Intelligent task allocation method and related equipment

Abstract

The invention discloses an intelligent task allocation method and related equipment. The method comprises the steps of receiving multi-format task description information, extracting multi-dimensional attributes to generate a task vector, constructing an executive vector containing multiple capability dimensions, matching an optimal main executive through a multi-layer perception model, and performing real-time monitoring and adjustment. According to the scheme, the existing defects are specifically solved: multi-format information is analyzed to extract multi-dimensional attributes, so that a system comprehensively understands the context of a task, and the understanding insufficiency of a traditional method is avoided; constructing a multi-dimensional executive capacity vector and combining a scoring model to realize accurate matching of a task and an executive, and solving the problem of unreasonable distribution caused by asymmetry of man-machine capacity; the execution state is monitored in real time, redistribution is triggered, a dynamic feedback closed loop is formed, and the defect that the dynamic performance of a labor division mechanism is insufficient is overcome, so that the task execution flexibility and reliability are improved, resource waste and decision errors are reduced, and the execution efficiency is improved.

Description

Technical Field

[0001] This application relates to the field of human-computer collaboration technology, and more specifically, to a method and related equipment for intelligent task allocation. Background Technology

[0002] With the rapid development of artificial intelligence technology, human-machine collaborative decision-making systems have been widely applied in complex scenarios such as command, intelligent manufacturing, disaster emergency response, and complex decision-making in specific fields. In these scenarios, the task environment often presents significant challenges such as unclear task objectives, incomplete information, and dynamic environmental changes, placing extremely high demands on the accuracy and adaptability of task allocation.

[0003] In existing technologies, traditional human-machine division of labor often relies on statically predefined strategies or experience-based manual allocation methods. These methods lack a deep understanding of the task context and dynamic adaptability. When faced with complex and ever-changing tasks, they are prone to unreasonable role allocation, leading to low task execution efficiency and potentially causing resource waste and decision-making errors. In recent years, although the development of technologies such as large language models and reinforcement learning has empowered intelligent agents with stronger cognitive and reasoning abilities, several key technical challenges remain in practical applications: First, tasks are highly complex, with multi-level, multi-stage, and multi-constraint structures, making it difficult to simply determine whether a task should be completed by a human or a machine; second, there is an asymmetry in human and machine capabilities. Humans possess strong experience and judgment, but are inefficient and prone to fatigue, while intelligent agents are efficient in structured tasks but lack the ability to handle ambiguity and define responsibility boundaries; third, the division of labor mechanism lacks dynamism, and existing systems struggle to adjust roles in real time during task execution.

[0004] Therefore, there is an urgent need for a new intelligent task allocation method to address the shortcomings of existing technologies. Summary of the Invention

[0005] This application provides a task intelligent allocation method and related equipment. By parsing multi-format task information, constructing multi-dimensional attribute and capability vectors, and matching the optimal executor through a model, combined with real-time monitoring and dynamic adjustment, it can solve the problems of insufficient understanding of tasks, unreasonable human-machine matching, and lack of dynamic division of labor in traditional methods. It can achieve precise matching and flexible adjustment of tasks and executors, thereby improving execution efficiency and reliability.

[0006] A method for intelligent task allocation, characterized in that it includes:

[0007] Receive task description information input by the user, the task description information including natural language text, structured data, and mixed formats;

[0008] The task description information is parsed to extract multi-dimensional task attribute features and generate a task attribute vector. The attribute dimensions included in the task attribute vector include at least three of the following: task determinism, computability, context dependence, urgency, security level requirements, and intensity of human-computer interaction requirements.

[0009] Construct capability vectors for each candidate executor. The executor includes human operators, physical execution devices, and intelligent agent systems. The capability dimensions included in each candidate executor capability vector include at least four of the following: perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range.

[0010] The task attribute vector and the capability vector of each candidate executor are input into the multilayer perceptual scoring model to calculate the fit score, and the executor with the highest output score is used as the main executor to assign tasks.

[0011] Monitor the task execution status of the main executor, collect response latency, execution deviation and abnormal event indicators, and trigger task interruption and reallocation if failure risk is detected.

[0012] Optionally, the task description information is parsed to extract multi-dimensional task attribute features and generate a task attribute vector, including:

[0013] If the task description information is natural language text, a large language model is invoked to perform semantic parsing, identify keywords and map them to preset attribute dimensions, and missing attribute dimensions are filled in by reasoning from the large language model.

[0014] If the task description information is structured data, directly extract the field values ​​to populate the task attribute vector and verify the field integrity;

[0015] If the task description information is in a mixed format, the structured and unstructured parts are separated and processed in parallel, and the conflict attributes are given priority based on the unstructured parsing results.

[0016] Optionally, construct the capability vector for each candidate executor, including:

[0017] For the ability vector of human operators, fatigue is calculated by collecting physiological indicators through wearable devices, and dynamically updated by combining historical task error rates and multi-task switching delays;

[0018] The capability vector of a physical execution device is generated by acquiring the mechanical state in real time through embedded sensors and fusing the device's basic performance parameters.

[0019] The capability vector of the intelligent agent system is obtained by acquiring inference confidence, resource utilization, and historical task response time through the API interface, and then updating it with time decay weight.

[0020] Optionally, the multi-layer perceptual scoring model is trained by constructing a training set based on historical task allocation records labeled with task attribute vectors, candidate execution entity capability vectors and true matching labels, and optimized by training with mean squared error as the loss function.

[0021] The loss function is:

[0022]

[0023] in, The value of the loss function. The total number of samples in the training set. Let θ be the model's prediction fit score for the i-th sample, and let θ be the set of model parameters. To obtain the i-th input feature by concatenating the task attribute vector and the candidate executor capability vector, Let be the true matching label of the i-th sample.

[0024] Optionally, the formula for calculating the fit score of the multi-layer perception scoring model is:

[0025]

[0026] in, Let be the fit score between the i-th candidate executor and the task. To obtain the i-th input feature by concatenating the task attribute vector and the candidate executor capability vector, It is the Sigmoid activation function. It is the ReLU activation function. This is the set of model parameters.

[0027] Optional, also includes:

[0028] Collect task execution log data, and statistically analyze the data deviation between the actual success rate of the task and the predicted success rate of the multilayer perception scoring model based on a sliding window. When the data deviation exceeds a preset number of stable thresholds, retraining of the multilayer perception scoring model is triggered.

[0029] A task intelligent allocation device, comprising:

[0030] The description information unit is used to receive task description information input by the user, wherein the task description information includes natural language text, structured data, and mixed formats;

[0031] The attribute vector unit is used to parse the task description information, extract multi-dimensional task attribute features, and generate a task attribute vector. The attribute dimensions included in the task attribute vector include at least three of the following: task determinism, computability, context dependence, urgency, security level requirements, and intensity of human-computer interaction requirements.

[0032] A capability vector unit is used to construct capability vectors for each candidate executor. The executor includes human operators, physical execution devices, and intelligent agent systems. The capability dimensions included in the capability vectors of each candidate executor include at least four of the following: perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range.

[0033] The adaptation and allocation unit is used to input the task attribute vector and the capability vector of each candidate executor into the multilayer perceptual scoring model, calculate the adaptation score, and assign the executor with the highest output score as the main executor to the task.

[0034] The monitoring and interruption unit is used to monitor the task execution status of the main executor, collect response latency, execution deviation and abnormal event indicators, and trigger task interruption and reallocation if failure risk is detected.

[0035] A task intelligent allocation device, comprising a memory and a processor;

[0036] The memory is used to store programs;

[0037] The processor is configured to execute the program to implement the steps of the task intelligent allocation method as described in any of the preceding claims.

[0038] A readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the task intelligent allocation method as described in any of the preceding claims.

[0039] A computer program product includes a computer program that, when executed by a processor, performs the steps of the task intelligent allocation method as described in any of the preceding claims.

[0040] As can be seen from the above technical solutions, the task intelligent allocation method and related equipment provided in this application include receiving task description information containing natural language text, structured data, and mixed formats; parsing the information to extract multi-dimensional task attribute features and generate a task attribute vector, the attribute dimensions of which include at least three types such as task determinism; constructing capability vectors for each candidate executor, including human operators, physical execution devices, and intelligent agent systems, the capability dimensions of which include at least four types such as perception capabilities; inputting the task attribute vector and the executor capability vector into a multi-layer perception scoring model, selecting the executor with the highest suitability as the main executor for task allocation; monitoring the task execution status of the main executor, and triggering task interruption and reassignment if a failure risk is detected.

[0041] This solution effectively addresses the shortcomings of existing technologies: Firstly, it addresses the lack of task context understanding in traditional methods by parsing multi-format task description information and extracting multi-dimensional attribute features, enabling the system to comprehensively grasp task characteristics. Secondly, it addresses the asymmetry between human and machine capabilities by constructing a multi-dimensional executor capability vector, which, combined with a multi-layered perception scoring model, achieves precise matching between tasks and executors, avoiding unreasonable role allocation. Thirdly, it addresses the lack of dynamism in the division of labor mechanism by monitoring execution status in real time and triggering reassignment when risks arise, forming a dynamically adjusted feedback loop. This improves the flexibility and reliability of task execution, reduces resource waste and decision-making errors, and significantly enhances task execution efficiency. Attached Figure Description

[0042] 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 only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0043] Figure 1 This is a flowchart of a task intelligent allocation method disclosed in an embodiment of this application;

[0044] Figure 2 This is a schematic diagram of a task intelligent allocation method disclosed in an embodiment of this application;

[0045] Figure 3 This is a schematic diagram of a task intelligent allocation device disclosed in an embodiment of this application;

[0046] Figure 4 This is a hardware structure block diagram of a task intelligent allocation device disclosed in an embodiment of this application. Detailed Implementation

[0047] 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, and 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.

[0048] This application can be used in a wide variety of general-purpose or special-purpose computing device environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor devices, distributed computing environments including any of the above devices, etc.

[0049] The following section introduces the solution proposed in this application. The technical solution is as follows, please refer to the text below for details.

[0050] Figure 1 This is a flowchart of a task intelligent allocation method disclosed in an embodiment of this application.

[0051] Figure 2 This is a schematic diagram of a task intelligent allocation method disclosed in an embodiment of this application.

[0052] like Figure 1 and Figure 2 As shown, the method may include:

[0053] Step S1: Receive the task description information input by the user.

[0054] Specifically, this step enables the receipt of user task description information, supporting users to submit task requirements in diverse formats, including natural language text, structured data, and mixed formats, providing basic data for subsequent task attribute parsing. Users can submit tasks through hardware such as management control terminals (e.g., touch screens, desktop control stations), mobile interactive devices (e.g., tablets, voice interaction headsets), or remote control servers. The input information is processed by a preprocessing mechanism: the task description information includes natural language text (e.g., "Dispatch manpower or intelligent agents to complete a security patrol of a certain area within 30 minutes, requiring video recognition to judge abnormal behavior"), structured data (e.g., a form filled out according to a preset template containing fields such as task objectives, time limits, and security levels), and mixed formats (e.g., task descriptions containing text descriptions and table attachments). First, the format parsing module determines the input type and calls the corresponding parser (NLP parser for natural language, form parser for structured data) to extract valid information; then, unstructured input is converted into system standard fields through field mapping, and field validation is performed (e.g., time format validity check, empty field completion); for natural language input, a large language model can also be connected for semantic enhancement (e.g., error correction, standardization processing) to ensure the integrity and standardization of task information.

[0055] The processing methods for task description information in different formats are as follows:

[0056] If the task description information is natural language text, a large language model is invoked for semantic parsing to identify keywords and map them to preset attribute dimensions. Missing attribute dimensions are filled in by reasoning from the large language model.

[0057] Specifically, if the task description information is natural language text, the system calls a large language model (such as GPT) to perform semantic parsing, extracts information through entity recognition and keyword matching (such as recognizing the time limit corresponding to "within 30 minutes" and the task type corresponding to "security patrol"), and maps the recognition results to preset attribute dimensions; for attribute dimensions missing in the parsing process (such as security level requirements not explicitly mentioned), the large language model completes them based on contextual reasoning to ensure the completeness of attribute information.

[0058] If the task description information is structured data, directly extract the field values ​​to populate the task attribute vector and verify the field integrity.

[0059] Specifically, if the task description information is structured data, the system directly extracts field values ​​from preset fields, fills them into the corresponding task attribute vector dimension, and checks the field integrity through the field validation module (such as validating the time format and whether required fields are empty), and prompts for missing fields to be filled in.

[0060] If the task description information is in a mixed format, the structured and unstructured parts are separated and processed in parallel, and the conflict attributes are given priority based on the unstructured parsing results.

[0061] Specifically, if the task description information is in a mixed format, the system first separates the structured part (such as table fields) and the unstructured part (such as text description), and processes them in parallel using structured data processing method and natural language text processing method respectively; if there is an attribute conflict between the two parsing results (such as the structured part being marked "low security level" while the text description mentions "high risk area"), the unstructured parsing result takes priority to ensure the semantic consistency of the task description.

[0062] Step S2: Parse the task description information, extract multi-dimensional task attribute features, and generate a task attribute vector.

[0063] Specifically, by performing structured parsing and feature extraction on the task description information received in step S1, the multi-source heterogeneous task information is transformed into a standardized task attribute vector, providing a quantitative basis for subsequent capability matching. The attribute dimensions included in the task attribute vector must cover at least three of the following: task determinism, computability, context dependence, urgency, security level requirements, and intensity of human-computer interaction needs. The meanings and extraction methods of each dimension are as follows:

[0064] Task certainty: Characterizes the degree of clarity of task objectives and execution paths (e.g., "high / medium / low"), determined by analyzing whether there are ambiguous conditions (e.g., "adjust as needed") or explicit constraints (e.g., "patrol along a fixed route") in the task description;

[0065] Computability: Identify whether the task can be completed automatically through algorithms or rules (e.g., numerical scores of 0-1 or "yes / no"), based on whether the description involves keywords such as "automatic recognition" or "programmed operation", or content such as "complex situational reasoning" that requires manual judgment;

[0066] Context-dependent: Determining whether a task depends on human understanding of the environment, cultural background, or implicit intentions (e.g., "yes / no") is determined by identifying whether the description contains expressions that require human contextual cognition, such as "in combination with the actual situation on site" or "judgment based on experience."

[0067] Urgency level: Reflects the response time requirements of the task (such as "high / medium / low"), extracted based on time constraint keywords such as "execute immediately" and "complete within 30 minutes";

[0068] Security level requirements: Characterize the level of personal safety, equipment risk, or data sensitivity involved in the task (such as "high / medium / low"), and are determined based on descriptions such as "high-risk area" and "processing of classified data";

[0069] Human-computer interaction demand intensity: reflects the necessity of human assistance, review, or multiple rounds of confirmation in task execution (e.g., "high / medium / low"), extracted through interaction requirements such as "requires human review" and "real-time reporting and waiting for instructions".

[0070] In specific processing, if the task description is natural language text, a large language model is used for semantic parsing, and an intermediate semantic structure is generated by combining it with preset keyword matching rules. This structure is then mapped to the aforementioned attribute dimensions. For ambiguous or missing dimensions, the large language model completes the data based on contextual reasoning. If the data is structured, values ​​are directly extracted from the corresponding fields and populated into the attribute vector, while verifying the completeness of the fields. If the data is in a mixed format, the structured and unstructured parts are separated and processed in parallel, with conflicting attributes taking priority based on the unstructured parsing results. The final generated task attribute vector is such as "[Task Determinism: Medium, Computability: 0.7, Urgency: High]", which achieves a quantitative expression of task features.

[0071] Step S3: Construct the capability vectors of each candidate executor.

[0072] Specifically, the capability dimensions included in each candidate executor capability vector include at least four of the following: perception capability (such as human ability to recognize changes in the scene, and the image recognition accuracy of the device), reasoning and judgment capability (such as human high-order logical reasoning capability, and the fuzzy task analysis capability of the intelligent agent), response speed (such as the average time to complete an operation, and the instruction response latency of the device), fault tolerance capability (such as error correction capability, and the self-recovery capability of the device), and algorithm adaptation range (such as the NLP, CV and other algorithm fields supported by the intelligent agent).

[0073] For three types of candidate executors—human operators, physical execution devices, and intelligent agent systems—capability vectors containing multi-dimensional capability features are constructed. The capability dimensions included in the capability vectors must cover at least four of the following: perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range, in order to achieve quantitative representation of the executor's capabilities.

[0074] Human operator capability vector construction: Physiological indicators (heart rate variability, blink frequency, etc.) are collected through wearable devices (such as heart rate monitoring bracelets and eye trackers). Real-time fatigue is calculated based on a preset algorithm, which is associated with fault tolerance and response speed. At the same time, error rate data (reflecting the accuracy of perception and reasoning ability) and operation delay data (characterizing response speed) in the historical task logs are extracted. The above parameters are integrated through a dynamic weighting algorithm to form a real-time updated capability vector that includes perception ability, reasoning ability, response speed, and fault tolerance.

[0075] Capability vector construction of physical execution devices: The device's embedded sensors (such as vibration sensors and temperature sensors) are used to collect mechanical operating status (such as vibration amplitude and component temperature) in real time. Combined with the device's basic performance parameters (such as camera resolution and robotic arm operation accuracy), the data fusion model is used to calculate the perception capability (such as image recognition accuracy), response speed (such as command execution latency), fault tolerance capability (such as fault self-diagnosis efficiency), and operability, etc., to generate the capability vector of the physical execution device.

[0076] Capability vector construction of the intelligent agent system: Real-time acquisition of intelligent agent operation data through API interface, including inference confidence (reflecting the model's accuracy level and related inference judgment ability), CPU / memory resource utilization (affecting response speed), and historical task response time (directly representing response speed); The time decay weighted algorithm is used to fuse historical data and real-time data, focusing on updating dimensions such as algorithm adaptation range (e.g., the expansion of supported task domains), response speed, inference judgment ability, and fault tolerance (e.g., the self-correction rate of abnormal inference), to form a dynamically updated intelligent agent capability vector.

[0077] Step S4: Input the task attribute vector and the capability vector of each candidate executor into the multilayer perceptual scoring model, calculate the fit score, and assign the executor with the highest output score as the main executor to the task.

[0078] Specifically, the task attribute vector generated in step S2 is integrated with the candidate executor capability vectors constructed in step S3 to form comprehensive input information that includes task requirements and executor capabilities. The task attribute vector covers multi-dimensional features such as task determinism and computability, while the executor capability vector includes capability dimensions such as perception and reasoning. The two are combined into the model's input data by concatenating vectors.

[0079] The multilayer perceptual scoring model uses a multilayer neural network structure for nonlinear calculations: the input layer receives the concatenated vector, the hidden layer performs feature transformation using the ReLU activation function, and the output layer outputs a fitness score using the Sigmoid activation function. This score quantifies the degree of matching between task attributes and the capabilities of the performer, and its value ranges from 0 to 1; a higher score indicates a higher degree of matching.

[0080] The model's parameters are optimized offline through supervised learning. The training data comes from historical task assignment records, with each record containing a task attribute vector, a candidate executor capability vector, and the corresponding best actual matching result. During training, the model continuously adjusts the network parameters to minimize the error between the predicted score and the actual matching result, thereby learning the complex mapping relationship between tasks and capabilities.

[0081] After the scoring is completed, the system sorts the fit scores of all candidate executors, selects the executor with the highest score as the main executor and assigns tasks; other candidate executors with high scores can be assigned auxiliary roles, such as being responsible for task monitoring or result verification, to improve the coordination and reliability of task execution.

[0082] Step S5: Monitor the task execution status of the main executor, collect response latency, execution deviation and abnormal event indicators, and trigger task interruption and reallocation if failure risk is detected.

[0083] Specifically, the first step is to activate a real-time status acquisition mechanism to continuously acquire operational data from the main executor, including but not limited to location information, task execution progress, feedback data from sensing devices (such as camera footage and sensor readings), and the executor's own status (such as human operator operation records, energy consumption of physical execution devices, and computing power usage of the intelligent agent system). Based on this, three key monitoring indicators are collected:

[0084] Response latency: refers to the time interval between the main executor receiving the task instruction and the start of execution, or the time consumed in responding to real-time scheduling instructions during execution, which is calculated by comparing timestamps;

[0085] Execution deviation: By comparing the actual execution path with the preset planned path, and the difference between the actual progress and the expected progress, the degree of deviation in task execution is quantified, such as the deviation of the motion trajectory of physical execution equipment and the gap in task completion rate of intelligent agent system;

[0086] Abnormal events include, but are not limited to, equipment malfunctions (such as sensor failure of physical execution equipment, program crash of intelligent agent system), operational errors (such as human operator misoperation), and unexpected environmental conditions (such as extreme weather affecting the task, signal interruption), etc.

[0087] The collected metrics are analyzed in real time to determine if there is a risk of failure. A risk of failure is determined when the following situations occur: response delay continuously exceeds the preset threshold (e.g., three consecutive delays exceeding 10 seconds), execution deviation reaches the warning value (e.g., path deviation exceeds the safe range), or a fatal abnormal event occurs (e.g., equipment failure, operational errors involving safety).

[0088] If a failure risk is detected, the system immediately triggers an interruption and reallocation mechanism: First, the execution of the current task is paused, and the task status is marked as suspended; then, the role rematching module is invoked, and based on the remaining task objectives, the current status of the main executor (such as excluding failed executors), and the real-time capability vectors of other candidate executors, the matching process of step S4 is re-executed to select a new optimal executor; finally, the remaining task sub-objectives are issued to the new main executor in the form of standardized instructions to continue advancing the task, ensuring that the task has fault-tolerant recovery capabilities in a dynamic environment.

[0089] Meanwhile, the system displays real-time monitoring data to human monitors through a visual interface, including an overview of task status, curves showing changes in various indicators, and risk warning information. It also provides a manual intervention entry point (such as manually triggering interruptions or adjusting redistribution strategies) to achieve closed-loop control through human-machine collaboration.

[0090] Furthermore, to improve the adaptation accuracy and long-term adaptability of the multi-layer perception scoring model, this application may also add a model adaptive optimization process, specifically including:

[0091] Collect task execution log data, and statistically analyze the data deviation between the actual success rate of the task and the predicted success rate of the multilayer perception scoring model based on a sliding window. When the data deviation exceeds a preset number of stable thresholds, retraining of the multilayer perception scoring model is triggered.

[0092] Specifically, task execution log data is collected. This log data includes task attribute vectors for each task, candidate executor capability vectors, the predicted success rate of the multilayer perceptron scoring model, the actual execution results of the task, and execution process parameters. Based on a sliding window (e.g., the window size is set to the most recent 50 task data points), the data deviation between the actual success rate of the task within this window and the model's predicted success rate is statistically analyzed. This data deviation can be quantified using methods such as absolute difference or mean squared error.

[0093] A preset stability threshold (e.g., a deviation threshold of 15%) and a preset number of iterations (e.g., 3 consecutive iterations) are set. When the deviation counted by the sliding window continuously exceeds the stability threshold and reaches the preset number of iterations, the model performance is considered to be degrading. At this point, the system automatically triggers retraining of the multilayer perceptual scoring model: the feedback collection and learning optimization module is invoked to extract the latest training samples (including task attributes, capability vectors, and actual execution labels) from the log data. The model parameters are updated using strategy fine-tuning or supervised learning to minimize prediction deviation and enable the model to continuously adapt to the dynamically changing task environment and the capability characteristics of the executor.

[0094] As can be seen from the above technical solutions, the task intelligent allocation method and related equipment provided in this application include receiving task description information containing natural language text, structured data, and mixed formats; parsing the information to extract multi-dimensional task attribute features and generate a task attribute vector, the attribute dimensions of which include at least three types such as task determinism; constructing capability vectors for each candidate executor, including human operators, physical execution devices, and intelligent agent systems, the capability dimensions of which include at least four types such as perception capabilities; inputting the task attribute vector and the executor capability vector into a multi-layer perception scoring model, selecting the executor with the highest suitability as the main executor for task allocation; monitoring the task execution status of the main executor, and triggering task interruption and reassignment if a failure risk is detected.

[0095] This solution effectively addresses the shortcomings of existing technologies: Firstly, it addresses the lack of task context understanding in traditional methods by parsing multi-format task description information and extracting multi-dimensional attribute features, enabling the system to comprehensively grasp task characteristics. Secondly, it addresses the asymmetry between human and machine capabilities by constructing a multi-dimensional executor capability vector, which, combined with a multi-layered perception scoring model, achieves precise matching between tasks and executors, avoiding unreasonable role allocation. Thirdly, it addresses the lack of dynamism in the division of labor mechanism by monitoring execution status in real time and triggering reassignment when risks arise, forming a dynamically adjusted feedback loop. This improves the flexibility and reliability of task execution, reduces resource waste and decision-making errors, and significantly enhances task execution efficiency.

[0096] In some embodiments of this application, the multi-layer perception scoring model in step S4 is described.

[0097] The training process for the multilayer perceptual scoring model:

[0098] The multi-layer perceptual scoring model is trained by constructing a training set based on historical task allocation records labeled with task attribute vectors, candidate executor capability vectors, and true matching labels, and optimized by using mean squared error as the loss function.

[0099] The loss function is:

[0100]

[0101] in, The value of the loss function. The total number of samples in the training set. Let θ be the model's prediction fit score for the i-th sample, and let θ be the set of model parameters. To obtain the i-th input feature by concatenating the task attribute vector and the candidate executor capability vector, Let be the true matching label of the i-th sample.

[0102] The formula for calculating the fit score of the multi-layer perception scoring model is as follows:

[0103]

[0104] in, Let be the fit score between the i-th candidate executor and the task. To obtain the i-th input feature by concatenating the task attribute vector and the candidate executor capability vector, It is the Sigmoid activation function. It is the ReLU activation function. This is the set of model parameters.

[0105] Specifically, a training set is constructed based on historical task allocation records. Each record contains three core types of information: a task attribute vector (a feature description of the task, such as complexity and timeliness); a candidate executor capability vector (capability dimension data of the executor, such as human fatigue and equipment precision); and a true match label (marking whether the executor is the actual main executor selected for the task, used to determine the accuracy of the prediction). Mean squared error is used as the loss function for model training. Its function is to quantify the difference between the model's predicted fitness score and the true match label. The square of the difference between the predicted score and the true label is calculated for each training sample. The average of the squared differences of all samples is taken to obtain the final loss value, where a smaller loss value indicates more accurate model predictions. By minimizing this loss value, the internal parameters of the model, such as the weights and biases of each layer, are optimized.

[0106] The model calculates the fit between the executor and the task through a multi-layer neural network, as follows:

[0107] Input concatenation: The task attribute vector and the candidate executor capability vector are concatenated to form the input features of the model;

[0108] Hidden layer processing: When the input data is processed through the hidden layer, nonlinearity is introduced through the ReLU activation function;

[0109] Output layer mapping: After the final calculation by the output layer, the result is compressed to between 0 and 1 by the Sigmoid activation function to obtain the fit score between the executor and the task (the higher the score, the higher the fit).

[0110] The following describes a task intelligent allocation device provided in the embodiments of this application. The task intelligent allocation device described below can be referred to in correspondence with the task intelligent allocation method described above.

[0111] See Figure 3 , Figure 3 This is a schematic diagram of a task intelligent allocation device disclosed in an embodiment of this application.

[0112] like Figure 3 As shown, the intelligent task allocation device may include:

[0113] The description information unit 110 is used to receive task description information input by the user, wherein the task description information includes natural language text, structured data, and mixed formats;

[0114] The attribute vector unit 120 is used to parse the task description information, extract multi-dimensional task attribute features, and generate a task attribute vector. The attribute dimensions included in the task attribute vector include at least three of the following: task determinism, computability, context dependence, urgency, security level requirements, and intensity of human-computer interaction requirements.

[0115] Capability vector unit 130 is used to construct capability vectors for each candidate executor. The executor includes human operators, physical execution devices, and intelligent agent systems. The capability dimensions included in the capability vectors of each candidate executor include at least four of the following: perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range.

[0116] The adaptation and allocation unit 140 is used to input the task attribute vector and the capability vector of each candidate executor into the multilayer perceptual scoring model, calculate the adaptation score, and assign the executor with the highest output score as the main executor to the task.

[0117] The monitoring interruption unit 150 is used to monitor the task execution status of the main executor, collect response delay, execution deviation and abnormal event indicators, and trigger task interruption and reallocation if a failure risk is detected.

[0118] As can be seen from the above technical solutions, the task intelligent allocation method and related equipment provided in this application include receiving task description information containing natural language text, structured data, and mixed formats; parsing the information to extract multi-dimensional task attribute features and generate a task attribute vector, the attribute dimensions of which include at least three types such as task determinism; constructing capability vectors for each candidate executor, including human operators, physical execution devices, and intelligent agent systems, the capability dimensions of which include at least four types such as perception capabilities; inputting the task attribute vector and the executor capability vector into a multi-layer perception scoring model, selecting the executor with the highest suitability as the main executor for task allocation; monitoring the task execution status of the main executor, and triggering task interruption and reassignment if a failure risk is detected.

[0119] This solution effectively addresses the shortcomings of existing technologies: Firstly, it addresses the lack of task context understanding in traditional methods by parsing multi-format task description information and extracting multi-dimensional attribute features, enabling the system to comprehensively grasp task characteristics. Secondly, it addresses the asymmetry between human and machine capabilities by constructing a multi-dimensional executor capability vector, which, combined with a multi-layered perception scoring model, achieves precise matching between tasks and executors, avoiding unreasonable role allocation. Thirdly, it addresses the lack of dynamism in the division of labor mechanism by monitoring execution status in real time and triggering reassignment when risks arise, forming a dynamically adjusted feedback loop. This improves the flexibility and reliability of task execution, reduces resource waste and decision-making errors, and significantly enhances task execution efficiency.

[0120] Optionally, the task description information is parsed to extract multi-dimensional task attribute features and generate a task attribute vector, including:

[0121] If the task description information is natural language text, a large language model is invoked to perform semantic parsing, identify keywords and map them to preset attribute dimensions, and missing attribute dimensions are filled in by reasoning from the large language model.

[0122] If the task description information is structured data, directly extract the field values ​​to populate the task attribute vector and verify the field integrity;

[0123] If the task description information is in a mixed format, the structured and unstructured parts are separated and processed in parallel, and the conflict attributes are given priority based on the unstructured parsing results.

[0124] Optionally, construct the capability vector for each candidate executor, including:

[0125] For the ability vector of human operators, fatigue is calculated by collecting physiological indicators through wearable devices, and dynamically updated by combining historical task error rates and multi-task switching delays;

[0126] The capability vector of a physical execution device is generated by acquiring the mechanical state in real time through embedded sensors and fusing the device's basic performance parameters.

[0127] The capability vector of the intelligent agent system is obtained by acquiring inference confidence, resource utilization, and historical task response time through the API interface, and then updating it with time decay weight.

[0128] Optionally, the multi-layer perceptual scoring model is trained by constructing a training set based on historical task allocation records labeled with task attribute vectors, candidate execution entity capability vectors, and true matching labels, and then optimized using mean squared error as the loss function.

[0129] The loss function is:

[0130]

[0131] in, The value of the loss function. The total number of samples in the training set. Let θ be the model's prediction fit score for the i-th sample, and let θ be the set of model parameters. To obtain the i-th input feature by concatenating the task attribute vector and the candidate executor capability vector, Let be the true matching label of the i-th sample.

[0132] Optionally, the formula for calculating the fit score of the multi-layer perception scoring model is:

[0133]

[0134] in, Let be the fit score between the i-th candidate executor and the task. To obtain the i-th input feature by concatenating the task attribute vector and the candidate executor capability vector, It is the Sigmoid activation function. It is the ReLU activation function. This is the set of model parameters.

[0135] Optional, also includes:

[0136] Collect task execution log data, and statistically analyze the data deviation between the actual success rate of the task and the predicted success rate of the multilayer perception scoring model based on a sliding window. When the data deviation exceeds a preset number of stable thresholds, retraining of the multilayer perception scoring model is triggered.

[0137] The task intelligent allocation device provided in this application embodiment can be applied to task intelligent allocation equipment. Figure 4 The hardware structure block diagram of the task intelligent allocation device is shown below. Figure 4The hardware structure of the task intelligent allocation device may include: at least one processor 1, at least one communication interface 2, at least one memory 3 and at least one communication bus 4;

[0138] In this embodiment of the application, the number of processor 1, communication interface 2, memory 3, and communication bus 4 is at least one, and processor 1, communication interface 2, and memory 3 communicate with each other through communication bus 4;

[0139] Processor 1 may be a central processing unit (CPU), an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement embodiments of the present invention.

[0140] Memory 3 may include high-speed RAM, and may also include non-volatile memory, such as at least one disk storage device;

[0141] The memory stores a program, which the processor can call. The program is used for:

[0142] Receive task description information input by the user, the task description information including natural language text, structured data, and mixed formats;

[0143] The task description information is parsed to extract multi-dimensional task attribute features and generate a task attribute vector. The attribute dimensions included in the task attribute vector include at least three of the following: task determinism, computability, context dependence, urgency, security level requirements, and intensity of human-computer interaction requirements.

[0144] Construct capability vectors for each candidate executor. The executor includes human operators, physical execution devices, and intelligent agent systems. The capability dimensions included in each candidate executor capability vector include at least four of the following: perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range.

[0145] The task attribute vector and the capability vector of each candidate executor are input into the multilayer perceptual scoring model to calculate the fit score, and the executor with the highest output score is used as the main executor to assign tasks.

[0146] Monitor the task execution status of the main executor, collect response latency, execution deviation and abnormal event indicators, and trigger task interruption and reallocation if failure risk is detected.

[0147] Optionally, the refined and extended functions of the program can be referred to the above description.

[0148] This application embodiment also provides a readable storage medium that can store a program suitable for execution by a processor, the program being used for:

[0149] Receive task description information input by the user, the task description information including natural language text, structured data, and mixed formats;

[0150] The task description information is parsed to extract multi-dimensional task attribute features and generate a task attribute vector. The attribute dimensions included in the task attribute vector include at least three of the following: task determinism, computability, context dependence, urgency, security level requirements, and intensity of human-computer interaction requirements.

[0151] Construct capability vectors for each candidate executor. The executor includes human operators, physical execution devices, and intelligent agent systems. The capability dimensions included in each candidate executor capability vector include at least four of the following: perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range.

[0152] The task attribute vector and the capability vector of each candidate executor are input into the multilayer perceptual scoring model to calculate the fit score, and the executor with the highest output score is used as the main executor to assign tasks.

[0153] Monitor the task execution status of the main executor, collect response latency, execution deviation and abnormal event indicators, and trigger task interruption and reallocation if failure risk is detected.

[0154] Optionally, the refined and extended functions of the program can be referred to the above description.

[0155] This application also provides a computer program product, including a computer program, wherein the computer program is executed by a processor using the following method:

[0156] Receive task description information input by the user, the task description information including natural language text, structured data, and mixed formats;

[0157] The task description information is parsed to extract multi-dimensional task attribute features and generate a task attribute vector. The attribute dimensions included in the task attribute vector include at least three of the following: task determinism, computability, context dependence, urgency, security level requirements, and intensity of human-computer interaction requirements.

[0158] Construct capability vectors for each candidate executor. The executor includes human operators, physical execution devices, and intelligent agent systems. The capability dimensions included in each candidate executor capability vector include at least four of the following: perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range.

[0159] The task attribute vector and the capability vector of each candidate executor are input into the multilayer perceptual scoring model to calculate the fit score, and the executor with the highest output score is used as the main executor to assign tasks.

[0160] Monitor the task execution status of the main executor, collect response latency, execution deviation and abnormal event indicators, and trigger task interruption and reallocation if failure risk is detected.

[0161] Optionally, the refined and extended functions of the program can be referred to the above description.

[0162] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0163] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. The same or similar parts between the various embodiments can be referred to each other.

[0164] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent task allocation, characterized in that, Comprising: receiving user input task description information, the task description information including natural language text, structured data and mixed format; parsing the task description information, extracting multi-dimensional task attribute features, and generating a task attribute vector, wherein the attribute dimensions contained in the task attribute vector include at least three of task certainty, computability, context dependence, urgency, safety level requirement and human-computer interaction demand intensity; constructing each candidate executor capability vector, the executor including human operators, physical execution devices and agent systems, the capability dimensions contained in the candidate executor capability vector including at least four of perception capability, reasoning and judgment capability, response speed, fault tolerance capability and algorithm adaptation range; inputting the task attribute vector and each candidate executor capability vector into a multi-layer perception scoring model, calculating an adaptation score, and assigning the executor with the highest output score as the main executor to assign the task; monitoring the task execution state of the main executor, collecting response delay, execution deviation and abnormal event indicators, and triggering task interruption and reassignment if a failure risk is detected.

2. The method of claim 1, wherein, Parsing the task description information, extracting multi-dimensional task attribute features, and generating a task attribute vector, comprising: if the task description information is natural language text, calling a large language model for semantic analysis, identifying keywords and mapping them to preset attribute dimensions, and missing attribute dimensions are inferred and completed by the large language model; if the task description information is structured data, directly extracting field values to fill the task attribute vector, and verifying field integrity; if the task description information is in mixed format, separate structured and unstructured parts for parallel processing, and conflict attributes are given priority to unstructured analysis results.

3. The method of claim 1, wherein, Constructing each candidate executor capability vector, comprising: for the capability vector of human operators, calculating fatigue degree by collecting physiological indicators through wearable devices, and dynamically updating based on historical task error rate and multi-task switching delay; for the capability vector of physical execution devices, real-time acquisition of mechanical state through embedded sensors, and fusion of device basic performance parameters to generate; for the capability vector of agent systems, acquiring reasoning confidence, resource occupation rate and historical task response time through API interface, and updating by time decay weighting.

4. The method of claim 1, wherein, The multi-layer perception scoring model is based on a training set constructed by historical task assignment records labeled with task attribute vectors, candidate executor capability vectors and real matching labels, and is optimized and trained by taking mean square error as a loss function; The loss function is: wherein, is a loss function value, is the total number of training set samples, is the predicted fitness score of the model for the i-th sample, and θ is the set of model parameters, is the i-th input feature based on the splicing of the task attribute vector and the candidate executor capability vector, is the true matching label of the i-th sample.

5. The method of claim 1, wherein, The adaptation score calculation formula of the multi-layer perception scoring model is: wherein, is the fitness score of the ith candidate executor to the task, is the ith input feature based on the concatenation of the task attribute vector and the candidate executor capability vector, is the Sigmoid activation function, is the ReLU activation function, is the set of model parameters.

6. The method of claim 1, wherein, Further comprising: collecting task execution log data, and based on a sliding window, calculating the data deviation between the actual success rate of the task and the predicted success rate of the multi-layer perception scoring model, and when the data deviation continuously exceeds the stable threshold for a preset number of times, triggering retraining of the multi-layer perception scoring model.

7. A task intelligent allocation apparatus, characterized by, Comprising: a description unit for receiving user input task description information, the task description information including natural language text, structured data and mixed format; An attribute vector unit is configured to parse the task description information, extract multi-dimensional task attribute features, and generate a task attribute vector, wherein the attribute dimensions included in the task attribute vector comprise at least three of task certainty, computability, context dependence, emergency level, safety level requirement, and human-computer interaction demand intensity. A capability vector unit is configured to construct a candidate execution body capability vector, wherein the execution body comprises a human operator, a physical execution device, and an agent system, and the capability dimensions included in the candidate execution body capability vector comprise at least four of perception capability, reasoning and judgment capability, response speed, fault tolerance capability, and algorithm adaptation range. An adaptation allocation unit is configured to input the task attribute vector and the candidate execution body capability vector into a multi-layer perception scoring model, calculate an adaptation score, and allocate the execution body with the highest output score as a main execution body to perform the task. A monitoring interruption unit is configured to monitor the task execution state of the main execution body, collect response delay, execution deviation, and abnormal event indicators, and trigger task interruption and reallocation if a failure risk is detected.

8. A task intelligent allocation device, characterized by, comprising a memory and a processor; the memory is configured to store a program; the processor is configured to execute the program to implement each step of the task intelligent allocation method according to any one of claims 1-6.

9. A readable storage medium, having stored thereon a computer program, characterized in that, The computer program, when executed by the processor, implements each step of the task intelligent allocation method according to any one of claims 1-6.

10. A computer program product comprising a computer program, characterized in that, The computer program, when executed by the processor, implements each step of the task intelligent allocation method according to any one of claims 1-6.

Images