Model-based task instruction management method and device, storage medium and equipment

Abstract

The invention discloses a model-based task instruction management method and device, a storage medium and equipment, and belongs to the technical field of deep learning. Obtaining text data of each task to be executed, wherein the text data comprises description information, demand information and supplementary information of the tasks and log information of historical tasks; extracting key information of each task from the text data by utilizing a large language model, wherein the key information comprises an emergency parameter, a target parameter, a resource demand parameter and a time limit parameter of the task; generating a corresponding task instruction based on the key information of each task by utilizing a generative model; combining similar task instructions in the plurality of task instructions; and performing priority ranking on the combined task instructions, and performing execution scheduling on the ranked task instructions. According to the method, the generation requirement of the high-complexity task instruction can be met, the execution efficiency of the instruction and the resource utilization rate are improved, and execution scheduling of the task instruction is adjusted in real time according to resources.

Description

Technical Field

[0001] This application relates to the field of deep learning technology, and particularly to a model-based task instruction management method, device, storage medium, and equipment. Background Art

[0002] Modern task instruction management systems are facing increasingly complex task scheduling, instruction generation, and execution environments. Especially in fields such as military, emergency response, and enterprise resource management, the quantity and complexity of task instructions are growing exponentially, and a more intelligent, efficient, and dynamic way is needed to generate, process, sort, and distribute task instructions.

[0003] Traditional task instruction management methods often have the following main problems when dealing with tasks with high complexity, large scale, and multi-source data:

[0004] 1. Information redundancy and duplicate instructions: Among a large number of task instructions, there are many similar or duplicate task instructions, resulting in waste of resources and reduced efficiency.

[0005] 2. Unintelligent task prioritization and resource scheduling: Traditional methods often rely on manually setting rules for task sorting and resource scheduling, lacking flexibility and dynamics, and it is difficult to adapt to real-time changing requirements.

[0006] 3. Inefficiency of instruction generation and execution: Traditional template- or rule-based instruction generation methods are difficult to meet the requirements of high-complexity tasks, and the execution is not flexible enough to adjust in a timely manner according to the actual situation. Summary of the Invention

[0007] This application provides a model-based task instruction management method, device, storage medium, and equipment for solving the problems of information redundancy of task instructions, unintelligent prioritization and resource scheduling, and low efficiency of execution generation and execution. The technical solutions are as follows:

[0008] According to the first aspect of this application, a model-based task instruction management method is provided. The method includes:

[0009] Obtain the text data of each task to be executed. The text data includes the description information, requirement information, supplementary information, and log information of historical tasks of the task. The supplementary information is the detailed information and target information of the task input by the user;

[0010] Use a large language model to extract the key information of each task from the text data. The key information includes the urgency parameter, target parameter, resource requirement parameter, and time limit parameter of the task;

[0011] Use a generative model to generate corresponding task instructions based on the key information of each task;

[0012] Combine similar task instructions among multiple task instructions;

[0013] Perform priority sorting on each of the combined task instructions, and perform execution scheduling on the sorted task instructions.

[0014] In an optional embodiment, the performing priority sorting on each of the combined task instructions includes:

[0015] Calculate the priority scores of each of the combined task instructions according to the weights corresponding to the key information, and perform priority sorting on each task instruction according to the priority scores;

[0016] Evaluate the long-term benefits of each of the sorted task instructions by using a reinforcement learning algorithm, and update the priority sorting of each task instruction according to the long-term benefits.

[0017] In an optional embodiment, the calculating the priority scores of each of the combined task instructions according to the weights corresponding to the key information includes:

[0018] For each task instruction, determine an urgency score, an importance score, a resource requirement score, and a time limit score according to the key information of the task instruction;

[0019] Obtain preset urgency weights, importance weights, resource requirement weights, and time limit weights;

[0020] Perform weighted calculations on the urgency score and the urgency weight, the importance score and the importance weight, the resource requirement score and the resource requirement weight, and the time limit score and the time limit weight to obtain the priority score of the task instruction.

[0021] In an optional embodiment, the evaluating the long-term benefits of each of the sorted task instructions by using a reinforcement learning algorithm includes:

[0022] Select the task instruction to be executed in the current state according to each of the sorted task instructions;

[0023] Calculate the Q value after executing the selected task instruction by using the Q value function in the reinforcement learning algorithm, where the Q value function represents the maximum expected return that can be obtained by executing a task instruction in a given state;

[0024] Update the priority sorting of each task instruction according to the objective of maximizing the Q value.

[0025] In an optional embodiment, the method further includes:

[0026] After generating multiple task instructions, optimize each task instruction using the domain knowledge and decision rules in the rules engine;

[0027] Use the ROUGE automated evaluation algorithm to evaluate the quality of each generated task instruction, and adjust the generation strategy of the task instruction by the generative model according to the evaluation results.

[0028] In an optional embodiment, the merging of similar task instructions among the multiple task instructions includes:

[0029] Calculate the semantic similarity between every two task instructions;

[0030] If the semantic similarity is greater than the similarity threshold, merge the two task instructions.

[0031] In an optional embodiment, the method further includes:

[0032] Real-time monitor the execution status and feedback information of each task instruction;

[0033] Adjust the priorities and resource allocations of each task instruction according to the execution status and the feedback information to optimize the execution scheduling of each task instruction.

[0034] According to the second aspect of the present application, there is provided a model-based task instruction management device, the device includes:

[0035] An acquisition module, configured to acquire the text data of each task to be executed, the text data includes the description information, requirement information, supplementary information, and log information of the historical tasks of the task, and the supplementary information is the detailed information and target information of the task input by the user;

[0036] An extraction module, configured to extract the key information of each task from the text data by using a large language model, and the key information includes the urgency parameter, target parameter, resource requirement parameter, and time limit parameter of the task;

[0037] A generation module, configured to generate corresponding task instructions based on the key information of each task by using a generative model;

[0038] A merging module, configured to merge similar task instructions among multiple task instructions;

[0039] A management module, configured to perform priority sorting on each merged task instruction and perform execution scheduling on the sorted task instructions.

[0040] According to a third aspect of the present application, there is provided a computer-readable storage medium storing at least one instruction, which is loaded and executed by a processor to implement the above-described model-based task instruction management method.

[0041] According to a fourth aspect of the present application, there is provided a computer device including the above-described model-based task instruction management device.

[0042] The beneficial effects of the technical solution provided by the present application at least include:

[0043] By using a large language model to extract key information from the text data of a task, and then using a generative model to generate task instructions based on the key information, it is possible to convert the text data of the task into task instructions through the large model. Compared with generating task instructions according to templates or rules, it can meet the generation requirements of highly complex task instructions; by merging similar task instructions among multiple task instructions, duplicate instructions and information redundancy can be reduced, and the execution efficiency of instructions and resource utilization rate can be improved; by sorting the priorities of the merged task instructions and performing execution scheduling on the sorted task instructions, it is possible to schedule task instructions according to priorities, which not only ensures the priority execution of important task instructions, but also can adjust the execution order of task instructions in real time according to resources, improving flexibility and dynamics. BRIEF DESCRIPTION OF THE DRAWINGS

[0044] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0045] Figure 1 is a flowchart of a model-based task instruction management method provided by an embodiment of the present application;

[0046] Figure 2 is a flowchart of a model-based task instruction management method provided by an embodiment of the present application;

[0047] Figure 3 is a structural block diagram of a model-based task instruction management device provided by an embodiment of the present application. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0048] To make the objectives, technical solutions, and advantages of the embodiments of the present application clearer, the following will further describe the embodiments of the present application in detail with reference to the drawings.

[0049] As Figure 1As shown, it shows a flowchart of a model-based task instruction management method provided by an embodiment of the present application. This model-based task instruction management method can be applied to a computer device. This model-based task instruction management method may include:

[0050] Step 101, obtain the text data of each task to be executed. The text data includes the description information, requirement information, supplementary information, and log information of historical tasks. The supplementary information is the detailed information and target information of the task input by the user.

[0051] The description information is the descriptive text of the task collected in real time, which may include the task objective, requirements, resource status, and may also include real-time data collected by the production management system, such as the current factory production line status, machine working conditions, worker attendance, etc.

[0052] The requirement information is the requirement description generated according to the task environment and scenario, usually provided by external personnel or systems.

[0053] The supplementary information is the task instruction input by the user through the interface or manually, including task details, target description, etc., and also includes production goals, task types, time limits, resource requirements, etc. Each task will include detailed production goals, such as "produce 1000 product A", required resources, such as "10 workers, 5 production machines", and task types (such as production tasks, maintenance tasks, etc.).

[0054] The log information of historical tasks is the relevant information extracted from the data of historical tasks, which helps to understand the past task execution situation, usually including task types, priorities, execution times, and also includes the completion situation of past production tasks, resource utilization efficiency, task delay records, etc.

[0055] After obtaining the text data, it is necessary to perform standardization processing on the text data to ensure data consistency and clarity. The purpose of data standardization is to convert the original text data into structured and unified format text, including: denoising: removing irrelevant information, noise data, duplicate content (such as advertisements, incorrect formats, etc.); format unification: ensuring consistent data formats and removing unnecessary characters, punctuation marks, spaces, etc.

[0056] Step 102, use a large language model to extract the key information of each task from the text data. The key information includes the urgency parameter, target parameter, resource requirement parameter, and time limit parameter of the task.

[0057] After obtaining the text data, a large language model can be used to transform the text data into high-dimensional semantic vectors. Among them, the large language model can be a Bidirectional Encoder Representations from Transformers (BERT) model. The BERT model is a pre-trained language model based on Transformer. After inputting the text data into the BERT model, the BERT model can capture the context information in the text data and transform the text data into a high-dimensional semantic vector (usually 512-dimensional or 768-dimensional). For example, the sentence "Produce 1000 product A, resources: 10 workers, 5 production machines" is transformed into a high-dimensional semantic vector through the BERT model.

[0058] Then, the key information of the task can be extracted according to the high-dimensional semantic vector. The key information includes an urgency parameter, a target parameter, a resource requirement parameter, and a time limit parameter. Among them, the urgency parameter judges the urgency of the task according to the keywords in the text data (such as "urgent", "immediately", etc.); the target parameter identifies the target of the task from the text data (such as "destroy the radar", "repair the equipment", etc.); the resource requirement parameter extracts the resource-related information in the text data (such as "5 vehicles are needed", "3 devices are needed", etc.); the time limit parameter extracts the time requirement of the task according to the time expression in the text data (such as "complete within one hour").

[0059] Step 103: Use a generative model to generate corresponding task instructions based on the key information of each task.

[0060] Instruction generation means considering the task environment and constraints, and generating a preliminary instruction text based on the task objective, resource availability, and time limit to ensure that the instruction is detailed and specific.

[0061] The generative model can be a Generative Pre-trained Transformer (GPT) model.

[0062] The GPT model can generate task instructions consistent with the key information by considering the task environment and constraints, etc., through the language patterns and context relationships learned during its training process. For example, after inputting "Produce 1000 product A, 10 workers, 5 machines, complete within 3 hours", the generated task instruction is "Dispatch 10 workers and 5 production machines to produce 1000 product A and complete it within 3 hours."

[0063] Step 104: Merge similar task instructions among multiple task instructions.

[0064] In this embodiment, redundancy in task instructions can be reduced and the task structure can be optimized through semantic similarity calculation and an automatic merging mechanism. The semantic similarity calculation and the automatic merging mechanism are described in detail below.

[0065] Step 105: Perform priority sorting on each of the merged task instructions, and perform execution scheduling on the sorted task instructions.

[0066] In this embodiment, the priority sorting includes two parts. One part is priority analysis, and the other part is reinforcement learning analysis. The specific analysis directions are described in detail below.

[0067] After determining the priority sorting, the task instructions can be dynamically split and executed. Specifically, according to the task priority and resource situation, the task instructions can be automatically distributed to the corresponding execution units. Then, the task scheduling is dynamically adjusted through real-time monitoring and feedback mechanisms to optimize resource allocation, ensuring that high-priority task instructions are executed in a timely manner and low-priority tasks are postponed or scheduled appropriately.

[0068] The task instruction management method provided in this embodiment is superior to traditional methods in terms of flexibility, automation, efficiency, real-time adjustment ability, etc. It can especially handle complex and changing task environments. Therefore, this task instruction management method can not only meet the urgent needs of high-efficiency and intelligent instruction management in fields such as military, emergency response, and enterprise scheduling, but also make up for the deficiencies of traditional methods in multi-task scheduling, resource optimization, etc., and promote the intelligent development of instruction management systems in various industries.

[0069] In summary, the model-based task instruction management method provided in the embodiments of this application extracts key information from the text data of tasks by using a large language model, and then uses a generative model to generate task instructions based on the key information. Thus, the text data of tasks can be converted into task instructions through the large model. Compared with generating task instructions according to templates or rules, it can meet the generation requirements of high-complexity task instructions; by merging similar task instructions among multiple task instructions, duplicate instructions and information redundancy can be reduced, and the execution efficiency of instructions and resource utilization rate can be improved; by performing priority sorting on each of the merged task instructions and performing execution scheduling on the sorted task instructions, the task instructions can be scheduled according to the priority, which not only ensures the priority execution of important task instructions, but also can adjust the execution order of task instructions in real time according to resources, improving flexibility and dynamics.

[0070] As Figure 2 shown, it shows a flowchart of the model-based task instruction management method provided in an embodiment of this application. The model-based task instruction management method can be applied to a computer device. The model-based task instruction management method may include:

[0071] Step 201: Obtain the text data of each task to be executed. The text data includes the description information, requirement information, supplementary information of the task, and the log information of historical tasks. The supplementary information is the detailed information and target information of the task input by the user.

[0072] Step 202: Use a large language model to extract the key information of each task from the text data. The key information includes the urgency parameter, target parameter, resource requirement parameter, and time limit parameter of the task.

[0073] Step 203: Use a generative model to generate corresponding task instructions based on the key information of each task.

[0074] For the implementation processes of Steps 201 - 203, please refer to the descriptions in Steps 101 - 103, which will not be elaborated here.

[0075] Step 204: After generating multiple task instructions, use the domain knowledge and decision rules in the rule engine to optimize each task instruction.

[0076] In this embodiment, after generating the task instructions, they can also be intelligently optimized. Among them, the intelligent optimization includes two parts. One part is the rule engine optimization, and the other part is the generation result evaluation and adjustment. Please refer to Step 205 for details.

[0077] The rule engine optimization refers to optimizing the task instructions by combining domain knowledge and decision rules. Here, the content and priority of the instructions can be adjusted through manually set rules to ensure the execution feasibility of the instructions in a complex environment.

[0078] The optimization rules mainly include the following two aspects: prioritize urgent tasks and schedule priority resources. That is, when allocating tasks, the content and priority of the instructions can be adjusted according to the urgency, task type, and resource availability. For example, for urgent tasks, the priority of the execution speed can be increased, and resource allocation can be increased; for low - priority tasks, resource allocation can be reduced.

[0079] Step 205: Use the ROUGE automatic evaluation algorithm to evaluate the quality of each generated task instruction, and adjust the generation strategy of the generative model for task instructions according to the evaluation results.

[0080] Quality assessment refers to using the ROUGE (Recall-Oriented Understudy for Gisting Evaluation) automated evaluation method to calculate the similarity between the generated instructions and the reference instructions, and adjusting the generation strategy according to the evaluation results. Among them, ROUGE focuses on recall rate and can measure whether the generated task instructions cover the key content. ROUGE mainly evaluates the similarity between the generated task instructions and the reference instructions through the following indicators:

[0081] ROUGE-N: Usually ROUGE-1 and ROUGE-2, which calculate the coincidence degree of n-grams. The higher the coincidence degree, the closer the generated instructions are to the target instructions. ROUGE-1: Calculate the coincidence degree of 1-gram (i.e., a single word). ROUGE-2: Calculate the coincidence degree of 2-grams (i.e., two consecutive words).

[0082] ROUGE-L: Calculate the similarity of task instructions based on the Longest Common Subsequence (LCS), and evaluate whether the generated task instructions are semantically similar to the reference instructions. Among them, LCS refers to the maximum length of the common element subsequence remaining after deleting some elements (while maintaining the relative order unchanged) in two sequences. The LCS similarity is a similarity measure calculated based on the length of the longest common subsequence. Assuming two given strings S1 and S2, LCS will find the longest common subsequence between them. For example, consider the following two strings:

[0083] S1 = "ABCBDAB"

[0084] S2 = "BDCABB"

[0085] The longest common subsequence is "BCAB" with a length of 4.

[0086] The calculation formula for LCS similarity is:

[0087] LCS Similarity = LCS length / min(length of S1, length of S2) (1)

[0088] Among them, LCS length is the longest common subsequence; length of S1 and length of S12 are the lengths of the two strings respectively, and min means taking the minimum value; LCS Similarity is the LCS similarity. The range of LCS similarity is from 0 to 1, and the larger the value, the more similar the two strings are. The LCS similarity of the above S1 and S2 is approximately 0.67.

[0089] Suppose there are two task instructions. Task instruction A is: mobilize 10 workers and 5 production machines to produce 1,000 product A and complete it within 3 hours. Task instruction B is: mobilize 10 workers and 5 production machines to produce 1,000 product A and it is expected to be completed within 3 hours.

[0090] The common subsequence of task instruction A and task instruction B is: mobilize, 10 workers, 5 production machines, produce 1,000 product A, complete within 3 hours. Among them, the longest common subsequence is: mobilize, 10 workers, 5 production machines, produce 1,000 product A, complete within 3 hours. The length of the LCS is 5 (that is, 5 words appear in both task instructions).

[0091] Calculate the LCS similarity: The length of task instruction A is 9 words, the length of task instruction B is 9 words, and the length of the LCS is 5 words. Then the LCS similarity is:

[0092] LCS Similarity = 5 / 9 ≈ 0.56 (2)

[0093] The above formula indicates that the LCS similarity between task instruction A and task instruction B is 56%, which is lower than the threshold of 0.8. This shows that the semantic consistency is insufficient, and the quality of the instructions generated by the generative model can be improved by adding details, optimizing the structure, and enhancing semantic understanding.

[0094] Step 206, merge similar task instructions among multiple task instructions.

[0095] In this embodiment, the redundancy in task instructions can be reduced and the task structure can be optimized through semantic similarity calculation and automatic merging mechanism.

[0096] The purpose of semantic similarity calculation is to identify which task instructions are similar in content, so as to merge these task instructions under appropriate circumstances, reduce redundancy, and optimize the task structure.

[0097] Specifically, merging similar task instructions among multiple task instructions includes: calculating the semantic similarity between every two task instructions; if the semantic similarity is greater than the similarity threshold, then merge the two task instructions.

[0098] When calculating semantic similarity, the task instructions and historical task instructions can be transformed into semantic vectors through the BERT model, and the cosine similarity is used to measure the angle between two semantic vectors. The closer the value of the cosine similarity is to 1, the higher the similarity of the two task instructions. The cosine similarity calculation formula is:

[0099] cosine_similarity(A, B) = (A·B) / (∥A∥∥B∥) (3)

[0100] where A and B are the semantic vectors of two task instructions, A·B represents the dot product of A and B, and ∥A∥∥B∥ represents the norms of A and B.

[0101] Then, compare the calculated cosine similarity with a set similarity threshold (0.8). If the cosine similarity of the two task instructions exceeds this similarity threshold, it is considered that the two task instructions are similar and can be merged.

[0102] The automatic merging mechanism refers to automatically extracting the key task actions and objectives in the task instructions and merging them into a simplified task instruction to ensure that the task description is clear and non-redundant. Here, we can use a generative model to merge the task instructions.

[0103] Specifically, similar task instructions can be input into the generative model. The generative model learns through training to semantically merge similar task instructions, ensuring that the generated task instructions are neither redundant nor lose semantic integrity, so that the generated task instructions are optimized and screened and merged into clear and concise task instructions.

[0104] For example, Task Instruction 1 is: Produce 1000 Product A, with 10 workers and 5 machines, to be completed within 3 hours; Task Instruction 2 is: Produce 500 Product B, with 5 workers and 3 machines, to be completed within 6 hours; the merged task instruction is: Mobilize 15 workers and 8 production machines to produce 1500 products and complete within 6 hours.

[0105] Step 207: Calculate the priority scores of the merged task instructions according to the weights corresponding to the key information, and perform priority sorting on the task instructions according to the priority scores.

[0106] In this embodiment, the priority sorting includes two parts. One part is priority analysis, and the other part is reinforcement learning analysis. See Step 208 for details.

[0107] Priority analysis refers to performing priority scoring on the task instructions according to the key information extracted from the instructions (including urgency parameters, task importance, time limit, resource requirements) to ensure that the task order meets the actual execution requirements.

[0108] Specifically, calculating the priority scores of each merged task instruction according to the weights corresponding to the key information may include: for each task instruction, determining the urgency score, importance score, resource requirement score, and time limit score according to the key information of the task instruction; obtaining the preset urgency weight, importance weight, resource requirement weight, and time limit weight; performing weighted calculations on the urgency score and urgency weight, importance score and importance weight, resource requirement score and resource requirement weight, and time limit score and time limit weight to obtain the priority score of the task instruction.

[0109] Assume that the urgency weight is 0.4, the importance weight is 0.3, the resource requirement weight is 0.2, and the time limit weight is 0.1. The following is a priority analysis of three task instructions:

[0110] (1) Urgency score

[0111] Task instruction A: Must be completed within 1 hour, and the urgency score is 1.0.

[0112] Task instruction B: To be completed within 4 hours, and the urgency score is 0.3.

[0113] Task instruction C: To be completed within 2 hours, and the urgency score is 0.6.

[0114] (2) Importance score

[0115] Task instruction A: Critical task, and the importance score is 1.0.

[0116] Task instruction B: Relatively important logistics task, and the importance score is 0.7.

[0117] Task instruction C: Inspection task, and the importance score is 0.5.

[0118] (3) Resource demand score

[0119] Task instruction A: The resource demand score is 0.8 (relatively more available resources).

[0120] Task instruction B: The resource demand score is 0.6.

[0121] Task instruction C: The resource demand score is 1.0 (the lowest resource requirement).

[0122] (4) Time constraint score

[0123] Task instruction A: To be completed within 1 hour, and the time constraint score is 1.0.

[0124] Task Instruction B: To be completed within 4 hours, with a time limit score of 0.4.

[0125] Task Instruction C: To be completed within 2 hours, with a time limit score of 0.7.

[0126] Based on the manually defined weights, calculate the task priority scores:

[0127] Task Instruction A: 0.4×1.0 + 0.3×1.0 + 0.2×0.8 + 0.1×1.0 = 0.96

[0128] Task Instruction B: 0.4×0.3 + 0.3×0.7 + 0.2×0.6 + 0.1×0.4 = 0.52

[0129] Task Instruction C: 0.4×0.6 + 0.3×0.5 + 0.2×1.0 + 0.1×0.7 = 0.66

[0130] Sort the priority scores to obtain the priority order as A, C, B.

[0131] Step 208, use the reinforcement learning algorithm to evaluate the long-term benefits of each sorted task instruction, and update the priority order of each task instruction according to the long-term benefits.

[0132] Reinforcement learning optimization refers to using the Q-Learning reinforcement learning algorithm to evaluate the long-term benefits of the priority order of task instructions, learn the optimal task splitting and scheduling strategies, and ensure that task instructions can be efficiently executed under limited resources.

[0133] Q-Learning evaluates the value of taking a certain action (a) in a certain state (s) by learning a Q-value function (i.e., action-value function). The Q-value function represents the maximum expected return that can be obtained by taking a certain action in a given state. The update formula of the Q-value function is as follows:

[0134] Q(s t ,a t ) = Q(s t ,a t ) + α[R t+1 + γa'maxQ(s t+1 ,a') - Q(s t ,a t )] (4)

[0135] Q(s t ,a t ) is the Q-value of taking action a t in the current state s t ; R t+1 is the reward for taking action a tThe immediate reward obtained later; γ is the discount factor, used to weigh the impact of current and future rewards; α is the learning rate, used to control the magnitude of Q-value update.

[0136] The states, actions, and rewards in the formula are described as follows:

[0137] (1) Define the state: The state can include the attributes of all current task instructions and the resource situation of the system. For example, the state can include information such as the urgency, importance, resource requirements, completion progress, and remaining resources of the task instructions. Each state is a specific combination of the task instruction set.

[0138] (2) Define the action: In the problem of task instruction sorting, the action is to select a task instruction and assign a priority to it (or select a position). For example, the system can choose to rank task instruction A before task instruction B, or arrange task instruction C as the first priority and task instruction D as the second priority, etc.

[0139] (3) Define the reward: The reward is the feedback signal after the execution of the task instruction, usually related to the efficiency, time, resource consumption, and the effect of priority selection of the task instruction execution. For example, when completing a high-priority task instruction, the system may give a higher reward, and a lower reward for delaying the task instruction. The reward can be dynamically adjusted according to the actual situation of task instruction completion:

[0140] A) If the task instruction is completed on time and meets the expected effect, the reward is high.

[0141] B) If the task instruction is delayed or resources are wasted, the reward is low.

[0142] After each task instruction sorting, the system updates the Q-value according to the actual execution result. By continuously updating the Q-value, the system can make better decisions based on the current state of the task, so that the sorting of the tasks is gradually optimized, and finally the optimal sorting strategy is obtained.

[0143] Specifically, using the reinforcement learning algorithm to evaluate the long-term benefits of each sorted task instruction can include: selecting the task instruction to be executed in the current state according to each sorted task instruction; calculating the Q-value after the execution of the selected task instruction using the Q-value function in the reinforcement learning algorithm. The Q-value function represents the maximum expected return that can be obtained by executing a task instruction in a given state; updating the priority sorting of each task instruction according to the goal of maximizing the Q-value.

[0144] In one example, the priority sorting of existing task instructions is as follows: Action 1: Task instruction A is executed first; Action 2: Task instruction C is executed first; Action 3: Task instruction B is executed first. Then, set the initial Q value record: Status: (Task A, Task C, Task B); Actions 1, 2, and 3 are all 0 points in the initial state. Assume the system selects Action 1 (task instruction A is executed first), and then the system will execute task instruction A and update the Q value table. If the task instruction is completed on time and resources are used efficiently, the system will receive a reward of +10. According to the Q value update formula, the system will update the corresponding Q value. By repeatedly executing this process, the system will gradually learn the optimal task sorting strategy, enabling the task instructions to be sorted most reasonably according to the priority and maximizing the long-term reward (i.e., using resources most effectively and completing tasks on time).

[0145] Step 209, perform execution scheduling on the sorted task instructions.

[0146] Execution scheduling is the diversion and execution of task instructions, and it automatically distributes task instructions to the corresponding execution queues according to the priority and resource situation of the task instructions.

[0147] Step 210, monitor the execution status and feedback information of each task instruction in real time.

[0148] The real-time monitoring and feedback mechanism is a key part of execution scheduling, which can help the system dynamically adjust the task execution strategy, optimize resource allocation, ensure that high-priority task instructions are executed in a timely manner, and low-priority tasks are postponed or scheduled appropriately.

[0149] Specifically, the monitoring system can monitor the execution status of each task instruction in real time, including task progress, resource consumption, remaining time, etc., obtain feedback information through data sources such as sensors, logs, and status reports, and judge whether the task instruction is executed as expected according to the feedback data. If problems such as delays and resource conflicts occur, the monitoring system will optimize the task scheduling by dynamically adjusting the priority of the task instruction, reallocating resources, or changing the execution queue.

[0150] Step 211, adjust the priority and resource allocation of each task instruction according to the execution status and feedback information to optimize the execution scheduling of each task instruction.

[0151] Dynamic adjustment of execution scheduling refers to the process of continuously improving task scheduling through real-time feedback and optimization algorithms. That is, when the task progress changes, the scheduling system automatically adjusts the task order. If a task instruction is completed ahead of schedule, the scheduling system can schedule other task instructions for execution; if a task instruction is completed late, the scheduling system can postpone it or re-evaluate its priority. When the resource status changes (such as equipment failure, personnel shortage, etc.), the scheduling system dynamically adjusts the task instruction allocation through an optimization algorithm to ensure that high-priority task instructions receive resource support. If a task instruction depends on the completion of other task instructions, the scheduling system will give priority to executing these dependent task instructions and start other task instructions in a timely manner after the dependent task instructions are completed.

[0152] For example, if task instruction A is completed ahead of schedule, the scheduling system can automatically promote task instruction C to a task instruction with priority for execution and adjust the resource allocation.

[0153] The task instruction management method provided in this embodiment is superior to traditional methods in terms of flexibility, automation, efficiency, real-time adjustment ability, etc. In particular, it can cope with complex and changeable task environments. Therefore, this task instruction management method can not only meet the urgent needs of efficient and intelligent instruction management in fields such as military, emergency response, and enterprise scheduling, but also make up for the deficiencies of traditional methods in multi-task scheduling, resource optimization, etc., and promote the intelligent development of instruction management systems in various industries.

[0154] In summary, the model-based task instruction management method provided in the embodiments of this application extracts key information from the text data of tasks by using a large language model, and then uses a generative model to generate task instructions based on the key information. Thus, it can convert the text data of tasks into task instructions through a large model. Compared with generating task instructions according to templates or rules, it can meet the generation requirements of high-complexity task instructions; by merging similar task instructions among multiple task instructions, it can reduce duplicate instructions and information redundancy, and improve the execution efficiency of instructions and resource utilization rate; by sorting the priorities of the merged task instructions and performing execution scheduling on the sorted task instructions, it can schedule task instructions according to priorities, not only ensuring the priority execution of important task instructions, but also being able to adjust the execution order of task instructions according to resources in real time, improving flexibility and dynamics.

[0155] As Figure 3 shown, it shows the structural block diagram of a model-based task instruction management device provided in an embodiment of this application. This model-based task instruction management device can be applied to a computer device. This model-based task instruction management device may include:

[0156] An acquisition module 310 for acquiring text data of each task to be executed, where the text data includes description information, requirement information, supplementary information, and log information of historical tasks, and the supplementary information is detailed information and target information of the task input by the user;

[0157] An extraction module 320 for extracting key information of each task from the text data by using a large language model, where the key information includes an urgency parameter, a target parameter, a resource requirement parameter, and a time limit parameter of the task;

[0158] A generation module 330 for generating corresponding task instructions based on the key information of each task by using a generative model;

[0159] A merging module 340 for merging similar task instructions among multiple task instructions;

[0160] A management module 350 for performing priority sorting on the merged task instructions and performing execution scheduling on the sorted task instructions.

[0161] In an optional embodiment, the management module 350 is further configured to:

[0162] Calculate a priority score of each merged task instruction according to the weight corresponding to the key information, and perform priority sorting on each task instruction according to the priority score;

[0163] Evaluate the long-term benefits of each sorted task instruction by using a reinforcement learning algorithm, and update the priority sorting of each task instruction according to the long-term benefits.

[0164] In an optional embodiment, the management module 350 is further configured to:

[0165] For each task instruction, determine an urgency score, an importance score, a resource requirement score, and a time limit score according to the key information of the task instruction;

[0166] Obtain preset urgency weights, importance weights, resource requirement weights, and time limit weights;

[0167] Perform weighted calculations on the urgency score and the urgency weight, the importance score and the importance weight, the resource requirement score and the resource requirement weight, and the time limit score and the time limit weight to obtain the priority score of the task instruction.

[0168] In an optional embodiment, the management module 350 is further configured to:

[0169] Select the task instruction to be executed in the current state according to each sorted task instruction;

[0170] Calculate the Q-value after the execution of the selected task instruction using the Q-value function in the reinforcement learning algorithm. The Q-value function represents the maximum expected return that can be obtained by executing a task instruction in a given state.

[0171] Update the priority ranking of each task instruction according to the goal of maximizing the Q-value.

[0172] In an optional embodiment, the generation module 330 is further configured to:

[0173] After generating multiple task instructions, optimize each task instruction using the domain knowledge and decision rules in the rule engine.

[0174] Use the ROUGE automatic evaluation algorithm to evaluate the quality of each generated task instruction, and adjust the generation strategy of the generative model for task instructions according to the evaluation results.

[0175] In an optional embodiment, the merging module 340 is further configured to:

[0176] Calculate the semantic similarity between every two task instructions.

[0177] If the semantic similarity is greater than the similarity threshold, merge the two task instructions.

[0178] In an optional embodiment, the management module 350 is further configured to:

[0179] Monitor the execution status and feedback information of each task instruction in real time.

[0180] Adjust the priority and resource allocation of each task instruction according to the execution status and feedback information to optimize the execution scheduling of each task instruction.

[0181] In summary, the model-based task instruction management device provided by the embodiments of the present application extracts key information from the text data of the task using the large language model, and then uses the generative model to generate task instructions based on the key information, so that the text data of the task can be converted into task instructions through the large model. Compared with generating task instructions according to templates or rules, it can meet the generation requirements of highly complex task instructions; by merging similar task instructions among multiple task instructions, duplicate instructions and information redundancy can be reduced, and the execution efficiency of instructions and resource utilization rate can be improved; by sorting the priorities of each merged task instruction and performing execution scheduling on the sorted task instructions, the task instructions can be scheduled according to the priorities, which not only ensures the priority execution of important task instructions, but also can adjust the execution order of task instructions in real time according to resources, improving flexibility and dynamicity.

[0182] An embodiment of the present application provides a computer-readable storage medium, in which at least one instruction is stored, and the at least one instruction is loaded and executed by a processor to implement the model-based task instruction management method as described above.

[0183] An embodiment of the present application provides a computer device, and the computer device includes any of the above model-based task instruction management devices.

[0184] It should be noted that when the model-based task instruction management device provided in the above embodiment performs model-based task instruction management, only the division of the above functional modules is used for illustration. In actual applications, the above functions can be assigned to different functional modules according to needs, that is, the internal structure of the model-based task instruction management device is divided into different functional modules to complete all or part of the functions described above. In addition, the model-based task instruction management device provided in the above embodiment and the embodiment of the model-based task instruction management method belong to the same concept. For the specific implementation process, please refer to the method embodiment and will not be elaborated here.

[0185] Those of ordinary skill in the art can understand that all or part of the steps to implement the above embodiment can be completed by hardware, or can be completed by instructing relevant hardware through a program. The program can be stored in a computer-readable storage medium. The storage medium mentioned above can be a read-only memory, a magnetic disk, an optical disc, etc.

[0186] The above does not intend to limit the embodiments of the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the embodiments of the present application shall be included in the protection scope of the embodiments of the present application.

Claims

1. A model-based task instruction management method, characterized in that, The method includes: Obtaining the text data of each task to be executed, where the text data includes the description information, requirement information, supplementary information, and log information of historical tasks of the task, and the supplementary information is the detailed information and target information of the task input by the user; Using a large language model to extract the key information of each task from the text data, where the key information includes the urgency parameter, target parameter, resource requirement parameter, and time limit parameter of the task; Using a generative model to generate corresponding task instructions based on the key information of each task; Merging similar task instructions among multiple task instructions; Performing priority sorting on the merged task instructions and performing execution scheduling on the sorted task instructions.

2. The method for managing task instructions based on a model according to claim 1, wherein The performing priority sorting on the merged task instructions includes: Calculating the priority scores of the merged task instructions according to the weights corresponding to the key information, and performing priority sorting on the task instructions according to the priority scores; Using a reinforcement learning algorithm to evaluate the long-term benefits of the sorted task instructions, and updating the priority sorting of the task instructions according to the long-term benefits.

3. The model-based task instruction management method according to claim 2, wherein The calculating the priority scores of the merged task instructions according to the weights corresponding to the key information includes: For each task instruction, determining the urgency score, importance score, resource requirement score, and time limit score according to the key information of the task instruction; Obtaining the preset urgency weight, importance weight, resource requirement weight, and time limit weight; Performing weighted calculations on the urgency score and the urgency weight, the importance score and the importance weight, the resource requirement score and the resource requirement weight, and the time limit score and the time limit weight to obtain the priority score of the task instruction.

4. The method for managing task instructions based on a model according to claim 2, wherein The using a reinforcement learning algorithm to evaluate the long-term benefits of the sorted task instructions includes: Selecting the task instructions to be executed in the current state according to the sorted task instructions; Using the Q-value function in the reinforcement learning algorithm to calculate the Q value after the selected task instruction is executed, where the Q-value function represents the maximum expected return that can be obtained by executing a task instruction in a given state; Updating the priority sorting of the task instructions according to the target that maximizes the Q value.

5. The model-based task instruction management method according to claim 1, wherein The method further includes: After generating multiple task instructions, optimizing each task instruction by using the domain knowledge and decision rules in the rule engine; Using the ROUGE automatic evaluation algorithm to evaluate the quality of each generated task instruction, and adjusting the generation strategy of the generative model for task instructions according to the evaluation results.

6. The model-based task instruction management method according to claim 1, wherein The merging similar task instructions among the multiple task instructions includes: Calculating the semantic similarity between every two task instructions; If the semantic similarity is greater than the similarity threshold, merging the two task instructions.

7. The model-based task instruction management method according to claim 1, wherein The method further includes: Real-time monitoring the execution status and feedback information of each task instruction; Adjusting the priorities and resource allocations of the task instructions according to the execution status and the feedback information to optimize the execution scheduling of the task instructions.

8. A model-based task instruction management device, characterized in that, The device includes: An acquisition module, configured to acquire text data of each task to be executed, where the text data includes description information, requirement information, supplementary information, and log information of historical tasks of the task, and the supplementary information is detailed information and target information of the task input by the user; An extraction module, configured to extract key information of each task from the text data by using a large language model, where the key information includes an urgency parameter, a target parameter, a resource requirement parameter, and a time limit parameter of the task; A generation module, configured to generate corresponding task instructions based on the key information of each task by using a generative model; A merging module, configured to merge similar task instructions among multiple task instructions; A management module, configured to perform priority sorting on the merged task instructions and perform execution scheduling on the sorted task instructions.

9. A computer-readable storage medium, characterized in that, At least one instruction is stored in the storage medium, and the at least one instruction is loaded and executed by a processor to implement the model-based task instruction management method according to any one of claims 1 to 7.

10. A computer device, characterized in that, The computer device includes: the model-based task instruction management device according to claim 8.

Images