Test data analysis method under complex working conditions

Abstract

The present invention discloses a method for analyzing test data under complex working conditions, comprising the following steps: S1, collecting multi-source time series data and preprocessing the multi-source time series data; S2, constructing a Fourier neural operator model and generating predicted data through Fourier transform; S3, introducing an adaptive ant colony memory optimization algorithm and dynamically optimizing memory resource allocation; S4, monitoring memory status in real time, reclaiming low-priority caches, and prioritizing memory allocation to critical tasks; S5, dynamically adjusting memory allocation during the prediction process based on an updated memory resource allocation model; S6, comparing the predicted results with actual data, triggering anomaly detection when a threshold is exceeded, and generating an early warning signal; S7, evaluating the performance of the Fourier neural operator model and dynamically updating the memory resource allocation model. The present invention combines the Fourier neural operator model with the adaptive ant colony memory optimization to achieve real-time intelligent analysis and prediction of test data under complex working conditions, thereby improving equipment early warning and resource utilization efficiency.

Description

Technical Field

[0001] The present invention relates to the technical field of industrial Internet of Things, and in particular to a test data analysis method under complex working conditions. Background Art

[0002] With the ongoing development of industrial automation and intelligent manufacturing, test data analysis has become a crucial tool for ensuring safe equipment operation, improving production efficiency, and reducing maintenance costs. With the rapid adoption of various sensor and IoT technologies, the amount of data generated on industrial sites has exploded. This data, often sourced from diverse sensors and monitoring systems, is diverse and complex in dimensions, with nonlinear, multimodal, and dynamically changing characteristics. Under complex operating conditions, equipment operating environments are often accompanied by uncertainties such as high temperature, high pressure, vibration, noise interference, and multivariate coupling. This presents numerous challenges for traditional test data analysis methods when processing this complex data.

[0003] Currently, commonly used data analysis technologies include data mining techniques based on statistical methods, traditional time series analysis methods, and prediction models based on machine learning. Although traditional statistical methods have certain advantages in processing regular data, they often lack sufficient adaptability to the nonlinear and sudden changes in data under complex working conditions. Time series analysis methods such as autoregressive models, moving average models, and autoregressive integrated moving average models, due to their relatively strict model assumptions, are difficult to accurately capture the diversity and dynamic changes in equipment operating states in complex environments. At the same time, prediction models based on machine learning, such as support vector machines, random forests, and long short-term memory networks, although they have improved prediction results to a certain extent, still have problems such as high computational complexity, slow response speed, and insufficient model robustness when faced with high-dimensional data, multi-source data fusion, and real-time analysis scenarios with limited resources.

[0004] In the field of deep learning, a number of advanced models have emerged in recent years, such as the Fourier neural operator model. This model uses the Fourier transform to map time series data into the frequency domain, extracts high-dimensional features through frequency-domain convolution operations, and combines graph neural networks to extract spatial features, providing new insights for processing test data under complex working conditions. Although the Fourier neural operator model excels in long-term time series prediction and feature extraction from complex data, it still suffers from deficiencies in memory resource management and computational efficiency in practical applications. In industrial field applications, equipment is often deployed on edge computing platforms or IoT nodes, which typically have very limited computing and memory resources. Therefore, achieving efficient, low-latency memory resource scheduling and optimization while ensuring model prediction accuracy has become a key issue that needs to be addressed.

[0005] In existing technologies, some research has begun to attempt to introduce swarm intelligence optimization algorithms into model parameter tuning, such as using ant colony algorithms to dynamically adjust model hyperparameters. However, these methods mostly focus on parameter optimization, and a systematic and comprehensive solution for the dynamic allocation and real-time control of memory resources has yet to be developed. Furthermore, traditional memory management strategies, such as the Least Recently Used (LRU) and Least Recently Frequent (LFU) algorithms, while widely used in general scenarios, often fail to meet practical needs during test data analysis under complex conditions due to complex data characteristics, high real-time requirements, and the model's sensitivity to memory resources. Their main drawback is the failure to adaptively allocate and optimize memory resources based on the actual load of different modules and the importance of the prediction task. This leads to the risk of uneven memory resource allocation, prediction delays, and even system crashes when critical prediction tasks are under high load.

[0006] Therefore, how to provide a test data analysis method under complex working conditions is an urgent problem that needs to be solved by those skilled in the art. Summary of the Invention

[0007] One purpose of the present invention is to propose a test data analysis method under complex working conditions. The present invention makes full use of advanced technologies such as Fourier neural operator models, adaptive ant colony memory optimization algorithms, and graph neural networks, and describes in detail how to convert multi-source test data from the time domain to the frequency domain, and realize high-dimensional feature extraction through multi-scale filtering and nonlinear activation. At the same time, the graph neural network is used to capture the spatial correlation information in the data to generate prediction data of the equipment operation status. In addition, the present invention innovatively introduces an adaptive ant colony memory optimization algorithm to realize dynamic monitoring of the memory resources of each module in the model, automatically reclaim low-priority task caches, and prioritize memory allocation to key prediction tasks, thereby effectively improving memory resource utilization and system response speed. The present invention has the significant advantages of high prediction accuracy, strong real-time performance, intelligent and efficient memory management, and excellent system stability under complex working conditions.

[0008] The test data analysis method under complex working conditions according to an embodiment of the present invention includes the following steps:

[0009] S1. Collect multi-source time series data under complex working conditions and pre-process the multi-source time series data;

[0010] S2. Construct a Fourier neural operator model, including a frequency domain feature extraction module, a spatial feature extraction module, and a prediction module. Input the preprocessed multi-source time series data into the Fourier neural operator model, map it to the frequency domain space through Fourier transform, and generate prediction data of the equipment operation status.

[0011] S3. Introduce an adaptive ant colony memory optimization algorithm to dynamically optimize the memory resources of the Fourier neural operator model, establish a memory resource allocation model, and dynamically adjust the memory allocation weight of the Fourier neural operator model based on the pheromone propagation mechanism;

[0012] S4. Real-time monitoring is performed based on the memory resource allocation model. When memory resources are tight, the cache of low-priority tasks is automatically recycled, and memory resources are preferentially allocated to key prediction tasks. The memory resource allocation model is updated through the pheromone volatilization mechanism.

[0013] S5. Use the Fourier neural operator model to predict the real-time input device operation data. During the prediction process, combine it with the updated memory resource allocation model to dynamically adjust the memory resource allocation strategy in real time.

[0014] S6. Compare the prediction results with the actual monitoring data. When the prediction error exceeds the preset threshold, the anomaly detection module is triggered to identify potential risks in the equipment operation status and generate an early warning signal, which is notified to the operation and maintenance personnel through interface alarms, SMS push or email;

[0015] S7. The Fourier neural operator model is evaluated through the fuzzy logic control algorithm, and the memory resource allocation model is dynamically updated using the pheromone data collected by the adaptive ant colony memory optimization algorithm.

[0016] Optionally, the S2 specifically includes:

[0017] S21, performing Fourier transform on the preprocessed multi-source time series data, converting the multi-source time series data from the time domain to the frequency domain, obtaining frequency domain feature data, and analyzing the frequency components and phase information of the multi-source time series data in the frequency domain;

[0018] S22. Construct a Fourier neural operator model, wherein the frequency domain feature extraction module is used to extract high-dimensional features from the frequency domain feature data, the spatial feature extraction module constructs a spatial correlation model between data based on a graph neural network, and the prediction module generates prediction data of the equipment operation status through nonlinear mapping;

[0019] S23. Input the frequency domain feature data into the frequency domain feature extraction module, and extract high-dimensional feature data from the frequency domain features through multi-layer frequency domain convolution operations:

[0020] ;

[0021] in, is high-dimensional feature data, and denote the Fourier transform and inverse Fourier transform, respectively. represents element-wise multiplication, For the frequency domain convolution kernel The weight coefficient of is the residual connection parameter, is the activation function, represents the total number of learnable frequency domain convolution kernels, Represents the preprocessed multi-source time series data, is a frequency variable;

[0022] S24. Input the high-dimensional feature data into the spatial feature extraction module. Use the graph neural network to perform feature propagation and aggregation operations on the nodes in the data. Use the node adjacency relationship to construct a spatial relationship graph of the data. In the graph neural network, the multi-head attention mechanism is used to further enhance the expression ability of spatial features:

[0023] ;

[0024] in, is the learnable weight matrix, is the scaling factor of the feature dimension, is the residual connection parameter, is the node adjacency matrix, is the spatial feature matrix, is the normalized node adjacency matrix, is the normalization operation;

[0025] S25. Input the spatial features into the prediction module and perform nonlinear mapping on the spatial features through a fully connected network. The prediction module includes a multi-layer perceptron structure. During the prediction process, normalization and regularization operations are introduced to generate prediction data of the equipment operation status.

[0026] Optionally, the S3 specifically includes:

[0027] S31, initialize the memory resource allocation model and set the module's initial memory allocation ratio vector , initialize the number of ant colonies And the memory allocation path of each ant ,in Indicates the The first ant The memory ratio allocated to each module, and the pheromone vector of the module is initialized at the same time ;

[0028] S32, using the global exploration mechanism, according to the pheromone concentration Calculate the probability of each ant selecting each module , determine the memory allocation path:

[0029] :

[0030] in, For the The current pheromone concentration of each module, is the pheromone importance factor, is the importance factor of the heuristic function, Indicates the The prediction error of each module, Indicates the Memory resource consumption of each module, is the memory consumption adjustment factor, Indicates the The prediction error of each module, Indicates the Memory resource consumption of each module;

[0031] S33. Use the local development mechanism to dynamically perturb and update the memory allocation path of each ant:

[0032] ;

[0033] in, For the current The memory allocation ratio of each module, is the adjustment coefficient, Assign weights to memory resources;

[0034] S34, implement adaptive adjustment mechanism, according to the prediction error of each module and memory resource consumption Update the pheromone vector:

[0035] ;

[0036] in, Indicates the current iteration number, is the pheromone volatilization rate, is the pheromone deposition coefficient, represents the updated pheromone concentration of the jth module in the next iteration, represents the pheromone concentration of the jth module at the tth iteration;

[0037] S35. Introduce a dynamic threshold adjustment mechanism to adaptively control each module based on the ratio of prediction error to memory consumption:

[0038] ;

[0039] in, For the current The memory allocation ratio of each module, To prevent division by zero for small positive numbers, is the proportional coefficient used to adjust the steepness of the logistic function, is the preset threshold parameter, is the adjustment coefficient, Assign weights to the updated memory resources, is an exponential function;

[0040] S36. Memory resource allocation model Applied to each module in the Fourier neural operator model, the memory resource allocation strategy of each module is adjusted in real time during operation.

[0041] Optionally, the S4 specifically includes:

[0042] S41. Based on the memory resource allocation model, monitor the memory resource allocation of each module in the Fourier neural operator model in real time, and collect memory consumption data, task priority, and model prediction task load of each module;

[0043] S42, determining a memory resource shortage state, and when detecting that the current system available memory is lower than a preset safety threshold, triggering a memory recycling mechanism and switching the system state to a memory resource protection mode;

[0044] S43, automatically reclaiming the cache of low-priority tasks. Based on the task priority list, cached data with high memory resource consumption and low task priority is reclaimed first, while temporary cache space for non-critical prediction tasks is released to ensure memory resource requirements for high-priority tasks.

[0045] S44, preferentially allocating recovered memory resources to key prediction tasks, dynamically adjusting the memory resource allocation model, and reallocating memory resources based on the task priorities and memory resource requirements of different modules;

[0046] S45. Update the memory resource allocation model through the pheromone volatilization mechanism. When memory resources are tight, accelerate the pheromone volatilization speed of low-priority modules, reduce the memory allocation weight, and increase the pheromone concentration of high-priority modules.

[0047] Optionally, the S5 specifically includes:

[0048] S51, real-time equipment operation data Input Fourier neural operator model, where Indicates time;

[0049] S52, real-time equipment operation data Perform Fourier transform to obtain frequency domain representation ,in represents a frequency variable;

[0050] S53. Use the frequency domain feature extraction module of the Fourier neural operator model to perform multi-scale filtering, residual connection and activation function processing on the frequency domain representation to generate real-time high-dimensional features ;

[0051] S54, real-time high-dimensional features Decomposition into subvectors by module , and combined with the updated memory resource allocation weight , nonlinear feature fusion is used to perform dynamic feature adjustment to generate adjusted features :

[0052] ;

[0053] in, Fusion regulatory factors, is the activation function, is the total number of modules, is a logarithmic function;

[0054] S55, the adjusted features Input prediction module, through full connection layer mapping, residual connection and activation function processing, to generate the final prediction data of the equipment operation status :

[0055] ;

[0056] in, and is the weight matrix of the fully connected layer, and is the bias vector, is the activation function.

[0057] Optionally, the S7 specifically includes:

[0058] S71. Collect performance index data from the Fourier neural operator model in real time to form a performance vector ,in represents the prediction accuracy, Indicates the response time, Indicates memory utilization;

[0059] S72, performance vector The indicators in the fuzzy process are processed and the membership degree of each performance indicator is calculated. , forming a fuzzy set ,in is the number of performance indicators:

[0060] ;

[0061] in, Indicates the The actual measured value of the performance indicator, For the The steepness parameter of the logistic function corresponding to the performance index, For the The central threshold parameter of the performance indicator, For the The standard deviation parameter of the performance index, is the weighting coefficient, is an exponential function;

[0062] S73, by integrating weighted arithmetic mean and weighted geometric mean, fully considering the contribution and nonlinear characteristics of each performance index, and using fuzzy comprehensive evaluation method to generate comprehensive performance index :

[0063] ;

[0064] in, Indicates the The evaluation scores of performance indicators, is the fusion adjustment coefficient, is the number of performance indicators;

[0065] S74, based on comprehensive performance indicators With preset reference value , using exponential decay function to calculate memory resource allocation control parameters :

[0066] ;

[0067] in, is the initial control parameter, is the adjustment coefficient;

[0068] S75, allocate the calculated memory resource control parameters Applied to the memory resource allocation model, the memory resource allocation weights of each module are dynamically updated.

[0069] The beneficial effects of the present invention are:

[0070] The present invention realizes deep feature extraction and efficient prediction of multi-source test data under complex working conditions by organically combining the Fourier neural operator model with the graph neural network and the adaptive ant colony memory optimization algorithm. The present invention first maps the original time series data to the frequency domain space through Fourier transform, and uses multi-scale filtering and residual connection technology to extract high-dimensional features, and then models the spatial correlation in the data through the graph neural network to obtain a richer feature representation. At the same time, the present invention innovatively introduces the adaptive ant colony memory optimization algorithm to perform real-time dynamic regulation of the memory resource allocation of each module within the model, thereby realizing efficient memory management and task scheduling under limited system resources. The organic integration of this series of technical means has not only significantly improved the system in terms of prediction accuracy and response speed, but also demonstrated extremely high intelligence and adaptability in terms of memory resource utilization, thereby effectively avoiding system delays or operation interruptions caused by uneven distribution of memory resources.

[0071] In addition, the comprehensive performance evaluation mechanism based on fuzzy logic control adopted by the present invention further enhances the overall system's ability to monitor and provide feedback on the equipment status, enabling the model to continuously optimize the memory allocation strategy based on the real-time collected performance indicators during operation, thereby ensuring long-term stable operation under complex working conditions. The entire system has achieved intelligence, dynamics, and adaptability in all aspects of data preprocessing, feature extraction, prediction, and memory management, so that the test data analysis not only has high prediction accuracy and real-time response capabilities, but also can ensure the efficient execution of key prediction tasks in scenarios with limited system resources. Therefore, the present invention shows excellent application prospects and practical benefits in the fields of industrial equipment health monitoring, intelligent operation and maintenance, and complex working condition data analysis. Its overall advantage is that it can greatly improve data processing efficiency, optimize memory resource utilization, and achieve accurate monitoring and prediction of equipment operating status, providing solid technical support for the realization of industrial automation and intelligent manufacturing. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0073] Figure 1 This is a flow chart of the test data analysis method under complex working conditions proposed by the present invention;

[0074] Figure 2 This is a flow chart of a real-time prediction and anomaly detection module for discrete patch traversal survey path planning under the influence of multiple factors proposed by the present invention. DETAILED DESCRIPTION

[0075] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0076] refer to Figure 1 and Figure 2 , the test data analysis method under complex working conditions includes the following steps:

[0077] S1. Collect multi-source time series data under complex working conditions and pre-process the multi-source time series data;

[0078] S2. Construct a Fourier neural operator model, including a frequency domain feature extraction module, a spatial feature extraction module, and a prediction module. Input the preprocessed multi-source time series data into the Fourier neural operator model, map it to the frequency domain space through Fourier transform, and generate prediction data of the equipment operation status.

[0079] S3. Introduce an adaptive ant colony memory optimization algorithm to dynamically optimize the memory resources of the Fourier neural operator model, establish a memory resource allocation model, and dynamically adjust the memory allocation weight of the Fourier neural operator model based on the pheromone propagation mechanism;

[0080] S4. Real-time monitoring is performed based on the memory resource allocation model. When memory resources are tight, the cache of low-priority tasks is automatically recycled, and memory resources are preferentially allocated to key prediction tasks. The memory resource allocation model is updated through the pheromone volatilization mechanism.

[0081] S5. Use the Fourier neural operator model to predict the real-time input device operation data. During the prediction process, combine it with the updated memory resource allocation model to dynamically adjust the memory resource allocation strategy in real time.

[0082] S6. Compare the prediction results with the actual monitoring data. When the prediction error exceeds the preset threshold, the anomaly detection module is triggered to identify potential risks in the equipment operation status and generate an early warning signal, which is notified to the operation and maintenance personnel through interface alarms, SMS push or email;

[0083] S7. The Fourier neural operator model is evaluated through the fuzzy logic control algorithm, and the memory resource allocation model is dynamically updated using the pheromone data collected by the adaptive ant colony memory optimization algorithm.

[0084] In this embodiment, S2 specifically includes:

[0085] S21, performing Fourier transform on the preprocessed multi-source time series data, converting the multi-source time series data from the time domain to the frequency domain, obtaining frequency domain feature data, and analyzing the frequency components and phase information of the multi-source time series data in the frequency domain;

[0086] S22. Construct a Fourier neural operator model, wherein the frequency domain feature extraction module is used to extract high-dimensional features from the frequency domain feature data, the spatial feature extraction module constructs a spatial correlation model between data based on a graph neural network, and the prediction module generates prediction data of the equipment operation status through nonlinear mapping;

[0087] S23. Input the frequency domain feature data into the frequency domain feature extraction module, and extract high-dimensional feature data from the frequency domain features through multi-layer frequency domain convolution operations:

[0088] ;

[0089] in, is high-dimensional feature data, and denote the Fourier transform and inverse Fourier transform, respectively. represents element-wise multiplication, For the frequency domain convolution kernel The weight coefficient of is the residual connection parameter, is the activation function, represents the total number of learnable frequency domain convolution kernels, Represents the preprocessed multi-source time series data, is a frequency variable;

[0090] S24. Input the high-dimensional feature data into the spatial feature extraction module. Use the graph neural network to perform feature propagation and aggregation operations on the nodes in the data. Use the node adjacency relationship to construct a spatial relationship graph of the data. In the graph neural network, the multi-head attention mechanism is used to further enhance the expression ability of spatial features:

[0091] ;

[0092] in, is the learnable weight matrix, is the scaling factor of the feature dimension, is the residual connection parameter, is the node adjacency matrix, is the spatial feature matrix, is the normalized node adjacency matrix, is the normalization operation;

[0093] S25. Input the spatial features into the prediction module and perform nonlinear mapping on the spatial features through a fully connected network. The prediction module includes a multi-layer perceptron structure. During the prediction process, normalization and regularization operations are introduced to generate prediction data of the equipment operation status.

[0094] In this embodiment, S3 specifically includes:

[0095] S31, initialize the memory resource allocation model and set the module's initial memory allocation ratio vector , initialize the number of ant colonies And the memory allocation path of each ant ,in Indicates the The first ant The memory ratio allocated to each module, and the pheromone vector of the module is initialized at the same time ;

[0096] S32, using the global exploration mechanism, according to the pheromone concentration Calculate the probability of each ant selecting each module , determine the memory allocation path:

[0097] :

[0098] in, For the The current pheromone concentration of each module, is the pheromone importance factor, is the importance factor of the heuristic function, Indicates the The prediction error of each module, Indicates the Memory resource consumption of each module, is the memory consumption adjustment factor, Indicates the The prediction error of each module, Indicates the Memory resource consumption of each module;

[0099] S33. Use the local development mechanism to dynamically perturb and update the memory allocation path of each ant:

[0100] ;

[0101] in, For the current The memory allocation ratio of each module, is the adjustment coefficient, Assign weights to memory resources;

[0102] S34, implement adaptive adjustment mechanism, according to the prediction error of each module and memory resource consumption Update the pheromone vector:

[0103] ;

[0104] in, Indicates the current iteration number, is the pheromone volatilization rate, is the pheromone deposition coefficient, represents the updated pheromone concentration of the jth module in the next iteration, represents the pheromone concentration of the jth module at the tth iteration;

[0105] S35. Introduce a dynamic threshold adjustment mechanism to adaptively control each module based on the ratio of prediction error to memory consumption:

[0106] ;

[0107] in, For the current The memory allocation ratio of each module, To prevent division by zero for small positive numbers, is the proportional coefficient used to adjust the steepness of the logistic function, is the preset threshold parameter, is the adjustment coefficient, Assign weights to the updated memory resources, is an exponential function;

[0108] S36. Memory resource allocation model Applied to each module in the Fourier neural operator model, the memory resource allocation strategy of each module is adjusted in real time during operation.

[0109] In this embodiment, the S4 specifically includes:

[0110] S41. Based on the memory resource allocation model, monitor the memory resource allocation of each module in the Fourier neural operator model in real time, and collect memory consumption data, task priority, and model prediction task load of each module;

[0111] S42, determining a memory resource shortage state, and when detecting that the current system available memory is lower than a preset safety threshold, triggering a memory recycling mechanism and switching the system state to a memory resource protection mode;

[0112] S43, automatically reclaiming the cache of low-priority tasks. Based on the task priority list, cached data with high memory resource consumption and low task priority is reclaimed first, while temporary cache space for non-critical prediction tasks is released to ensure memory resource requirements for high-priority tasks.

[0113] S44, preferentially allocating recovered memory resources to key prediction tasks, dynamically adjusting the memory resource allocation model, and reallocating memory resources based on the task priorities and memory resource requirements of different modules;

[0114] S45. Update the memory resource allocation model through the pheromone volatilization mechanism. When memory resources are tight, accelerate the pheromone volatilization speed of low-priority modules, reduce the memory allocation weight, and increase the pheromone concentration of high-priority modules.

[0115] In this embodiment, the S5 specifically includes:

[0116] S51, real-time equipment operation data Input Fourier neural operator model, where Indicates time;

[0117] S52, real-time equipment operation data Perform Fourier transform to obtain frequency domain representation ,in represents a frequency variable;

[0118] S53. Use the frequency domain feature extraction module of the Fourier neural operator model to perform multi-scale filtering, residual connection and activation function processing on the frequency domain representation to generate real-time high-dimensional features ;

[0119] S54, real-time high-dimensional features Decomposition into subvectors by module , and combined with the updated memory resource allocation weight , nonlinear feature fusion is used to perform dynamic feature adjustment to generate adjusted features :

[0120] ;

[0121] in, Fusion regulatory factors, is the activation function, is the total number of modules, is a logarithmic function;

[0122] S55, the adjusted features Input prediction module, through full connection layer mapping, residual connection and activation function processing, to generate the final prediction data of the equipment operation status :

[0123] ;

[0124] in, and is the weight matrix of the fully connected layer, and is the bias vector, is the activation function.

[0125] In this embodiment, the S7 specifically includes:

[0126] S71. Collect performance index data from the Fourier neural operator model in real time to form a performance vector ,in represents the prediction accuracy, Indicates the response time, Indicates memory utilization;

[0127] S72, performance vector The indicators in the fuzzy process are processed and the membership degree of each performance indicator is calculated. , forming a fuzzy set ,in is the number of performance indicators:

[0128] ;

[0129] in, Indicates the The actual measured value of the performance indicator, For the The steepness parameter of the logistic function corresponding to the performance index, For the The central threshold parameter of the performance indicator, For the The standard deviation parameter of the performance index, is the weighting coefficient, is an exponential function;

[0130] S73, by integrating weighted arithmetic mean and weighted geometric mean, fully considering the contribution and nonlinear characteristics of each performance index, and using fuzzy comprehensive evaluation method to generate comprehensive performance index :

[0131] ;

[0132] in, Indicates the The evaluation scores of performance indicators, is the fusion adjustment coefficient, is the number of performance indicators;

[0133] S74, based on comprehensive performance indicators With preset reference value , using exponential decay function to calculate memory resource allocation control parameters :

[0134] ;

[0135] in, is the initial control parameter, is the adjustment coefficient;

[0136] S75, allocate the calculated memory resource control parameters Applied to the memory resource allocation model, the memory resource allocation weights of each module are dynamically updated.

[0137] Example 1:

[0138] In order to verify the feasibility of the present invention in implementation, the present invention was applied to a large manufacturing enterprise, in which a number of CNC machine tools, automated assembly robots and intelligent detection equipment are installed in the production workshop of the enterprise. Each device is equipped with a variety of sensors such as temperature, vibration, pressure, and current for real-time collection of equipment operation data. However, when processing complex working condition data, traditional test data analysis methods have problems such as insufficient prediction accuracy, response delays and uneven allocation of memory resources, resulting in untimely equipment failure warnings, affecting production efficiency and equipment safety. To solve these problems, the present invention proposes a test data analysis method under complex working conditions, and in actual applications, performs real-time prediction of equipment operating status and fault warnings, thereby effectively ensuring the continuity and safety of the production process.

[0139] In this embodiment, first, data such as temperature, vibration, pressure, and current are collected in real time by sensors installed on each device, and after data cleaning, missing value filling, outlier removal, and normalization, the sensor data are integrated into multi-source time series data in a unified format. Then, the Fourier neural operator model constructed in the present invention is used to perform Fourier transform on the processed time series data, converting the data from the time domain to the frequency domain, and then generating high-dimensional frequency domain features through multi-scale filtering, residual connection, and activation function processing. Subsequently, the model uses a graph neural network to extract spatial features from the frequency domain features, forming a rich feature representation, and finally generating prediction data for the device operating status. At the same time, in order to cope with the challenge of limited memory resources in the edge computing platform, the present invention introduces an adaptive ant colony memory optimization algorithm to dynamically optimize the memory resource allocation of each module of the model. By real-time monitoring of memory resource status, automatic recovery of low-priority task caches, and prioritization of memory resources to key prediction tasks, the system response speed and memory utilization efficiency are significantly improved.

[0140] In actual application, from December 2023 to May 2024, the system implemented the method of the present invention in a key workshop of a manufacturing enterprise to continuously monitor and predict the status of equipment. The implementation results show that the implementation plan of the present invention significantly improves the accuracy and timeliness of equipment failure warnings. After the improvement, the accuracy of equipment failure warnings increased from the original 82% to 95%, the average system response time decreased from 450 milliseconds to 310 milliseconds, and the memory resource utilization rate increased from 65% to 82%. In addition, the number of failure warnings for key equipment increased from an average of 5 times per month to an average of 9 times per month, and the average advance warning time was extended from 1.5 hours to 4.2 hours, effectively avoiding production interruptions and safety accidents caused by equipment failures. According to statistics, during the six consecutive months of monitoring, the cumulative downtime caused by equipment failures was reduced by nearly 48 hours, and the maintenance cost was reduced by about 30%, which greatly improved production efficiency and enterprise economic benefits.

[0141] Table 1 Key performance indicator comparison table

[0142] ;

[0143] Based on the data in the table, we can clearly see the significant advantages of the present invention in analyzing test data under complex working conditions. In terms of prediction accuracy, after implementing the method of the present invention, the system gradually improved from 82% in December 2019 to 95% in May 2020, achieving an increase of 13 percentage points. This shows that the combination of the Fourier neural operator model and the adaptive ant colony memory optimization algorithm effectively enhances the ability to predict the operating status of the equipment, greatly reduces the prediction error, and significantly improves the stability and reliability of the system.

[0144] In terms of system response speed, the method of the present invention reduces the average response time from 450 milliseconds to 310 milliseconds, improving the overall real-time performance of the system. In industrial production scenarios, rapid response is crucial for timely warning of equipment anomalies and ensuring continuous operation of production lines. Through intelligent memory resource allocation strategies, the present invention enables the system to maintain rapid response even under high load conditions, effectively reducing security risks caused by prediction delays.

[0145] Memory resource utilization has also significantly improved, rising from 65% to 82%. This improvement stems from the adaptive ant colony memory optimization algorithm used in this invention. By monitoring memory resource usage in real time, dynamically reclaiming the cache of low-priority tasks, and prioritizing memory resources for critical prediction tasks, this algorithm avoids memory waste and memory overflow, thereby improving overall resource utilization efficiency.

[0146] The number of equipment failure warnings increased from an average of 5 to 9 per month, and the average lead time increased from 1.5 hours to 4.2 hours. This demonstrates that the method not only increases the frequency of warnings but also extends their lead time, giving operations and maintenance personnel more time to implement countermeasures and reducing the economic losses caused by unplanned equipment downtime. The data in the table shows that monthly equipment downtime has been reduced from 12 hours to 7 hours, and maintenance costs have also been reduced from no significant savings with traditional methods to a maximum of 30%. These data further demonstrate the significant economic benefits of the present invention in practical applications.

[0147] In summary, the comparative data in the above table demonstrates that the proposed method demonstrates significant improvements in prediction accuracy, response speed, memory resource utilization, warning frequency, and maintenance costs. This not only demonstrates the technical advantages of the present invention in predicting equipment operating status under complex operating conditions, but also provides strong support for intelligent operation and maintenance and equipment management in actual industrial scenarios, contributing to the safer, more efficient, and more intelligent production process.

[0148] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. The test data analysis method under complex working conditions is characterized by: The steps include: S1. Collect multi-source time series data under complex working conditions and pre-process the multi-source time series data; S2. Construct a Fourier neural operator model, including a frequency domain feature extraction module, a spatial feature extraction module, and a prediction module. Input the preprocessed multi-source time series data into the Fourier neural operator model, map it to the frequency domain space through Fourier transform, and generate prediction data of the equipment operation status. S3. Introduce an adaptive ant colony memory optimization algorithm to dynamically optimize the memory resources of the Fourier neural operator model, establish a memory resource allocation model, and dynamically adjust the memory allocation weight of the Fourier neural operator model based on the pheromone propagation mechanism; S4. Real-time monitoring is performed based on the memory resource allocation model. When memory resources are tight, the cache of low-priority tasks is automatically recycled, and memory resources are preferentially allocated to key prediction tasks. The memory resource allocation model is updated through the pheromone volatilization mechanism. S5. Use the Fourier neural operator model to predict the real-time input device operation data. During the prediction process, combine it with the updated memory resource allocation model to dynamically adjust the memory resource allocation strategy in real time. S6. Compare the prediction results with the actual monitoring data. When the prediction error exceeds the preset threshold, the anomaly detection module is triggered to identify potential risks in the equipment operation status and generate an early warning signal, which is notified to the operation and maintenance personnel through interface alarms, SMS push or email; S7. The Fourier neural operator model is evaluated through the fuzzy logic control algorithm, and the memory resource allocation model is dynamically updated using the pheromone data collected by the adaptive ant colony memory optimization algorithm.

2. The test data analysis method under complex working conditions according to claim 1, characterized in that: The S2 specifically includes: S21, performing Fourier transform on the preprocessed multi-source time series data, converting the multi-source time series data from the time domain to the frequency domain, obtaining frequency domain feature data, and analyzing the frequency components and phase information of the multi-source time series data in the frequency domain; S22. Construct a Fourier neural operator model, wherein the frequency domain feature extraction module is used to extract high-dimensional features from the frequency domain feature data, the spatial feature extraction module constructs a spatial correlation model between data based on a graph neural network, and the prediction module generates prediction data of the equipment operation status through nonlinear mapping; S23. Input the frequency domain feature data into the frequency domain feature extraction module, and extract high-dimensional feature data from the frequency domain features through multi-layer frequency domain convolution operations: ； in, is high-dimensional feature data, and denote the Fourier transform and inverse Fourier transform, respectively. represents element-wise multiplication, For the frequency domain convolution kernel The weight coefficient of is the residual connection parameter, is the activation function, represents the total number of learnable frequency domain convolution kernels, Represents the preprocessed multi-source time series data, is a frequency variable; S24. Input the high-dimensional feature data into the spatial feature extraction module. Use the graph neural network to perform feature propagation and aggregation operations on the nodes in the data. Use the node adjacency relationship to construct a spatial relationship graph of the data. In the graph neural network, the multi-head attention mechanism is used to further enhance the expression ability of spatial features: ； in, is the learnable weight matrix, is the scaling factor of the feature dimension, is the residual connection parameter, is the node adjacency matrix, is the spatial feature matrix, is the normalized node adjacency matrix, is the normalization operation; S25. Input the spatial features into the prediction module and perform nonlinear mapping on the spatial features through a fully connected network. The prediction module includes a multi-layer perceptron structure. During the prediction process, normalization and regularization operations are introduced to generate prediction data of the equipment operation status.

3. The test data analysis method under complex working conditions according to claim 1, characterized in that: The S3 specifically includes: S31, initialize the memory resource allocation model and set the module's initial memory allocation ratio vector , initialize the number of ant colonies And the memory allocation path of each ant ,in Indicates the The first ant The memory ratio allocated to each module, and the pheromone vector of the module is initialized at the same time ; S32, using the global exploration mechanism, according to the pheromone concentration Calculate the probability of each ant selecting each module , determine the memory allocation path: ： in, For the The current pheromone concentration of each module, is the pheromone importance factor, is the importance factor of the heuristic function, Indicates the The prediction error of each module, Indicates the Memory resource consumption of each module, is the memory consumption adjustment factor, Indicates the The prediction error of each module, Indicates the Memory resource consumption of each module; S33. Use the local development mechanism to dynamically perturb and update the memory allocation path of each ant: ； in, For the current The memory allocation ratio of each module, is the adjustment coefficient, Assign weights to memory resources; S34, implement adaptive adjustment mechanism, according to the prediction error of each module and memory resource consumption Update the pheromone vector: ； in, Indicates the current iteration number, is the pheromone volatilization rate, is the pheromone deposition coefficient, represents the updated pheromone concentration of the jth module in the next iteration, represents the pheromone concentration of the jth module at the tth iteration; S35. Introduce a dynamic threshold adjustment mechanism to adaptively control each module based on the ratio of prediction error to memory consumption: ； in, For the current The memory allocation ratio of each module, To prevent division by zero for small positive numbers, is the proportional coefficient used to adjust the steepness of the logistic function, is the preset threshold parameter, is the adjustment coefficient, Assign weights to the updated memory resources, is an exponential function; S36. Memory resource allocation model Applied to each module in the Fourier neural operator model, the memory resource allocation strategy of each module is adjusted in real time during operation.

4. The method for analyzing test data under complex working conditions according to claim 1, characterized in that: The S4 specifically includes: S41. Based on the memory resource allocation model, monitor the memory resource allocation of each module in the Fourier neural operator model in real time, and collect memory consumption data, task priority, and model prediction task load of each module; S42, determining a memory resource shortage state, and when detecting that the current system available memory is lower than a preset safety threshold, triggering a memory recycling mechanism and switching the system state to a memory resource protection mode; S43, automatically reclaiming the cache of low-priority tasks. Based on the task priority list, cached data with high memory resource consumption and low task priority is reclaimed first, while temporary cache space for non-critical prediction tasks is released to ensure memory resource requirements for high-priority tasks. S44, preferentially allocating recovered memory resources to key prediction tasks, dynamically adjusting the memory resource allocation model, and reallocating memory resources based on the task priorities and memory resource requirements of different modules; S45. Update the memory resource allocation model through the pheromone volatilization mechanism. When memory resources are tight, accelerate the pheromone volatilization speed of low-priority modules, reduce the memory allocation weight, and increase the pheromone concentration of high-priority modules.

5. The test data analysis method under complex working conditions according to claim 1, characterized in that: The S5 specifically includes: S51, real-time equipment operation data Input Fourier neural operator model, where Indicates time; S52, real-time equipment operation data Perform Fourier transform to obtain frequency domain representation ,in represents a frequency variable; S53. Use the frequency domain feature extraction module of the Fourier neural operator model to perform multi-scale filtering, residual connection and activation function processing on the frequency domain representation to generate real-time high-dimensional features ; S54, real-time high-dimensional features Decomposition into subvectors by module , and combined with the updated memory resource allocation weight , nonlinear feature fusion is used to perform dynamic feature adjustment to generate adjusted features : ； in, Fusion regulatory factors, is the activation function, is the total number of modules, is a logarithmic function; S55, the adjusted features Input prediction module, through full connection layer mapping, residual connection and activation function processing, to generate the final prediction data of the equipment operation status : ； in, and is the weight matrix of the fully connected layer, and is the bias vector, is the activation function.

6. The method for analyzing test data under complex working conditions according to claim 1, characterized in that: The S7 specifically includes: S71. Collect performance index data from the Fourier neural operator model in real time to form a performance vector ,in represents the prediction accuracy, Indicates the response time, Indicates memory utilization; S72, performance vector The indicators in the fuzzy process are processed and the membership degree of each performance indicator is calculated. , forming a fuzzy set ,in is the number of performance indicators: ； in, Indicates the The actual measured value of the performance indicator, For the The steepness parameter of the logistic function corresponding to the performance index, For the The central threshold parameter of the performance indicator, For the The standard deviation parameter of the performance index, is the weighting coefficient, is an exponential function; S73, by integrating weighted arithmetic mean and weighted geometric mean, fully considering the contribution and nonlinear characteristics of each performance index, and using fuzzy comprehensive evaluation method to generate comprehensive performance index : ； in, Indicates the The evaluation scores of performance indicators, is the fusion adjustment coefficient, is the number of performance indicators; S74, based on comprehensive performance indicators With preset reference value , using exponential decay function to calculate memory resource allocation control parameters : ； in, is the initial control parameter, is the adjustment coefficient; S75, allocate the calculated memory resource control parameters Applied to the memory resource allocation model, the memory resource allocation weights of each module are dynamically updated.

Images