Process planning and arrangement system and method based on intention recognition in dynamic environment, storage medium and equipment

Abstract

The invention belongs to the technical field of artificial intelligence, and discloses a process planning and arranging system and method based on intention recognition in a dynamic environment, a storage medium and equipment. The system comprises a sensing module, a decision module and an execution module. The sensing module is used for collecting multi-source data and extracting a feature vector set from the multi-source data; the decision module is used for identifying the intention of the user according to the feature vector set and planning an optimal task execution strategy conforming to the current intention of the user and the detection scene; and the execution module is used for executing each subtask in the task execution strategy and generating an equipment detection report. According to the method and the device, the problem of low accuracy of final report generation caused by inaccurate identification of user requirements due to only dependence on user interaction instructions in a traditional scheme is solved, and an efficient and accurate scheme is provided for generation of the equipment detection report.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a process planning and orchestration system, method, storage medium and device based on intent recognition in a dynamic environment. Background Technology

[0002] During live-line testing of equipment, generating accurate and comprehensive test reports is a crucial step in assessing equipment status and guiding maintenance decisions. Current report generation methods typically involve operators inputting interactive commands to generate a report. The system then uses preset keywords or a static rule base to recognize the intent of the commands, inferring the user's intention and calling upon the appropriate report generation template to output a structured test report.

[0003] However, while current report generation methods can automate the generation of inspection reports to some extent, they still have the following shortcomings: The intent recognition process of existing technologies is limited to user-provided interactive data, failing to effectively correlate and integrate dynamically changing environmental information at the inspection site and changes in key physical quantity data collected in real time by multi-source sensing units. In complex inspection scenarios, such as when equipment experiences partial discharge anomalies or when ambient temperature and humidity fluctuate drastically, the user's true intent is often closely related to the specific abnormal state of the equipment or environmental conditions. However, because the system's intent recognition lacks perception and understanding of this dynamic information, it is difficult to accurately capture and analyze the user's deeper needs based solely on isolated interactive commands, leading to inaccurate recognition and reduced report generation accuracy. Furthermore, requiring users to issue multiple, repetitive commands to obtain the expected report also negatively impacts the user experience. Summary of the Invention

[0004] Therefore, embodiments of this application provide a process planning and orchestration system, method, storage medium, and device based on intent recognition in a dynamic environment, which can improve the accuracy and efficiency of report generation.

[0005] Firstly, this application provides a process planning and orchestration system based on intent recognition in a dynamic environment.

[0006] This application is achieved through the following technical solution: A process planning and orchestration system based on intent recognition in a dynamic environment includes a perception module, a decision-making module, and an execution module; The perception module includes a data acquisition unit and a feature extraction unit; The data acquisition unit is used to acquire multi-source data in real time, including equipment detection data of the electrical equipment, user interaction data, and environmental detection data. The feature extraction unit is used to receive the multi-source data, extract speech feature vectors and behavioral feature vectors from the user interaction data through hierarchical feature encoding, extract scene perception features from the environment detection data and device detection data; and determine the detection scene corresponding to the current state according to the multi-source data and preset matching rules, and adjust the weights of the speech feature vectors, behavioral feature vectors and scene perception feature vectors according to the detection scene through a cross-modal attention weight dynamic allocation algorithm, and fuse them to form a feature vector set. The decision-making module includes an intent recognition unit and a task planning unit; The intent recognition unit is used to perform intent reasoning through a multi-task learning model and a reinforcement learning model based on the feature vector set output by the feature extraction unit, and obtain the intent recognition result. The task planning unit is used to call the NSGA-Ⅲ multi-objective optimization algorithm to optimize and solve the report generation task corresponding to the detection scenario based on the intent recognition result and the constraints of the detection scenario, and output the optimal task execution strategy. The execution module includes an interactive data parsing unit, a data population unit, a data validation unit, and a report generation unit; An interactive data parsing unit is used to receive user interaction data from the data acquisition unit and output structured text instructions. The data population unit is used to extract a set of structured fields that meet the requirements of the report template fields from the environmental detection data and the equipment detection data; A data verification unit is used to verify the set of structured fields; The report generation unit is used to receive the structured text instructions and the validated set of structured fields, fill them into the preset report module, and generate a device test report.

[0007] In a preferred embodiment of this application, the environmental detection data may be further configured to include at least temperature data, humidity data, and electromagnetic field strength data, and the device detection data may include at least partial discharge quantity, discharge frequency, ultrasonic signal, contact ultrasonic data, ultra-high frequency signal, operating current, and surface temperature.

[0008] In a preferred example of this application, the feature extraction unit can be further configured to: The user-input speech data is pre-emphasized, framed, and windowed. MFCC features are extracted from the processed speech data, and the MFCC features are input into a convolutional neural network to obtain a speech feature vector. The user's input trajectory is interpolated, and an LSTM network is used to extract behavioral feature vectors. Environmental monitoring data and equipment monitoring data are standardized to obtain standardized perception data. The perception data is then input into a fully connected network to obtain scene perception feature vectors.

[0009] In a preferred embodiment of this application, the intent recognition unit is further configured to: The feature vector set is received and input into the multi-task learning module. The main task classification head outputs the preliminary intent classification probability, and the sub-task classification head outputs the device status prediction result. The preliminary intent classification probability and device state prediction results are concatenated into a reinforcement learning state. The reinforcement learning state is then input into the reinforcement learning model. After processing through the state input layer, intermediate layer, and output layer, 10 user intents and their corresponding credibility scores are obtained. The user intent with the highest credibility score is taken as the final intent recognition result.

[0010] In a preferred example of this application, it can be further set as follows: The perception module further includes a weight adjustment unit, used for: Based on at least one dimension of the noise level, data integrity, and data stability of the data source corresponding to each feature vector, the confidence of the speech feature vector, behavior feature vector, and scene-aware feature vector is evaluated to obtain three data confidence scores. The weights of the speech feature vector, behavior feature vector, and scene-aware feature vector are then adjusted based on the data confidence scores.

[0011] In a preferred embodiment of this application, the execution module may further be configured to include a feedback unit, used for: The execution status of the interactive data parsing unit, data filling unit, data verification unit, and report generation unit is collected in real time, and the execution status is fed back to the decision module.

[0012] In a preferred embodiment of this application, the decision module may further include a strategy optimization unit, used for: The execution status feedback from the execution module is obtained, and combined with the device status prediction result output by the intent recognition unit, a strategy optimization instruction is generated based on the execution status and the device status prediction result through an online fine-tuning algorithm. The strategy optimization instruction is used to coordinately adjust the model parameters of the intent recognition unit and the parameters of the task planning unit.

[0013] Secondly, this application provides a process planning and orchestration method based on intent recognition in a dynamic environment.

[0014] This application is achieved through the following technical solution: A process planning and orchestration method based on intent recognition in a dynamic environment, executed using the process planning and orchestration system based on intent recognition in a dynamic environment described in the first aspect above, includes: Real-time acquisition of multi-source data, including equipment detection data of electrical equipment, user interaction data, and environmental detection data; The system receives the multi-source data, extracts voice feature vectors and behavior feature vectors from the interaction data through hierarchical feature encoding, and extracts scene perception feature vectors from the environmental detection data and device detection data. Based on the multi-source data and preset matching rules, the detection scenario corresponding to the current state is determined. Through a cross-modal attention weight dynamic allocation algorithm, the weights of the speech feature vector, behavior feature vector and scene perception feature vector are adjusted according to the detection scenario and fused to form a feature vector set. Based on the set of feature vectors, intent reasoning is performed through a multi-task learning model and a reinforcement learning model to obtain the intent recognition result. Based on the intent recognition results and the constraints of the detection scenario, the NSGA-Ⅲ multi-objective optimization algorithm is invoked to optimize and solve the report generation task corresponding to the detection scenario, and the optimal task execution strategy is output. Structured text instructions are parsed from the user interaction data, and environmental detection data and equipment detection data are transformed into a set of structured fields that meet the requirements of the report template fields; Based on the structured text instructions and the validated set of structured fields, a device testing report is generated by filling the preset report module.

[0015] Thirdly, this application provides a computer-readable storage medium.

[0016] This application is achieved through the following technical solution: A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described process planning and orchestration method based on intent recognition in any of the dynamic environments.

[0017] Fourthly, this application is achieved through the following technical solution: A computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of any of the above-described dynamic environment-based process planning and orchestration methods.

[0018] In summary, compared with the prior art, the beneficial effects of the technical solution provided in this application include at least the following: This application constructs a "scene context" by introducing sensor and environmental data. This data provides precise semantic anchors and scene boundaries for ambiguous commands, dissolving ambiguity and improving the accuracy of user intent recognition. By utilizing a hierarchical feature attention mechanism and reinforcement learning, it overcomes the problems of isolated understanding and lack of physical world anchoring caused by traditional methods relying solely on interaction data, significantly improving the accuracy of intent recognition and scene fit. Secondly, compared to traditional single-agent fixed-schedule tasks, this solution constructs a "perception-decision-execution" multi-agent collaborative architecture, achieving intelligent dynamic adaptation throughout the entire process and improving report generation efficiency. Finally, this application also monitors the execution status of the execution module and dynamically feeds back the execution status to the decision module, optimizing the task execution strategy. In summary, this solution solves the problem of inaccurate user demand recognition and low report accuracy caused by relying solely on user interaction commands in traditional solutions. It provides efficient, accurate, and highly adaptive technical support for the intelligent generation of equipment inspection reports. Attached Figure Description

[0019] Figure 1 This is a flowchart illustrating a process planning and orchestration method based on intent recognition in a dynamic environment, as provided in an embodiment of this application. Detailed Implementation

[0020] This specific embodiment is merely an explanation of this application and is not intended to limit it. After reading this specification, those skilled in the art can make modifications to this embodiment without contributing any inventive step, but such modifications are protected by patent law as long as they fall within the scope of the claims of this application.

[0021] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0022] Furthermore, the term "and / or" in this application is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this application, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0023] In this application, the terms "first," "second," etc., are used to distinguish identical or similar items with essentially the same function. It should be understood that there is no logical or temporal dependency between "first," "second," and "nth," nor are there any restrictions on quantity or execution order.

[0024] In the embodiments of this application, the terms "exemplary" or "for example" are used to indicate that something is an example, illustration, or description. Any embodiment or design that is described as "exemplary" or "for example" in the embodiments of this application should not be construed as being more preferred or advantageous than other embodiments or design. Specifically, the use of the terms "exemplary" or "for example" is intended to present the relevant concepts in a specific manner.

[0025] The embodiments of this application will now be described in further detail with reference to the accompanying drawings.

[0026] The first exemplary embodiment of this application provides a process planning and orchestration system based on intent recognition in a dynamic environment, including a perception module, a decision module, and an execution module; The perception module includes a data acquisition unit and a feature extraction unit. The data acquisition unit is used to collect multi-source data in real time, including equipment detection data of live equipment, user interaction data, and environmental monitoring data.

[0027] In practical implementation, the data acquisition unit connects to various devices through specific interfaces to achieve real-time acquisition of multi-source data. For example, it connects to user terminals (such as detection tablets and smartphones) to collect interactive data in the form of voice input by the user. Voice data includes the user's input commands; behavioral data includes the user's click or swipe input trajectories on the interactive interface. Environmental detection data includes multiple environmental parameters of the detection area, and device detection data includes the device parameters of the device under test detected by multiple sensors. It should be noted that user interaction data can also be text data input by the user through the interactive interface.

[0028] The feature extraction unit receives multi-source data, extracts speech feature vectors and behavioral feature vectors from interaction data through hierarchical feature encoding, and extracts scene perception feature vectors from environmental detection data and device detection data. Based on the multi-source data and preset matching rules, it determines the detection scene corresponding to the current state, and adjusts the weights of speech feature vectors, behavioral feature vectors and scene perception feature vectors according to the detection scene through a cross-modal attention weight dynamic allocation algorithm, and fuses them to form a feature vector set. The decision-making module includes an intent recognition unit and a task planning unit; The intent recognition unit is used to perform intent reasoning through a multi-task learning model and a reinforcement learning model based on the feature vector set output by the feature extraction unit, and obtain the intent recognition result. The intent recognition result includes at least the user intent and the corresponding credibility score. The task planning unit is used to optimize the report generation task corresponding to the detection scenario by calling the NSGA-Ⅲ multi-objective optimization algorithm based on the intent recognition results and the constraints of the detection scenario, and output the optimal task execution strategy. The execution module includes an interactive data parsing unit, a data population unit, a data validation unit, and a report generation unit; The interactive data parsing unit is used to receive user interaction data from the data acquisition unit and output structured text instructions. The data population unit is used to extract a set of structured fields that meet the requirements of the report template from environmental monitoring data and equipment monitoring data; The data validation unit is used to validate the structured field set; The report generation unit receives structured text instructions and a validated set of structured fields, fills them into a pre-set report module, and generates a device test report.

[0029] The environmental monitoring data collected by the data acquisition unit includes at least temperature data, humidity data, and electromagnetic field strength data, while the equipment monitoring data includes at least partial discharge quantity, discharge frequency, ultrasonic signal, contact ultrasonic data, ultra-high frequency signal, operating current, and surface temperature.

[0030] Specifically, the data acquisition unit connects to the temperature and humidity sensors of the environmental monitoring equipment to acquire real-time temperature and humidity data of the detection area. It also connects to an electromagnetic field strength sensor to acquire real-time electromagnetic field strength data of the detection area. Furthermore, it connects to a TEV (Transient Earth Voltage) partial discharge detection sensor to collect the amount and number of partial discharges. For example, it connects to a non-contact ultrasonic detection sensor to acquire ultrasonic signals; to a contact ultrasonic detection sensor to acquire contact ultrasonic data; to a UHF detection sensor to acquire UHF signals; to a current detection sensor to acquire the equipment's operating current; and to an infrared detection sensor to acquire the equipment's surface temperature. The data acquisition cycle is uniformly set to 100ms to ensure real-time data transmission.

[0031] In some preferred embodiments, after acquiring the raw data, an outlier detection algorithm is used to verify the acquired raw data. When outlier data is detected, a data re-acquisition mechanism is triggered, with a maximum of 3 re-acquisitions. If the data is still outlier after re-acquisition, the data outlier level is marked, which is divided into mild, moderate, and severe, and a hypothetical cause of the outlier is added. For example, if the current data exceeds the rated current of the device by ±20%, it is marked as mild outlier, and the hypothetical cause of the outlier is poor sensor contact.

[0032] In some embodiments, the feature extraction unit employs a hierarchical feature attention mechanism to perform hierarchical encoding and feature extraction on the collected multi-source data, specifically: The user-input speech data is pre-emphasized, framed, and windowed. MFCC features are extracted from the processed speech data and then input into a convolutional neural network to obtain a speech feature vector. The user's input trajectory is interpolated, and an LSTM network is used to extract behavioral feature vectors. Environmental monitoring data and equipment monitoring data are standardized to obtain standardized perception data. The perception data is then input into a fully connected network to obtain scene perception feature vectors.

[0033] In the specific implementation process of this embodiment, the speech data is first pre-emphasized with a pre-emphasis coefficient of 0.97, then framed, with a frame length of 20ms and a frame shift of 10ms. The framed data is then windowed to extract MFCC features, which are then input into a CNN network to further extract speech feature vectors. The CNN network has a convolution kernel size of 3×3, a number of 64, and a stride of 1.

[0034] The behavioral data was interpolated, and the behavioral feature vector was extracted using an LSTM network. The number of hidden layer units in the LSTM network was set to 128, the number of layers was set to 2, and the dropout probability was set to 0.2.

[0035] Environmental and equipment detection data were standardized and mapped to the [0,1] interval. Scene-aware feature vectors were extracted using a fully connected network. The number of hidden layer nodes in the fully connected network was set to 64, and ReLU was used as the activation function.

[0036] Subsequently, the weights of the speech feature vector, behavior feature vector, and scene perception feature vector are adjusted according to the detection scenario using a cross-modal attention weight dynamic allocation algorithm, and then fused to form a feature vector set.

[0037] In some implementations, if the user input interaction data is in text format, the text data is encoded to extract semantic feature vectors. This can be done by inputting the text data into a pre-trained BERT model, encoding the text data, and extracting semantic feature vectors. In this case, the hidden layer dimension of the BERT model is set to 768, the training batch size is set to 32, and the learning rate is set to 2e. -5 .

[0038] In some embodiments, the intent recognition unit is specifically used for: Receive a set of feature vectors and input the set of feature vectors into the multi-task learning module. The main task classification head outputs the preliminary intent classification probability, and the sub-task classification head outputs the device status prediction result. The task classification head and device state prediction results are concatenated into a reinforcement learning state. The reinforcement learning state is then input into the reinforcement learning model. After processing through the state input layer, intermediate layer, and output layer, 10 user intentions and their corresponding credibility scores are obtained. The user intention with the highest credibility score is taken as the final intention recognition result.

[0039] In the specific implementation of this embodiment, the main task of the multi-task learning module is intent classification, using cross-entropy loss as the loss function. The sub-task is device status prediction, predicting the device's operating status, such as normal or abnormal, using MSE as the loss function, and the sub-task weight is set to 0.2. Reinforcement learning uses the DQN algorithm, where the reward function is: R = αR_acc + βR_env + γR_time, where R_acc is the intention recognition accuracy reward (1 point for correct recognition, -0.5 points for incorrect recognition), and α is 0.5; R_env is the environment adaptability reward (1 point for matching the intention with the device status and sensor data, -0.3 points for mismatch), and β is 0.3; R_time is the response speed reward (0.5 points for response time less than 50ms, 0.2 points for response time between 50ms and 100ms, 0 points for response time greater than 100ms), and γ is 0.2. The learning rate is set to 1e. -4 The experience replay buffer size is 10,000, and the target network update cycle is 100 steps.

[0040] In some implementations, the intent recognition unit is also used to construct an evolutionary graph of intent-environment-historical interaction, using the GNN algorithm to mine the evolutionary pattern of intent, and output the probability distribution of user intent within a preset time period based on past interaction data and real-time sensor data. When the intent probability is greater than a preset threshold, a pre-preparation instruction is sent to the execution module in advance.

[0041] In some implementations, the sensing module further includes a weight adjustment unit for: Based on at least one dimension of the noise level, data integrity, and data stability of the data source corresponding to each feature vector, the confidence of the speech feature vector, behavior feature vector, and scene-aware feature vector is evaluated, and three data confidence scores are obtained respectively. The weights of the speech feature vector, behavior feature vector, and scene-aware feature vector are adjusted based on the data confidence scores.

[0042] In the specific implementation of this embodiment, the perception module pre-sets the basic weights of each feature vector under different detection scenarios, forming a basic weight matrix W_base, for example: In the detection scenario, which is a routine inspection scenario with scenario ID S1, the weight of the voice feature vector is 0.3, the weight of the behavior feature vector is 0.3, and the weight of the scenario perception feature vector is 0.4. In the detection scenario of partial discharge anomaly, the scenario ID is S2, the weight of the speech feature vector is 0.15, the weight of the behavior feature vector is 0.15, and the weight of the scenario perception feature vector is 0.7. In the detection scenario of device overheating, the scenario ID is S3, the weight of the voice feature vector is 0.15, the weight of the behavior feature vector is 0.15, and the weight of the scenario perception feature vector is 0.7. In the detection scenario of strong electromagnetic interference, the scenario ID is S4, the weight of the speech feature vector is 0.4, the weight of the behavior feature vector is 0.1, and the weight of the scenario perception feature vector is 0.5. In the detection scenario of emergency repair, the scenario ID is S5, the weight of the voice feature vector is 0.2, the weight of the behavior feature vector is 0.2, and the weight of the scene perception feature vector is 0.6.

[0043] Furthermore, to avoid the static limitations of basic weights, this application also introduces a correction factor based on data confidence, which dynamically adjusts the weights according to the real-time quality of each modality of data, ensuring that data with high confidence receive higher weights.

[0044] In some preferred embodiments, the sensing module further includes a weight adjustment unit for: Based on at least one dimension of the noise level, data integrity, and data stability of the data source corresponding to each feature vector, the confidence of the speech feature vector, behavior feature vector, and scene-aware feature vector is evaluated, and three data confidence scores are obtained respectively. The weights of the speech feature vector, behavior feature vector, and scene-aware feature vector are adjusted based on the data confidence scores.

[0045] In practical implementation, the confidence level of the speech data is determined based on the signal-to-noise ratio (SNR). For example, when the SNR exceeds 30 dB, the weight of the speech feature vector is adjusted to 1.2 of the original base weight; when the SNR is between 20 and 30 dB, the weight is 1.0; and when the SNR is below 20 dB, the weight is adjusted to 0.8. In another implementation, the weight of the speech feature vector can also be adjusted based on the speech recognition accuracy.

[0046] The confidence level of behavioral data is determined based on the completeness of the behavioral trajectory. For example, if there are no breakpoints in the behavioral trajectory, the weight of the behavioral feature vector is adjusted to 1.2. If the number of breakpoints is less than the preset breakpoint threshold, the weight of the behavioral feature vector is adjusted to 1.0. If the number of breakpoints is greater than the preset breakpoint threshold, the weight of the behavioral feature vector is adjusted to 0.8.

[0047] The confidence level of the scene perception feature vector is assessed based on the data stability of environmental monitoring data and equipment monitoring data. The data stability is determined based on the data standard deviation. For example, when the data standard deviation is below 5%, the weight of the scene perception feature vector is adjusted to 1.2; when the data standard deviation is between 5% and 10%, the weight of the scene perception feature vector is adjusted to 1.0; and when the data standard deviation is greater than 10%, the weight of the scene perception feature vector is adjusted to 0.8.

[0048] Finally, the weights of the data for each modality are: W_i=(W_base_i×α_i) / Σ(W_base_j×α_j), where i=1,2,3; j=1,2,3.

[0049] The perception module collects environmental and sensor data to determine the detection scenario. Based on the detection scenario, it matches basic weights and calculates the confidence scores of the three modalities. Based on the confidence scores, it adjusts the weights of the three modalities and substitutes the adjusted weights into the correction formula to obtain the final weights of each modality. The final weights are then embedded into a cross-modal attention mechanism, which allocates attention resources according to the weights during feature fusion. The higher the weight, the greater the influence of the feature in the fusion process.

[0050] In some preferred embodiments, the execution module further includes a feedback unit for: The execution status of the voice command parsing unit, data filling unit, data verification unit, and report generation unit is collected in real time, and the execution status is fed back to the decision module.

[0051] In some preferred embodiments, the decision-making module further includes a strategy optimization unit, used for: The execution status feedback from the execution module is obtained, and combined with the device status prediction result output by the intent recognition unit, a strategy optimization instruction is generated based on the execution status and device status prediction result through an online fine-tuning algorithm. The strategy optimization instruction is used to coordinately adjust the model parameters of the intent recognition unit and the task execution strategy of the task planning unit.

[0052] Specifically, the online fine-tuning algorithm is the PPO algorithm, and the learning rate is set to 5e. -5 The batch size is set to 64. When the information filling error rate of one of the testing devices exceeds 1%, the process parameters are automatically adjusted, such as adding a secondary verification step for the data.

[0053] In the specific implementation of this application, the execution module is responsible for performing the specific report generation operation. The functional module is divided into a task execution sub-module, which includes an interactive data parsing unit, a data filling unit, a data validation unit, and a report generation unit, specifically: The interactive data parsing unit employs a CTC-Attention hybrid model, with a CNN and BiLSTM encoder and an Attention mechanism decoder. Its vocabulary includes 5000 power detection technical terms and common instruction templates, achieving high parsing accuracy. The input data for the interactive data parsing unit consists of raw speech data collected by the perception module in common audio file formats, such as WAV. The auxiliary constraint information is the detection scene. The output data is structured text instructions, including the target device, instruction type, and operation parameters.

[0054] For example, the system first receives a fill instruction from the decision module, determining the report template and fields to be filled. It then queries the database for inherent device parameters (model, ratings, etc.) and receives environmental and device detection data from the sensing module. The system matches the names of the fields to be filled with the database fields, environmental and device detection data using a rule-based and similarity-based algorithm; if the similarity exceeds a preset threshold, the data is automatically associated. Successfully mapped data is then filled into a structured field set. Missing data is marked as needing to be supplemented.

[0055] The data verification unit loads targeted verification rules from the rule base based on the current detection scenario and verifies the structured field set populated in the report to ensure data accuracy and consistency. Simultaneously, the data verification unit adds cross-sensor data consistency verification. For example, it compares TEV partial discharge data with UHF data; when the difference reaches 30%, a manual review request is triggered. For each field, a pass or fail verification result is given. For fields that fail, intelligent modification suggestions are provided based on historical data. After the verification results are displayed, the current verification result is entered into the historical database.

[0056] The modules in the aforementioned dynamic environment-based intent-based process planning and orchestration system can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can invoke and execute the corresponding operations of each module.

[0057] Another embodiment of this application also provides a process planning and orchestration method based on intent recognition in a dynamic environment, used to execute the above-described method, such as... Figure 1 As shown, the method includes: S1: Real-time acquisition of multi-source data, including equipment detection data of live equipment, user interaction data, and environmental detection data; S2: Receive multi-source data, extract speech feature vectors and behavioral feature vectors from the interaction data through hierarchical feature encoding, and extract scene perception feature vectors from environmental detection data and device detection data; S3: Based on multi-source data and preset matching rules, determine the detection scene corresponding to the current state. Through a cross-modal attention weight dynamic allocation algorithm, adjust the weights of speech feature vectors, behavioral feature vectors and scene perception feature vectors according to the detection scene, and fuse them to form a feature vector set. S4: Based on the feature vector set, intention reasoning is performed through a multi-task learning model and a reinforcement learning model to obtain the intention recognition result; S5: Based on the intent recognition results and the constraints of the detection scenario, the NSGA-Ⅲ multi-objective optimization algorithm is invoked to optimize and solve the report generation task corresponding to the detection scenario, and the optimal task execution strategy is output. S6: Parse structured text instructions from user interaction data and transform environmental monitoring data and equipment monitoring data into a set of structured fields that meet the requirements of the report template fields; S7: Based on structured text instructions and a validated set of structured fields, populate the preset report module to generate a device test report.

[0058] The specific limitations of the process planning and orchestration method based on intent recognition in a dynamic environment provided in this embodiment can be found in the embodiment of a process planning and orchestration system based on intent recognition in a dynamic environment described above, and will not be repeated here.

[0059] This application provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the computer program implements the steps of a process planning and orchestration method based on intent recognition in a dynamic environment, as described in any of the above embodiments. The computer-readable storage medium refers to a data storage medium, which may include, but is not limited to, floppy disks, optical disks, hard disks, flash memory, USB flash drives, and / or Memory Sticks. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices.

[0060] The working process, working details, and technical effects of the computer-readable storage medium provided in this embodiment can be found in the embodiment of a process planning and orchestration method based on intent recognition in a dynamic environment described above, and will not be repeated here.

[0061] This application provides a computer device that may include a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system, computer programs, and a database. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage medium. The network interface communicates with external terminals via a network connection. When the computer program is executed by the processor, it causes the processor to perform steps of a process planning and orchestration method based on intent recognition in a dynamic environment, as described in any of the above embodiments.

[0062] The working process, working details, and technical effects of the computer device provided in this embodiment can be found in the embodiment of a process planning and orchestration method based on intent recognition in a dynamic environment described above, and will not be repeated here.

[0063] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0064] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0065] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the above-described division of functional units and modules is used as an example. In practical applications, the above functions can be assigned to different functional units and modules as needed, that is, the internal structure of the system described in this application can be divided into different functional units or modules to complete all or part of the functions described above.

Claims

1. A process planning and orchestration system based on intent recognition in a dynamic environment, characterized in that, It includes a perception module, a decision-making module, and an execution module; The perception module includes a data acquisition unit and a feature extraction unit; The data acquisition unit is used to acquire multi-source data in real time, including equipment detection data of the electrical equipment, user interaction data, and environmental detection data. The feature extraction unit is used to receive the multi-source data, extract speech feature vectors and behavioral feature vectors from the user interaction data through hierarchical feature encoding, extract scene perception features from the environment detection data and device detection data; and determine the detection scene corresponding to the current state according to the multi-source data and preset matching rules, and adjust the weights of the speech feature vectors, behavioral feature vectors and scene perception feature vectors according to the detection scene through a cross-modal attention weight dynamic allocation algorithm, and fuse them to form a feature vector set. The decision-making module includes an intent recognition unit and a task planning unit; The intent recognition unit is used to perform intent reasoning through a multi-task learning model and a reinforcement learning model based on the feature vector set output by the feature extraction unit, and obtain the intent recognition result. The task planning unit is used to call the NSGA-Ⅲ multi-objective optimization algorithm to optimize and solve the report generation task corresponding to the detection scenario based on the intent recognition result and the constraints of the detection scenario, and output the optimal task execution strategy. The execution module includes an interactive data parsing unit, a data population unit, a data validation unit, and a report generation unit; An interactive data parsing unit is used to receive user interaction data from the data acquisition unit and output structured text instructions. The data population unit is used to extract a set of structured fields that meet the requirements of the report template fields from the environmental detection data and the equipment detection data; A data verification unit is used to verify the set of structured fields; The report generation unit is used to receive the structured text instructions and the validated set of structured fields, fill them into the preset report module, and generate a device test report.

2. The process planning and orchestration system based on intent recognition in a dynamic environment according to claim 1, characterized in that, The environmental monitoring data includes at least temperature data, humidity data, and electromagnetic field strength data, and the equipment monitoring data includes at least partial discharge quantity, discharge frequency, ultrasonic signal, contact ultrasonic data, ultra-high frequency signal, operating current, and surface temperature.

3. The process planning and orchestration system based on intent recognition in a dynamic environment according to claim 1, characterized in that, The feature extraction unit is specifically used for: The user-input speech data is pre-emphasized, framed, and windowed. MFCC features are extracted from the processed speech data, and the MFCC features are input into a convolutional neural network to obtain a speech feature vector. The user's input trajectory is interpolated, and an LSTM network is used to extract behavioral feature vectors. Environmental monitoring data and equipment monitoring data are standardized to obtain standardized perception data. The perception data is then input into a fully connected network to obtain scene perception feature vectors.

4. The process planning and orchestration system based on intent recognition in a dynamic environment according to claim 1, characterized in that, The intent recognition unit is specifically used for: The feature vector set is received and input into the multi-task learning module. The main task classification head outputs the preliminary intent classification probability, and the sub-task classification head outputs the device status prediction result. The preliminary intent classification probability and device state prediction results are concatenated into a reinforcement learning state. The reinforcement learning state is then input into the reinforcement learning model. After processing through the state input layer, intermediate layer, and output layer, 10 user intents and their corresponding credibility scores are obtained. The user intent with the highest credibility score is taken as the final intent recognition result.

5. The process planning and orchestration system based on intent recognition in a dynamic environment according to claim 1, characterized in that, The perception module further includes a weight adjustment unit, used for: Based on at least one dimension of the noise level, data integrity, and data stability of the data source corresponding to each feature vector, the confidence of the speech feature vector, behavior feature vector, and scene-aware feature vector is evaluated to obtain three data confidence scores. The weights of the speech feature vector, behavior feature vector, and scene-aware feature vector are then adjusted based on the data confidence scores.

6. The process planning and orchestration system based on intent recognition in a dynamic environment according to claim 4, characterized in that, The execution module further includes a feedback unit, used for: The execution status of the interactive data parsing unit, data filling unit, data verification unit, and report generation unit is collected in real time, and the execution status is fed back to the decision module.

7. The process planning and orchestration system based on intent recognition in a dynamic environment according to claim 6, characterized in that, The decision-making module further includes a strategy optimization unit, used for: The execution status feedback from the execution module is obtained, and combined with the device status prediction result output by the intent recognition unit, a strategy optimization instruction is generated based on the execution status and the device status prediction result through an online fine-tuning algorithm. The strategy optimization instruction is used to coordinately adjust the model parameters of the intent recognition unit and the parameters of the task planning unit.

8. A process planning and orchestration method based on intent recognition in a dynamic environment, characterized in that, include: Real-time acquisition of multi-source data, including equipment detection data of electrical equipment, user interaction data, and environmental detection data; The system receives the multi-source data, extracts voice feature vectors and behavior feature vectors from the interaction data through hierarchical feature encoding, and extracts scene perception feature vectors from the environmental detection data and device detection data. Based on the multi-source data and preset matching rules, the detection scenario corresponding to the current state is determined. Through a cross-modal attention weight dynamic allocation algorithm, the weights of the speech feature vector, behavior feature vector and scene perception feature vector are adjusted according to the detection scenario and fused to form a feature vector set. Based on the set of feature vectors, intent reasoning is performed through a multi-task learning model and a reinforcement learning model to obtain the intent recognition result. Based on the intent recognition results and the constraints of the detection scenario, the NSGA-Ⅲ multi-objective optimization algorithm is invoked to optimize and solve the report generation task corresponding to the detection scenario, and the optimal task execution strategy is output. Structured text instructions are parsed from the user interaction data, and environmental detection data and equipment detection data are transformed into a set of structured fields that meet the requirements of the report template fields; Based on the structured text instructions and the validated set of structured fields, a device testing report is generated by filling the preset report module.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method of claim 8.

10. A computer device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method of claim 8.

Images