Intelligent control method and device based on processing process management and chip

Abstract

The invention relates to the technical field of intelligent control, and discloses an intelligent control method, device and chip based on processing process management, and the method comprises the steps: setting an anomaly detection instruction through an intelligent control chip, writing the anomaly detection instruction into a control state register, and creating an anomaly detection infrastructure; constructing a data acquisition protocol, and executing feature extraction operation on the processing process data to obtain multi-dimensional feature training data; constructing first anomaly detection sub-models of a plurality of processing stages, and writing the pre-training parameters into the first anomaly detection sub-models to obtain a plurality of initialized anomaly detection models; deploying the plurality of initialized anomaly detection models to the intelligent control chip for model parallel training to obtain a plurality of second anomaly detection sub-models; and calculating sub-model weight coefficients of the plurality of second anomaly detection sub-models, and performing optimization integration and incremental learning to obtain a machining anomaly detection model, thereby ensuring continuous optimization and performance improvement of the anomaly detection model for machining process management.

Description

Technical Field

[0001] The present invention relates to the field of intelligent control technology, and in particular to an intelligent control method, device and chip based on machining process management. Background Art

[0002] With the development of intelligent industrial manufacturing, processing management has put forward higher requirements for the accuracy and real-time performance of anomaly detection. Traditional anomaly detection methods mainly rely on a single monitoring model, which is difficult to cope with complex and changing processing environments, and the computing resource utilization efficiency is low, which cannot meet the needs of real-time monitoring.

[0003] In the actual production process, due to the complexity of the processing steps and the diversity of process parameters, it is difficult for a single detection model to fully capture the abnormal characteristics of different processing stages. At the same time, the existing detection methods lack an effective parameter sharing mechanism, resulting in low model training efficiency and difficulty in adapting to the dynamically changing processing environment. Summary of the invention

[0004] The present invention provides an intelligent control method, device and chip based on machining process management, which ensures continuous optimization and performance improvement of anomaly detection models in machining process management.

[0005] In a first aspect, the present invention provides an intelligent control method based on machining process management, the intelligent control method based on machining process management comprising: Set anomaly detection instructions through the intelligent control chip and write them into the control status register to create anomaly detection infrastructure; Constructing a data collection protocol according to the anomaly detection infrastructure, and performing feature extraction operations on the processing data to obtain multi-dimensional feature training data; Building a first anomaly detection sub-model of multiple processing stages based on the multi-dimensional feature training data, writing pre-trained parameters into the first anomaly detection sub-model, and obtaining multiple initialized anomaly detection models; Deploying the multiple initialized anomaly detection models to the intelligent control chip for model parallel training to obtain multiple second anomaly detection sub-models; The sub-model weight coefficients of the plurality of second anomaly detection sub-models are calculated, and optimization integration and incremental learning are performed to obtain a processing anomaly detection model.

[0006] In a second aspect, the present invention provides an intelligent control device based on machining process management, the intelligent control device based on machining process management comprising: Create a module to set anomaly detection instructions through the intelligent control chip and write to the control status register to create anomaly detection infrastructure; A feature extraction module is used to construct a data acquisition protocol according to the anomaly detection infrastructure and perform feature extraction operations on the processing data to obtain multi-dimensional feature training data; A pre-training module, used for constructing a first anomaly detection sub-model of multiple processing stages based on the multi-dimensional feature training data, writing pre-training parameters into the first anomaly detection sub-model, and obtaining a plurality of initialized anomaly detection models; A parallel training module, used for deploying the multiple initialized anomaly detection models to the intelligent control chip for model parallel training to obtain multiple second anomaly detection sub-models; The optimization integration module is used to calculate the sub-model weight coefficients of the multiple second abnormality detection sub-models, and perform optimization integration and incremental learning to obtain a processing abnormality detection model.

[0007] A third aspect of the present invention provides a chip, in which instructions are stored, and when the chip is run on a computer, the computer is enabled to execute the above-mentioned intelligent control method based on machining process management.

[0008] In the technical solution provided by the present invention, by setting anomaly detection instructions in the intelligent control chip and configuring anomaly detection circuits, an infrastructure supporting two detection modes, data redundancy and time redundancy, is constructed, thereby improving the reliability and flexibility of anomaly detection. A multi-stage anomaly detection sub-model is constructed using multi-dimensional feature training data, and a shared parameter layer is set, thereby reducing the redundancy of model parameters and optimizing the utilization efficiency of computing resources. The training strategies of data parallelism and model parallelism are introduced, and the anomaly detection model is given a stronger generalization ability by dynamically adjusting the training parameters and the cross-validation mechanism. Based on the integrated strategy of resource allocation and energy efficiency optimization, combined with the dynamic adjustment mechanism of weight coefficients, the reasonable allocation of computing resources and the optimization of energy efficiency are realized. Through incremental learning and parameter update mechanisms, a complete model optimization and performance evaluation system is established, thereby ensuring the continuous optimization and performance improvement of the anomaly detection model. The hierarchical storage and parameter index management method is adopted to improve the access efficiency of model parameters and facilitate the dynamic update and maintenance of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0010] Figure 1 A schematic diagram of the steps of an intelligent control method based on machining process management in an embodiment of the present invention; Figure 2It is a schematic diagram of the structure of an intelligent control device based on machining process management in an embodiment of the present invention. DETAILED DESCRIPTION

[0011] Embodiments of the present invention provide an intelligent control method, device and chip based on processing management. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0012] For ease of understanding, the specific process of the embodiment of the present invention is described below. Figure 1 , an embodiment of the intelligent control method based on processing management in the embodiment of the present invention includes: Step S1, setting an abnormality detection instruction through the intelligent control chip and writing it into the control status register to create an abnormality detection infrastructure; It is understandable that the execution subject of the present invention may be an intelligent control device based on processing management, or a terminal or a server, which is not limited here. The embodiment of the present invention is described by taking the server as the execution subject as an example.

[0013] Specifically, an abnormality detection instruction set and a control status register address space are added to the intelligent control chip, and the instruction system architecture of the existing chip is expanded to support the functional requirements of abnormality detection. By expanding the instruction system architecture, a new instruction encoding is created for the abnormality detection task, the function of each instruction is clarified, and sufficient address space is allocated to the control status register to accommodate all necessary configuration data. After the instruction system architecture is expanded, the data redundancy detection threshold and the time redundancy detection threshold are set according to the new architecture to form an abnormality detection parameter matrix. The generation of this matrix needs to comprehensively consider the data characteristics in the processing process and the distribution law of potential abnormalities to ensure that the set threshold has practical significance and detection capability. The abnormality detection parameter matrix is ​​written into the control status register to form abnormality detection configuration data. These configuration data not only cover all detection thresholds in the parameter matrix, but also include the definition of parameters such as detection task priority and enabling conditions in the control status register. In order to adapt to different processing scenarios and requirements, the abnormality detection configuration data is divided into operation modes, specifically divided into data redundancy mode and time redundancy mode. In the data redundancy mode, attention is paid to the consistency of the processing data between different sampling points. By configuring multiple sets of data comparison logic, the real-time collected data is compared and analyzed to detect the existing abnormalities. In the time redundancy mode, the focus is on the time continuity of the processing process. By configuring the timing comparison logic, whether there is an unreasonable time deviation between the operation and the result is detected. After completing the division of the operation mode, the operation mode configuration parameters are generated based on these modes, and the data comparison unit and the result output unit are configured. The core function of the data comparison unit is to perform real-time comparison of the data collected during the processing process, including comparing the differences between different sampling points of the same data through hardware logic, or comparing the redundant relationship of different data channels. The task of the result output unit is to process and output the results generated by the comparison logic. Whether it is through the signal feedback control chip or through the data interface to the upper system, its design must meet the requirements of real-time and accuracy. The data comparison unit and the result output unit are cascaded to form an abnormality detection circuit. The abnormality detection circuit is functionally verified. The process of functional verification includes simulating various possible abnormal situations in different processing scenarios to detect whether the abnormality detection circuit can accurately identify and output the expected results. The results of the functional verification are fed back to the intelligent control chip to optimize the abnormality detection instruction set and the configuration parameters of the abnormality detection circuit. If the verification results show that the detection performance is insufficient or there are misjudgments, the relevant parameters or logic are adjusted, such as redefining the detection threshold, optimizing the comparison logic, or expanding the coverage of the operation mode. Through the above iterative optimization process, the anomaly detection infrastructure is finally formed.

[0014] Step S2: construct a data collection protocol according to the anomaly detection infrastructure, and perform feature extraction operations on the processing data to obtain multi-dimensional feature training data; Specifically, the parameters related to data sampling are set based on the anomaly detection infrastructure, including key indicators such as sampling frequency, sampling time window length and sampling accuracy. By analyzing the dynamic change characteristics of the processing process, the sampling frequency is reasonably set to ensure that the collected data has sufficient time resolution, while the length of the sampling time window determines the required observation time range, and the sampling accuracy is related to the balance between sensor capabilities and data volume. After completing the parameter setting, the sampling parameters are written into the data acquisition control unit in the intelligent control chip to form the basic configuration of data acquisition. Constraints are set for the basic configuration of data acquisition. By combining the actual needs and process characteristics of the processing process, the scope, frequency and accuracy of data acquisition are graded and constrained, and data acquisition levels are divided for different types of processing parameters. High-priority key parameters, such as temperature, pressure or speed, require higher sampling frequency and accuracy, while secondary parameters adopt lower sampling requirements to reduce system load. After completing the acquisition level division, a data filtering threshold is set for each acquisition level to eliminate abnormal values ​​or noise data that appear, ensuring that the collected data has high quality and availability. These constraints and level divisions are finally integrated into the data acquisition protocol, which clarifies the acquisition method, priority and preliminary rules for data processing of various parameters. The data acquisition protocol is imported into the data acquisition module of the intelligent control chip, and the multi-dimensional sampling of the processing data is completed through the intelligent control chip to form a preliminary raw data matrix. The implementation of multi-dimensional sampling includes the simultaneous acquisition of data signals from multiple sensors, such as temperature sensors, vibration sensors, and flow meters, to ensure comprehensive recording of various dynamic information of the processing process. The raw data matrix is ​​subjected to denoising and standardization processing to obtain preprocessed data. Denoising processing uses filtering or signal smoothing algorithms to eliminate high-frequency noise or sudden abnormal values ​​in the data, while standardization processing unifies the data into a dimensional range that is easy to compare, such as mapping all parameters to the same order of magnitude through normalization or Z-Score conversion. The time domain features, frequency domain features, and statistical features of the preprocessed data are extracted respectively. Time domain feature extraction focuses on the change pattern of data in the time dimension, such as calculating the average, maximum, minimum, mean square error, etc., to capture the amplitude characteristics and change trends of the signal. Frequency domain feature extraction uses Fourier transform or wavelet transform to convert time signals into frequency signals, from which features such as spectrum density, main frequency, and frequency band energy are extracted. It is suitable for analyzing the dynamic characteristics of vibration or periodic change data. Statistical feature extraction focuses on the distribution characteristics of data, such as skewness, kurtosis, and probability density distribution, to help identify whether the signal has abnormal distribution or atypical change patterns. Time domain features, frequency domain features, and statistical features are combined to form multi-dimensional feature training data.

[0015] Step S3, constructing a first anomaly detection sub-model of multiple processing stages based on the multi-dimensional feature training data, writing the pre-training parameters into the first anomaly detection sub-model, and obtaining a plurality of initialized anomaly detection models; Specifically, according to the order of the processing procedures, the multi-dimensional feature training data is reasonably divided, and the data of different stages are assigned to the corresponding processing stage data sets to form multiple processing stage feature data sets. Consider the uniqueness of each processing stage, for example, some processes focus on the changes in pressure and temperature, while other processes focus on the monitoring of vibration signals and processing time. By allocating multi-dimensional feature data in a targeted manner, it is ensured that the anomaly detection model constructed subsequently can be optimized according to the characteristics of each processing stage. In the process of data division, the integrity and balance of the data set are ensured to avoid the degradation of model performance due to uneven data distribution or excessively skewed feature distribution. The structure of the sub-model is determined for each processing stage feature data based on the computing resources of the intelligent control chip. According to the complexity and computing requirements of the feature data of each stage, the number of network layers and neurons of the sub-model are dynamically adjusted. For processing stages with higher feature dimensions or faster dynamic changes, a deeper network structure is designed to capture complex nonlinear relationships; while for more stable stages, a shallower network structure is selected to reduce computing overhead. Through resource-adaptive sub-model design, the computing power of the intelligent control chip is maximized, while ensuring that the structure of each sub-model adapts to actual needs, and multiple first anomaly detection sub-models are generated. Parameter space planning is performed for multiple first anomaly detection sub-models, and the parameter space is divided into an independent parameter area and a shared parameter area. The independent parameter area is responsible for processing the specific features of each processing stage, while the shared parameter area is used to capture the correlation between adjacent processing stages. A parameter sharing channel is established between the first anomaly detection sub-models of adjacent processing stages, so that these stages can effectively collaborate in the parameter space. Through this design, the knowledge of the previous stage is partially transferred to the next stage by utilizing the continuity of the processing steps, thereby improving the overall efficiency and accuracy of anomaly detection. Based on the partition results of the parameter space, a parameter space distribution map is generated. A shared parameter layer is constructed according to the parameter space distribution map, and the shared parameter layer is inserted into the network structure of multiple first anomaly detection sub-models. The core function of the shared parameter layer is to establish a parameter transfer matrix to achieve efficient parameter sharing between adjacent stages. The parameter transfer matrix defines the specific transfer method and weight ratio of the shared parameters, such as through a specific linear combination or nonlinear mapping mechanism, so that the shared parameters can retain the characteristics of the previous stage and adapt to the feature distribution of the current stage. The pre-trained parameters of the anomaly detection model with a shared layer are initialized. The goal of pre-training is to provide a good initial state for the model based on the feature distribution of each processing stage and accelerate the subsequent training process. By analyzing the distribution characteristics of the feature data at each stage, the pre-training parameter values ​​of each shared layer and independent parameter area are determined, and the pre-training parameter matrix is ​​generated based on these values. The setting of pre-training parameters is based on empirical rules, data statistical analysis, or the migration results of the pre-training model to ensure that it has strong adaptability to the anomaly detection tasks at each stage.The pre-trained parameter matrix is ​​written into the anomaly detection model with a shared layer to complete the parameter loading and model compilation process. When loading parameters, the independent and shared parts of the parameter matrix are mapped to the corresponding network layers respectively, and the correctness and adaptability of the parameters are verified. During the model compilation process, the calculation graph and execution logic of the model are optimized so that it can run efficiently on the intelligent control chip. After the optimization and loading operations, multiple initialized anomaly detection models are formed.

[0016] Step S4: deploy multiple initialized anomaly detection models to the intelligent control chip for model parallel training to obtain multiple second anomaly detection sub-models; Specifically, the training tasks of multiple initialized anomaly detection models are reasonably divided. According to the amount of training data and computational complexity required by each model, the training data is allocated to different computing units in batches. At the same time, training parameters are set for each computing unit, including initial learning rate, gradient update rule and batch size, to form a data parallel training scheme. Based on the data parallel training scheme, different batches of training data are processed in parallel. Each computing unit independently processes the assigned data and completes the calculation of forward propagation and error back propagation in a local range. After each computing unit completes the training task of the current batch of data, the gradient information of all computing units is aggregated, and the training results of each sub-model are synchronized through the global gradient update method to achieve the consistency and global optimization of the sub-model parameters. In this process, the gradient aggregation algorithm weights and integrates the gradient information of different computing units to ensure that the direction and amplitude of the parameter update can correctly guide the optimization of the model. Through multiple rounds of parallel processing and gradient aggregation, a data parallel training model is formed. The model structure of the data parallel training model is split to make full use of the hardware architecture characteristics of the intelligent control chip. According to the network structure of each sub-model, sub-models with different structures are assigned to independent computing units, so that each sub-model can complete efficient training tasks on independent hardware resources. At the same time, in order to ensure the collaborative work between sub-models, communication links are established between sub-models, and real-time exchange of parameters or intermediate results is achieved through high-speed communication channels. For example, if sub-models in adjacent stages need to share certain parameter update information or gradient feedback, the communication link is required to support low-latency, high-bandwidth data transmission. After the model structure is split and the communication link is established, a model parallel training scheme is formed. Based on the model parallel training scheme, specific training tasks of the sub-model are executed. The learning rate and batch size of each sub-model are dynamically adjusted to adapt to dynamic changes in the training process. For example, when the performance indicators of the model fluctuate at a certain stage, the learning rate is reduced to avoid gradient oscillation, and when the model is close to convergence, the batch size is gradually increased to accelerate the convergence process. At the same time, the performance indicators during the training process, such as training loss, accuracy, and gradient norm, are recorded in real time. By dynamically adjusting the training parameters and monitoring the training process, it is ensured that each sub-model maintains an efficient and stable optimization state during the training process, thereby improving the overall performance of the model. After the sub-model training is completed, the model training data generated during the training process is cross-validated. Cross-validation effectively detects overfitting or underfitting problems by evaluating the performance of the model on an independent data set. According to the cross-validation results, the training parameters are optimized and adjusted, including resetting the learning rate, updating the regularization parameters, and re-dividing the distribution scheme of the training data. The optimized parameters are written into each sub-model to improve the model's anomaly detection ability and generalization performance. Through the optimization and verification process, multiple second anomaly detection sub-models are finally generated.

[0017] Step S5: Calculate sub-model weight coefficients of multiple second anomaly detection sub-models, and perform optimization integration and incremental learning to obtain a processing anomaly detection model.

[0018] Specifically, a resource allocation strategy is formulated according to the operation status data of multiple second anomaly detection sub-models. By monitoring the operation status data of each sub-model, combined with the utilization rate of computing resources and the occupancy rate of storage resources, the real-time load of each sub-model is analyzed, and an adaptive resource weight is allocated to it. In this process, the resource allocation ratio is dynamically adjusted to ensure the efficient utilization of computing resources and the reasonable occupancy of storage resources, and a sub-model resource configuration scheme is generated. Based on the sub-model resource configuration scheme, the computing load of multiple second anomaly detection sub-models is analyzed. By quantitatively analyzing the energy consumption index and computing resource consumption index of each sub-model, its operation efficiency is evaluated. These indicators are input into the energy efficiency evaluation function to calculate the energy efficiency optimization parameters of each sub-model. According to the energy efficiency optimization parameters, the integrated weight calculation of multiple second anomaly detection sub-models is performed. Comprehensively considering the detection accuracy and energy efficiency optimization parameters of the sub-model, these factors are input into the weight calculation matrix, and the weight distribution table of the sub-model is obtained by normalization. Based on the weight distribution table, an optimization integration strategy is established to combine multiple second anomaly detection sub-models into an integrated detection model. According to the order of the processing stages, the sub-models are combined in stages so that the anomaly detection task of each processing stage can be undertaken by the most suitable sub-model. At the same time, in order to improve the dynamic adaptability of the integrated model, a model switching mechanism is established and trigger conditions are set. When there are stage changes in the processing process or the anomaly detection requirements are adjusted, the switching mechanism dynamically selects the optimal sub-model for the detection task according to the trigger conditions to achieve the flexibility and robustness of the integrated model. After the dynamic integrated model is built, it is incrementally learned. By analyzing the feature distribution of the newly added data samples, a parameter update strategy is formulated to enable the model to adapt to new changes in the processing process. In the incremental learning process, the model parameters of the newly added samples are gradually updated, and the performance of the updated model is evaluated in real time to ensure that the model maintains high detection accuracy and generalization ability when processing new data. The introduction of incremental learning enables the dynamic integrated model to have adaptive capabilities, so that it can continuously optimize its detection capabilities as the processing process changes. The updated anomaly detection model is divided into storage units. The structural parameters and weight parameters of the model are written into the fixed storage area and the dynamic storage area respectively to form a clear parameter storage structure. The fixed storage area is used to store the stable model structure, while the dynamic storage area is used to store the weight parameters that are continuously updated with incremental learning. At the same time, by establishing a parameter index table, the required model parameters can be quickly located and accessed to improve the management efficiency of the model. Finally, a complete processing anomaly detection model is generated.

[0019] The model parameters of the dynamic integration model are divided into two categories according to their functions: structural parameters and weight parameters. Structural parameters mainly define the topological structure of the model, such as the number of layers of the neural network, the node connection method, etc. These parameters have a decisive influence on the overall framework of the model. The weight parameters are specific values ​​learned by the model during the training process, which are used to transform and predict the input data. After the parameters are classified, all parameters are sorted according to their importance according to the sub-model weight allocation table to generate a parameter update priority table. This priority table is the core basis for model update. By giving priority to updating parameters that have a greater impact on model performance, the efficiency of incremental learning is significantly improved, and the interference of irrelevant parameter adjustments on model performance is reduced. After completing the parameter sorting, a parameter update strategy is constructed based on the parameter update priority table. The newly added data samples are preprocessed and classified and labeled, and appropriate labels are assigned to each sample according to the characteristics of the processing stage. The labeled data samples are allocated to the corresponding second anomaly detection sub-model according to the processing stage to form an incremental training data set. The construction of this data set needs to consider the balance and representativeness of the sample distribution to ensure that each sub-model can obtain sufficient training data while avoiding excessive bias towards certain specific processing stages. After the data allocation is completed, the incremental training data is input into the corresponding sub-model for parameter update. Based on the incremental training data set, the model parameters are updated layer by layer in the order of parameter importance. During the update process, the change amount of each parameter and the change index of the model performance are recorded to comprehensively evaluate the impact of parameter adjustment on model performance. For example, the performance change is quantified by calculating the detection accuracy, precision, recall rate and other indicators of the updated model, and the change in resource consumption is recorded to evaluate the calculation cost of the update. These records constitute the parameter update results. After the parameter update is completed, the parameter update results are evaluated for performance. The performance evaluation calculates the difference in detection accuracy before and after the update to determine whether the model achieves the expected effect on the newly added data. And evaluate the changes in resource consumption of the updated model, such as computing resource usage, storage resource occupancy, and runtime delay, to ensure that the model update does not cause excessive resource overhead while improving performance. Based on the model evaluation results, parameter rollback judgment is performed. By comparing the performance evaluation index with the preset threshold, it is identified which parameter adjustments cause performance degradation. If it is found that the update of some parameters makes the detection performance of the model lower than the expected threshold, these parameters are rolled back to restore them to the state before the update. The implementation of parameter rollback combines the parameter changes and performance change indicators saved in the historical records to ensure that the rollback operation can accurately locate the problematic parameters and effectively solve the performance degradation problem. After completing the parameter rollback, a parameter optimization plan is generated, and the optimized parameters are written into the dynamic integration model to improve the stability and adaptability of the model. Through the above process, an updated anomaly detection model is finally obtained.The model can achieve continuous performance improvement with the addition of new data samples, and maintain high reliability and resource utilization with the support of performance evaluation and parameter optimization.

[0020] In the embodiment of the present invention, by setting anomaly detection instructions in the intelligent control chip and configuring anomaly detection circuits, an infrastructure supporting two detection modes, data redundancy and time redundancy, is constructed, thereby improving the reliability and flexibility of anomaly detection. A multi-stage anomaly detection sub-model is constructed using multi-dimensional feature training data, and a shared parameter layer is set, thereby reducing the redundancy of model parameters and optimizing the utilization efficiency of computing resources. The training strategies of data parallelism and model parallelism are introduced, and the anomaly detection model is given a stronger generalization ability by dynamically adjusting the training parameters and the cross-validation mechanism. Based on the integrated strategy of resource allocation and energy efficiency optimization, combined with the dynamic adjustment mechanism of weight coefficients, the reasonable allocation of computing resources and the optimization of energy efficiency are achieved. Through incremental learning and parameter update mechanisms, a complete model optimization and performance evaluation system is established, thereby ensuring the continuous optimization and performance improvement of the anomaly detection model. The hierarchical storage and parameter index management method is adopted to improve the access efficiency of model parameters and facilitate the dynamic update and maintenance of the model.

[0021] In a specific embodiment, the process of executing step S1 may specifically include the following steps: Add an anomaly detection instruction set and a control status register address space to the intelligent control chip to obtain an extended instruction system architecture, and set a data redundancy detection threshold and a time redundancy detection threshold according to the extended instruction system architecture to generate an anomaly detection parameter matrix; Writing the anomaly detection parameter matrix into the control state register to obtain anomaly detection configuration data, and dividing the anomaly detection configuration data into operation modes to obtain a data redundancy mode and a time redundancy mode; Configure multiple sets of data comparison logic for the data redundancy mode, and configure timing comparison logic for the time redundancy mode to obtain operation mode configuration parameters; Based on the operation mode configuration parameters, a data comparison unit and a result output unit are configured, and the data comparison unit and the result output unit are cascade-connected to obtain an abnormality detection circuit; The anomaly detection circuit is functionally verified to obtain a circuit verification result, which is then fed back to the intelligent control chip. The anomaly detection instructions and configuration parameters of the anomaly detection circuit are adjusted according to the circuit verification result to obtain an anomaly detection infrastructure.

[0022] Specifically, add an exception detection instruction set to the intelligent control chip. The exception detection instruction set is the core of the intelligent chip control system. By extending the instruction system architecture, it provides dedicated instruction support for the exception detection function. For example, add a new instruction DETECT_ERR, which is used to trigger data comparison and exception judgment operations. And expand the address space of the control status register to store the configuration parameters and real-time status information of the exception detection. Assume that the address range of the expanded control status register is ,in and Respectively represent the starting address and ending address of the register, then the capacity of the expanded register is: ; in, Indicates the capacity of the register, which can accommodate all anomaly detection parameters and operating status data. According to the extended instruction system architecture, set the data redundancy detection threshold and time redundancy detection threshold. The function of these thresholds is to define the specific judgment conditions for anomaly detection. For example, the data redundancy detection threshold Used to determine the consistency between multiple sets of redundant data. The calculation formula is: ; in, Indicates Sample data, Indicates reference data, Indicates the size of the data group. When the set threshold is exceeded, it is determined that there is data anomaly. Similarly, the time redundancy detection threshold Defined as the time interval error between multiple operations: ; in, Represents the time interval between two consecutive operations. By combining these thresholds, the anomaly detection parameter matrix is ​​generated : ; matrix Contains all possible detection threshold combinations. After writing the control status register, the abnormality detection configuration data is obtained. In order to optimize the detection efficiency, the abnormality detection configuration data is divided into operation modes, specifically into data redundancy mode and time redundancy mode. In the data redundancy mode, the consistency of multiple sets of data is compared by configuring multiple sets of data comparison logic. For example, for Group Redundancy Data , the comparison logic is defined as: In the time redundancy mode, the time interval deviation is compared, and the timing comparison logic is defined as: ; in, Represents the preset threshold for time redundancy detection. Based on the operation mode configuration parameters, the data comparison logic and result output logic are configured to the data comparison unit and the result output unit, and the two are cascaded through hardware connection to form an anomaly detection circuit. For example, the data comparison unit implements real-time comparison of input data through Boolean logic, and the result output unit is responsible for feeding back the comparison result to the control chip in the form of a signal. When the input data meets the detection condition, the output unit triggers an alarm signal to indicate that the system is abnormal. After the design of the anomaly detection circuit is completed, it is functionally verified. Functional verification is performed through simulation tests or actual operation tests. For example, a set of preset test data is input, some of which are deliberately set to abnormal values, and the accuracy and reliability of the circuit are verified by observing whether the anomaly detection circuit can accurately output the expected abnormal signal. If the verification results show that the detection accuracy is insufficient or there is a false alarm, the anomaly detection instruction set and configuration parameters are adjusted according to the feedback results. For example, the threshold range of data comparison is redefined or the sensitivity of the timing comparison logic is optimized to improve the circuit performance. After multiple rounds of verification and adjustment, the anomaly detection infrastructure is finally formed.

[0023] In a specific embodiment, the process of executing step S2 may specifically include the following steps: Set data sampling parameters based on the anomaly detection infrastructure, and write the sampling frequency, sampling time window length and sampling accuracy of the data sampling parameters into the data acquisition control unit to obtain the data acquisition basic configuration; Set constraints on the basic configuration of data collection, divide the data collection levels according to the processing parameter types, set the data filtering threshold of each data collection level, and obtain the data collection protocol; The data acquisition protocol is imported into the data acquisition module, and the processing data is sampled in multiple dimensions to generate the original data matrix, and the original data matrix is ​​subjected to denoising and standardization processing to obtain the preprocessed data; The time domain features, frequency domain features and statistical features of the preprocessed data are extracted respectively, and the time domain features, frequency domain features and statistical features are combined into multi-dimensional feature training data.

[0024] Specifically, data sampling parameters are set based on the anomaly detection infrastructure to ensure that the collected processing data can fully reflect the dynamic behavior of the system. Data sampling parameters include key indicators such as sampling frequency, sampling time window length, and sampling accuracy. Sampling frequency It refers to the number of samples collected per second, which determines the highest signal frequency that the system can capture. According to the Nyquist theorem, the sampling frequency needs to satisfy is the highest frequency of the signal. Sampling time window length It refers to the time range of each sampling, which determines the length of the time series that the system can analyze. The calculation formula is: ; in The number of samples to obtain for each sampling. Represents the resolution of the sampled data. For example, the sensor output is 12-bit or 16-bit accuracy. The higher the accuracy, the more subtle signal changes can be captured, but at the same time the data storage requirements will increase. After these sampling parameters are written into the data acquisition control unit, the basic configuration of data acquisition is formed. Constraints are set for the basic configuration of data acquisition to optimize acquisition efficiency and improve data quality. Based on the parameter characteristics in the processing process, the processing parameters are classified according to importance and dynamics, and divided into different data acquisition levels. For example, key parameters such as temperature, pressure, and vibration signals are set to high priority, with higher sampling frequency and accuracy; while some static or slowly changing parameters, such as ambient humidity or power consumption, are set to low priority, with lower sampling frequency and accuracy. After completing the acquisition level division, set a data filtering threshold for each level to eliminate noise or outliers. For example, using the Z score method, the calculation formula is: ; in, Representation sample The standardized score of is the data mean, is the standard deviation of the data. If the sample exceeds the set threshold (such as 3), it is judged as an outlier and filtered out. Through these operations, a complete data acquisition protocol including sampling levels and filtering rules is formed. After the designed data acquisition protocol is imported into the data collection module, the processing data is sampled in multiple dimensions to generate the original data matrix. The shape is ,in Indicates the type of parameters collected. Represents the number of samples for each parameter. For example, for a system containing three parameters: temperature, vibration, and pressure, each parameter is sampled with 1000 samples, and the original data matrix is ​​represented as: ; in, , and Represents temperature, vibration and pressure in the The raw data is de-noised and standardized to generate high-quality pre-processed data. De-noising uses methods such as low-pass filtering or wavelet transform. For example, a low-pass filter is used to eliminate high-frequency noise. The filter output is: ; in, is the filter coefficient, is the input signal, is the output signal after filtering. Standardization processing uses normalization or Z-score transformation to map all parameters to a uniform numerical range. For example, for normalization operation, the standardized data is: ; in, and are the minimum and maximum values ​​of the data respectively. Through this step, the preprocessed data matrix is ​​obtained. On the basis of the preprocessed data, time domain features, frequency domain features and statistical features are extracted, and these features are combined into multi-dimensional feature training data. Time domain feature extraction mainly analyzes the time series changes of the signal, such as calculating the mean, variance, peak, skewness and kurtosis. Taking the mean as an example, its calculation formula is: ; Frequency domain feature extraction converts the signal from the time domain to the frequency domain through fast Fourier transform, and extracts spectrum features such as main frequency, frequency band energy, etc. It is expressed as: ; Statistical feature extraction includes the distribution characteristics of data, such as median, quantile, etc. Combine time domain features, frequency domain features and statistical features to form a multi-dimensional feature training data matrix : ; in, , and Represent time domain, frequency domain and statistical characteristics respectively.

[0025] In a specific embodiment, the process of executing step S3 may specifically include the following steps: According to the processing sequence, the multi-dimensional feature training data is distributed to different processing stage data sets to obtain multiple processing stage feature data; Determine the sub-model structure for the characteristic data of each processing stage based on the computing resources of the intelligent control chip, set the number of network layers and the number of neurons of the sub-model, and obtain multiple first abnormality detection sub-models; Perform parameter space planning on multiple first anomaly detection sub-models, divide the parameter space into an independent parameter area and a shared parameter area, establish a parameter sharing channel between the first anomaly detection sub-models in adjacent processing stages, and obtain a parameter space distribution map; Constructing a shared parameter layer according to the parameter space distribution diagram, inserting the shared parameter layer into the network structure of the plurality of first anomaly detection sub-models, establishing a parameter transfer matrix, and obtaining an anomaly detection model with a shared layer; The pre-trained parameters of the anomaly detection model with a shared layer are initialized, the pre-trained parameter values ​​are determined according to the feature distribution of each processing stage, a pre-trained parameter matrix is ​​generated, and the pre-trained parameter matrix is ​​written into the anomaly detection model with a shared layer for parameter loading and model compilation to obtain multiple initialized anomaly detection models.

[0026] Specifically, according to the order of the processing steps, the multi-dimensional feature training data is distributed to different processing stage data sets so as to build an anomaly detection model for each stage separately. stages, and the multi-dimensional feature training data of each stage is recorded as ), the overall training data is decomposed into: ; in, Indicates The characteristic data set of each stage includes the time domain characteristics, frequency domain characteristics and statistical characteristics of a specific processing stage. Based on the computing resources of the intelligent control chip, the sub-model structure is determined for the characteristic data of each processing stage to ensure the adaptability of the model complexity and chip resources. In the design of the neural network model, the number of network layers and the number of neurons in each layer are two key parameters. Assume that The dimension of the characteristic data of the stage is , then the number of neurons in the input layer is initially set to , the number of neurons in the output layer is 1 (used for binary classification to judge abnormalities), the number of hidden layers and the number of neurons Dynamically adjust based on computing resources, for example: ; in, is the maximum number of neurons allowed by chip resources, Indicates the number of neurons that decreases layer by layer. Taking the refinement stage as an example, assuming that its feature dimension , the maximum number of neurons allowed by resources , then the hidden layer is set to 3 layers, with the number of neurons being 32, 16 and 8 respectively, forming the After completing the sub-model design, parameter space planning is performed on multiple first anomaly detection sub-models, and their parameter spaces are divided into independent parameter areas and shared parameter areas, so as to establish parameter sharing channels between sub-models in adjacent processing stages. The parameter set of the stage sub-model is ,in represents independent parameters, Represents shared parameters. Shared parameters capture the correlation between stages by sharing with adjacent stages, for example: ; in, Indicates In this way, a parameter space distribution diagram is generated to describe the independence and sharing relationship of the parameters in each stage. According to the parameter space distribution diagram, a shared parameter layer is constructed and inserted into the network structure of multiple first anomaly detection sub-models. The role of the shared parameter layer is to pass the matrix through the parameter Realize the sharing of parameters between adjacent stages. Assume that the parameter transfer matrix The dimension is ,in is the number of shared parameters, then the output of the shared layer is expressed as: ; in, and Respectively represent and The output of the shared layer in each stage is connected to transfer the knowledge of the previous stage to the next stage, thus improving the overall coordination ability of anomaly detection. The pre-trained parameters of the anomaly detection model with a shared layer are initialized to improve the initial performance of the model. The pre-trained parameter initialization determines the initial parameter value according to the feature distribution of each processing stage. For example, by counting the mean of each feature and standard deviation , normalize and initialize the parameters: ; in, Indicates is the mean, is the normal distribution of variance. After writing the generated pre-trained parameter matrix into the anomaly detection model with shared layers, the parameter loading and operating environment optimization are completed through model compilation. The compilation process includes calculation graph construction, gradient optimization rule setting, etc., to ensure that the model can run efficiently on the intelligent control chip. Through the above steps, multiple initialized anomaly detection models are finally obtained.

[0027] Among them, the first anomaly detection sub-model includes: dividing the multidimensional feature training data into input data stream and label data stream according to the time series, and constructing a feature extraction layer; the feature extraction layer includes 5 convolution units, each of which is composed of a two-dimensional convolution layer, a batch normalization layer and a ReLU activation function. The number of input channels of the first convolution unit is the feature dimension of the multidimensional feature training data, the number of output channels is 64, the convolution kernel size is 3×3, and the step size is 1; the number of input channels of the second to fifth convolution units is doubled successively, and the number of output channels is 128, 256, 512 and 1024 respectively, and the convolution kernel size is 3×3, with a step size of 2; a time series modeling layer is constructed based on the output of the feature extraction layer. The time series modeling layer includes 3 bidirectional long short-term memory network units, each unit contains 256 hidden layer nodes, the input dimension of the first unit is 1024, the output dimension is 512, the input dimension of the second unit is 512, the output dimension is 256, and the input dimension of the third unit is 256, and the output dimension is 128; the output of the time series modeling layer is sent to the attention layer. The attention layer adopts a multi-head self-attention mechanism, which includes 8 attention heads. The query dimension, key dimension, and value dimension of each attention head are 64. By scaling the dot product attention The attention weight matrix is ​​calculated to perform weighted aggregation on the time series features; the anomaly detection layer is constructed based on the output of the attention layer. The anomaly detection layer includes three fully connected layers. The input dimension of the first fully connected layer is 128, the output dimension is 64, and the ReLU activation function is used; the input dimension of the second fully connected layer is 64, the output dimension is 32, and the ReLU activation function is used; the input dimension of the third fully connected layer is 32, the output dimension is 2, and the Softmax activation function is used; the output of the anomaly detection layer is sent to the loss calculation layer, and the loss calculation layer uses the cross entropy loss function to calculate the difference between the prediction result and the label data stream The overall loss function is constructed in combination with the L2 regularization term for model parameter optimization. A back propagation layer is constructed based on the overall loss function, and the back propagation layer uses the Adam optimizer to update parameters. The initial learning rate is set to 0.001, and the learning rate is reduced to the original 0.1 after every 50 training rounds. The momentum parameters β1 are set to 0.9, and β2 is set to 0.999. Residual connections are added between the feature extraction layer and the timing modeling layer, between the timing modeling layer and the attention layer, and between the attention layer and the anomaly detection layer, and layer normalization is added after each residual connection to obtain the complete first anomaly detection sub-model structure.

[0028] In a specific embodiment, the process of executing step S4 may specifically include the following steps: Divide the training tasks of multiple initialized anomaly detection models, distribute the training data to different computing units in batches, set the training parameters of each computing unit, and obtain a data parallel training plan; Based on the data parallel training scheme, different batches of training data are processed in parallel, the training results of each computing unit are gradient aggregated, and the sub-model parameters are updated to obtain the data parallel training model; The data parallel training model is split into model structures, sub-models with different structures are assigned to independent computing units, and communication links between sub-models are established to obtain a model parallel training solution. Perform sub-model training based on the model parallel training scheme, dynamically adjust the learning rate and batch size of each sub-model, record the performance indicators during the training process, and obtain model training data; The model training data is cross-validated to obtain a cross-validation result, and the training parameters are optimized and adjusted according to the cross-validation result to obtain adjusted parameters, and the adjusted parameters are written into each sub-model to obtain multiple second anomaly detection sub-models.

[0029] Specifically, the training tasks of multiple initialized anomaly detection models are divided, and the training data are reasonably distributed to different computing units in batches to maximize the parallel processing capabilities of hardware resources. Assume that the total amount of training data is , whose dimensions are ,in represents the number of data samples, Represents the feature dimension of each sample. In order to facilitate parallel processing, the data is divided into batches Subset ,in Indicates The data set of batches meets the following conditions: ; After the data is divided, assign corresponding training tasks to each computing unit and set training parameters, such as learning rate. , batch size And optimization algorithm. For example, using stochastic gradient descent (SGD) optimization, the parameter update rule of each computing unit is expressed as: ; in, For the The model parameters at the iteration, For the The loss function on batches of data is Represents the gradient of the loss function. Based on the above data parallel training scheme, different computing units simultaneously process the batch data assigned to them and independently calculate the gradient for each batch. Assume that the system has The gradient calculation result is , these results are aggregated into global gradients through the gradient aggregation mechanism: ; Then use the global gradient to update the model parameters: ; The gradient aggregation method ensures the global consistency of parameter updates while fully utilizing the efficiency of parallel computing. After completing the data parallel training model, the model structure is split. Since the structures and computational complexity of different sub-models are significantly different, they are assigned to independent computing units to optimize resource utilization. For example, assuming The neural network structure of each sub-model is composed of The number of parameters in each layer is ( ), the total number of parameters of the model is: ; According to the storage and computing capabilities of each computing unit, sub-models with larger parameters are assigned to high-performance computing units, while smaller sub-models are assigned to low-power devices. Communication links are established between sub-models to exchange parameters or intermediate results. The bandwidth and latency of the communication link must meet the following conditions: ; in, represents the communication bandwidth, Indicates the amount of parameters exchanged, Represents the communication cycle, For communication delay, is the delay tolerance threshold. Based on the model parallel training scheme, when training each sub-model, its learning rate and batch size are dynamically adjusted to adapt to the current training status. For example, a larger learning rate is used to accelerate convergence in the early stage of training, and the learning rate is gradually reduced in the later stage to improve accuracy: ; in, is the initial learning rate, is the attenuation coefficient, is the current iteration number. During training, performance indicators such as training loss are recorded. , Accuracy and the gradient norm , these data are used for subsequent cross-validation and optimization adjustments. After the training is completed, the model training data is cross-validated to evaluate the generalization ability of the sub-model. The validation results are obtained by dividing the data into a training set and a validation set, and evaluating the model performance on the validation set. For example, the validation accuracy is expressed as: ; in, is the indicator function, and are the predicted value and the true value respectively, is the number of samples in the validation set. According to the cross-validation results, the training parameters are optimized and adjusted. If the performance of some sub-models is found to be substandard, improvements can be made by adjusting the learning rate, reallocating data, or expanding the model structure. For example, if Validation loss of each sub-model If the threshold is exceeded, the performance is improved by increasing the training rounds of the model or optimizing the initial values ​​of its parameters. The optimized parameters are written into each sub-model to obtain multiple second anomaly detection sub-models.

[0030] In a specific embodiment, the process of executing step S5 may specifically include the following steps: Formulate a resource allocation strategy based on the operation status data of the multiple second anomaly detection sub-models, and perform resource weight allocation on the multiple second anomaly detection sub-models in combination with the computing resource utilization rate and the storage resource occupancy rate to obtain a sub-model resource configuration plan; Performing computational load analysis on multiple second anomaly detection sub-models based on the sub-model resource configuration scheme, inputting the energy consumption index and computational resource consumption index of each second anomaly detection sub-model into the energy efficiency evaluation function, and obtaining the sub-model energy efficiency optimization parameters; Perform integrated weight calculation on multiple second anomaly detection sub-models, input the detection accuracy and sub-model energy efficiency optimization parameters into the weight calculation matrix to normalize the weight coefficients, and obtain a sub-model weight distribution table; Establish an optimized integration strategy according to the sub-model weight distribution table, combine multiple second anomaly detection sub-models into an integrated detection model according to the processing stage sequence, establish a model switching mechanism and set trigger conditions to obtain a dynamic integrated model; Perform incremental learning on the dynamic integration model, determine the parameter update strategy, and update the model parameters and evaluate the performance of the newly added data samples to obtain the updated anomaly detection model; The updated anomaly detection model is divided into storage units, the model structure parameters and weight parameters are written into the fixed storage area and the dynamic storage area respectively, a parameter index table is established, and the processing anomaly detection model is obtained.

[0031] Specifically, a resource allocation strategy is formulated based on the operating status data of the plurality of second anomaly detection sub-models. The operating status data includes computing resource utilization and storage resource utilization Assume that The computing resources and storage resource usage of each sub-model are and , and the total system resources are and , then the resource utilization is expressed as: ; in, and Respectively represent The computing resource requirements and storage resource requirements of each sub-model are analyzed comprehensively, and resource weights are assigned to each sub-model based on the importance indicators of each sub-model (such as detection frequency or key task priority). , the following conditions are met: ; in, Indicates After the resource allocation strategy is formulated, the computational load analysis is performed based on the sub-model resource configuration scheme to evaluate the energy efficiency performance of each sub-model. The energy consumption of each sub-model during operation is , the computing resource consumption is , then the energy efficiency ratio Defined as: ; in, For the The detection accuracy of each sub-model is The higher the value, the higher the detection efficiency of the sub-model under unit energy and computing resources. The energy consumption and computing resource consumption of all sub-models are input into the energy efficiency evaluation function to obtain the energy efficiency optimization parameters of each sub-model, which are used to guide the subsequent model integration optimization. According to the energy efficiency optimization parameters, the integration weights of multiple second anomaly detection sub-models are calculated to determine their contribution to the integrated model. The detection accuracy of each sub-model is calculated. and energy efficiency optimization parameters Input the weight calculation matrix and get the weight distribution table after normalization. The weight calculation formula is: ; in, Indicates The normalized weights of the sub-models, and is the coefficient for adjusting the importance of detection accuracy and energy efficiency ratio, and the strategic tendency of weight allocation is controlled by adjusting them. Based on the weight allocation table, an optimized integration strategy is established and multiple second anomaly detection sub-models are combined into an integrated detection model according to the order of processing stages. Assume that the processing process is divided into Each stage is responsible for one or more sub-models. The design of the integrated model organizes the sub-models in a phased order, for example: ; in, Indicates In order to adapt to the changes in the dynamic environment, a model switching mechanism is established for the integrated model and trigger conditions are set. When the trigger condition is met, the optimal sub-model is dynamically selected to switch to the current task, for example: ; in, is the sub-model under the current conditions After the dynamic integration model is built, incremental learning is performed to improve performance and adaptability. The core of incremental learning is to dynamically update model parameters based on the feature distribution of the newly added data samples. Assume that the newly added data samples are , by calculating the gradient Update model parameters: ; in, is the learning rate, is the loss function. At the same time, the performance of the updated model is evaluated, and the optimization effect of incremental learning is determined by comparing the detection accuracy and resource consumption changes before and after the update. In order to facilitate the long-term storage and management of the updated anomaly detection model, the storage unit is divided, and the model structure parameters and weight parameters are written into the fixed storage area and the dynamic storage area respectively. For example, the network structure and optimization algorithm of the model are fixed and stored in the fixed storage area, while the weight parameters updated with incremental learning are stored in the dynamic storage area. In order to quickly retrieve model parameters, a parameter index table is established , where each entry records the parameter name, storage address, and version number, for example: ; Quickly locate and load the latest parameters of the model through the parameter index table.

[0032] In a specific embodiment, the execution step performs incremental learning on the dynamic integration model, determines the parameter update strategy, and updates the model parameters and performs performance evaluation on the newly added data samples. The process of obtaining the updated anomaly detection model may specifically include the following steps: The model parameters of the dynamic integration model are divided into two categories: structural parameters and weight parameters, and the parameter importance is sorted according to the sub-model weight distribution table to obtain the parameter update priority table; Construct a parameter update strategy based on the parameter update priority table, classify and label the newly added data samples according to the processing stage, obtain labeled data samples, and assign the labeled data samples to the corresponding second anomaly detection sub-model to obtain an incremental training data set; Update the model parameters layer by layer based on the incremental training data set and the order of parameter importance, record the parameter change amount and performance change index, obtain the parameter update results, and perform performance evaluation on the parameter update results, calculate the difference in detection accuracy and resource consumption before and after the update, and generate the model evaluation results; Based on the model evaluation results, parameter rollback judgment is performed, the performance evaluation indicators are compared with the preset thresholds, and the rollback operation is performed on the parameters with degraded performance to obtain the parameter optimization plan. The parameter optimization plan is written into the dynamic integration model to optimally update the model parameters to obtain the updated anomaly detection model.

[0033] Specifically, the model parameters of the dynamic integration model are divided into structural parameters and weight parameters. The structural parameters define the basic architecture of the model, such as the number of layers in the network. , the number of neurons in each layer , activation function etc., which have a global impact on the overall performance of the model. Weight parameters are specific values ​​learned by the model during training, such as the weight matrix of each layer and the bias vector . Assume that the total parameter set of the model is , then it is expressed as: ; in, Represents a set of structural parameters, Represents a set of weighted parameters. After completing parameter classification, all parameters are sorted by importance according to the sub-model weight allocation table to generate a parameter update priority table. The weight of each sub-model in the sub-model weight allocation table Indicates its importance in the dynamic integration model, and assigns update priority to the parameters of each sub-model according to its weight. The weight parameters of each sub-model , its importance is defined as: ; in, is the loss function, Represents the sensitivity of the parameter to the loss. Parameters with higher importance are updated first, so as to maximize the effect of incremental learning. A parameter update strategy is constructed based on the parameter update priority table, and the newly added data samples are classified and labeled according to the processing stage to form an incremental training data set. The labeling of the newly added data samples is based on the feature distribution of the processing stage. For example, the samples in the rough processing stage mainly contain vibration and pressure features, while the samples in the fine processing stage contain temperature and cutting force features. Assume that the newly added data sample set is Through feature extraction and classification annotation, it is divided into Data subsets for each stage: ; in, Indicates The labeled data samples are assigned to the corresponding second anomaly detection sub-model for incremental training. Based on the incremental training data set and the order of parameter importance, the model parameters are updated layer by layer. and bias The gradient descent method is used for updating, and the update rule is: ; in, is the learning rate, and are the gradients of weight and bias respectively. During the update process, the parameter changes are recorded And performance change indicators, such as detection accuracy The amount of change and resource consumption changes After the update is completed, the performance of the parameter update results is evaluated. By comparing the detection accuracy and resource consumption changes before and after the update, the model evaluation results are generated. For example, the performance evaluation is performed by calculating the accuracy improvement ratio and resource efficiency improvement ratio To measure: ; in, and are the accuracy after updating and the accuracy before updating respectively, and The resource consumption after the update and the resource consumption before the update are respectively. According to the model evaluation results, parameter rollback judgment is performed. When the update of certain parameters causes performance degradation and is lower than the preset threshold, a rollback operation is performed to restore these parameters to the state before the update. For example, when it is detected that or When the corresponding parameter recover: ; in, and are the thresholds for accuracy and resource efficiency, respectively. The rollback operation can effectively avoid performance degradation caused by improper parameter updates. After completing the parameter rollback, a parameter optimization plan is generated, and the optimized parameters are written into the dynamic integration model to complete the optimal update of the model parameters. The updated anomaly detection model is divided into storage units, and the fixed structural parameters are written into the fixed storage area, while the dynamically changing weight parameters are written into the dynamic storage area, and a parameter index table is established for efficient retrieval. For example, the index table records the storage address and update timestamp of each parameter: ; Through the index table, the latest parameter version can be quickly located and loaded to obtain the updated anomaly detection model.

[0034] In this embodiment, obtaining the processing anomaly detection model also includes: designing a distributed architecture for the processing anomaly detection model, dividing the model structure into a training module and an execution module, deploying the training module on a cloud server, deploying the execution module on an edge computing node, establishing a parameter synchronization mechanism between modules, and obtaining a cloud-edge collaborative anomaly detection architecture; constructing a feature importance evaluation unit based on the cloud-edge collaborative anomaly detection architecture, performing mutual information calculation and contribution analysis on the feature data of each processing stage, generating a feature importance ranking table, screening key features according to the importance threshold, and obtaining a reduced-dimensional feature subspace; inputting the reduced-dimensional feature subspace into a multi-agent system, configuring a corresponding detection agent for each processing stage, the detection agent including a state perception module, an action selection module, and a reward calculation module, establishing a collaboration mechanism between agents, and obtaining a distributed detection network; performing domain randomization training on the distributed detection network, generating a variety of training scenarios by perturbing processing parameters, environmental parameters, and model parameters, constructing a domain randomized data set, and inputting the domain randomized data set into a cloud server to obtain a robustness training. data; a hybrid multi-agent reinforcement learning network is constructed based on robust training data. The reinforcement learning network includes a policy network and a value network. An independent network parameter space and a shared experience pool are set for each detection agent. The policy parameters are optimized through the soft Actor-Critic algorithm to obtain a reinforcement learning model; the reinforcement learning model is deployed to the cloud-edge collaborative anomaly detection architecture, strategy optimization and parameter update are performed on the cloud, real-time detection and action execution are performed on the edge, a cloud-edge two-way communication channel is established, and dynamic adjustment of the detection strategy is achieved to obtain a dynamic optimization system; the operating status of the dynamic optimization system is monitored in real time, performance indicators such as detection accuracy, response delay, and resource utilization are collected, a performance evaluation matrix is ​​constructed, and the cloud training strategy and edge execution strategy are dynamically adjusted based on the evaluation results to obtain a performance optimization plan; the performance optimization plan is written into the cloud-edge collaborative anomaly detection architecture, cloud-edge resource allocation, communication bandwidth, and computing load are optimized, and robustness verification experiments are performed to verify the detection performance of the system in an uncertain environment to obtain a processing anomaly detection system with adaptive capabilities.

[0035] The above describes the intelligent control method based on processing management in the embodiment of the present invention. The following describes the intelligent control device based on processing management in the embodiment of the present invention. Figure 2 In one embodiment of the present invention, an intelligent control device based on processing management includes: Create a module to set anomaly detection instructions through the intelligent control chip and write to the control status register to create anomaly detection infrastructure; A feature extraction module is used to construct a data collection protocol based on the anomaly detection infrastructure and perform feature extraction operations on the processing data to obtain multi-dimensional feature training data; A pre-training module, used to construct a first anomaly detection sub-model of multiple processing stages based on multi-dimensional feature training data, write pre-training parameters into the first anomaly detection sub-model, and obtain multiple initialized anomaly detection models; A parallel training module is used to deploy multiple initialized anomaly detection models to the intelligent control chip for model parallel training to obtain multiple second anomaly detection sub-models; The optimization integration module is used to calculate the sub-model weight coefficients of multiple second anomaly detection sub-models, and perform optimization integration and incremental learning to obtain a processing anomaly detection model.

[0036] Through the cooperation of the above components, by setting anomaly detection instructions in the intelligent control chip and configuring anomaly detection circuits, an infrastructure supporting data redundancy and time redundancy detection modes is constructed, which improves the reliability and flexibility of anomaly detection. Multi-stage anomaly detection sub-models are constructed using multi-dimensional feature training data, and a shared parameter layer is set to reduce the redundancy of model parameters and optimize the utilization efficiency of computing resources. The training strategies of data parallelism and model parallelism are introduced, and the anomaly detection model is given stronger generalization ability by dynamically adjusting training parameters and cross-validation mechanisms. Based on the integrated strategy of resource allocation and energy efficiency optimization, combined with the dynamic adjustment mechanism of weight coefficients, the reasonable allocation of computing resources and the optimization of energy efficiency are achieved. Through incremental learning and parameter update mechanisms, a complete model optimization and performance evaluation system is established to ensure the continuous optimization and performance improvement of the anomaly detection model. The hierarchical storage and parameter index management method is adopted to improve the access efficiency of model parameters and facilitate the dynamic update and maintenance of the model.

[0037] An embodiment of the present invention further provides a chip on which a computer program is stored, and the computer program implements the above method when executed by a processor.

[0038] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0039] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art or the whole or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program code.

[0040] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features thereof may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent control method based on machining process management, characterized in that: The method comprises: Set anomaly detection instructions through the intelligent control chip and write them into the control status register to create anomaly detection infrastructure; Constructing a data collection protocol according to the anomaly detection infrastructure, and performing feature extraction operations on the processing data to obtain multi-dimensional feature training data; Building a first anomaly detection sub-model of multiple processing stages based on the multi-dimensional feature training data, writing pre-trained parameters into the first anomaly detection sub-model, and obtaining multiple initialized anomaly detection models; Deploying the multiple initialized anomaly detection models to the intelligent control chip for model parallel training to obtain multiple second anomaly detection sub-models; The sub-model weight coefficients of the plurality of second anomaly detection sub-models are calculated, and optimization integration and incremental learning are performed to obtain a processing anomaly detection model.

2. The intelligent control method based on machining process management according to claim 1 is characterized in that: The method of setting an abnormality detection instruction through an intelligent control chip and writing it into a control status register to create an abnormality detection infrastructure includes: Adding an abnormality detection instruction set and a control status register address space to the intelligent control chip to obtain an extended instruction system architecture, and setting a data redundancy detection threshold and a time redundancy detection threshold according to the extended instruction system architecture to generate an abnormality detection parameter matrix; Writing the anomaly detection parameter matrix into a control state register to obtain anomaly detection configuration data, and dividing the anomaly detection configuration data into operation modes to obtain a data redundancy mode and a time redundancy mode; Configuring multiple groups of data comparison logic for the data redundancy mode, and configuring timing comparison logic for the time redundancy mode, to obtain operation mode configuration parameters; Based on the operation mode configuration parameters, a data comparison unit and a result output unit are configured, and the data comparison unit and the result output unit are cascade-connected to obtain an abnormality detection circuit; Functional verification is performed on the abnormality detection circuit to obtain a circuit verification result, and the circuit verification result is fed back to the intelligent control chip. The abnormality detection instruction and the configuration parameters of the abnormality detection circuit are adjusted according to the circuit verification result to obtain an abnormality detection infrastructure.

3. The intelligent control method based on machining process management according to claim 1 is characterized in that: The data collection protocol is constructed according to the abnormality detection infrastructure, and feature extraction operations are performed on the processing data to obtain multi-dimensional feature training data, including: Setting data sampling parameters based on the anomaly detection infrastructure, and writing the sampling frequency, sampling time window length and sampling accuracy of the data sampling parameters into a data acquisition control unit to obtain a data acquisition basic configuration; Setting constraints on the data acquisition basic configuration, dividing the data acquisition levels according to the processing parameter types, setting the data filtering threshold of each data acquisition level, and obtaining the data acquisition protocol; Importing the data acquisition protocol into a data acquisition module, performing multi-dimensional sampling on the processing data to generate an original data matrix, and performing denoising and standardization processing on the original data matrix to obtain pre-processed data; The time domain features, frequency domain features and statistical features of the preprocessed data are extracted respectively, and the time domain features, the frequency domain features and the statistical features are combined into multi-dimensional feature training data.

4. The intelligent control method based on machining process management according to claim 1 is characterized in that: The first abnormality detection sub-models of multiple processing stages are constructed based on the multi-dimensional feature training data, and the pre-trained parameters are written into the first abnormality detection sub-model to obtain multiple initialized abnormality detection models, including: According to the processing sequence, the multi-dimensional feature training data is distributed to different processing stage data sets to obtain a plurality of processing stage feature data; Determine the sub-model structure for the characteristic data of each processing stage based on the computing resources of the intelligent control chip, set the number of network layers and the number of neurons of the sub-model, and obtain multiple first abnormality detection sub-models; Performing parameter space planning on the multiple first anomaly detection sub-models, dividing the parameter space into an independent parameter area and a shared parameter area, establishing a parameter sharing channel between the first anomaly detection sub-models in adjacent processing stages, and obtaining a parameter space distribution map; Constructing a shared parameter layer according to the parameter space distribution graph, inserting the shared parameter layer into the network structure of the plurality of first anomaly detection sub-models, establishing a parameter transfer matrix, and obtaining an anomaly detection model with a shared layer; The pre-trained parameters of the anomaly detection model with a shared layer are initialized, the pre-trained parameter values ​​are determined according to the feature distribution of each processing stage, a pre-trained parameter matrix is ​​generated, and the pre-trained parameter matrix is ​​written into the anomaly detection model with a shared layer for parameter loading and model compilation to obtain multiple initialized anomaly detection models.

5. The intelligent control method based on machining process management according to claim 1 is characterized in that: The step of deploying the multiple initialized anomaly detection models to the intelligent control chip for model parallel training to obtain multiple second anomaly detection sub-models includes: Dividing the training tasks of the multiple initialized anomaly detection models, allocating the training data to different computing units in batches, setting the training parameters of each computing unit, and obtaining a data parallel training scheme; Based on the data parallel training scheme, different batches of training data are processed in parallel, the training results of each computing unit are gradient aggregated, and sub-model parameters are updated to obtain a data parallel training model; The data parallel training model is split into model structures, sub-models with different structures are allocated to independent computing units, communication links between sub-models are established, and a model parallel training scheme is obtained; Execute sub-model training based on the model parallel training scheme, dynamically adjust the learning rate and batch size of each sub-model, record the performance indicators during the training process, and obtain model training data; The model training data is cross-validated to obtain a cross-validation result, and the training parameters are optimized and adjusted according to the cross-validation result to obtain adjusted parameters, and the adjusted parameters are written into each sub-model to obtain multiple second anomaly detection sub-models.

6. The intelligent control method based on machining process management according to claim 1 is characterized in that: The calculating of the sub-model weight coefficients of the plurality of second anomaly detection sub-models, and performing optimization integration and incremental learning to obtain a processing anomaly detection model includes: Formulate a resource allocation strategy according to the operation status data of the plurality of second anomaly detection sub-models, and perform resource weight allocation on the plurality of second anomaly detection sub-models in combination with the computing resource utilization rate and the storage resource occupancy rate to obtain a sub-model resource configuration plan; Performing computational load analysis on the plurality of second anomaly detection sub-models based on the sub-model resource configuration scheme, inputting the energy consumption index and the computational resource consumption index of each second anomaly detection sub-model into an energy efficiency evaluation function, and obtaining sub-model energy efficiency optimization parameters; Performing integrated weight calculation on the multiple second anomaly detection sub-models, inputting the detection accuracy and the sub-model energy efficiency optimization parameters into a weight calculation matrix to perform weight coefficient normalization processing, and obtaining a sub-model weight allocation table; Establishing an optimization integration strategy according to the sub-model weight distribution table, combining the multiple second anomaly detection sub-models into an integrated detection model in the order of processing stages, establishing a model switching mechanism and setting trigger conditions to obtain a dynamic integrated model; Performing incremental learning on the dynamic integrated model, determining a parameter update strategy, and updating model parameters and evaluating performance on newly added data samples to obtain an updated anomaly detection model; The updated anomaly detection model is divided into storage units, the model structure parameters and weight parameters are written into the fixed storage area and the dynamic storage area respectively, a parameter index table is established, and the processing anomaly detection model is obtained.

7. The intelligent control method based on machining process management according to claim 6 is characterized in that: The step of performing incremental learning on the dynamic integrated model, determining a parameter update strategy, and performing model parameter update and performance evaluation on newly added data samples to obtain an updated anomaly detection model includes: The model parameters of the dynamic integrated model are divided into two categories: structural parameters and weight parameters, and the parameter importance is sorted according to the sub-model weight allocation table to obtain a parameter update priority table; Constructing a parameter update strategy based on the parameter update priority table, classifying and labeling the newly added data samples according to the processing stage to obtain labeled data samples, and assigning the labeled data samples to the corresponding second anomaly detection sub-model to obtain an incremental training data set; Based on the incremental training data set and the order of parameter importance, the model parameters are updated layer by layer, the parameter changes and performance change indicators are recorded, the parameter update results are obtained, and the performance evaluation is performed on the parameter update results, the detection accuracy difference and resource consumption change before and after the update are calculated, and the model evaluation results are generated; Based on the model evaluation result, parameter rollback judgment is performed, the performance evaluation index is compared with the preset threshold, and a rollback operation is performed on the parameters with degraded performance to obtain a parameter optimization plan, and the parameter optimization plan is written into the dynamic integration model to perform optimal update of the model parameters to obtain an updated anomaly detection model.

8. An intelligent control device based on processing management, characterized in that: Used to execute the intelligent control method based on machining process management as described in any one of claims 1 to 7, the intelligent control device based on machining process management comprises: Create a module to set anomaly detection instructions through the intelligent control chip and write to the control status register to create anomaly detection infrastructure; A feature extraction module is used to construct a data acquisition protocol according to the anomaly detection infrastructure and perform feature extraction operations on the processing data to obtain multi-dimensional feature training data; A pre-training module, used for constructing a first anomaly detection sub-model of multiple processing stages based on the multi-dimensional feature training data, writing pre-training parameters into the first anomaly detection sub-model, and obtaining a plurality of initialized anomaly detection models; A parallel training module, used for deploying the multiple initialized anomaly detection models to the intelligent control chip for model parallel training to obtain multiple second anomaly detection sub-models; The optimization integration module is used to calculate the sub-model weight coefficients of the multiple second abnormality detection sub-models, and perform optimization integration and incremental learning to obtain a processing abnormality detection model.

9. A chip, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the processor is enabled to execute the intelligent control method based on machining process management as claimed in any one of claims 1 to 7.

Images