Predictive maintenance model tuning method and device based on large language model, equipment and storage medium

Abstract

The application discloses a predictive maintenance model optimization method and device based on a large language model, equipment and a storage medium, relates to the technical field of model optimization and artificial intelligence, and comprises the following steps: obtaining diagnosis information of a target optimizer and a predictive maintenance model; inputting the diagnosis information of the predictive maintenance model into a large language model for analysis to obtain an optimization signal; determining a correction score of a historical candidate hyperparameter based on the target optimizer and the optimization signal; and performing evaluation optimization on the predictive maintenance model based on the correction score of the historical candidate hyperparameter and the target optimizer, and outputting a target predictive maintenance model. Through semantic analysis of the diagnosis information of the predictive maintenance model in the large language model, semantic information can be combined in the optimization process, so that the efficiency of model optimization is improved.

Description

Technical Field

[0001] This application relates to the fields of model tuning and artificial intelligence technology, and in particular to predictive maintenance model tuning methods, apparatus, devices and storage media based on large language models. Background Technology

[0002] With the rapid development of industrial internet, intelligent manufacturing, and digital equipment operation and maintenance technologies, complex equipment and industrial production systems are gradually moving towards automation, intelligence, and continuous operation. Predictive maintenance technology has been widely applied in various scenarios.

[0003] Existing predictive maintenance model tuning mainly relies on manual experience or grid search. However, current manual experience-based tuning depends on algorithm engineers' understanding of equipment mechanisms, data characteristics, and model structure, resulting in long tuning cycles and poor reusability. Grid search, on the other hand, tends to generate a large number of invalid experiments in high-dimensional hyperparameter spaces, leading to low sample utilization and consequently low model tuning efficiency. Summary of the Invention

[0004] The main purpose of this application is to provide a method, apparatus, device and storage medium for predictive maintenance model tuning based on a large language model, which aims to solve the technical problem that the current predictive maintenance model tuning relies on human experience, resulting in low model tuning efficiency.

[0005] To achieve the above objectives, this application proposes a method for tuning a predictive maintenance model based on a large language model. This method includes: Obtain diagnostic information from the objective optimizer and predictive maintenance model; The diagnostic information from the predictive maintenance model is input into the large language model for parsing to obtain the tuning signal; The corrected scores of historical candidate hyperparameters are determined based on the target optimizer and the tuning signal. Based on the corrected scores of the historical candidate hyperparameters and the objective optimizer, the predictive maintenance model is evaluated and tuned, and the target predictive maintenance model is output.

[0006] In one embodiment, the step of determining the corrected score of historical candidate hyperparameters based on the target optimizer and the tuning signal includes: The evaluation metrics are determined by training and validation using historical hyperparameter combinations in the target optimizer. Obtain the optimization target, and determine the comprehensive weight corresponding to each evaluation indicator based on the optimization target; The basic score is determined based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination. The adjustment direction and confidence level in the optimization signal are converted into semantic feedback correction terms; The corrected scores for historical candidate hyperparameters are determined based on the base score, preset weights, and semantic feedback correction terms.

[0007] In one embodiment, the step of determining the basic score based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination includes: The comprehensive score of the historical hyperparameter combination is calculated based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination. Based on the comprehensive score, the historical hyperparameter combinations are divided into a first sample distribution and a second sample distribution; A first sample density model is established based on the first sample distribution and the second sample distribution; The base score is determined based on the first sample density model.

[0008] In one embodiment, the step of determining each evaluation metric by training and validating through historical hyperparameter combinations in the target optimizer includes: The fitted predictive maintenance model and training log are obtained by configuring and training through historical hyperparameter combinations in the target optimizer. Obtain validation set data, and use the validation set data to validate and monitor the fitted predictive maintenance model to obtain validation results and monitoring data; The confusion matrix is ​​obtained based on the verification results; Each evaluation metric is determined based on the training log, the confusion matrix, and the monitoring data.

[0009] In one embodiment, prior to the step of obtaining diagnostic information for the target optimizer and the predictive maintenance model, the method further includes: Obtain information on predictive maintenance tasks, preset optimizers, and tuning targets; The predictive maintenance task information and the optimization target are input into a large language model for semantic parsing to obtain the search space; The preset optimizer is initialized based on the search space to obtain the target optimizer.

[0010] In one embodiment, after the step of inputting the predictive maintenance task information and the optimization target into a large language model for semantic parsing to obtain the search space, the method further includes: Determine the value range, type, training interface, hyperparameter dependencies, and resource consumption of each hyperparameter in the search space; When the range of the hyperparameter values ​​is not within the preset range, the range of the hyperparameter values ​​is corrected to obtain the first corrected search space; When the hyperparameter type does not match the training interface, the hyperparameter type is converted and corrected to obtain a second corrected search space; When the hyperparameter dependencies have dependency conflicts, the hyperparameters corresponding to the hyperparameter dependencies are corrected to obtain a third corrected search space; When the resource consumption situation is resource over-limit, the range of the target hyperparameters is compressed to obtain the fourth corrected search space, wherein the target hyperparameters include at least the number of network layers, the number of hidden units, and the batch size; The target search space is determined based on at least one of the first modified search space, the second modified search space, the third modified search space, and the fourth modified search space; Update the search space to the target search space.

[0011] In one embodiment, the step of evaluating and tuning the predictive maintenance model based on the corrected scores of the historical candidate hyperparameters and the target optimizer, and outputting the target predictive maintenance model, includes: Based on the corrected scores of the historical candidate hyperparameters, a second sample density model is established through the objective optimizer; Several candidate hyperparameter combinations are generated based on the second sample density model; The candidate parameters are combined and assigned to different computing nodes to train and evaluate the predictive maintenance model, and the evaluation results are obtained. Based on the evaluation results, a target predictive maintenance model is output.

[0012] Furthermore, to achieve the above objectives, this application also proposes a predictive maintenance model tuning device based on a large language model, the predictive maintenance model tuning device based on a large language model comprising: The acquisition module is used to acquire diagnostic information from the target optimizer and the predictive maintenance model; The processing module is used to input the diagnostic information of the predictive maintenance model into the large language model for parsing to obtain the tuning signal; The processing module is also used to determine the correction score of historical candidate hyperparameters based on the target optimizer and the tuning signal; The output module is used to evaluate and tune the predictive maintenance model based on the corrected scores of the historical candidate hyperparameters and the target optimizer, and output the target predictive maintenance model.

[0013] Furthermore, to achieve the above objectives, this application also proposes a predictive maintenance model tuning device based on a large language model. The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the predictive maintenance model tuning method based on a large language model as described above.

[0014] In addition, to achieve the above objectives, this application also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the steps of the predictive maintenance model tuning method based on a large language model as described above.

[0015] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the predictive maintenance model tuning method based on a large language model as described above.

[0016] One or more technical solutions proposed in this application have at least the following technical effects: 1) By introducing diagnostic information from the predictive maintenance model and obtaining "tuning clues" from the diagnostic information, and then parsing these "tuning clues" through a large language model to transform them into tuning signals, the diagnostic information can be fully utilized to effectively reduce trial and error costs and improve model tuning efficiency.

[0017] 2) By combining the tuning signal with the objective optimizer, the scores of historical candidate hyperparameters are corrected to obtain corrected scores of historical candidate hyperparameters. Based on the corrected scores of historical candidate hyperparameters and the objective optimizer, the predictive maintenance model is evaluated and tuned, thereby outputting the optimal predictive maintenance model in the tuning process, thus achieving the accuracy of the output tuning results. Attached Figure Description

[0018] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0019] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 This is a flowchart illustrating an embodiment of the predictive maintenance model tuning method based on a large language model in this application. Figure 2 This is a schematic diagram of the overall framework of the predictive maintenance model tuning system based on a large language model, provided in an embodiment of the predictive maintenance model tuning method based on a large language model of this application. Figure 3 This is a schematic diagram of the overall framework of the asynchronous parallelized TPE scheduling module provided in an embodiment of the predictive maintenance model tuning method based on a large language model in this application. Figure 4 This is a flowchart illustrating Embodiment 2 of the predictive maintenance model tuning method based on a large language model in this application. Figure 5 This is a flowchart illustrating Embodiment 3 of the predictive maintenance model tuning method based on a large language model in this application. Figure 6 This is a schematic diagram of the scoring correction process provided in an embodiment of the predictive maintenance model tuning method based on a large language model in this application. Figure 7 This is a schematic diagram of the module structure of the predictive maintenance model tuning device based on a large language model according to an embodiment of this application; Figure 8 This is a schematic diagram of the hardware operating environment involved in the predictive maintenance model tuning method based on a large language model in the embodiments of this application.

[0021] The purpose, features, and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0022] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0023] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0024] The main solution of this application embodiment is: to obtain diagnostic information of the target optimizer and the predictive maintenance model; to input the diagnostic information of the predictive maintenance model into a large language model for parsing to obtain a tuning signal; to determine the correction score of historical candidate hyperparameters based on the target optimizer and the tuning signal; to evaluate and tune the predictive maintenance model based on the correction score of the historical candidate hyperparameters and the target optimizer, and to output the target predictive maintenance model.

[0025] Existing technologies rely on manual experience-based tuning or grid search for model optimization. Manual experience-based tuning depends on algorithm engineers' understanding of equipment mechanisms, data characteristics, and model structure, resulting in long tuning cycles and poor reusability. Grid search, on the other hand, tends to generate a large number of invalid experiments in high-dimensional hyperparameter spaces, leading to low sample utilization and consequently low model optimization efficiency.

[0026] This application provides a solution that introduces diagnostic information from a predictive maintenance model and obtains "tuning clues" from the diagnostic information. These "tuning clues" are then parsed using a large language model and transformed into tuning signals. By combining the tuning signals with the objective optimizer, the scores of historical candidate hyperparameters are corrected to obtain corrected scores. Based on the corrected scores of the historical candidate hyperparameters and the objective optimizer, the predictive maintenance model is evaluated and tuned, thereby outputting the optimal predictive maintenance model during the tuning process.

[0027] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, such as a predictive maintenance model tuning system based on a large language model. The following description uses a predictive maintenance model tuning system based on a large language model as an example to illustrate this embodiment and the subsequent embodiments. All actions involving the acquisition of signals, information, or data in this application are performed in accordance with the relevant data protection regulations of the country where the application is located and with authorization from the owner of the corresponding device.

[0028] Based on this, embodiments of this application provide a predictive maintenance model tuning method based on a large language model, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the predictive maintenance model tuning method based on a large language model in this application.

[0029] In this embodiment, the predictive maintenance model tuning method based on a large language model includes steps S10 to S40: Step S10: Obtain diagnostic information for the objective optimizer and predictive maintenance model.

[0030] like Figure 2 As shown, Figure 2This diagram illustrates the overall framework of a predictive maintenance model tuning system based on a large language model. It includes a data and task input layer, a large model semantic parsing and search space layer, a parallelized TPE (Tree-structured Parzen Estimator) optimization execution layer, and a multi-objective evaluation and result output layer. The data and task input layer acquires information such as equipment operation data, sensor time series, fault records, maintenance logs, historical tuning results, metrics and parameters, deployment constraints, latency / resources, tuning objectives, and multi-metric requirements. The large model semantic parsing and search space layer processes the information acquired from the data and task input layer (including large model semantic parsing, adaptive search space generation, and knowledge base and rule verification), and finally outputs a verified structured search space. After obtaining the structured search space, the TPE can be initialized, resulting in an initialized TPE optimizer (i.e., the objective optimizer). The parallelized TPE optimization execution layer and the multi-objective evaluation and result output layer calculate the correction score of historical candidate hyperparameters in the objective optimizer (which can be calculated by the ratio of good sample density to poor sample density + ...). The process involves semantic feedback correction to obtain a corrected score, which in turn yields candidate parameter combinations. A central scheduler then deduplicates and manages versions of the task queue to ensure the stability of these combinations. The processed candidate parameter combinations are then distributed to different Worker nodes (computing nodes) for parallel training and evaluation. (During training and evaluation, candidate parameter combinations can also be sent to the initialized TPE optimizer for updates.) The evaluation results are stored in an experimental database. Finally, the evaluation results are output to the multi-objective evaluation and result output layer in the parallelized TPE optimization execution layer and the multi-objective evaluation and result output layer. This multi-objective evaluation and result output layer calculates the optimal hyperparameters and the optimal predictive maintenance model, and summarizes the process records, semantic interpretations, and deployment suggestions into a tuning report (which can also be fed back to the initialized TPE optimizer), thus completing the tuning of the predictive maintenance model. It is understandable that the tuning process of the predictive maintenance model is iterative. During iteration, the system continuously updates the TPE sample density model in the target optimizer based on the evaluation results and repeatedly executes the candidate parameter generation, parallel training and evaluation, and multi-objective evaluation processes until the stopping condition is met. The stopping conditions can be reaching the maximum number of iterations, reaching a preset performance index, performance improvement being less than a threshold for multiple consecutive rounds, or reaching the upper limit of computing resources. Ultimately, the system output includes at least the optimal hyperparameter combination, the predictive maintenance model corresponding to the optimal hyperparameter combination, model evaluation metrics, tuning process records, semantic feedback explanations, and a tuning report with deployment suggestions.

[0031] It should be noted that the objective optimizer is typically the TPE optimizer initialized with the search space. Diagnostic information for predictive maintenance models can usually come from at least one of the following: operations and maintenance experts, algorithm engineers, the system's automatic evaluation module, and model diagnostic reports. For example, diagnostic information for predictive maintenance models might include "high false alarm rate under high-frequency noise," "insufficient early fault identification capability," "excessive inference time, unsuitable for edge deployment," "numerous missed faults in a certain type of bearing," and "the model exhibits overfitting tendencies."

[0032] Step S20: Input the diagnostic information of the predictive maintenance model into the large language model for parsing to obtain the tuning signal.

[0033] It should be noted that large language models are typically deep learning models with a very large parameter scale, based on the Transformer architecture. Through self-supervised pre-training on massive amounts of text data, they learn the syntax, semantics, common sense, and reasoning abilities of a language, enabling them to generate coherent and logical text based on input prompts. Optimization signals should at least include question type, association evaluation metrics, association hyperparameters, adjustment direction, adjustment strength, and confidence level.

[0034] In a specific embodiment, when the diagnostic information of the predictive maintenance model is "excessive inference time," the associated metrics in the optimization signal output by the large model are inference latency and resource consumption, the associated hyperparameters are the number of network layers and the number of hidden units, and the adjustment direction is to reduce model complexity. When the diagnostic information of the predictive maintenance model is "high false alarm rate under high-frequency noise," the system first performs semantic parsing on the feedback text to extract the problem object, problem type, and influencing scenario. A high false alarm rate is identified as a classification performance problem in anomaly detection or fault diagnosis tasks, with corresponding associated evaluation metrics being false alarm rate, accuracy, and precision. "High-frequency noise" is identified as a data quality and signal preprocessing problem, with corresponding associated hyperparameters including time window length, filtering parameters, feature selection ratio, regularization coefficient, dropout ratio, and classification threshold. Based on a preset feedback semantic-evaluation metric-adjustable hyperparameter-adjustment direction mapping table, the system transforms the diagnostic information of the predictive maintenance model into a structured optimization signal, such as: reducing the false alarm rate, improving accuracy, enhancing noise resistance, appropriately increasing regularization strength, adjusting the filtering window, or expanding the time window length.

[0035] Step S30: Determine the correction score of historical candidate hyperparameters based on the target optimizer and the tuning signal.

[0036] It should be noted that historical candidate hyperparameters are hyperparameters that the objective optimizer has suggested in previous iterations and that have already been trained and evaluated. The corrected score of the historical candidate hyperparameters is calculated by converting the tuning signal into a correction term and combining it with the base score of the historical candidate hyperparameters.

[0037] Step S40: Evaluate and fine-tune the predictive maintenance model based on the corrected scores of the historical candidate hyperparameters and the target optimizer, and output the target predictive maintenance model.

[0038] It should be noted that predictive maintenance models can be a type of machine learning model that uses real-time sensor data (such as vibration, temperature, current, etc.) to predict the occurrence of failures or estimate the remaining service life. Evaluation and tuning are usually systematic evaluation and optimization processes centered around predictive maintenance models.

[0039] In one embodiment, the step of evaluating and tuning the predictive maintenance model based on the corrected scores of the historical candidate hyperparameters and the target optimizer, and outputting the target predictive maintenance model, includes: establishing a second sample density model using the target optimizer based on the corrected scores of the historical candidate hyperparameters; generating several candidate hyperparameter combinations based on the second sample density model; allocating the several candidate parameter combinations to different computing nodes to train and evaluate the predictive maintenance model, obtaining evaluation results; and outputting the target predictive maintenance model based on the evaluation results.

[0040] Specifically, based on the corrected scores of historical candidate hyperparameters after modification and updating, a second sample density model (including a second good sample density model and a second bad sample density model) is established through the objective optimizer. Based on the second good sample density model and the second bad sample density model, candidate hyperparameter combinations that conform to both historical optimization trends and semantic feedback directions based on diagnostic information from the predictive maintenance model are generated. These candidate parameter combinations are then distributed to different computing nodes to perform parallel training and evaluation of the predictive maintenance model, yielding evaluation results that include at least the hyperparameter combinations, model performance metrics, training status, failure reasons, and resource consumption. Based on the evaluation results, the target hyperparameter combination and the corresponding target predictive maintenance model are output.

[0041] This embodiment re-establishes the sample density model by correcting the scores of historical candidate hyperparameters, and generates new candidate hyperparameter combinations accordingly. The new candidate hyperparameter combinations are then assigned to different computing nodes for model training and evaluation, thereby obtaining the target predictive maintenance model. By introducing correcting scores, the reliability of the predictive maintenance model tuning is improved.

[0042] In addition, to avoid duplicate evaluations and probabilistic model update conflicts during parallel tuning and evaluation, a task deduplication mechanism and a versioning update mechanism can be set up. The task deduplication mechanism is used to determine whether the newly generated candidate hyperparameter combinations are the same as or similar to those already completed, currently being executed, or retried after failure. If duplicates or high similarities exist, candidate parameters are regenerated. The versioning update mechanism is used to record the version number of each sample density model update. Worker nodes carry version information (at least including the result timestamp and model version) when receiving tasks and returning results. The system merges the experimental results based on the result timestamp and model version, thereby ensuring the stability of sample density model updates in asynchronous parallel scenarios. For hyperparameter combinations that fail to train, exceed resource limits, fail to converge, or exceed inference latency limits, they can be recorded as negative samples in the experimental database, reducing the probability of the corresponding parameter region in subsequent TPE sampling, thereby reducing invalid experiments and wasted computational resources.

[0043] In one implementation, the parallel optimization and evaluation process can be implemented using an asynchronous parallelized TPE scheduling module, such as... Figure 3 As shown, Figure 3 This diagram illustrates the overall framework of the asynchronous parallel TPE scheduling module. The asynchronous parallel TPE scheduling module includes at least a central scheduler, a global TPE probabilistic model manager, a candidate parameter generator, an asynchronous task distribution module, multiple training workers, a task deduplication and version control module, and an experiment record database. The asynchronous parallel TPE scheduling module is divided into a scheduling control layer, an optimization modeling layer, a distributed execution layer, and a data management layer. The central scheduler generates multiple sets of candidate hyperparameter combinations based on the current TPE probabilistic model and assigns these combinations to different workers for parallel model training and evaluation. After completing their training tasks, each worker sends back the hyperparameter combinations, model performance metrics, training status, failure reasons, and resource consumption to the experiment record database, triggering an asynchronous update of the TPE probabilistic model. The experiment record database stores at least performance metric records, training status records, failure sample records, and resource consumption records.

[0044] This embodiment provides a method for tuning a predictive maintenance model based on a large language model. It involves acquiring diagnostic information from the target optimizer and the predictive maintenance model; inputting the diagnostic information of the predictive maintenance model into the large language model for parsing to obtain a tuning signal; determining correction scores for historical candidate hyperparameters based on the target optimizer and the tuning signal; evaluating and tuning the predictive maintenance model based on the correction scores of the historical candidate hyperparameters and the target optimizer, and outputting the target predictive maintenance model. By inputting the diagnostic information of the predictive maintenance model into the large language model for semantic parsing, semantic information can be incorporated into the tuning process, thereby improving the efficiency of model tuning.

[0045] Based on the first embodiment of this application, in the second embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 4 Step S10 includes steps A11 to A13: Step A11: Obtain predictive maintenance task information, preset optimizers, and tuning targets.

[0046] It should be noted that predictive maintenance task information includes at least the following: equipment type, equipment operating condition, sensor type, sampling frequency, data scale, fault category, historical fault text, maintenance records, model type, historical tuning results, and deployment constraints. The preset optimizer is typically an uninitialized TPE optimizer. The tuning objective is a target determined based on the user's tuning requirements; it refines vague user needs into a calculable and optimizable mathematical function, ensuring the normal operation of the model tuning process.

[0047] Step A12: Input the predictive maintenance task information and the optimization target into the large language model for semantic parsing to obtain the search space.

[0048] It should be noted that the search space is typically a complete definition of the legal range of all hyperparameters to be tuned, the possible combinations of candidate hyperparameters, and the logical relationships between them. The search space is the "map" and "boundary" of the TPE optimizer's search, determining in which configurations the TPE optimizer can find the optimal solution.

[0049] Specifically, predictive maintenance task information and optimization objectives are input into a large language model. The large language model uses semantic parsing to identify the characteristics of the current predictive maintenance task, such as whether the task belongs to fault classification, anomaly detection, or remaining life prediction; whether the data is vibration signal, temperature sequence, current signal, or multi-source fusion data; and whether the predictive maintenance model is suitable for traditional machine learning methods or deep time-series models. Finally, the large language model outputs a structured search space. This search space includes at least candidate model types, hyperparameter names, parameter types, continuous parameter value ranges, discrete parameter candidate sets, conditional dependencies, and resource constraints. For example, for fault diagnosis tasks based on time-series sensor data, the search space may include parameters such as window length, learning rate, number of network layers, number of hidden units, batch size, regularization coefficient, and dropout ratio; for fault classification tasks based on tree models, the search space may include parameters such as tree depth, number of trees, sampling ratio, feature selection ratio, and minimum number of samples per leaf node.

[0050] Furthermore, to avoid the search space generated by large models being too wide, too narrow, or not conforming to engineering constraints, this application further establishes a search space verification mechanism. This mechanism performs boundary checks, type checks, dependency checks, and resource constraint checks on the search space generated by large models by predictively maintaining a domain knowledge base, historical optimization results, and rule constraints.

[0051] In a specific embodiment, after the step of inputting the predictive maintenance task information and the optimization target into a large language model for semantic parsing to obtain a search space, the method further includes: determining the value range of each hyperparameter, the type of each hyperparameter, the training interface, the hyperparameter dependency relationship, and the resource consumption in the search space; when the value range of the hyperparameter is not within a preset range, correcting the range of the hyperparameter value to obtain a first corrected search space; when the hyperparameter type does not match the training interface, converting and correcting the hyperparameter type to obtain a second corrected search space; when there is a dependency conflict in the hyperparameter dependency relationship, correcting the hyperparameter corresponding to the hyperparameter dependency relationship to obtain a third corrected search space; when the resource consumption is beyond the limit, compressing the range of the target hyperparameter to obtain a fourth corrected search space, wherein the target hyperparameter includes at least the number of network layers, the number of hidden units, and the batch size; determining the target search space based on at least one of the first corrected search space, the second corrected search space, the third corrected search space, and the fourth corrected search space; and updating the search space to the target search space.

[0052] Specifically, firstly, the range of values ​​for each hyperparameter, the type of each hyperparameter, the training interface, the hyperparameter dependencies, and the resource consumption in the search space are determined; then, a boundary check is performed on the search space to determine whether the range of values ​​for each hyperparameter is reasonable: (1) the learning rate should be limited to a preset order of magnitude; (2) the dropout ratio should be between 0 and 1; (3) the number of hidden units, batch size, and other hyperparameters should not exceed the range allowed by engineering experience or the historical best interval. When the hyperparameter range is too wide, the upper and lower bounds are narrowed according to historical tuning results and domain rules; when the hyperparameter range is too narrow, the search interval is appropriately expanded according to the historical results of similar tasks. After the boundary check, the first corrected search space is obtained.

[0053] The search space undergoes a type check to determine whether the data type and value form of each hyperparameter meet the model requirements: (1) the number of network layers and batch size should be integer parameters; (2) the learning rate, regularization coefficient, and sampling ratio should be continuous parameters; and (3) the optimizer type, activation function type, and model structure type should be discrete enumeration parameters. When the hyperparameter type is inconsistent with the actual training interface, the system will automatically convert it to an executable format or mark the hyperparameter as invalid and feed it back to the large language model for regeneration. After the type check, a second corrected search space is obtained.

[0054] The search space is checked for dependencies. The dependency check is used to determine whether there are logical conflicts or conditional constraints between different hyperparameters: (1) In the training of the neural network model, when the optimizer type is Adam, the search space can include parameters such as learning rate, β1, β2, and weight decay coefficient; (2) When the optimizer type is SGD (Stochastic Gradient Descent), the search space can include parameters such as learning rate, momentum coefficient, and weight decay coefficient, but should not include β1 and β2 parameters that are only applicable to the Adam optimizer; (3) When the regularization method is L1 regularization, the L1 coefficient search range is enabled; (4) When the regularization method is Dropout, the dropout ratio search range is enabled, and its value is checked to be between 0 and 1. For the parameters with conditional dependencies, the system can establish a rule table of "parameter options - applicable hyperparameters - constraints" and automatically enable, disable or correct the relevant parameters according to the current model configuration. After the dependency check, the third corrected search space is obtained.

[0055] The search space undergoes a resource constraint check to determine if it meets engineering constraints such as training time, GPU / RAM, CPU / GPU resources, and deployment scale. When training resources are limited, the system restricts the maximum number of network layers, hidden units, trees, and batch size. When the model needs to be deployed to edge devices, the system restricts the number of model parameters, inference latency, and memory usage. If the search space causes the expected training or deployment cost to exceed a threshold, the system compresses the relevant hyperparameter ranges according to resource constraint priority. After the resource constraint check, a fourth revised search space is obtained. Finally, the target search space is determined based on at least one of the first, second, third, and fourth revised search spaces; the original search space is then updated to the target search space.

[0056] This embodiment performs boundary checks, type checks, dependency checks, and resource constraint checks on the search space, which can compress the invalid search range of the search space, ensure the logical consistency and structural integrity of the search space, and thus improve the efficiency of model tuning.

[0057] Step A13: Initialize the preset optimizer based on the search space to obtain the target optimizer.

[0058] It should be noted that the target optimizer is the optimizedr after initialization. After obtaining the search space, the internal state of the preset optimizer can be initialized based on the search space, clarifying the structure and boundaries of the search domain that the optimizer is responsible for. The initialized optimizer (i.e., the target optimizer) becomes a specific optimizer dedicated to the current tuning task.

[0059] This embodiment obtains predictive maintenance task information, a preset optimizer, and a tuning objective; inputs the predictive maintenance task information and the tuning objective into a large language model for semantic parsing to obtain a search space; initializes the preset optimizer based on the search space to obtain a target optimizer; and automatically generates a structured search space by performing semantic parsing of the predictive maintenance task and tuning objective through the large language model, reducing reliance on human experience and improving the rationality of the search space.

[0060] Based on the first embodiment of this application, in the third embodiment of this application, the content that is the same as or similar to that in the first embodiment described above can be referred to the above description, and will not be repeated hereafter. Based on this, please refer to... Figure 5 Step S30 includes steps S301 to S305: Step S301: Train and verify using historical hyperparameter combinations in the target optimizer to determine each evaluation index.

[0061] It should be noted that evaluation metrics typically refer to specific numerical values ​​used to quantitatively measure the performance of a predictive maintenance model trained with a set of hyperparameters across multiple dimensions, including predictive capability, operational efficiency, and resource consumption. These are the sole objective criteria used by the TPE optimizer to judge the quality of hyperparameters, update the probabilistic model, and ultimately select the optimal model.

[0062] In one embodiment, the step of determining each evaluation index by training and validating through historical hyperparameter combinations in the target optimizer includes: performing configuration training through historical hyperparameter combinations in the target optimizer to obtain a fitted predictive maintenance model and training logs; obtaining validation set data and using the validation set data to monitor and validate the fitted predictive maintenance model to obtain validation results and monitoring data; obtaining a confusion matrix based on the validation results; and determining each evaluation index based on the training logs, the confusion matrix, and the monitoring data.

[0063] Specifically, historical hyperparameter combinations are extracted from the target optimizer and fed into the training interface. A predictive maintenance model is configured and trained based on the historical hyperparameter combinations to obtain a fitted predictive maintenance model and training logs. Validation set data is obtained and used to validate the fitted predictive maintenance model. Hardware monitoring data is collected simultaneously to obtain validation results and monitoring data. A confusion matrix is ​​constructed based on the predicted and true labels in the validation results. Finally, the training logs, confusion matrix, and monitoring data are combined to calculate and determine each evaluation metric.

[0064] In addition, evaluation metrics include positive and negative metrics. Positive metrics typically include accuracy, F1 score, recall, and AUC (Area Under the Curve), with higher values ​​for positive metrics indicating better model performance. Negative metrics typically include false positive rate, false negative rate, inference latency, training time, model size, CPU / GPU resource usage, and memory usage, with lower values ​​for negative metrics indicating better model performance.

[0065] Step S302: Obtain the optimization target and determine the comprehensive weight of each evaluation indicator based on the optimization target.

[0066] It should be noted that the optimization goal is a target determined based on the user's optimization needs. It refines the vague user needs into a calculable and optimizable mathematical function, ensuring the normal operation of the model optimization process.

[0067] Understandably, the system can normalize evaluation metrics with different dimensions, converting them into uniform scoring values. Then, it assigns corresponding comprehensive weights to each evaluation metric based on the requirements of the optimization goals. Comprehensive weights are typically used to represent the importance of different metrics in the current optimization task. It is necessary to identify user needs within the optimization goals and determine the priority objectives of the optimization task. A task scenario-evaluation metric-weight configuration mapping table can also be pre-established to automatically provide initial comprehensive weights based on different task scenarios. For example, for safety-critical equipment, increasing the comprehensive weights for recall and false negative penalties can reduce the risk of missed fault detections; for edge-deployed devices, increasing the comprehensive weights for inference latency, model size, and resource consumption can ensure model deployability; for scenarios with high false positive costs, increasing the comprehensive weights for false positive penalties can reduce unnecessary maintenance alarms.

[0068] Step S303: Determine the basic score based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination.

[0069] In one embodiment, the step of determining the basic score based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination includes: calculating the comprehensive score of the historical hyperparameter combination based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination; dividing the historical hyperparameter combination into a first sample distribution and a second sample distribution based on the comprehensive score; establishing a first sample density model based on the first sample distribution and the second sample distribution; and determining the basic score based on the first sample density model.

[0070] In practice, the comprehensive objective function can be determined based on the evaluation indicators, their corresponding comprehensive weights, and historical hyperparameter combinations. The basic score is then calculated based on the comprehensive objective function, which is expressed as follows:

[0071] In the above formula, x represents the historical hyperparameter combination, and S i (x) represents the normalized score value of the i-th evaluation indicator, W i This represents the comprehensive weight corresponding to the i-th evaluation indicator, and satisfies... The comprehensive score for each historical hyperparameter combination is calculated based on the comprehensive objective function, and the score results are used as the evaluation basis in the TPE optimization process. During optimization, the system first generates several historical hyperparameter combinations from the TPE and completes model training, validation, and index calculation respectively; then, the comprehensive score for each historical hyperparameter combination is calculated based on the comprehensive objective function; and then, the candidate combinations are divided into a good sample set (i.e., the first sample distribution) and a poor sample set (i.e., the second sample distribution) according to the score. The good sample set represents the parameter region with good overall performance under the current weight preference, and the poor sample set represents the parameter region with poor overall performance or that does not meet the engineering constraints. Then, a first sample density model (including the first good sample density model l(x) and the first poor sample density model g(x)) is established based on the good sample set and the poor sample set, and the final base score is determined as l(x) / g(x).

[0072] This embodiment introduces a comprehensive weight, which is not only used to calculate the overall score of candidate models, but also directly affects the division between excellent and poor samples, thereby influencing the update direction of the sample density model. The system can thus achieve a more reasonable trade-off between model accuracy, fault recall, false alarm control, inference speed, and resource consumption based on different application scenarios.

[0073] Step S304: Convert the adjustment direction and confidence level in the tuning signal into semantic feedback correction terms.

[0074] It is understandable that the adjustment direction and confidence level in the tuning signal are converted into semantic feedback correction terms, and then the semantic feedback correction terms are introduced into the calculation of the correction score. The score of candidate parameters that conform to the optimization direction is increased, and the score of candidate parameters that do not conform to the optimization direction is decreased, thereby guiding TPE to sample hyperparameter regions that are more in line with actual engineering needs.

[0075] Step S305: Determine the corrected score of the historical candidate hyperparameters based on the basic score, the preset weight, and the semantic feedback correction term.

[0076] In one embodiment, a candidate hyperparameter correction scoring function can be constructed based on the base score, preset weights, and semantic feedback correction terms, wherein the candidate hyperparameter correction scoring function is expressed as follows:

[0077] In the above formula, This is the base score obtained based on historical hyperparameter combinations. To preset weights, This is a semantic feedback correction term. TPE tends to select candidate parameters with higher base scores, i.e., parameters that are closer to historically high-performing samples and farther from historically low-performing samples. Based on this, this invention introduces a semantic feedback correction term to improve the scores of candidate parameters that conform to the direction of semantic optimization.

[0078] In one implementation, the scoring correction process can be achieved using semantic feedback and numerical evaluation results, such as... Figure 6 As shown, Figure 6 The flowchart illustrates the scoring correction process. Natural language feedback (diagnostic information for predictive maintenance models) is input into the semantic feedback module for semantic parsing, outputting a structured tuning signal. Semantic feedback correction terms are generated based on the indicators, parameters, directions, and weights in the tuning signal. Simultaneously, a comprehensive target value (basic score) obtained from historical numerical evaluation indicators is acquired. The semantic feedback correction terms and the comprehensive target value are jointly input into the scoring correction module for scoring correction, resulting in corrected scores for historical candidate hyperparameters. The TPE probability model is updated based on the corrected scores of historical candidate hyperparameters, thereby changing the sampling direction of subsequent hyperparameters.

[0079] This embodiment uses historical hyperparameter combinations in the target optimizer for training and verification to determine each evaluation index; obtains the tuning target and determines the comprehensive weight corresponding to each evaluation index based on the tuning target; determines the basic score according to the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combinations; converts the adjustment direction and confidence level in the tuning signal into semantic feedback correction terms; determines the correction score of historical candidate hyperparameters according to the basic score, preset weights, and the semantic feedback correction terms; by utilizing the adjustment direction and confidence level in the tuning signal based on diagnostic information from a predictive maintenance model, semantic feedback correction terms are obtained, and the semantic feedback correction terms are used to correct the score of historical hyperparameter combinations with the basic score, thereby improving the targeting of subsequent tuning directions.

[0080] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the predictive maintenance model tuning method based on large language models in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0081] This application also provides a predictive maintenance model tuning device based on a large language model, please refer to... Figure 7 The predictive maintenance model tuning device based on a large language model includes: Module 10 is used to acquire diagnostic information from the target optimizer and the predictive maintenance model.

[0082] The processing module 20 is used to input the diagnostic information of the predictive maintenance model into the large language model for parsing to obtain the tuning signal.

[0083] The processing module 20 is also used to determine the correction score of historical candidate hyperparameters based on the target optimizer and the tuning signal.

[0084] Output module 30 is used to evaluate and tune the predictive maintenance model based on the corrected scores of the historical candidate hyperparameters and the target optimizer, and output the target predictive maintenance model.

[0085] The predictive maintenance model tuning device based on a large language model provided in this application employs the predictive maintenance model tuning method based on a large language model in the above embodiments, which can solve the technical problem of low model tuning efficiency caused by the reliance on manual experience in current predictive maintenance model tuning. Compared with the prior art, the beneficial effects of the predictive maintenance model tuning device based on a large language model provided in this application are the same as those of the predictive maintenance model tuning method based on a large language model provided in the above embodiments, and other technical features in the predictive maintenance model tuning device based on a large language model are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0086] In one embodiment, the processing module 20 is further configured to: perform training and verification using historical hyperparameter combinations in the target optimizer to determine each evaluation index; obtain the tuning target and determine the comprehensive weight corresponding to each evaluation index based on the tuning target; determine the basic score according to the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combinations; convert the adjustment direction and confidence level in the tuning signal into semantic feedback correction terms; and determine the correction score of the historical candidate hyperparameters according to the basic score, the preset weight, and the semantic feedback correction terms.

[0087] In one embodiment, the processing module 20 is further configured to calculate a comprehensive score of the historical hyperparameter combination based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination; divide the historical hyperparameter combination into a first sample distribution and a second sample distribution based on the comprehensive score; establish a first sample density model based on the first sample distribution and the second sample distribution; and determine a basic score based on the first sample density model.

[0088] In one embodiment, the processing module 20 is further configured to perform configuration training using historical hyperparameter combinations in the target optimizer to obtain a fitted predictive maintenance model and training logs; acquire validation set data and perform validation monitoring on the fitted predictive maintenance model using the validation set data to obtain validation results and monitoring data; obtain a confusion matrix based on the validation results; and determine various evaluation metrics based on the training logs, the confusion matrix, and the monitoring data.

[0089] In one embodiment, the acquisition module 10 is further configured to acquire predictive maintenance task information, a preset optimizer, and a tuning target; input the predictive maintenance task information and the tuning target into a large language model for semantic parsing to obtain a search space; and initialize the preset optimizer based on the search space to obtain a target optimizer.

[0090] In one embodiment, the acquisition module 10 is further configured to determine the value range of each hyperparameter, the type of each hyperparameter, the training interface, the hyperparameter dependency relationship, and the resource consumption in the search space; when the value range of the hyperparameter is not within a preset range, the value range of the hyperparameter is corrected to obtain a first corrected search space; when the hyperparameter type does not match the training interface, the hyperparameter type is converted and corrected to obtain a second corrected search space; when there is a dependency conflict in the hyperparameter dependency relationship, the hyperparameter corresponding to the hyperparameter dependency relationship is corrected to obtain a third corrected search space; when the resource consumption is beyond the limit, the range of the target hyperparameter is compressed to obtain a fourth corrected search space, wherein the target hyperparameter includes at least the number of network layers, the number of hidden units, and the batch size; the target search space is determined based on at least one of the first corrected search space, the second corrected search space, the third corrected search space, and the fourth corrected search space; and the search space is updated to the target search space.

[0091] In one embodiment, the output module 30 is further configured to: establish a second sample density model based on the corrected scores of the historical candidate hyperparameters through the target optimizer; generate several candidate hyperparameter combinations based on the second sample density model; allocate the several candidate parameter combinations to different computing nodes to train and evaluate the predictive maintenance model, and obtain evaluation results; and output the target predictive maintenance model based on the evaluation results.

[0092] This application provides a predictive maintenance model tuning device based on a large language model. The predictive maintenance model tuning device based on a large language model includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the predictive maintenance model tuning method based on a large language model in the above embodiment 1.

[0093] The following is for reference. Figure 8 This document illustrates a structural schematic diagram of a predictive maintenance model tuning device based on a large language model suitable for implementing embodiments of this application. The predictive maintenance model tuning device based on a large language model in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), tablets, PMPs (Portable Media Players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 8The predictive maintenance model tuning device based on a large language model shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0094] like Figure 8 As shown, the predictive maintenance model tuning device based on a large language model may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the predictive maintenance model tuning device based on a large language model. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the large language model-based predictive maintenance model tuning device to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a large language model-based predictive maintenance model tuning device with various systems, it should be understood that it is not required to implement or have all the systems shown. More or fewer systems can be implemented alternatively.

[0095] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0096] The predictive maintenance model tuning device based on a large language model provided in this application employs the predictive maintenance model tuning method based on a large language model in the above embodiments, which can solve the technical problem of low model tuning efficiency caused by the reliance on manual experience in current predictive maintenance model tuning. Compared with the prior art, the beneficial effects of the predictive maintenance model tuning device based on a large language model provided in this application are the same as those of the predictive maintenance model tuning method based on a large language model provided in the above embodiments, and other technical features in this predictive maintenance model tuning device based on a large language model are the same as those disclosed in the previous embodiment method, and will not be repeated here.

[0097] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0098] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0099] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, the computer-readable program instructions being used to execute the predictive maintenance model tuning method based on a large language model in the above embodiments.

[0100] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory or Flash Memory), optical fibers, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0101] The aforementioned computer-readable storage medium may be included in a predictive maintenance model tuning device based on a large language model; or it may exist independently and not be assembled into a predictive maintenance model tuning device based on a large language model.

[0102] The aforementioned computer-readable storage medium carries one or more programs. When these programs are executed by a predictive maintenance model tuning device based on a large language model, the predictive maintenance model tuning device based on a large language model performs the following: acquiring diagnostic information of the target optimizer and the predictive maintenance model; inputting the diagnostic information of the predictive maintenance model into the large language model for parsing to obtain a tuning signal; determining correction scores for historical candidate hyperparameters based on the target optimizer and the tuning signal; evaluating and tuning the predictive maintenance model based on the correction scores of the historical candidate hyperparameters and the target optimizer, and outputting the target predictive maintenance model.

[0103] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0104] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0105] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0106] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., computer programs) for executing the above-described predictive maintenance model tuning method based on a large language model. This solves the technical problem that current predictive maintenance model tuning relies on human experience, leading to low model tuning efficiency. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as those of the predictive maintenance model tuning method based on a large language model provided in the above embodiments, and will not be repeated here.

[0107] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the predictive maintenance model tuning method based on a large language model as described above.

[0108] The computer program product provided in this application can solve the technical problem that the current predictive maintenance model tuning relies on human experience, resulting in low model tuning efficiency. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the predictive maintenance model tuning method based on large language models provided in the above embodiments, and will not be repeated here.

[0109] The above description is only a part of the embodiments of this application and does not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A predictive maintenance model tuning method based on a large language model, characterized in that, The method includes: Obtain diagnostic information from the objective optimizer and predictive maintenance model; The diagnostic information from the predictive maintenance model is input into the large language model for parsing to obtain the tuning signal; The corrected scores of historical candidate hyperparameters are determined based on the target optimizer and the tuning signal. Based on the corrected scores of the historical candidate hyperparameters and the objective optimizer, the predictive maintenance model is evaluated and tuned, and the target predictive maintenance model is output.

2. The method as described in claim 1, characterized in that, The step of determining the corrected score of historical candidate hyperparameters based on the target optimizer and the tuning signal includes: The evaluation metrics are determined by training and validation using historical hyperparameter combinations in the target optimizer. Obtain the optimization target, and determine the comprehensive weight corresponding to each evaluation indicator based on the optimization target; The basic score is determined based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination. The adjustment direction and confidence level in the optimization signal are converted into semantic feedback correction terms; The corrected scores for historical candidate hyperparameters are determined based on the base score, preset weights, and semantic feedback correction terms.

3. The method as described in claim 2, characterized in that, The step of determining the basic score based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination includes: The comprehensive score of the historical hyperparameter combination is calculated based on the evaluation index, the comprehensive weight corresponding to the evaluation index, and the historical hyperparameter combination. Based on the comprehensive score, the historical hyperparameter combinations are divided into a first sample distribution and a second sample distribution; A first sample density model is established based on the first sample distribution and the second sample distribution; The base score is determined based on the first sample density model.

4. The method as described in claim 2, characterized in that, The step of determining each evaluation index by training and validating through historical hyperparameter combinations in the target optimizer includes: The fitted predictive maintenance model and training log are obtained by configuring and training through historical hyperparameter combinations in the target optimizer. Obtain validation set data, and use the validation set data to validate and monitor the fitted predictive maintenance model to obtain validation results and monitoring data; The confusion matrix is ​​obtained based on the verification results; Each evaluation metric is determined based on the training log, the confusion matrix, and the monitoring data.

5. The method as described in claim 1, characterized in that, Before the step of obtaining diagnostic information for the target optimizer and predictive maintenance model, the method further includes: Obtain information on predictive maintenance tasks, preset optimizers, and tuning targets; The predictive maintenance task information and the optimization target are input into a large language model for semantic parsing to obtain the search space; The preset optimizer is initialized based on the search space to obtain the target optimizer.

6. The method as described in claim 5, characterized in that, After the step of inputting the predictive maintenance task information and the optimization target into a large language model for semantic parsing to obtain the search space, the method further includes: Determine the value range, type, training interface, hyperparameter dependencies, and resource consumption of each hyperparameter in the search space; When the range of the hyperparameter values ​​is not within the preset range, the range of the hyperparameter values ​​is corrected to obtain the first corrected search space; When the hyperparameter type does not match the training interface, the hyperparameter type is converted and corrected to obtain a second corrected search space; When the hyperparameter dependencies have dependency conflicts, the hyperparameters corresponding to the hyperparameter dependencies are corrected to obtain a third corrected search space; When the resource consumption situation is resource over-limit, the range of the target hyperparameters is compressed to obtain the fourth corrected search space, wherein the target hyperparameters include at least the number of network layers, the number of hidden units, and the batch size; The target search space is determined based on at least one of the first modified search space, the second modified search space, the third modified search space, and the fourth modified search space; Update the search space to the target search space.

7. The method as described in claim 1, characterized in that, The step of evaluating and tuning the predictive maintenance model based on the corrected scores of the historical candidate hyperparameters and the objective optimizer, and outputting the objective predictive maintenance model, includes: Based on the corrected scores of the historical candidate hyperparameters, a second sample density model is established through the objective optimizer; Several candidate hyperparameter combinations are generated based on the second sample density model; The candidate parameters are combined and assigned to different computing nodes to train and evaluate the predictive maintenance model, and the evaluation results are obtained. Based on the evaluation results, a target predictive maintenance model is output.

8. A predictive maintenance model tuning device based on a large language model, characterized in that, The predictive maintenance model tuning device based on a large language model includes: The acquisition module is used to acquire diagnostic information from the target optimizer and the predictive maintenance model; The processing module is used to input the diagnostic information of the predictive maintenance model into the large language model for parsing to obtain the tuning signal; The processing module is also used to determine the correction score of historical candidate hyperparameters based on the target optimizer and the tuning signal; The output module is used to evaluate and tune the predictive maintenance model based on the corrected scores of the historical candidate hyperparameters and the target optimizer, and output the target predictive maintenance model.

9. A predictive maintenance model tuning device based on a large language model, characterized in that, The device includes: a memory, a processor, and a computer program stored in the memory and executable on the processor, the computer program being configured to implement the steps of the predictive maintenance model tuning method based on a large language model as described in any one of claims 1 to 7.

10. A storage medium, characterized in that, The storage medium is a computer-readable storage medium, and a computer program is stored on the storage medium. When the computer program is executed by a processor, it implements the steps of the predictive maintenance model tuning method based on a large language model as described in any one of claims 1 to 7.

Images