Industrial time series data analysis method, device and equipment and storage medium

Abstract

The invention discloses an industrial time series data analysis method and device, equipment and a storage medium. Continuous industrial time series data is converted into a data set containing a high-dimensional vector, an adaptive data format is provided for subsequent model processing, and rich original information support is provided for model learning. An architecture model with Lma as a base is built, global dependency relationships of different time points in industrial time sequence data can be captured in parallel, and the defect that a traditional model processes long-term time sequence dependency is overcome. Through supervision and fine tuning of the time series data set, the model can be specialized in an industrial safety parameter prediction task, and the prediction accuracy of the model in a target industrial scene is improved. Real-time data is analyzed through the safety early warning model, a safety analysis result is rapidly output, real-time monitoring and potential risk identification of the operation state of the industrial system are achieved, a timely and effective decision basis is provided for industrial safety early warning, and industrial production safety is guaranteed.

Description

Technical Field

[0001] This invention relates to the field of hydrogen safety, and in particular to an industrial time-series data analysis method, apparatus, equipment, and storage medium. Background Technology

[0002] Against the backdrop of the integrated development of artificial intelligence and industrial intelligence, industrial time-series data has become a core basis for reflecting the operational status of industrial systems and identifying potential risks. Accurate analysis and prediction of industrial time-series data are crucial for achieving industrial safety early warning and ensuring production safety. Therefore, how to efficiently process industrial time-series data to build a reliable early warning system has become an important direction for the intelligent development of the industrial sector.

[0003] In existing technologies, solutions for processing time-series data are mainly divided into two categories: one is traditional linear models such as ARIMA and exponential smoothing, and the other is deep learning models such as RNN and LSTM. Meanwhile, in the field of natural language processing, large models with the Transformer architecture, represented by Llama, have shown advantages in sequence data processing due to their self-attention mechanism, providing a reference for cross-domain applications.

[0004] Traditional linear models cannot capture the nonlinear characteristics of industrial time-series data, resulting in low prediction accuracy. RNNs and LSTMs are prone to gradient vanishing or exploding problems when processing ultra-long time-series data, making it difficult to capture long-term dependencies. Furthermore, existing models have poor generalization ability and require retraining for different industrial scenarios, leading to high development costs. At the same time, the advantages of large Llama-type models have not been fully integrated into industrial time-series prediction, making it difficult to meet the demand for high-precision and highly adaptable industrial safety early warning. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes an industrial time series data analysis method, apparatus, equipment, and storage medium, which solves the technical problems of poor generalization ability, the need for retraining in different scenarios, and high development costs of existing time series prediction models.

[0006] In a first aspect, embodiments of this disclosure provide an industrial time-series data analysis method, the method comprising: Acquire industrial time-series data and generate a time-series dataset based on the industrial time-series data. The time-series dataset includes various high-dimensional vectors. An architecture model based on Llama was built, and the architecture model was trained under supervision using a time series dataset to generate a security early warning model. Acquire real-time time-series data and perform security analysis on the real-time time-series data through a security early warning model.

[0007] Optionally, industrial time-series data can be acquired, including: identifying each target hydrogen energy base and acquiring standardized data exported from the local data terminal of each target hydrogen energy base; removing outliers from the standardized data to generate industrial time-series data.

[0008] Optionally, a time-series dataset is generated based on industrial time-series data, including: cutting the industrial time-series data into fixed-length segments using a preset sliding window and mapping each segment to a high-dimensional vector; and integrating the high-dimensional vectors to generate a time-series dataset for model training.

[0009] Optionally, an architecture model based on Llama is constructed, including: using a converter architecture base model based on Llama to generate an initial model framework adapted to industrial time series forecasting tasks; obtaining time series forecasting components, wherein the time series forecasting components include position encoding, multi-head self-attention layer, feedforward neural network and residual connection and normalization module; and combining the time series forecasting components with the initial model framework to generate an architecture model.

[0010] Optionally, supervised fine-tuning training of the architecture model is performed using a time-series dataset to generate a security warning model. This includes: dividing the time-series dataset into a training set and a validation set; using the training set to perform supervised fine-tuning training of the architecture model to generate a pre-fine-tuned model; verifying the performance of the pre-fine-tuned model using the validation set and determining whether the model performance meets the preset accuracy requirements. If so, a security warning model is generated; otherwise, the model parameters are readjusted and supervised fine-tuning training continues until a security warning model that meets the accuracy requirements is generated.

[0011] Optionally, security analysis is performed on real-time time series data using a security early warning model, including: cutting real-time time series data into fixed-length segments using a preset sliding window, mapping each segment to a high-dimensional vector to form a real-time sequence; inputting the real-time sequence into the security early warning model for inference calculation, and outputting a predicted sequence; and performing security analysis based on the predicted sequence.

[0012] Optionally, security analysis can be performed based on the predicted sequence, including: comparing the predicted values ​​of each parameter in the predicted sequence with the corresponding thresholds in each time period; if the predicted value is within the threshold range, it is determined that there is no security risk in the target time period; otherwise, it is determined that there is a security risk in the target time period and an early warning signal is triggered.

[0013] Secondly, embodiments of this disclosure also provide an industrial time-series data analysis device, the device comprising: The time series dataset generation module is used to acquire industrial time series data and generate a time series dataset based on the industrial time series data. The time series dataset includes various high-dimensional vectors. The security warning model production module is used to build an architecture model based on Llama, and to perform supervised fine-tuning training of the architecture model using time series datasets to generate a security warning model. The data analysis module is used to acquire real-time time-series data and perform security analysis on the real-time time-series data through a security early warning model.

[0014] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein, when the memory stores a computer program executable by the at least one processor, the computer program is executed by the at least one processor to enable the at least one processor to perform an industrial time-series data analysis method as described in any embodiment of this disclosure.

[0015] Fourthly, embodiments of this disclosure provide a computer storage medium having a computer program stored thereon, which, when executed by a processor, implements an industrial time-series data analysis method as described in any embodiment of this disclosure.

[0016] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of the present invention, nor is it intended to limit the scope of the invention. Other features of the invention will become readily apparent from the following description.

[0017] Therefore, the present invention has the following beneficial effects: 1. Transforming continuous industrial time-series data into a dataset containing high-dimensional vectors can transform the industrial time-series prediction problem into a sequence generation problem similar to natural language processing, providing a suitable data format for subsequent model processing. At the same time, high-dimensional vectors can effectively carry the complex features in industrial time-series data, providing rich raw information support for model learning.

[0018] 2. The architecture model based on Llama has a self-attention mechanism that can capture global dependencies at different time points in industrial time series data in parallel, solving the shortcomings of traditional models in handling long-term time series dependencies. Through supervised fine-tuning of time series datasets, the model can be specialized for industrial safety parameter prediction tasks, significantly improving the prediction accuracy of the model in target industrial scenarios, and forming a dedicated model adapted to the needs of industrial safety early warning.

[0019] 3. By analyzing real-time data through a safety early warning model, the system can quickly output safety analysis results, enabling real-time monitoring of the operational status of industrial systems and identification of potential risks. This provides timely and effective decision-making support for industrial safety early warning, ensuring industrial production safety. Attached Figure Description

[0020] Figure 1This is a flowchart of an industrial time-series data analysis method provided in Embodiment 1 of the present invention; Figure 2 This is a schematic diagram of an industrial time-series data analysis device according to Embodiment 2 of the present invention; Figure 3 This is a schematic diagram of the structure of an electronic device provided according to Embodiment 3 of the present invention. Detailed Implementation

[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, and not all embodiments. The components of the embodiments of this application described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0022] In the description of this application, it should be noted that the terms "inner" and "outer," etc., indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings, or the orientation or positional relationship commonly used when the product is in use. They are used only for the convenience of describing this application and for simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application. Furthermore, the terms "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0023] In the description of this application, it should also be noted that, unless otherwise expressly specified and limited, the terms "setup" and "connection" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection of two components. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0024] The present invention will now be described in detail with reference to the accompanying drawings. Example 1

[0025] Figure 1 This document provides a flowchart of an industrial time-series data analysis method according to Embodiment 1 of the present invention. This embodiment is applicable to industrial safety early warning scenarios. The method can be executed by the industrial time-series data analysis device provided in this disclosure. This device can be implemented in software and / or hardware and is generally integrated into a computer device. The method of this disclosure specifically includes: S110: Acquire industrial time-series data and generate a time-series dataset based on the industrial time-series data, wherein the time-series dataset includes various high-dimensional vectors.

[0026] Industrial time-series data refers to continuous data generated over time from industrial production and operation processes. In this application, it specifically refers to time-series data related to safety parameters from multiple industrial scenarios such as hydrogen energy bases. A time-series dataset is a dataset generated from acquired industrial time-series data through processing, containing various high-dimensional vectors. High-dimensional vectors refer to time-series data tokens, which are vectors obtained by mapping each segment after cutting continuous industrial time-series data into fixed-length fragments. Their function is to transform the time-series prediction problem into a sequence generation problem similar to natural language processing.

[0027] Optionally, industrial time-series data can be acquired, including: identifying each target hydrogen energy base and acquiring standardized data exported from the local data terminal of each target hydrogen energy base; removing outliers from the standardized data to generate industrial time-series data.

[0028] Among them, the target hydrogen energy base refers to the hydrogen energy bases selected from numerous industrial scenarios that require safety parameter prediction and safety early warning as data collection objects. The local data terminal is the equipment in the hydrogen energy base used to store and record various data during the production and operation process. The exported standardized data refers to data processed in a unified format, whose format meets the requirements of subsequent data processing and model input, avoiding obstacles to subsequent steps due to inconsistent data formats. The standardized data contains various time-series information related to safety during the operation of the hydrogen energy base and is the initial data for generating industrial time-series data.

[0029] It is known that in actual industrial production, the data recorded by local data terminals may contain abnormal values ​​due to factors such as equipment failure and external interference, which may affect the accuracy of subsequent model training. In order to ensure the quality of industrial time series data used for model training, this application will identify and remove abnormal data in the standardized data, and finally obtain industrial time series data that meets the quality requirements and can accurately reflect the operating status of the hydrogen energy base.

[0030] Optionally, a time-series dataset is generated based on industrial time-series data, including: cutting the industrial time-series data into fixed-length segments using a preset sliding window and mapping each segment to a high-dimensional vector; and integrating the high-dimensional vectors to generate a time-series dataset for model training.

[0031] It's important to note that the Transformer architecture typically handles discrete sequential data in natural language processing, while industrial time-series data is continuous. Therefore, a sliding window is needed to transform the continuous time-series data into discrete segments of uniform length. The pre-defined sliding window refers to setting the window length and sliding step size in advance. The window length determines the number of time-series data points contained in each segment, and the sliding step size determines the distance the window moves each time. This sliding segmentation ensures that multiple segments of consistent length with complete information are extracted from the continuous industrial time-series data, preparing for subsequent mapping to high-dimensional vectors. The controller then maps each segment to a high-dimensional vector. Vector representation better captures the feature information of the time-series data within a segment, while also meeting the Transformer model's requirements for processing high-dimensional inputs. Through this mapping process, the temporal trends, numerical relationships, and other features contained in each fixed-length segment are transformed into a high-dimensional numerical form that the model can recognize and process, converting the originally continuous time-series information into a discrete vector form that meets the input requirements of the Transformer architecture. Finally, the high-dimensional vectors are integrated to generate a time-series dataset for model training, providing a structured and standardized source of training data for the model. Since this application will subsequently perform supervised fine-tuning of the Llama-based Transformer model, and model training requires a large amount of ordered and uniformly formatted data, integrating the previously obtained high-dimensional vectors according to chronological order or data logic to form a complete time-series dataset ensures that the dataset contains sufficient training samples while maintaining the temporal correlation of the time-series data itself, meeting the model's requirements for processing the sequential nature of time-series data. Furthermore, this dataset will be used for supervised fine-tuning of the model in hydrogen-related scenarios, allowing the model to learn the characteristic patterns of industrial time-series data in hydrogen-related scenarios during training, thereby improving the model's prediction accuracy in these scenarios.

[0032] Optionally, a dynamic dimensionality adjustment mechanism can be added when mapping each segment to a high-dimensional vector. This means that the dimensionality of the high-dimensional vector is automatically adjusted based on the feature complexity of the industrial time-series data. For example, hydrogen energy base data contains multiple parameters such as hydrogen concentration, pressure, and temperature, resulting in high feature complexity; while single-device vibration data has low feature complexity. When feature complexity is high, the vector dimension is increased to ensure complete representation of feature information; when feature complexity is low, the vector dimension is reduced to decrease the computational load of the model. Simultaneously, feature weight labels are added to the high-dimensional vectors, assigning higher weights to the vector dimensions corresponding to key safety parameters. This allows the model to prioritize core features during training, improving the model's predictive sensitivity for key safety indicators.

[0033] S120: Build an architecture model based on Llama, use time series datasets to supervise and fine-tune the architecture model, and generate a security early warning model.

[0034] The Llama-based architecture model refers to a model built using the Llama model as its foundation and employing the Transformer architecture. Leveraging the powerful self-attention mechanism of the Transformer architecture, this model can capture global dependencies between different time points in industrial time-series data in parallel, laying the architectural foundation for subsequent industrial time-series prediction. It requires supervised fine-tuning with a time-series dataset before it can be used for safety early warning. Supervised fine-tuning training refers to the supervised training process of the Llama-based architecture model using a time-series dataset containing high-dimensional vectors. During training, the pre-trained Llama-based model is first transformed into a time-series tokenized model, then training is conducted using the time-series dataset to set prediction task objectives. Performance is evaluated using a validation set; if the accuracy requirements are met, training is complete. The aim is to specialize the model for industrial safety parameters, such as predicting safety parameters in hydrogen-related scenarios, thereby improving the model's prediction accuracy in this field. The safety early warning model refers to the model generated after supervised fine-tuning training with a time-series dataset on the Llama-based architecture model. The safety early warning model has the ability to process industrial time-series data and predict future sequences. It can receive real-time time-series data and perform safety analysis, ultimately providing support for industrial safety early warning. In this application, it can be used for safety early warning in industrial scenarios such as hydrogen-related scenarios.

[0035] Optionally, an architecture model based on Llama is constructed, including: using a converter architecture base model based on Llama to generate an initial model framework adapted to industrial time series forecasting tasks; obtaining time series forecasting components, wherein the time series forecasting components include position encoding, multi-head self-attention layer, feedforward neural network and residual connection and normalization module; and combining the time series forecasting components with the initial model framework to generate an architecture model.

[0036] Llama is a large, pre-trained model in the field of natural language processing (NLP) with powerful sequence processing capabilities. This application leverages the advantages of the Transformer architecture to solve industrial time-series prediction problems. Therefore, Llama was chosen as the base model, utilizing its existing pre-trained feature extraction capabilities to reduce the cost of building a model from scratch. Furthermore, since Llama was originally designed for NLP tasks, it needs to be adapted first to give the initial model framework the basic ability to process time-series data, rather than directly adopting the structure of NLP scenarios. This lays the groundwork for later integration with time-series prediction components, ensuring that the initial framework initially matches the needs of industrial time-series prediction tasks. Additionally, because the Transformer architecture itself lacks the ability to capture temporal order when processing sequence data, and the temporal correlation of industrial time-series data is crucial, a position encoding module is needed to add temporal position information to each input time-series segment, enabling the model to recognize the temporal order of the data. The multi-head self-attention layer solves the gradient vanishing or exploding problem in traditional LSTM and other models when handling ultra-long temporal dependencies by calculating attention weights between data at different time points and capturing global dependencies in parallel. The feedforward neural network further nonlinearly transforms the features output by the multi-head self-attention layer, enhancing the model's ability to fit complex features of temporal data. The residual connection and normalization module alleviates the training difficulties caused by the increase in model depth by passing gradients through residual connections and stabilizing the training process through normalization.

[0037] In summary, while the initial Llama framework possesses the basic feature extraction capabilities for large models, it lacks specific components for processing time-series data, thus failing to directly meet the needs of industrial time-series forecasting. By integrating components such as position encoding and multi-head self-attention layers with the Llama framework, the generalization ability of the Llama model based on massive pre-training is preserved, while also endowing the model with the ability to capture the temporal order and global dependencies of time-series data, enabling the model to adapt to industrial time-series forecasting tasks.

[0038] Optionally, supervised fine-tuning training of the architecture model is performed using a time-series dataset to generate a security warning model. This includes: dividing the time-series dataset into a training set and a validation set; using the training set to perform supervised fine-tuning training of the architecture model to generate a pre-fine-tuned model; verifying the performance of the pre-fine-tuned model using the validation set and determining whether the model performance meets the preset accuracy requirements. If so, a security warning model is generated; otherwise, the model parameters are readjusted and supervised fine-tuning training continues until a security warning model that meets the accuracy requirements is generated.

[0039] The training set provides ample learning samples for the architecture model, allowing it to learn the characteristic patterns of time-series data in hydrogen-related scenarios, such as the trends and correlations of safety parameters at hydrogen energy bases over time. The validation set, on the other hand, evaluates the model's generalization ability during or after training, preventing overfitting due to excessive learning of training data features and ensuring good predictive performance even on unseen hydrogen-related scenario time-series data. Furthermore, while the Llama model possesses powerful feature extraction capabilities after pre-training on massive amounts of general data, it is not well-suited for hydrogen-related scenario time-series prediction tasks. Therefore, supervised fine-tuning is necessary to adapt the model to this scenario. During training, high-dimensional vectors of hydrogen-related time series data from the training set are input into the model. The prediction of future safety parameter time series is used as the supervision target. The model parameters are continuously adjusted through backpropagation, so that the model gradually learns the time series dependencies of safety parameters in hydrogen-related scenarios, such as the correlation between certain safety indicators of hydrogen energy bases at different time points. This allows the model to change from a general pre-trained state to a state that is initially adapted to the hydrogen-related safety prediction task, resulting in a pre-fine-tuned model.

[0040] In one specific implementation, when performing performance verification, high-dimensional vectors of hydrogen-related time series data from the validation set can be input into the initially fine-tuned model to obtain the predicted sequence output by the model. The predicted sequence is then compared with the actual time series data from the validation set. The model's predictive performance is measured using preset accuracy evaluation metrics, such as prediction error rate and accuracy, and must meet the accuracy requirements for industrial safety early warning. If the model performance meets the preset accuracy requirements, it indicates that the model has fully learned the characteristics of time series data in hydrogen-related scenarios and can accurately predict safety parameters in that scenario. In this case, it can be identified as a safety early warning model for subsequent real-time safety analysis. However, if the model performance does not meet the accuracy requirements, it may be due to reasons such as the model parameters not being optimally adjusted or insufficient features learned during training. In this case, the model parameters need to be readjusted, and supervised fine-tuning training using the training set should be performed again. This process is repeated until the model's performance on the validation set reaches the preset accuracy requirements, ultimately generating a reliable safety early warning model for hydrogen-related scenarios.

[0041] Optionally, when using the training set to perform supervised fine-tuning training of the architecture model, an adaptive learning rate algorithm based on the rate of change of loss can be employed. When the model's training loss decreases rapidly, the learning rate is automatically reduced to avoid overfitting. When the loss decreases stagnates, the learning rate is automatically increased to help the model escape local optima. Compared to a fixed learning rate, this accelerates model convergence and improves the final model accuracy.

[0042] S130: Acquire real-time time series data and perform security analysis on the real-time time series data through a security early warning model.

[0043] Real-time time-series data refers to continuous data that is generated in real time during industrial production and operation and changes dynamically over time. This data is consistent with the source scenarios of the industrial time-series data used for training and is primarily used as input to the safety early warning model for real-time safety analysis, enabling timely industrial safety warnings. Safety analysis involves inputting the acquired real-time time-series data into the safety early warning model. The model first tokenizes the real-time time-series data and then outputs a future prediction sequence through internal inference. Based on this prediction sequence, the model determines the safety status of the industrial scenario.

[0044] Optionally, a real-time sequence preprocessing acceleration module can be added before inputting real-time sequences into the safety early warning model for inference calculations. This module employs hardware-level parallel computing to accelerate the segmentation and vector mapping processes of real-time time-series data. It also includes a pre-defined feature cache pool, allowing features highly similar to historical data in real-time data to be directly retrieved from the cache, eliminating the need for repeated mapping calculations. This ensures the model can quickly respond to real-time data, meeting the real-time requirements of industrial safety early warning systems.

[0045] Optionally, security analysis is performed on real-time time series data using a security early warning model, including: cutting real-time time series data into fixed-length segments using a preset sliding window, mapping each segment to a high-dimensional vector to form a real-time sequence; inputting the real-time sequence into the security early warning model for inference calculation, and outputting a predicted sequence; and performing security analysis based on the predicted sequence.

[0046] Specifically, the controller can use a preset sliding window to cut real-time time series data into fixed-length segments and map each segment into a high-dimensional vector to form a real-time sequence. This step is consistent with the tokenization processing logic when generating time series datasets, both of which are to adapt to the input requirements of the security warning model.

[0047] Furthermore, the safety early warning model is based on Llama and fine-tuned under the supervision of hydrogen-related time-series data. It contains components such as a multi-head self-attention layer and position encoding. After inputting the real-time sequence, the position encoding adds time and position information to each high-dimensional vector to ensure that the model captures the temporal order of the real-time data. The multi-head self-attention layer calculates the dependencies between different time segments in the real-time sequence in parallel to accurately identify the temporal features in the real-time data. Then, the features are nonlinearly transformed by a feedforward neural network. Combined with the prediction rules of safety parameters in hydrogen-related scenarios learned during model fine-tuning, inference calculations are performed to finally output the predicted sequence of safety parameters for a period of time in the future.

[0048] Optionally, security analysis can be performed based on the predicted sequence, including: comparing the predicted values ​​of each parameter in the predicted sequence with the corresponding thresholds in each time period; if the predicted value is within the threshold range, it is determined that there is no security risk in the target time period; otherwise, it is determined that there is a security risk in the target time period and an early warning signal is triggered.

[0049] Specifically, the predicted sequence output by the safety early warning model is essentially a sequence of key safety parameters for hydrogen-related scenarios over a future period, such as pressure, concentration, and temperature—parameters related to hydrogen safety—changing over time. Each time period corresponds to a predicted result for a specific parameter. The corresponding thresholds are pre-set based on safety standards for hydrogen-related industrial scenarios, equipment operating limits, and historical safety accident data. Different safety parameters correspond to different safety critical values. Time-by-time comparison ensures that parameter changes in any time interval are not missed, avoiding misjudgments of safety risks due to time jumps.

[0050] Then, the controller will determine whether there is a safety risk during the target period based on the comparison results. If the predicted value is within the threshold range, it is determined that there is no safety risk during the target period. This indicates that the operating status of the hydrogen-related scenario during this period meets safety standards, and all key parameters have not exceeded the safety range that the equipment or process can withstand, and no safety accident will occur. However, if the predicted value exceeds the threshold range, it is determined that there is a safety risk during the target period and an early warning signal is triggered. Triggering an early warning signal can promptly inform relevant personnel of the potential risk, buy time for subsequent emergency response measures, and prevent a safety accident from occurring.

[0051] The technical solution of this invention transforms continuous industrial time-series data into a dataset containing high-dimensional vectors, providing a suitable data format for subsequent model processing and rich raw information support for model learning. An architecture model based on Llama is built, which can capture global dependencies at different time points in industrial time-series data in parallel, overcoming the shortcomings of traditional models in handling long-term time-series dependencies. Supervised fine-tuning through time-series datasets allows the model to specialize in industrial safety parameter prediction tasks, improving the prediction accuracy in target industrial scenarios. By analyzing real-time data through a safety early warning model, rapid output of safety analysis results is achieved, enabling real-time monitoring of the operational status of industrial systems and identification of potential risks. This provides timely and effective decision-making basis for industrial safety early warning, ensuring industrial production safety. Example 2

[0052] Figure 2 This is a schematic diagram of an industrial time-series data analysis device according to Embodiment 5 of the present invention. This device can be implemented using software and / or hardware, and is generally integrated into an electronic device that performs the method. For example... Figure 2As shown, the device includes: a time series dataset generation module 210, used to acquire industrial time series data and generate a time series dataset based on the industrial time series data, wherein the time series dataset includes various high-dimensional vectors; The safety warning model production module 220 is used to build an architecture model based on Llama, and to perform supervised fine-tuning training on the architecture model using time series datasets to generate a safety warning model. The data analysis module 230 is used to acquire real-time time series data and perform security analysis on the real-time time series data through a security early warning model.

[0053] Optionally, the time series dataset generation module 210 specifically includes: an industrial time series data acquisition unit, used to: determine each target hydrogen energy base and acquire standardized data exported from the local data terminal of each target hydrogen energy base; remove outliers from the standardized data to generate industrial time series data.

[0054] Optionally, the time series dataset generation module 210 specifically includes: a time series dataset generation unit, used to: cut industrial time series data into fixed-length segments through a preset sliding window, and map each segment to a high-dimensional vector; integrate the high-dimensional vectors to generate a time series dataset for model training.

[0055] Optionally, the safety early warning model production module 220 specifically includes: an architecture model building unit, used to: generate an initial model framework adapted to industrial time series prediction tasks by adopting a converter architecture base model based on Llama; obtain time series prediction components, wherein the time series prediction components include position encoding, multi-head self-attention layer, feedforward neural network and residual connection and normalization module; and combine the time series prediction components with the initial model framework to generate an architecture model.

[0056] Optionally, the safety warning model production module 220 specifically includes: a safety warning model generation unit, used to: divide the time series dataset into a training set and a validation set; use the training set to perform supervised fine-tuning training on the architecture model to generate a pre-fine-tuned model; perform performance verification on the pre-fine-tuned model through the validation set, and determine whether the model performance meets the preset accuracy requirements. If so, generate a safety warning model; otherwise, readjust the model parameters and continue supervised fine-tuning training until a safety warning model that meets the accuracy requirements is generated.

[0057] Optionally, the data analysis module 230 is specifically used for: cutting real-time time series data into fixed-length segments through a preset sliding window, mapping each segment to a high-dimensional vector to form a real-time sequence; inputting the real-time sequence into a security early warning model for inference calculation, and outputting a predicted sequence; and performing security analysis based on the predicted sequence.

[0058] Optionally, the data analysis module 230 specifically includes: a security analysis subunit, used to: compare the predicted values ​​of each parameter in the prediction sequence with the corresponding thresholds in each time period; if the predicted value is within the threshold range, determine that there is no security risk in the target time period; otherwise, determine that there is a security risk in the target time period and trigger an early warning signal.

[0059] The technical solution of this invention transforms continuous industrial time-series data into a dataset containing high-dimensional vectors, providing a suitable data format for subsequent model processing and rich raw information support for model learning. An architecture model based on Llama is built, which can capture global dependencies at different time points in industrial time-series data in parallel, overcoming the shortcomings of traditional models in handling long-term time-series dependencies. Supervised fine-tuning through time-series datasets allows the model to specialize in industrial safety parameter prediction tasks, improving the prediction accuracy in target industrial scenarios. By analyzing real-time data through a safety early warning model, rapid output of safety analysis results is achieved, enabling real-time monitoring of the operational status of industrial systems and identification of potential risks. This provides timely and effective decision-making basis for industrial safety early warning, ensuring industrial production safety.

[0060] The industrial time-series data analysis device provided in this embodiment of the invention can execute an industrial time-series data analysis method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method execution. Example 3

[0061] Figure 3 A schematic diagram of an electronic device 10 that can be used to implement embodiments of the present invention is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices (e.g., helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0062] like Figure 3As shown, the electronic device 10 includes at least one processor 11 and a memory, such as a read-only memory (ROM) 12 or a random access memory (RAM) 13, communicatively connected to the at least one processor 11. The memory stores computer programs executable by the at least one processor. The processor 11 can perform various appropriate actions and processes based on the computer program stored in the ROM 12 or loaded from storage unit 18 into the RAM 13. The RAM 13 can also store various programs and data required for the operation of the electronic device 10. The processor 11, ROM 12, and RAM 13 are interconnected via a bus 14. An input / output (I / O) interface 15 is also connected to the bus 14.

[0063] Multiple components in electronic device 10 are connected to I / O interface 15, including: input unit 16, such as keyboard, mouse, etc.; output unit 17, such as various types of displays, speakers, etc.; storage unit 18, such as disk, optical disk, etc.; and communication unit 19, such as network card, modem, wireless transceiver, etc. Communication unit 19 allows electronic device 10 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0064] Processor 11 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 11 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various processors running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. Processor 11 performs the various methods and processes described above, such as an industrial time-series data analysis method.

[0065] In some embodiments, an industrial time-series data analysis method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 18. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 10 via ROM 12 and / or communication unit 19. When the computer program is loaded into RAM 13 and executed by processor 11, one or more steps of the industrial time-series data analysis method described above may be performed. Alternatively, in other embodiments, processor 11 may be configured to perform an industrial time-series data analysis method by any other suitable means (e.g., by means of firmware).

[0066] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0067] Computer programs used to implement the methods of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be performed. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0068] In the context of this invention, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination thereof. Alternatively, a computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0069] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0070] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or middleware components (e.g., application servers), or frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0071] A computing system can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a hosting product within the cloud computing service system. It addresses the shortcomings of traditional physical hosts and VPS servers, such as high management difficulty and weak business scalability.

[0072] It should be understood that the various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this invention can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this invention can be achieved, and this is not limited herein.

[0073] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. An industrial time series data analysis method, characterized by, The method comprises the following steps: acquiring industrial time series data, generating a time series data set according to the industrial time series data, wherein each high-dimensional vector is included in the time series data set; building an architecture model based on Llama, and performing supervised fine-tuning training on the architecture model using the time series data set to generate a safety warning model; acquiring real-time time series data, and performing safety analysis on the real-time time series data through the safety warning model.

2. The method of claim 1, wherein, The acquisition of industrial time series data comprises: determining each target hydrogen energy base, and acquiring standardized data derived by a local data terminal of each target hydrogen energy base; performing outlier rejection on the standardized data to generate industrial time series data.

3. The method of claim 1, wherein, The generation of the time series data set according to the industrial time series data comprises: cutting the industrial time series data into fixed-length segments through a preset sliding window, and mapping each segment into a high-dimensional vector; integrating each high-dimensional vector to generate a time series data set for model training.

4. The method of claim 3, wherein, The building of the architecture model based on Llama comprises: adopting a transformer architecture base model based on Llama to generate an initial model framework adapted to an industrial time series prediction task; acquiring a time series prediction component, wherein the time series prediction component comprises position encoding, multi-head self-attention layer, feedforward neural network, and residual connection and normalization module; combining the time series prediction component with the initial model framework to generate an architecture model.

5. The method of claim 1, wherein, The supervised fine-tuning training of the architecture model using the time series data set to generate a safety warning model comprises: dividing the time series data set into a training set and a validation set; performing supervised fine-tuning training on the architecture model using the training set to generate a preliminarily fine-tuned model; performing performance verification on the preliminarily fine-tuned model through the validation set, and determining whether the model performance meets a preset precision requirement, if yes, generating a safety warning model; otherwise, readjusting model parameters and continuing the supervised fine-tuning training until a safety warning model meeting the precision requirement is generated.

6. The method of claim 5, wherein, The safety analysis of the real-time time series data through the safety warning model comprises: cutting the real-time time series data into fixed-length segments through a preset sliding window, and mapping each segment into a high-dimensional vector to form a real-time sequence; inputting the real-time sequence into the safety warning model for inference calculation to output a predicted sequence; performing safety analysis based on the predicted sequence.

7. The method of claim 6, wherein, The safety analysis based on the predicted sequence comprises: comparing the predicted value of each parameter in the predicted sequence with a corresponding threshold value in each time period, and if the predicted value is within the threshold value range, it is determined that there is no safety risk in the target time period; otherwise, it is determined that there is a safety risk in the target time period, and a warning signal is triggered.

8. An industrial time series data analysis apparatus characterized by comprising: The method comprises the following steps: a time series data set generation module is configured to acquire industrial time series data, and generate a time series data set according to the industrial time series data, wherein each high-dimensional vector is included in the time series data set; a safety warning model production module is configured to build an architecture model based on Llama, and perform supervised fine-tuning training on the architecture model using the time series data set to generate a safety warning model; A data analysis module is configured to acquire real-time time-series data and perform safety analysis on the real-time time-series data by using the safety warning model.

9. An electronic device, comprising: The electronic device includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores a computer program capable of being executed by the at least one processor, and the computer program is executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A computer storage medium, characterized in that, The computer storage medium stores computer instructions for causing the processor to implement the method of any one of claims 1-7 when executed.

Images