks

Abstract

This application provides a method and apparatus for dynamic power allocation in a hybrid energy storage system for intelligent robots with complex tasks. It constructs a task state vector through multi-source energy storage parameter acquisition and state estimation, and obtains dual-path target power commands by combining a task power prediction model with multi-constraint rolling optimization. Closed-loop power execution is achieved through control frame synchronous triggering and deviation compensation. This effectively solves the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and command execution compensation, providing technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios for intelligent robots.

Description

Technical Field

[0001] This application relates to the field of data processing, specifically to a dynamic power allocation method and device for a hybrid energy storage system for intelligent robots designed for complex tasks. Background Technology

[0002] Existing power allocation methods for intelligent robot hybrid energy storage systems have significant shortcomings. Traditional systems perform poorly in multi-source energy storage state perception, typically lacking a unified acquisition and state estimation mechanism for terminal voltage, output current, and temperature parameters of lithium battery packs and supercapacitor packs. They also fail to effectively integrate task type identifiers and task stage identifiers to construct a complete state vector and write it into a circular buffer. This results in insufficient perception of the dynamic power demand of the energy storage system under complex robot task scenarios, making it difficult to support subsequent refined power allocation decisions.

[0003] Furthermore, existing technologies face bottlenecks in rolling optimization of power allocation decisions. Most systems lack the ability to construct rolling optimization problems based on the power demand sequence output by the task power prediction model. They cannot simultaneously set power balance constraints, power limit constraints, and state-of-charge boundary constraints, and solve for allocation weighting factors. This results in the generation of target power commands for lithium battery packs and supercapacitor packs lacking coordinated consideration of energy storage state boundaries and dynamic task requirements, affecting the lifespan and energy utilization efficiency of hybrid energy storage systems.

[0004] Existing systems have technical shortcomings in power command execution and deviation compensation. They lack a closed-loop execution mechanism that encapsulates the target power command into a control frame to synchronously trigger the bidirectional converter, receives the actual output power and compares it with the target command for deviation, and triggers command compensation when exceeding limits. This affects the power output accuracy and response reliability of hybrid energy storage systems under complex dynamic load conditions. Solving these problems is of great significance for improving the intelligent level of dynamic power allocation in intelligent robot hybrid energy storage systems. Summary of the Invention

[0005] To address the problems in existing technologies, this application provides a dynamic power allocation method and device for a hybrid energy storage system for intelligent robots in complex tasks. This method effectively solves the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and instruction execution compensation, providing technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios for intelligent robots.

[0006] To solve at least one of the above problems, this application provides the following technical solution: In a first aspect, this application provides a dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks, comprising: The task status information is obtained by acquiring the task type identifier and task stage identifier through the robot main controller. The terminal voltage, output current and temperature of the lithium battery pack and supercapacitor pack are collected by the energy storage management unit to obtain the sampling data set. The state of charge of the lithium battery pack, the state of health of the lithium battery pack and the state of charge of the supercapacitor pack are obtained by performing state estimation on the sampling data set. The task status information and each state of charge and health are organized into a state vector and written into a circular buffer. The state vector sequence is read from the circular buffer and fed into the task power prediction model to obtain the power demand sequence. Based on the power demand sequence and the energy storage state, a rolling optimization problem is constructed and power balance constraints, power limit constraints and state of charge boundary constraints are set. The rolling optimization problem is solved to obtain the allocation weight factor. Based on the allocation weight factor, the target power command of the lithium battery pack and the target power command of the supercapacitor pack are calculated. The target power command of the lithium battery pack and the target power command of the supercapacitor pack are encapsulated into a control frame and sent to the corresponding bidirectional converter to perform synchronous triggering. The actual output power reported by the converter is received and compared with the target power command. When the deviation exceeds a preset threshold condition, command compensation is triggered.

[0007] Furthermore, it also includes: the robot main controller obtains raw task data by subscribing to task instruction frames issued by the task scheduling module through the internal bus, and obtains task status information by parsing the task type identifier and task stage identifier from the raw task data. The task type identifier uses predefined encoding to represent four types of task forms: movement, handling, operation and standby. The task stage identifier represents the start, continuation and termination status of task execution. The energy storage management unit synchronously collects the terminal voltage, output current, and surface temperature of the lithium battery pack and the terminal voltage and output current of the supercapacitor pack at fixed intervals through a simulated sampling channel to obtain raw sampling data. Median filtering is performed on the raw sampling data to remove pulse interference to obtain a sampling data set. The task status information and the sampling data set are aligned with a unified timestamp and then written into a circular buffer.

[0008] Furthermore, it also includes: extracting the terminal voltage, output current, and surface temperature of the lithium battery pack from the sampled data set as inputs for lithium battery pack state estimation; calculating the state of charge and health of the lithium battery pack using an extended Kalman filter algorithm based on the lithium battery pack state estimation inputs; extracting the terminal voltage and output current of the supercapacitor group from the sampled data set as inputs for supercapacitor group state estimation; and calculating the state of charge of the supercapacitor group using a voltage integration method based on the supercapacitor group state estimation inputs. The task status information, the state of charge of the lithium battery pack, the health status of the lithium battery pack, the temperature of the lithium battery pack, the state of charge of the supercapacitor pack, the bus voltage, and the total load current are organized into a state vector according to a preset field order. After attaching a unified timestamp to the state vector, it is written into a circular buffer using a double buffering mechanism. The depth of the circular buffer is configured according to the historical window length required by the task power prediction model.

[0009] Furthermore, it also includes: reading the state vector sequence within the most recent historical window from the circular buffer, extracting the historical value of total load power and the task type identifier sequence from the state vector sequence as input to the task power prediction model, wherein the task power prediction model uses a lightweight recurrent neural network structure to perform inference to obtain the power demand sequence within the future prediction window; Based on the power demand sequence, the state of charge (SOC) of the lithium battery pack, the health state of the lithium battery pack, and the SOC of the supercapacitor pack, a rolling optimization problem is constructed. A power balance constraint is set, requiring that the sum of the power of the lithium battery pack and the power of the supercapacitor pack equals the value of the power demand sequence at the corresponding time. A power limit constraint is set, which dynamically adjusts the upper and lower limits of the lithium battery pack power according to the health state of the lithium battery pack. A SOC boundary constraint is set, which limits the SOC of the supercapacitor pack to be between a preset upper and lower boundary.

[0010] Furthermore, it also includes: using a quadratic programming solver to perform iterative solutions to the rolling optimization problem in each control cycle; when the solution converges within a limited number of iterations, outputting the allocation weight factor at the first moment of the prediction window; when the solution fails to converge, using the allocation result of the previous cycle as the degraded output and setting an alarm flag. The target power command for the lithium battery pack is obtained by multiplying the weighting factor with the power demand value at the first moment of the prediction window in the power demand sequence. The difference between the power demand value and the target power command for the lithium battery pack is used as the target power command for the supercapacitor pack.

[0011] Furthermore, it also includes: encapsulating the target power command of the lithium battery pack and the target power command of the supercapacitor pack into control frames respectively, and sending the control frames to the bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack through the isolated communication interface after adding a sequence number and a unified timestamp. The bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack are synchronously triggered at the beginning edge of the same control cycle according to the timestamp in the control frame, update the power reference and enter the current inner loop tracking, and report the actual output power and bus voltage sampling value at a fixed feedback period.

[0012] Furthermore, it also includes: the energy storage management unit receives the actual output power and bus voltage sample values ​​reported by the bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack, calculates the difference between the actual output power and the corresponding target power command to obtain the power deviation value, and compares the bus voltage sample value with the preset allowable range to obtain the voltage deviation indicator; When the power deviation value exceeds the preset deviation threshold, the instruction compensation is triggered and a compensation power instruction is generated and sent to the corresponding bidirectional converter. When the voltage deviation indicator indicates that the bus voltage exceeds the allowable range, the emergency power redistribution logic is started, and the allocation result and execution status are written to the log storage area according to the control cycle.

[0013] Secondly, this application provides a dynamic power distribution device for a hybrid energy storage system for intelligent robots designed for complex tasks, comprising: The battery data acquisition module is used to obtain task status information by acquiring task type identifier and task stage identifier through the robot main controller, and to obtain sampled data set by acquiring terminal voltage, output current and temperature of lithium battery pack and supercapacitor pack through energy storage management unit. The module performs state estimation on the sampled data set to obtain the state of charge of lithium battery pack, the state of health of lithium battery pack and the state of charge of supercapacitor pack. The task status information and each state of charge and health state are organized into a state vector and written into a circular buffer. The weight allocation determination module is used to read the state vector sequence from the circular buffer and send it into the task power prediction model to obtain the power demand sequence. Based on the power demand sequence and the energy storage state, a rolling optimization problem is constructed and power balance constraints, power limit constraints and charge state boundary constraints are set. The weight allocation factor is obtained by solving the rolling optimization problem. Based on the weight allocation factor, the target power command of the lithium battery pack and the target power command of the supercapacitor pack are calculated. The dynamic allocation module is used to encapsulate the target power command of the lithium battery pack and the target power command of the supercapacitor pack into a control frame and send it to the corresponding bidirectional converter for synchronous triggering. It receives the actual output power reported by the converter and compares it with the target power command. When the deviation exceeds a preset threshold condition, it triggers command compensation.

[0014] Thirdly, this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the steps of the dynamic power allocation method for the intelligent robot hybrid energy storage system for complex tasks.

[0015] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the dynamic power allocation method for the intelligent robot hybrid energy storage system for complex tasks.

[0016] Fifthly, this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps of the dynamic power allocation method for the intelligent robot hybrid energy storage system for complex tasks.

[0017] As can be seen from the above technical solution, this application provides a dynamic power allocation method and device for a hybrid energy storage system for intelligent robots oriented towards complex tasks. It constructs a task state vector by collecting and estimating multi-source energy storage parameters, obtains dual-path target power commands by combining a task power prediction model and multi-constraint rolling optimization, and achieves closed-loop power execution through synchronous triggering and deviation compensation of control frames. It effectively solves the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and command execution compensation, and provides technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios of intelligent robots. Attached Figure Description

[0018] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0019] Figure 1 This is a flowchart illustrating the dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks, as described in this application embodiment. Figure 2 This is a structural diagram of the dynamic power distribution device of the intelligent robot hybrid energy storage system for complex tasks in the embodiments of this application. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0021] The acquisition, storage, use, and processing of data in this application comply with relevant laws and regulations.

[0022] In view of the problems existing in the prior art, this application provides a dynamic power allocation method and device for a hybrid energy storage system for intelligent robots with complex tasks. The method constructs a task state vector by collecting and estimating multi-source energy storage parameters, and obtains dual-path target power commands by combining a task power prediction model and multi-constraint rolling optimization. Closed-loop power execution is achieved through control frame synchronous triggering and deviation compensation. This method effectively solves the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and command execution compensation, and provides technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios of intelligent robots.

[0023] To effectively address the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and command execution compensation, and to provide technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios for intelligent robots, this application provides an embodiment of a dynamic power allocation method for hybrid energy storage systems of intelligent robots for complex tasks. See [link to embodiment]. Figure 1 The dynamic power allocation method for the intelligent robot hybrid energy storage system for complex tasks specifically includes the following: Step S101: Obtain task status information by acquiring task type identifier and task stage identifier through robot main controller; collect terminal voltage, output current and temperature of lithium battery pack and supercapacitor pack through energy storage management unit to obtain sampling data set; perform state estimation on the sampling data set to obtain lithium battery pack state of charge, lithium battery pack health state and supercapacitor pack state of charge; organize the task status information and each state of charge and health state into a state vector and write it into a circular buffer. In this embodiment, the robot's main controller subscribes to task instruction frames issued by the task scheduling module via an internal bus, and parses the task type identifier and task stage identifier from these frames. The task type identifier uses predefined encoding to represent four task types: movement, handling, operation, and standby. The task stage identifier represents the start, continuation, and termination states of task execution. This embodiment organizes the parsed task type identifier and task stage identifier into task status information, which serves as the basis for determining load characteristics in the subsequent power prediction stage.

[0024] While acquiring the task status information, this embodiment synchronously collects the terminal voltage, output current, and surface temperature of the lithium battery pack, as well as the terminal voltage and output current of the supercapacitor pack, at fixed intervals through the simulated sampling channel of the energy storage management unit. The acquisition process performs synchronous triggering on each channel to ensure time consistency. The acquired raw sampling data is processed by median filtering to remove pulse interference, forming a sampling data set. Each data item in the sampling data set is organized according to the sampling channel identifier for subsequent state estimation.

[0025] Accordingly, this embodiment extracts the terminal voltage, output current, and surface temperature of the lithium battery pack from the sampled data set as inputs for lithium battery pack state estimation. The state estimation process employs the Extended Kalman Filter (EKF) algorithm, using the equivalent circuit model of the lithium battery pack as the basis for the state equation, and using the terminal voltage as an observation to jointly estimate the state of charge and the state of health. In each sampling period, the EKF algorithm updates the predicted state value based on the output current, performs state correction based on the terminal voltage observation, and outputs the state of charge and the state of health of the lithium battery pack. The state of health reflects the degree of battery capacity degradation and will be used as the basis for dynamic adjustment of power limit constraints in subsequent step S102.

[0026] After the lithium battery pack state estimation is completed, this embodiment extracts the terminal voltage and output current of the supercapacitor bank from the sampled data set as inputs for the supercapacitor bank state estimation. Since the voltage and state of charge of the supercapacitor bank have an approximately linear relationship, this embodiment uses the voltage integration method to calculate the supercapacitor bank's state of charge. Specifically, this embodiment determines the current state of charge percentage based on the position of the terminal voltage relative to the rated voltage range. This supercapacitor bank's state of charge will be used as the input for determining the state of charge boundary constraints in subsequent step S102.

[0027] Based on the aforementioned state estimation results, this embodiment organizes the task state information, the state of charge of the lithium battery pack, the health state of the lithium battery pack, the temperature of the lithium battery pack, the state of charge of the supercapacitor pack, the bus voltage, and the total load current into a state vector according to a preset field order. The bus voltage is obtained from the voltage sampling channel of the common DC bus of the hybrid energy storage system, and the total load current is obtained by adding the output current of the lithium battery pack and the output current of the supercapacitor pack. In this embodiment, after attaching a unified timestamp to the state vector, a double-buffering mechanism is used to write it into a circular buffer. The double-buffering mechanism ensures consistent reading of the state vector through the separation of read and write pointers.

[0028] The depth of the circular buffer is configured according to the required historical window length of the task power prediction model. For example, when the historical window covers several recent control cycles, the depth of the circular buffer is set to the corresponding number of state vector storage units. After the state vectors are written into the circular buffer, in the subsequent step S102, the power prediction module reads the state vector sequence within the most recent historical window as the input to the task power prediction model, thereby generating a power demand sequence and constructing a rolling optimization problem.

[0029] Step S102: Read the state vector sequence from the circular buffer and send it into the task power prediction model to obtain the power demand sequence. Based on the power demand sequence and the energy storage state, construct a rolling optimization problem and set power balance constraints, power limit constraints and charge state boundary constraints. Solve the rolling optimization problem to obtain the allocation weight factor. Based on the allocation weight factor, calculate the target power command of the lithium battery pack and the target power command of the supercapacitor pack. In this embodiment, the state vector sequence within the most recent historical window is read from the circular buffer written in step S101. The historical total load power value and task type identifier sequence are extracted from the state vector sequence and used as input to the task power prediction model. The task power prediction model employs a lightweight recurrent neural network structure. In the offline phase, it is trained using power time-series samples collected under typical robot task scenarios. The training objective is to minimize the mean square deviation between the predicted and actual power values ​​within the prediction window. After the model parameters are solidified, they are deployed in the inference engine of the energy storage management unit. In the online phase, the model receives the state vector sequence and outputs the power demand sequence for each sampling time within the future prediction window.

[0030] Based on the power demand sequence, this embodiment constructs a rolling optimization problem by combining the state of charge (SOC) of the lithium battery pack, the health status of the lithium battery pack, and the SOC of the supercapacitor pack output in step S101. The decision variables of the rolling optimization problem are the power of the lithium battery pack and the power of the supercapacitor pack at each time point within the prediction window. The optimization objective is to minimize the combined cost of the fluctuation range of the lithium battery pack power and the deviation of the supercapacitor pack SOC from the median range within the prediction window. This embodiment expresses the optimization objective as a weighted summation. The first term is the sum of the squares of the differences in lithium battery pack power at adjacent time points, and the second term is the square of the deviation of the supercapacitor pack SOC from the median. The two terms are multiplied by their respective weight coefficients and then summed to obtain the objective function value.

[0031] Accordingly, this embodiment sets a power balance constraint for the rolling optimization problem, requiring that the sum of the power of the lithium battery pack and the power of the supercapacitor pack equals the value at the corresponding moment of the power demand sequence. The power balance constraint ensures that the total output power of the hybrid energy storage system strictly matches the load demand, avoiding fluctuations in the bus voltage due to power mismatch. This embodiment also sets a power limit constraint, setting upper and lower limits for the lithium battery pack power. These limits are dynamically adjusted based on the health status and temperature of the lithium battery pack. When the health status of the lithium battery pack is below a preset health threshold, the upper power limit is tightened to reduce the output stress of aging batteries. When the temperature of the lithium battery pack exceeds a preset temperature threshold, the upper power limit is further tightened to suppress the risk of thermal runaway.

[0032] Based on the power limit constraint, this embodiment sets a state of charge (SOC) boundary constraint to limit the SOC of the supercapacitor bank to between a preset upper and lower boundary. When the SOC of the supercapacitor bank approaches the upper boundary, this embodiment adds a penalty term to the optimization objective to guide the allocation weights to shift towards the lithium battery pack, thereby reducing the charging amount of the supercapacitor bank. When the SOC of the supercapacitor bank approaches the lower boundary, the penalty term guides the allocation weights to recharge the supercapacitor bank to restore its power throughput margin. The introduction of the SOC boundary constraint avoids the supercapacitor bank from operating in the overcharge or over-discharge range for extended periods.

[0033] Based on the aforementioned constraints, this embodiment employs a quadratic programming solver to iteratively solve the rolling optimization problem in each control cycle. The solver takes the quadratic structure of the objective function and linear constraints as input, and searches for the optimal solution within the feasible region using the interior-point method or the effective set method. When the solution converges within a limited number of iterations, this embodiment extracts the optimal power allocation result at the first moment of the prediction window and calculates the allocation weight factor, which represents the proportion of load power borne by the lithium battery pack. When the solution fails to converge within a limited number of iterations, this embodiment uses the allocation result from the previous control cycle as the degraded output and sets an alarm flag.

[0034] Based on the weighting factor, this embodiment multiplies the weighting factor with the power demand value at the first moment of the prediction window in the power demand sequence to obtain the target power command for the lithium battery pack. This embodiment uses the difference between the power demand value and the target power command for the lithium battery pack as the target power command for the supercapacitor pack, thereby ensuring that the sum of the two target power commands is strictly equal to the power demand value at the current moment. The target power command for the lithium battery pack and the target power command for the supercapacitor pack will be encapsulated into a control frame in subsequent step S103 and sent to the corresponding bidirectional converter for execution.

[0035] Step S103: After encapsulating the target power command of the lithium battery pack and the target power command of the supercapacitor pack into a control frame, send it to the corresponding bidirectional converter to perform synchronous triggering, receive the actual output power reported by the converter and compare it with the target power command. When the deviation exceeds the preset threshold condition, trigger command compensation.

[0036] In this embodiment, the target power commands for the lithium battery pack and the supercapacitor pack output in step S102 are encapsulated into control frames. During the encapsulation process, a sequence number and a unified timestamp are added to each target power command. The sequence number identifies the transmission order of the control frames to support idempotency verification at the receiving end, and the unified timestamp indicates the synchronization trigger time of the two converters. In this embodiment, the encapsulated control frames are sent to the bidirectional converters of the lithium battery pack and the supercapacitor pack through an isolated communication interface. The isolated communication interface employs an electrical isolation design to suppress interference coupling between the energy storage circuit and the control circuit.

[0037] After the control frame is sent, the bidirectional converters of the lithium battery pack and the supercapacitor pack perform synchronous triggering according to the timestamp in the control frame. Both converters read the target power command from their respective control frames at the start edge of the same control cycle and update their internal power references, then enter the current inner loop tracking mode. The current inner loop uses the current reference value calculated from the power reference as the tracking target, and drives the switching devices to adjust the output current through a proportional-integral regulator. The synchronous triggering mechanism ensures that the two converters complete the power reference switching at the same time, avoiding transient fluctuations in bus voltage and circulating current problems caused by execution timing deviations.

[0038] Accordingly, the bidirectional converters of the lithium battery pack and the supercapacitor pack report the actual output power and bus voltage sampling values ​​to the energy storage management unit at a fixed feedback period. The actual output power is obtained by multiplying the voltage sampling value and the current sampling value inside the converter, and the bus voltage sampling value is obtained from the voltage detection circuit of the common DC bus. In this embodiment, the energy storage management unit receives the feedback data reported by the two converters, and calculates the power deviation value by performing a difference calculation between the actual output power and the corresponding target power command.

[0039] Based on the power deviation value, this embodiment compares the absolute value of the power deviation value with a preset deviation threshold condition. When the absolute value of the power deviation value does not exceed the preset deviation threshold condition, this embodiment determines that the power tracking of the current control cycle is in a normal state and continues to execute the rolling optimization of the next cycle. When the absolute value of the power deviation value exceeds the preset deviation threshold condition, this embodiment triggers the instruction compensation logic, generates a compensation power instruction based on the sign and amplitude of the power deviation value, and sends it to the corresponding bidirectional converter. The compensation power instruction is superimposed on the current power reference, driving the converter to adjust the output to eliminate the tracking deviation.

[0040] Simultaneously with the deviation comparison, this embodiment compares the sampled bus voltage value with a preset allowable range to obtain a voltage deviation indicator. When the bus voltage is within the allowable range, the voltage deviation indicator is set to normal; when the bus voltage exceeds the upper or lower boundary of the allowable range, the voltage deviation indicator is set to abnormal. This embodiment initiates emergency power redistribution logic when an abnormal voltage deviation indicator is detected, temporarily adjusting the power output ratio of the two converters to prioritize restoring bus voltage stability. Once the bus voltage returns to the allowable range, the regular rolling optimization process resumes.

[0041] This embodiment writes the allocation results and execution status of each control cycle into the log storage area. Recorded fields include timestamp, task identifier, power requirement value, allocation weight factor, target power command for the lithium battery pack, target power command for the supercapacitor pack, actual output power, state of charge, bus voltage, and anomaly flag. The log storage area employs a cyclic overwrite strategy to manage storage space, and critical anomaly events are synchronously written to persistent storage to support post-event auditing and fault tracing. The log records provide a complete data link for model iteration and parameter calibration during the offline phase.

[0042] As can be seen from the above description, the dynamic power allocation method for hybrid energy storage systems of intelligent robots for complex tasks provided in this application can construct a task state vector by collecting and estimating multi-source energy storage parameters, obtain dual-path target power commands by combining a task power prediction model and multi-constraint rolling optimization, and achieve closed-loop power execution through control frame synchronous triggering and deviation compensation. This effectively solves the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and command execution compensation, and provides technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios of intelligent robots.

[0043] In one embodiment of the dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks according to this application, the method may further include the following: Step S201: The robot main controller obtains the original task data by subscribing to the task instruction frame issued by the task scheduling module through the internal bus. It then parses the task type identifier and task stage identifier from the original task data to obtain the task status information. The task type identifier uses predefined encoding to represent four types of task forms: movement, handling, operation, and standby. The task stage identifier represents the start, continuation, and termination status of task execution. Step S202: The energy storage management unit synchronously collects the terminal voltage, output current, and surface temperature of the lithium battery pack and the terminal voltage and output current of the supercapacitor pack at fixed intervals through the simulated sampling channel to obtain raw sampling data. The raw sampling data is then subjected to median filtering to remove pulse interference and obtain a sampling data set. The task status information and the sampling data set are aligned with a unified timestamp and written into a circular buffer.

[0044] In this embodiment, the robot's main controller initiates a subscription request to the task scheduling module via the internal bus. When the task status changes, the task scheduling module publishes a task instruction frame to the internal bus. Upon receiving the task instruction frame, the robot's main controller caches it as raw task data. This raw task data includes a task description field and a timing flag field encapsulated by the task scheduling module. The internal bus employs a publish-subscribe mechanism to achieve loosely coupled communication between the main controller and the task scheduling module, allowing the main controller to receive immediate notifications of task changes without polling.

[0045] Based on the original task data, this embodiment parses the task type identifier and task stage identifier. The parsing process locates the byte offset positions of the task type field and task stage field according to the predefined format of the task instruction frame, extracts the encoded values ​​from the corresponding positions, and maps them to internal representations. The task type identifier uses predefined encoding to represent four task modes: movement, transport, operation, and standby. For example, encoding value zero represents standby mode, encoding value one represents movement mode, encoding value two represents transport mode, and encoding value three represents operation mode. The task stage identifier represents the start, continuation, and termination states of task execution. The start state indicates that the task has just been triggered, the continuation state indicates that the task is in execution, and the termination state indicates that the task has been completed or interrupted.

[0046] Accordingly, this embodiment organizes the parsed task type identifier and task stage identifier into task status information. The task status information is stored in a structured data format, including a task type field, a task stage field, and a time-series marker field inherited from the original task data. This task status information will subsequently be aligned with the execution timestamps of the sampled data set to form part of the state vector.

[0047] While acquiring the task status information, this embodiment utilizes an energy storage management unit to perform multi-channel synchronous data acquisition via simulated sampling channels. The data acquisition targets include the terminal voltage, output current, and surface temperature of the lithium battery pack, as well as the terminal voltage and output current of the supercapacitor pack. Each sampling channel is driven by the same trigger signal at a fixed period to ensure consistency in sampling timing. The acquired channel values ​​constitute the raw sampling data, which is organized according to channel identifiers and marked with a sampling time.

[0048] Based on the original sampled data, this embodiment performs median filtering on the data of each channel to remove impulse interference. The median filtering process maintains a sliding window for each channel, sorting the sampled values ​​within the window by numerical value and taking the middle value as the filtered output. Median filtering suppresses isolated outliers caused by electromagnetic interference or switching noise, while maintaining the signal's step edges without smoothing them. This embodiment obtains a sampled data set after performing median filtering on all channels, where each data item retains the same channel identifier and time marker as the original sampled data.

[0049] This embodiment aligns the task status information with the sampled data set using a unified timestamp. The alignment process uses the sampling time marker of the sampled data set as a reference, pairing the record in the task status information whose time sequence marker is closest to this reference with the sampled data set. After alignment, this embodiment writes the pairing result into a circular buffer, including the task status information field, the sampled data set field, and the unified timestamp field. The data in the circular buffer will be read in subsequent step S102 to construct a state vector sequence and feed it into the task power prediction model.

[0050] In one embodiment of the dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks according to this application, the method may further include the following: Step S301: Extract the terminal voltage, output current, and surface temperature of the lithium battery pack from the sampled data set as the state estimation input of the lithium battery pack. Based on the state estimation input of the lithium battery pack, use the extended Kalman filter algorithm to calculate the state of charge and health of the lithium battery pack. Extract the terminal voltage and output current of the supercapacitor group from the sampled data set as the state estimation input of the supercapacitor group. Based on the state estimation input of the supercapacitor group, calculate the state of charge of the supercapacitor group using the voltage integration method. Step S302: Organize the task status information, the state of charge of the lithium battery pack, the health status of the lithium battery pack, the temperature of the lithium battery pack, the state of charge of the supercapacitor pack, the bus voltage, and the total load current into a state vector according to a preset field order. After attaching a unified timestamp to the state vector, write it into a circular buffer using a double buffering mechanism. The depth of the circular buffer is configured according to the historical window length required by the task power prediction model.

[0051] In this embodiment, the terminal voltage, output current, and surface temperature of the lithium battery pack are extracted from the sampled data set output in step S202 above as inputs for lithium battery pack state estimation. The extraction process locates the three sampled values ​​corresponding to the lithium battery pack based on the channel identifier, and organizes the terminal voltage, output current, and surface temperature values ​​into a lithium battery pack state estimation input vector. In this input vector, the terminal voltage is used as an observation for filtering and correction, the output current is used as a driving quantity for state prediction, and the surface temperature is used as an auxiliary quantity for temperature compensation of the model parameters.

[0052] Based on the state estimation input of the lithium battery pack, this embodiment uses the Extended Kalman Filter (EKF) algorithm to calculate the state of charge (SOC) and health state of the lithium battery pack. The EKF algorithm uses the equivalent circuit model of the lithium battery pack as the basis for the state equations, and the state variables include three components: SOC, health state, and polarization voltage. In the prediction phase, the state variables are recursively calculated one step based on the output current. During the recursion, the change in SOC is determined by the integral contribution of the output current to the rated capacity, and the health state is assumed to remain slowly changing over short periods. In the correction phase, the predicted terminal voltage output from the equivalent circuit model is compared with the actual sampled terminal voltage, and the state variables are corrected based on the deviation value and the Kalman gain. This embodiment outputs the SOC and health state of the lithium battery pack after performing one prediction and correction iteration in each sampling period.

[0053] Accordingly, this embodiment extracts the terminal voltage and output current of the supercapacitor bank from the sampled data set as inputs for supercapacitor bank state estimation. Because the supercapacitor bank's energy storage mechanism is electrostatic energy storage, there is an approximately linear mapping relationship between the terminal voltage and the state of charge (SOC). This embodiment calculates the SOC of the supercapacitor bank based on the SOC state estimation input using the voltage integration method. The calculation process maps the current terminal voltage value to a relative position within the rated voltage range; this relative position represents the percentage of the supercapacitor bank's SOC. The voltage integration method is characterized by its simplicity and rapid response in supercapacitor bank applications. The SOC of the supercapacitor bank will be used as input for the SOC boundary constraints in subsequent step S102.

[0054] Based on the state estimation results, this embodiment obtains the bus voltage and total load current to complete the fields required for the state vector. The bus voltage is read from the voltage detection channel of the common DC bus of the hybrid energy storage system, and the bus voltage reflects the system-level power balance state. The total load current is obtained by adding the output current of the lithium battery pack and the output current of the supercapacitor pack, and the product of the total load current and the bus voltage represents the total load power at the current moment. This embodiment also reads the task type identifier and task stage identifier fields from the task status information output in the aforementioned step S201.

[0055] In this embodiment, the task status information, the state of charge of the lithium battery pack, the health status of the lithium battery pack, the temperature of the lithium battery pack, the state of charge of the supercapacitor pack, the bus voltage, and the total load current are organized into a state vector according to a preset field order. The preset field order is determined based on the input interface specification of the task power prediction model, and each field occupies a fixed offset position in the state vector for subsequent parsing. In this embodiment, a unified timestamp is added to the state vector, which is inherited from the sampling time marker of the sampled data set to ensure timing consistency.

[0056] After the state vector is encapsulated, this embodiment employs a double-buffering mechanism to write it into a circular buffer. The double-buffering mechanism maintains two independent buffer areas, with write and read operations operating on different areas respectively. After writing, pointer switching is used to interchange the roles of these areas. This mechanism ensures that the reading end always accesses complete and consistent state vector data, avoiding data tearing caused by concurrent read and write operations. The depth of the circular buffer is configured based on the required historical window length of the task power prediction model. When the historical window covers several control cycles, the circular buffer depth is set to the corresponding number of storage units. The state vector sequence will be read by the power prediction module and sent to the task power prediction model in subsequent step S102.

[0057] In one embodiment of the dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks according to this application, the method may further include the following: Step S401: Read the state vector sequence within the most recent historical window from the circular buffer, extract the historical value of total load power and the task type identifier sequence from the state vector sequence as input to the task power prediction model, and use a lightweight recurrent neural network structure to perform inference to obtain the power demand sequence within the future prediction window; Step S402: Based on the power demand sequence, the state of charge of the lithium battery pack, the health state of the lithium battery pack, and the state of charge of the supercapacitor pack, a rolling optimization problem is constructed. A power balance constraint is set, requiring that the sum of the power of the lithium battery pack and the power of the supercapacitor pack equals the value of the power demand sequence at the corresponding time. A power limit constraint is set, dynamically adjusting the upper and lower limits of the lithium battery pack power according to the health state of the lithium battery pack. A state of charge boundary constraint is set, limiting the state of charge of the supercapacitor pack to be between a preset upper and lower boundary.

[0058] This embodiment reads the state vector sequence within the most recent history window from the circular buffer written in step S302. The reading process involves tracing back the number of storage units corresponding to the length of the history window based on the current write pointer position of the circular buffer, organizing all state vectors within the tracing range into a state vector sequence in timestamp order. Each state vector in the state vector sequence maintains the field order defined in step S302, covering fields such as task status information, the state of charge and health status of each energy storage unit, bus voltage, and total load current.

[0059] Based on the state vector sequence, this embodiment extracts the historical total load power value and the task type identifier sequence as inputs to the task power prediction model. The historical total load power value is obtained by multiplying the bus voltage field and the total load current field in each state vector. In this embodiment, this multiplication operation is performed on all state vectors within the historical window to form a power time series array. The task type identifier sequence is obtained by arranging the task type identifier fields in each state vector in chronological order. The task type identifier sequence provides context information for task mode switching for the task power prediction model.

[0060] Accordingly, this embodiment feeds the historical total load power value and the task type identifier sequence into the task power prediction model for inference. The task power prediction model employs a lightweight recurrent neural network structure, consisting of an input layer, a hidden layer, and an output layer. The input layer receives the power time series array and the task type identifier sequence and performs feature encoding. The hidden layer uses gated recurrent units to capture temporal dependencies and maintain internal memory states. The output layer maps the hidden layer output to the power prediction values ​​at each sampling time within the future prediction window. In this embodiment, the task power prediction model is invoked once in each control cycle to complete a forward inference. The output result is organized as a power demand sequence, where each element corresponds to the expected load power value at a discrete time within the prediction window.

[0061] Based on the power demand sequence, this embodiment constructs a rolling optimization problem by combining the state of charge (SOC) of the lithium battery pack, the health status of the lithium battery pack, and the SOC of the supercapacitor pack output in step S301. The rolling optimization problem uses the power of the lithium battery pack and the power of the supercapacitor pack at each time point within the prediction window as decision variables. The optimization objective is to minimize the combined cost of the fluctuation amplitude of the lithium battery pack power and the deviation of the supercapacitor pack's SOC from the median. This embodiment sets the optimization objective as a weighted sum of two costs: the first term accumulates the square of the change in lithium battery pack power at adjacent time points to suppress drastic power fluctuations; the second term calculates the square of the deviation of the supercapacitor pack's SOC from the center of the median interval to guide the SOC to remain at an intermediate level.

[0062] Based on the aforementioned optimization objective, this embodiment sets a power balance constraint as an equality constraint condition for the rolling optimization problem. The power balance constraint requires that the sum of the lithium battery pack power and the supercapacitor pack power at each moment within the prediction window equals the value at the corresponding moment in the power demand sequence. This constraint ensures that the total output power of the hybrid energy storage system strictly matches the load demand at each moment. The power balance constraint enters the optimization problem in the form of a linear equation, and the number of constraints is equal to the number of discrete moments within the prediction window.

[0063] This embodiment also sets power limit constraints as inequality constraints for the rolling optimization problem. The power limit constraints set upper and lower limits for the lithium battery pack power, with the upper limit dynamically adjusted based on the health status of the lithium battery pack. When the health status of the lithium battery pack is higher than a preset health threshold, the upper limit takes the rated power value; when the health status is lower than the preset health threshold, the upper limit shrinks proportionally to reduce the output stress of aging batteries. This embodiment also sets upper and lower limits for the supercapacitor pack power, with the boundary values ​​determined based on the rated power capability of the supercapacitor pack.

[0064] Based on the power limit constraint, this embodiment sets a state of charge (SOC) boundary constraint to limit the SOC of the supercapacitor bank to between a preset upper and lower boundary. The SOC boundary constraint incorporates the evolution equation of the supercapacitor bank's SOC into the optimization problem. The evolution equation establishes a recursive relationship between time steps based on the integral contribution of the supercapacitor bank's power to the SOC. This embodiment applies upper and lower boundary constraints to the predicted SOC values ​​of the supercapacitor bank at each time step within the prediction window. When the SOC approaches the boundary, the constraint conditions will restrict the power distribution in the corresponding direction. After the rolling optimization problem is constructed, it will be fed into the solver for iterative solution in subsequent step S501.

[0065] In one embodiment of the dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks according to this application, the method may further include the following: Step S501: Use a quadratic programming solver to perform iterative solution on the rolling optimization problem in each control cycle. When the solution converges within a limited number of iterations, output the allocation weight factor at the first moment of the prediction window. When the solution does not converge, use the allocation result of the previous cycle as the degraded output and set the alarm flag. Step S502: Perform a product operation based on the weighting factor and the power demand value at the first moment of the prediction window in the power demand sequence to obtain the target power command of the lithium battery pack, and take the difference between the power demand value and the target power command of the lithium battery pack as the target power command of the supercapacitor pack.

[0066] This embodiment employs a quadratic programming solver to iteratively solve the rolling optimization problem constructed in step S402. The quadratic programming solver takes the quadratic coefficient matrix of the objective function and the coefficient matrix of the linear constraints as input, searching within the feasible region for the decision variable values ​​that minimize the objective function. At the start of each control cycle, the solver reads the parameter configuration of the current rolling optimization problem, which includes the power demand sequence values, power limit boundaries, state of charge boundaries, and the coefficient matrices of each constraint. This embodiment stores the upper limit of the solver's iteration count and convergence accuracy as configurable parameters in the parameter area. The upper limit of the iteration count is determined based on the time margin of the control cycle to ensure that the solution process is completed within the cycle.

[0067] After the solver is started, this embodiment monitors the convergence status of the solution process. After each iteration, the solver calculates the change between the current solution and the solution from the previous iteration. When the norm of the change is lower than the convergence accuracy threshold, the solution is considered converged and the iteration is terminated. When the solution converges within a limited number of iterations, this embodiment extracts the power values ​​of the lithium battery pack and the supercapacitor pack at the first moment of the prediction window from the optimal decision variables output by the solver. This embodiment calculates the ratio of the lithium battery pack power value to the power demand value at the first moment as an allocation weighting factor. The allocation weighting factor represents the proportion of the total load power borne by the lithium battery pack at the current moment, and its value ranges between zero and one.

[0068] Accordingly, this embodiment sets up degradation processing logic for cases where the solution fails to converge. When the solver fails to meet the convergence accuracy threshold within a limited number of iterations, this embodiment determines that the current cycle has failed and initiates a degradation output mechanism. The degradation output mechanism reads the allocation weight factor from the buffer of the previous control cycle as the output value of the current cycle, ensuring the continuity of power allocation instructions. This embodiment sets an alarm flag while using degradation output. The alarm flag is written to the log storage area and reported to the robot's main controller for maintenance personnel to investigate the cause of the solution failure. Setting the alarm flag does not affect the normal solution process of subsequent control cycles; the alarm flag is automatically reset when the next cycle's solution converges.

[0069] Based on the weighting factor, this embodiment extracts the power demand value at the first moment of the prediction window from the power demand sequence output in step S401. The power demand value represents the target power supply on the load side within the current control cycle. This embodiment multiplies the weighting factor and the power demand value to obtain the target power command for the lithium battery pack. For example, when the weighting factor is 0.6 and the power demand value is 100 watts, the target power command for the lithium battery pack is calculated to be 60 watts. The target power command for the lithium battery pack indicates the power reference value that the bidirectional converter of the lithium battery pack should output.

[0070] Based on the target power command of the lithium battery pack, this embodiment performs a difference operation between the power demand value and the target power command of the lithium battery pack to obtain the target power command of the supercapacitor pack. The difference operation ensures that the sum of the target power command of the lithium battery pack and the target power command of the supercapacitor pack is strictly equal to the power demand value, thereby meeting the power balance constraint requirements. The target power command of the supercapacitor pack indicates the power reference value that the bidirectional converter of the supercapacitor pack should output. When the value is positive, the supercapacitor pack is in discharge mode supplying power to the load; when the value is negative, the supercapacitor pack is in charging mode absorbing power from the bus. The target power command of the lithium battery pack and the target power command of the supercapacitor pack will be encapsulated into a control frame in the subsequent step S601 and sent to the corresponding bidirectional converter for execution.

[0071] In one embodiment of the dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks according to this application, the method may further include the following: Step S601: Encapsulate the target power command of the lithium battery pack and the target power command of the supercapacitor pack into control frames respectively. After attaching a sequence number and a unified timestamp to the control frames, send them to the bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack through the isolated communication interface. Step S602: The bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack are synchronously triggered at the beginning edge of the same control cycle according to the timestamp in the control frame, update the power reference and enter the current inner loop tracking, and report the actual output power and bus voltage sampling value at a fixed feedback period.

[0072] In this embodiment, the target power command for the lithium battery pack and the target power command for the supercapacitor pack output in step S502 are encapsulated into control frames. The encapsulation process constructs an independent control frame data structure for each target power command. The control frame includes a target power command field, a target converter identifier field, and a frame type identifier field. The target power command field stores the power value calculated in step S502, the target converter identifier field distinguishes between the bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack, and the frame type identifier field indicates that the control frame is a power reference setting command.

[0073] Based on the control frame, this embodiment adds a sequence number and a unified timestamp. The sequence number is generated by an incrementing counter maintained by the energy storage management unit. The counter increments once after each control frame is sent. The sequence number is used by the receiving end to verify the arrival order and integrity of the control frames. The unified timestamp is read from the system clock of the energy storage management unit. Both control frames use the same clock value as the timestamp to ensure time base consistency. This embodiment writes the sequence number into the sequence number field of the control frame and the unified timestamp into the timestamp field of the control frame, completing the complete encapsulation of the control frame.

[0074] Accordingly, this embodiment sends the encapsulated control frame to the lithium battery pack bidirectional converter and the supercapacitor pack bidirectional converter via an isolated communication interface. The isolated communication interface establishes independent communication links between the energy storage management unit and each bidirectional converter. The interface circuit uses opto-isolators to achieve electrical isolation between the control side and the power side. The electrical isolation design suppresses the propagation of high-frequency switching noise and common-mode interference in the energy storage loop to the control loop, ensuring the signal integrity of the control frame transmission. In this embodiment, the lithium battery pack control frame is sent to the communication port of the lithium battery pack bidirectional converter, and the supercapacitor pack control frame is sent to the communication port of the supercapacitor pack bidirectional converter, based on the target converter identification field.

[0075] After the control frame arrives at the bidirectional converter, both the lithium battery pack bidirectional converter and the supercapacitor pack bidirectional converter parse the received control frame and extract the timestamp field. Based on the timestamp in the control frame, the two converters calculate the remaining waiting time until the target trigger time, and synchronously execute the power reference update when the waiting time reaches zero. This synchronous triggering mechanism ensures that the two converters complete the power reference switching at the beginning of the same control cycle, avoiding bus power mismatch and voltage transient fluctuations caused by execution timing deviations.

[0076] After the synchronization trigger is completed, both the lithium battery pack bidirectional converter and the supercapacitor pack bidirectional converter write the target power command from the control frame into their internal power reference registers. The value in the power reference register serves as the reference input for the current inner loop controller, which converts the power reference into the corresponding current reference value and calculates the deviation with the actual output current. The current inner loop controller uses a proportional-integral (PI) regulation strategy to adjust the pulse width modulation duty cycle based on the deviation value, driving the switching devices to change the output current to track the current reference value. In this embodiment, both converters enter the current inner loop tracking mode immediately after the power reference is updated, continuously performing closed-loop regulation.

[0077] During the current inner-loop tracking process, the bidirectional converters of the lithium battery pack and the supercapacitor pack report the actual output power and bus voltage sampling values ​​to the energy storage management unit at a fixed feedback period. The actual output power is obtained by multiplying the output voltage sampling value and the output current sampling value inside the converter, while the bus voltage sampling value is obtained from the voltage detection circuit of the common DC bus. In this embodiment, the actual output power and bus voltage sampling values ​​are encapsulated into a feedback frame and transmitted back to the energy storage management unit through an isolated communication interface. The feedback frame is appended with a converter identifier and a sampling time marker so that the energy storage management unit can distinguish the data source and timing. The feedback frame will be received by the energy storage management unit in the subsequent step S701 and used for deviation comparison and command compensation judgment.

[0078] In one embodiment of the dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks according to this application, the method may further include the following: Step S701: The energy storage management unit receives the actual output power and bus voltage sample values ​​reported by the bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack. It calculates the power deviation value by performing a difference calculation between the actual output power and the corresponding target power command. At the same time, it compares the bus voltage sample value with the preset allowable range to obtain a voltage deviation indicator. Step S702: When the power deviation value exceeds the preset deviation threshold, the instruction compensation is triggered and a compensation power instruction is generated and sent to the corresponding bidirectional converter. When the voltage deviation indicator indicates that the bus voltage exceeds the allowable range, the emergency power redistribution logic is started, and the allocation result and execution status are written to the log storage area according to the control cycle.

[0079] In this embodiment, the energy storage management unit receives the feedback frames reported by the bidirectional converters of the lithium battery pack and the supercapacitor pack in step S602. The receiving process reads the feedback frame data from the receive buffer of the isolated communication interface, distinguishing between the lithium battery pack feedback data and the supercapacitor pack feedback data based on the converter identification field in the feedback frame. This embodiment extracts the actual output power field and the bus voltage sample value field from each feedback frame, and organizes the two feedback data streams into a feedback data group based on the sampling time marker.

[0080] Based on the feedback data set, this embodiment performs a difference calculation between the actual output power and the corresponding target power command. The difference calculation process reads the corresponding values ​​from the lithium battery pack target power command and the supercapacitor pack target power command output in step S502. The lithium battery pack power deviation value is obtained by subtracting the lithium battery pack target power command from the actual output power of the lithium battery pack, and the supercapacitor pack power deviation value is obtained by subtracting the supercapacitor pack target power command from the actual output power of the supercapacitor pack. The power deviation value reflects the power tracking accuracy of the bidirectional converter; a positive value indicates that the actual output is higher than the command value, and a negative value indicates that the actual output is lower than the command value.

[0081] Accordingly, this embodiment compares the bus voltage sample value with a preset allowable range to obtain a voltage deviation indicator. The preset allowable range consists of an upper boundary voltage value and a lower boundary voltage value. The boundary values ​​are determined based on the rated bus voltage of the hybrid energy storage system and the voltage tolerance of the load equipment and are stored in the parameter area. The comparison process determines whether the bus voltage sample value is between the upper and lower boundaries. When the bus voltage is within the allowable range, the voltage deviation indicator is set to a normal status code; when the bus voltage is higher than the upper boundary, the voltage deviation indicator is set to an overvoltage status code; and when the bus voltage is lower than the lower boundary, the voltage deviation indicator is set to an undervoltage status code.

[0082] Based on the power deviation value, this embodiment compares the absolute value of the power deviation value with a preset deviation threshold condition to determine whether to trigger command compensation. The preset deviation threshold condition is stored as a configurable parameter in the parameter area, and its value is determined based on the steady-state tracking accuracy of the converter and the system power margin. When the absolute value of the power deviation value of the lithium battery pack or the supercapacitor pack exceeds the preset deviation threshold condition, this embodiment triggers the command compensation logic. The command compensation logic calculates the compensation power command based on the sign and magnitude of the power deviation value. The value of the compensation power command is the opposite of the power deviation value to offset the tracking deviation. In this embodiment, the compensation power command is encapsulated into a compensation control frame and sent to the corresponding bidirectional converter through an isolated communication interface. After receiving the compensation control frame, the converter adds the compensation amount to the current power reference to perform correction.

[0083] Based on the voltage deviation indicator, this embodiment determines whether to activate the emergency power redistribution logic. When the voltage deviation indicator indicates that the bus voltage exceeds the allowable range, this embodiment determines that the system power balance is abnormal and activates the emergency power redistribution logic. The emergency power redistribution logic temporarily suspends the regular rolling optimization process and adjusts the power output ratio of the two converters according to the direction of voltage deviation. When the bus voltage is higher than the upper boundary, it increases the power absorption on the load side or reduces the power output on the energy storage side to lower the bus voltage; when the bus voltage is lower than the lower boundary, it increases the power output on the energy storage side to raise the bus voltage. This embodiment deactivates the emergency state and resumes the regular rolling optimization process after the bus voltage returns to the allowable range.

[0084] In this embodiment, the allocation results and execution status of each control cycle are written to the log storage area according to a preset field format. The fields include timestamp, task identifier, power requirement value, allocation weight factor, target power command for the lithium battery pack, target power command for the supercapacitor pack, actual output power of the lithium battery pack, actual output power of the supercapacitor pack, state of charge of the lithium battery pack, state of charge of the supercapacitor pack, bus voltage, power deviation value, voltage deviation identifier, and anomaly marker. The log storage area employs a circular overwrite strategy to manage storage space; when the storage area is full, new records overwrite the oldest historical records. Critical abnormal events are simultaneously written to the persistent storage area while being written to the log storage area. Records in the persistent storage area do not participate in the circular overwrite to support post-event fault tracing. The log records provide a complete data link for model iteration and parameter calibration during the offline phase.

[0085] To effectively address the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and instruction execution compensation, and to provide technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios for intelligent robots, this application provides an embodiment of a dynamic power allocation device for a hybrid energy storage system for intelligent robots in complex task scenarios, which implements all or part of the aforementioned dynamic power allocation method for the hybrid energy storage system for intelligent robots in complex task scenarios. See [link to embodiment]. Figure 2 The dynamic power distribution device for the intelligent robot hybrid energy storage system for complex tasks specifically includes the following components: The battery data acquisition module 10 is used to obtain task status information by acquiring task type identifier and task stage identifier through the robot main controller, to acquire terminal voltage, output current and temperature of lithium battery pack and supercapacitor pack through energy storage management unit to obtain sampling data set, to perform state estimation on the sampling data set to obtain lithium battery pack state of charge, lithium battery pack health state and supercapacitor pack state of charge, and to organize the task status information and each state of charge and health state into a state vector and write it into a circular buffer. The weight allocation determination module 20 is used to read the state vector sequence from the circular buffer and send it into the task power prediction model to obtain the power demand sequence. Based on the power demand sequence and the energy storage state, a rolling optimization problem is constructed and power balance constraints, power limit constraints and charge state boundary constraints are set. The weight allocation factor is obtained by solving the rolling optimization problem. Based on the weight allocation factor, the target power command of the lithium battery pack and the target power command of the supercapacitor pack are calculated. The dynamic allocation module 30 is used to encapsulate the target power command of the lithium battery pack and the target power command of the supercapacitor pack into a control frame and send it to the corresponding bidirectional converter to perform synchronous triggering. It receives the actual output power reported by the converter and compares it with the target power command. When the deviation exceeds a preset threshold condition, it triggers command compensation.

[0086] As can be seen from the above description, the dynamic power allocation device for a hybrid energy storage system for intelligent robots with complex tasks provided in this application can construct a task state vector by collecting and estimating multi-source energy storage parameters, obtain dual-path target power commands by combining a task power prediction model and multi-constraint rolling optimization, and achieve closed-loop power execution through control frame synchronous triggering and deviation compensation. This effectively solves the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision-making, and command execution compensation, and provides technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios for intelligent robots.

[0087] This invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks.

[0088] This invention also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks.

[0089] This invention also provides a computer program product, which includes a computer program that, when executed by a processor, implements the above-described dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks.

[0090] In this embodiment of the invention, a task state vector is constructed by collecting and estimating multi-source energy storage parameters. The dual-path target power command is obtained by combining the task power prediction model and multi-constraint rolling optimization solution. Closed-loop power execution is achieved through control frame synchronous triggering and deviation compensation. This effectively solves the shortcomings of traditional technologies in multi-source energy storage state perception, rolling optimization power allocation decision and command execution compensation, and provides technical support for the dynamic power allocation and stable operation of hybrid energy storage systems in complex task scenarios of intelligent robots.

[0091] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0092] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0093] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0094] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0095] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above descriptions are merely specific embodiments of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A dynamic power allocation method for a hybrid energy storage system for intelligent robots designed for complex tasks, characterized in that, The method includes: The task status information is obtained by acquiring the task type identifier and task stage identifier through the robot main controller. The terminal voltage, output current and temperature of the lithium battery pack and supercapacitor pack are collected by the energy storage management unit to obtain the sampling data set. The state of charge of the lithium battery pack, the state of health of the lithium battery pack and the state of charge of the supercapacitor pack are obtained by performing state estimation on the sampling data set. The task status information and each state of charge and health are organized into a state vector and written into a circular buffer. The state vector sequence is read from the circular buffer and fed into the task power prediction model to obtain the power demand sequence. Based on the power demand sequence and the energy storage state, a rolling optimization problem is constructed and power balance constraints, power limit constraints and state of charge boundary constraints are set. The rolling optimization problem is solved to obtain the allocation weight factor. Based on the allocation weight factor, the target power command of the lithium battery pack and the target power command of the supercapacitor pack are calculated. The target power command of the lithium battery pack and the target power command of the supercapacitor pack are encapsulated into a control frame and sent to the corresponding bidirectional converter to perform synchronous triggering. The actual output power reported by the converter is received and compared with the target power command. When the deviation exceeds a preset threshold condition, command compensation is triggered.

2. The dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks according to claim 1, characterized in that, The task status information is obtained by acquiring the task type identifier and task stage identifier through the robot's main controller, and the sampled data set is obtained by collecting the terminal voltage, output current and temperature of the lithium battery pack and supercapacitor pack through the energy storage management unit, including: The robot main controller obtains raw task data by subscribing to task instruction frames issued by the task scheduling module through the internal bus. It then parses the task type identifier and task stage identifier from the raw task data to obtain task status information. The task type identifier uses predefined encoding to represent four types of task modes: movement, handling, operation, and standby. The task stage identifier represents the start, continuation, and termination status of task execution. The energy storage management unit synchronously collects the terminal voltage, output current, and surface temperature of the lithium battery pack and the terminal voltage and output current of the supercapacitor pack at fixed intervals through a simulated sampling channel to obtain raw sampling data. Median filtering is performed on the raw sampling data to remove pulse interference to obtain a sampling data set. The task status information and the sampling data set are aligned with a unified timestamp and then written into a circular buffer.

3. The dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks according to claim 1, characterized in that, The process of performing state estimation on the sampled data set to obtain the state of charge (SOC) of the lithium battery pack, the health state of the lithium battery pack, and the SOC of the supercapacitor pack, and then organizing the task state information and each SOC and health state into a state vector and writing it into a circular buffer, includes: The terminal voltage, output current, and surface temperature of the lithium battery pack are extracted from the sampled data set as inputs for lithium battery pack state estimation. Based on the lithium battery pack state estimation inputs, the extended Kalman filter algorithm is used to calculate the state of charge and health of the lithium battery pack. The terminal voltage and output current of the supercapacitor pack are extracted from the sampled data set as inputs for supercapacitor pack state estimation. Based on the supercapacitor pack state estimation inputs, the state of charge of the supercapacitor pack is calculated using the voltage integration method. The task status information, the state of charge of the lithium battery pack, the health status of the lithium battery pack, the temperature of the lithium battery pack, the state of charge of the supercapacitor pack, the bus voltage, and the total load current are organized into a state vector according to a preset field order. After attaching a unified timestamp to the state vector, it is written into a circular buffer using a double buffering mechanism. The depth of the circular buffer is configured according to the historical window length required by the task power prediction model.

4. The dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks according to claim 1, characterized in that, The process of reading the state vector sequence from the circular buffer and feeding it into the task power prediction model to obtain the power demand sequence, constructing a rolling optimization problem based on the power demand sequence and energy storage state, and setting power balance constraints, power limit constraints, and state of charge boundary constraints includes: The system reads the state vector sequence from the most recent historical window from the circular buffer, extracts the historical value of total load power and the task type identifier sequence from the state vector sequence as input to the task power prediction model, and the task power prediction model uses a lightweight recurrent neural network structure to perform inference to obtain the power demand sequence in the future prediction window. Based on the power demand sequence, the state of charge (SOC) of the lithium battery pack, the health state of the lithium battery pack, and the SOC of the supercapacitor pack, a rolling optimization problem is constructed. A power balance constraint is set, requiring that the sum of the power of the lithium battery pack and the power of the supercapacitor pack equals the value of the power demand sequence at the corresponding time. A power limit constraint is set, which dynamically adjusts the upper and lower limits of the lithium battery pack power according to the health state of the lithium battery pack. A SOC boundary constraint is set, which limits the SOC of the supercapacitor pack to be between a preset upper and lower boundary.

5. The dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks according to claim 1, characterized in that, The process of solving the rolling optimization problem to obtain the allocation weighting factor, and calculating the target power command for the lithium battery pack and the target power command for the supercapacitor pack based on the allocation weighting factor, includes: A quadratic programming solver is used to iteratively solve the rolling optimization problem in each control cycle. When the solution converges within a limited number of iterations, the allocation weight factor at the first moment of the prediction window is output. When the solution fails to converge, the allocation result of the previous cycle is used as the degraded output and an alarm flag is set. The target power command for the lithium battery pack is obtained by multiplying the weighting factor with the power demand value at the first moment of the prediction window in the power demand sequence. The difference between the power demand value and the target power command for the lithium battery pack is used as the target power command for the supercapacitor pack.

6. The dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks according to claim 1, characterized in that, The step of encapsulating the target power command of the lithium battery pack and the target power command of the supercapacitor pack into a control frame and then sending it to the corresponding bidirectional converter for synchronous triggering includes: The target power command for the lithium battery pack and the target power command for the supercapacitor pack are encapsulated into control frames. After adding a sequence number and a unified timestamp to the control frames, they are sent to the bidirectional converters of the lithium battery pack and the bidirectional converters of the supercapacitor pack through an isolated communication interface. The bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack are synchronously triggered at the beginning edge of the same control cycle according to the timestamp in the control frame, update the power reference and enter the current inner loop tracking, and report the actual output power and bus voltage sampling value at a fixed feedback period.

7. The dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks according to claim 1, characterized in that, The actual output power reported by the receiving converter is compared with the target power command. When the deviation exceeds a preset threshold, command compensation is triggered, including: The energy storage management unit receives the actual output power and bus voltage sample values ​​reported by the bidirectional converter of the lithium battery pack and the bidirectional converter of the supercapacitor pack. It calculates the power deviation value by performing a difference calculation between the actual output power and the corresponding target power command, and compares the bus voltage sample value with a preset allowable range to obtain a voltage deviation indicator. When the power deviation value exceeds the preset deviation threshold, the instruction compensation is triggered and a compensation power instruction is generated and sent to the corresponding bidirectional converter. When the voltage deviation indicator indicates that the bus voltage exceeds the allowable range, the emergency power redistribution logic is started, and the allocation result and execution status are written to the log storage area according to the control cycle.

8. A dynamic power distribution device for a hybrid energy storage system for intelligent robots designed for complex tasks, characterized in that, The device includes: The battery data acquisition module is used to obtain task status information by acquiring task type identifier and task stage identifier through the robot main controller, and to obtain sampled data set by acquiring terminal voltage, output current and temperature of lithium battery pack and supercapacitor pack through energy storage management unit. The module performs state estimation on the sampled data set to obtain the state of charge of lithium battery pack, the state of health of lithium battery pack and the state of charge of supercapacitor pack. The task status information and each state of charge and health state are organized into a state vector and written into a circular buffer. The weight allocation determination module is used to read the state vector sequence from the circular buffer and send it into the task power prediction model to obtain the power demand sequence. Based on the power demand sequence and the energy storage state, a rolling optimization problem is constructed and power balance constraints, power limit constraints and charge state boundary constraints are set. The weight allocation factor is obtained by solving the rolling optimization problem. Based on the weight allocation factor, the target power command of the lithium battery pack and the target power command of the supercapacitor pack are calculated. The dynamic allocation module is used to encapsulate the target power command of the lithium battery pack and the target power command of the supercapacitor pack into a control frame and send it to the corresponding bidirectional converter for synchronous triggering. It receives the actual output power reported by the converter and compares it with the target power command. When the deviation exceeds a preset threshold condition, it triggers command compensation.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the steps of the dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps of the dynamic power allocation method for a hybrid energy storage system for intelligent robots oriented towards complex tasks as described in any one of claims 1 to 7.

Images