A digital terminal data processing method, electronic equipment and digital terminal

Abstract

The present invention provides a digital terminal data processing method, an electronic device and a digital terminal, which relate to the field of data processing technology. The method collects multimodal data in real time, generates a reconstructed data set after processing, performs anomaly detection on the extracted feature data based on an anomaly detection model, marks anomalies and anomaly features, classifies anomalies through a classification model, identifies the anomaly type of anomalies, automatically triggers anomaly detection model retraining, detects multimodal data collected in real time based on an updated anomaly detection model, compares the prediction result with a preset warning threshold, and if the threshold is exceeded, issues a warning signal and generates a fault report. The digital terminal data processing method and system of the present invention can significantly improve the efficiency and accuracy of data processing, enhance the ability of anomaly detection and fault management, and provide users with a more intelligent, efficient and reliable data processing solution.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a digital terminal data processing method, electronic equipment and a digital terminal. Background Art

[0002] With the widespread application of IoT devices and smart terminals, a large amount of data is collected by sensors and transmitted to the background system for analysis. In modern industry, smart manufacturing and other scenarios, the normal operation of equipment is crucial, so real-time monitoring of equipment operation status, early detection and early warning of faults are of great significance. However, the data processing method of digital terminals in the prior art has the following problems:

[0003] The data is not real-time enough and cannot respond to abnormalities quickly;

[0004] The accuracy of anomaly detection is not high, and there are omissions and false detections;

[0005] The lack of effective fault prediction models makes it difficult to provide timely warnings.

[0006] In response to the above problems, there is an urgent need to provide a data processing method based on digital terminals that can realize real-time monitoring, anomaly detection, fault prediction and early warning functions, and effectively improve the safety and stability of the equipment. Summary of the invention

[0007] In view of this, in order to overcome the problems of insufficient real-time performance, low detection accuracy and inaccurate prediction and warning in the prior art, the present invention proposes a digital terminal data processing method, an electronic device and a digital terminal, which can perform efficient data processing on the digital terminal to realize the equipment's abnormality detection, fault prediction and warning functions.

[0008] The present invention is implemented by the following technical solutions:

[0009] In a first aspect, the present invention provides a method for processing digital terminal data, the method comprising the following steps:

[0010] Collect multimodal data in real time based on sensors installed on terminal devices, and receive multimodal data processing requests sent from terminal devices;

[0011] Allocate computing resources according to the priority of multimodal data processing requests and dynamically adjust the request queue order;

[0012] Parse the multimodal data processing request, discretize the multimodal data obtained by the analysis, convert the continuous data into discrete format, and interpolate the missing data to obtain discrete approximate data;

[0013] Extract and fuse features through discrete approximation data, standardize the fused feature data, and generate a reconstructed data set;

[0014] Based on the anomaly detection model, the extracted feature data is subjected to anomaly detection, anomalies and anomaly feature marks are marked, anomalies are classified through the classification model, and the anomaly type of the anomaly is identified;

[0015] Based on historical data and anomalies detected in real time, the anomaly detection model is automatically retrained. The multimodal data collected in real time is detected based on the updated anomaly detection model, and the prediction results are compared with the preset warning threshold. If the threshold is exceeded, a warning signal is issued and a fault report is generated.

[0016] As a further solution of the present invention, when allocating computing resources according to the priority of the multimodal data processing request and dynamically adjusting the request queue order, the following steps are included:

[0017] Receive and parse multimodal data processing requests, and perform weighted round-robin scheduling based on the priority and urgency of the requests;

[0018] Dynamically adjust the priority queue sorting in real time according to the priority of multimodal data processing requests, and dynamically allocate computing resources;

[0019] Assign a weight value to each request based on its priority and urgency, process the request according to the weight value in each round of scheduling, and set the processing times based on the weight;

[0020] During each scheduling, poll the requests in the queue and process the weighted number of each request;

[0021] The order and weight of requests in the request queue are dynamically adjusted based on the processing of requests and the addition of new requests.

[0022] As a further solution of the present invention, generating a reconstructed data set includes the following steps:

[0023] Parse and obtain continuous time series data collected from sensors;

[0024] Continuous data is sampled at predetermined time intervals and converted into discrete data points;

[0025] Map the data values ​​to the integer level range of discrete numerical values ​​to obtain discrete data;

[0026] Detect missing points in discrete data and fill in the missing data by linear interpolation;

[0027] Feature extraction and fusion are performed, and the fused feature data is standardized to generate a reconstructed data set.

[0028] As a further solution of the present invention, the reconstructed data set is a complete data set including all original data points, interpolated and filled data points, and feature vectors.

[0029] As a further solution of the present invention, when obtaining discrete approximation data, it also includes processing the discretized data through a normalization formula and fitting the normalized data using piecewise approximation; wherein: when fitting the normalized data using piecewise approximation, the following steps are included:

[0030] Segment the normalized discrete data according to the change points;

[0031] Apply cubic spline to each discrete data segment for piecewise approximation fitting, and use the least square method to determine the parameters of the fitting function;

[0032] The fitting results of all segments are synthesized to smooth the function values ​​and derivatives at the connection points to generate a complete discrete approximation to the data.

[0033] As a further solution of the present invention, before performing anomaly detection on the extracted feature data based on the anomaly detection model, it also includes anomaly detection model training, the steps are as follows:

[0034] Map the feature vectors in the reconstructed dataset to the mean and variance of the latent space through the encoder network, and use the mean and variance to generate samples of the latent space;

[0035] Decode the latent space samples into sample data through the decoder network;

[0036] Construct the loss function of VAE model training, calculate the reconstruction error as the basic indicator of anomaly detection based on the generated sample data, record the reconstruction error of each sample data, and set the anomaly threshold based on the reconstruction error distribution of the training data;

[0037] When the reconstruction error of the sample data exceeds the abnormal threshold, it is marked as an abnormal point.

[0038] As a further solution of the present invention, the classification model training includes:

[0039] Extract outlier features from anomaly detection models through reconstruction errors;

[0040] Manually label the types of outliers and divide the labeled data into training sets and test sets;

[0041] The extracted outlier features are used as input, the category of the outlier is used as the output label, the classification model is trained using the training set, and the model parameters are adjusted through cross-validation to obtain a trained classification model;

[0042] Use the trained classification model to classify the outliers and identify the anomaly types of the outliers.

[0043] As a further solution of the present invention, in the digital terminal data processing method, when multimodal data is collected in real time based on sensors installed on the terminal device, the data is processed based on the collaboration mechanism of edge computing nodes and cloud platforms; wherein an edge computing node for receiving and processing locally collected multimodal data is deployed on the terminal device;

[0044] Run anomaly detection models on edge nodes to perform real-time anomaly detection on preprocessed data and mark abnormal data;

[0045] Compress and filter the data after anomaly detection, and send the compressed anomaly events and filtered data to the cloud platform;

[0046] The classification model deployed on the cloud platform classifies the abnormal points of abnormal events and identifies the abnormal types of the abnormal points.

[0047] In a second aspect, the present invention further includes a digital terminal data processing system, the system comprising:

[0048] Multimodal data acquisition module: configured on the terminal device, used to receive multimodal data collected by the sensor in real time and receive modal data processing requests sent from the terminal device;

[0049] Computing resource allocation module: used to allocate computing resources according to the priority of multimodal data processing requests and dynamically adjust the request queue order;

[0050] Data processing module: used to parse multimodal data processing requests and obtain reconstructed data sets after processing multimodal data;

[0051] Anomaly detection module: used to perform anomaly detection on the extracted feature data through anomaly detection model, marking anomalies and anomaly features;

[0052] Classification and recognition module: used to classify abnormal points through classification models and identify the abnormal types of abnormal points;

[0053] Early warning monitoring module: It is used to detect multimodal data collected in real time based on the updated anomaly detection model, compare the prediction results with the preset early warning threshold, and if the threshold is exceeded, a warning signal is issued and a fault report is generated.

[0054] As a further solution of the present invention, the data processing module includes:

[0055] The data parsing submodule is used to parse multimodal data processing requests, perform discretization processing, convert continuous data into discrete format, and interpolate missing data to generate discrete approximate data;

[0056] The extraction and fusion submodule is used to perform feature extraction, feature fusion and standardization on discrete approximation data to generate a reconstructed data set.

[0057] As a further solution of the present invention, the digital terminal data processing system further includes:

[0058] The model update retraining module is used to automatically trigger the retraining of the anomaly detection model and update the model parameters based on historical data and anomalies detected in real time.

[0059] The present invention also includes a computer device, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor executes the digital terminal data processing method.

[0060] The present invention also includes a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute the digital terminal data processing method.

[0061] Compared with the prior art, the digital intelligent terminal data processing method, electronic device and digital intelligent terminal provided by the present invention have the following beneficial effects:

[0062] 1. Improved the ability of real-time data processing and response. The present invention uses real-time multimodal data and dynamically allocates computing resources based on the execution priority of multimodal data processing requests and adjusts the processing order of request queues, which significantly reduces the delay in data transmission to the cloud, improves the system's processing efficiency for multimodal data, realizes dynamic allocation of resources, ensures that key requests are processed in a timely manner, and optimizes the system's response time and resource utilization.

[0063] 2. Data quality optimization is achieved through discretization and interpolation processing, as well as feature extraction and fusion. The present invention converts continuous data into discrete format and interpolates missing data, reducing the impact of data loss and improving data integrity and accuracy; through feature extraction and fusion, the generated reconstructed data set has higher representativeness and processing accuracy, which helps to improve the analysis and prediction accuracy of subsequent models.

[0064] 3. Enhanced anomaly detection and classification capabilities, achieving accurate anomaly detection and automatic model update. The present invention combines anomaly detection models with classification models, which can timely identify anomalies in data and accurately classify anomaly types, thereby improving the sensitivity and accuracy of fault detection. Automatically triggering model retraining based on real-time and historical data ensures that the accuracy of the anomaly detection model continues to improve as data changes, enhancing the system's ability to adapt to new anomaly patterns.

[0065] 4. Possess early warning and fault management capabilities. By comparing the prediction results with the preset early warning threshold, the present invention can issue early warning signals when an abnormality occurs, helping users to identify potential problems in advance; the generated fault report provides detailed information on the abnormality, helping users to quickly locate and solve the problem, improving the system maintenance efficiency and user experience.

[0066] To sum up, the digital terminal data processing method and system of the present invention can significantly improve the efficiency and accuracy of data processing, enhance the capabilities of anomaly detection and fault management, and provide users with more intelligent, efficient and reliable data processing solutions.

[0067] These and other aspects of the present invention will be more clearly understood in the following description of the embodiments. It should be understood that the above general description and the following detailed description are only exemplary and explanatory and cannot limit the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or related technologies, the following briefly introduces the drawings required for use in the exemplary embodiments or related technical descriptions. The 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 to the present invention. In the drawings:

[0069] Figure 1 The present invention is a flowchart of a method for processing digital terminal data according to an embodiment of the present invention.

[0070] Figure 2 The present invention provides a flowchart for allocating computing resources and dynamically adjusting the order of request queues in the digital terminal data processing method according to an embodiment of the present invention.

[0071] Figure 3 The present invention is a flowchart of fitting normalized data using piecewise approximation in the digital terminal data processing method according to an embodiment of the present invention.

[0072] Figure 4 The present invention provides a flowchart for generating a reconstructed data set in a digital terminal data processing method according to an embodiment of the present invention.

[0073] Figure 5The present invention is a flowchart of anomaly detection model training in the digital terminal data processing method according to an embodiment of the present invention.

[0074] Figure 6 The present invention is a flowchart of the classification model training in the digital terminal data processing method according to an embodiment of the present invention. DETAILED DESCRIPTION

[0075] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0076] In some of the processes described in the specification and claims of the present invention and the above-mentioned figures, multiple operations that appear in a specific order are included, but it should be clearly understood that these operations may not be executed in the order in which they appear in this article or executed in parallel. The serial numbers of the operations, such as 101, 102, etc., are only used to distinguish different operations, and the serial numbers themselves do not represent any execution order. In addition, these processes may include more or fewer operations, and these operations may be executed in sequence or in parallel. It should be noted that the descriptions of "first", "second", etc. in this article are used to distinguish different messages, devices, modules, etc., do not represent the order of precedence, and do not limit the "first" and "second" to be different types.

[0077] The following will be combined with the drawings in the exemplary embodiments of the present invention to clearly and completely describe the technical solutions in the exemplary embodiments of the present invention. Obviously, the exemplary embodiments described are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0078] In order to solve the problems existing in the data processing methods of digital terminals in the prior art, such as insufficient data real-time performance, inability to quickly respond to anomalies; low anomaly detection accuracy, and the existence of missed detections and false detections; lack of effective fault prediction models, and difficulty in providing timely warnings, the present invention provides a digital terminal data processing method, electronic device, and digital terminal, which can realize real-time monitoring, anomaly detection, fault prediction and warning functions, and effectively improve the safety and stability of the equipment.

[0079] The technical solution of the present invention is further described below in conjunction with specific embodiments:

[0080] See also Figure 1 As shown, Figure 1A flowchart of a digital terminal data processing method provided by the present invention. A digital terminal data processing method provided in one embodiment of the present invention includes the following steps:

[0081] Step S10: collecting multimodal data in real time based on sensors installed on the terminal device, and receiving a multimodal data processing request sent from the terminal device.

[0082] In this step, the multimodal data obtained includes but is not limited to:

[0083] Temperature data: Temperature changes of the environment or equipment.

[0084] Humidity data: ambient humidity.

[0085] Vibration data: The vibration status of the device.

[0086] Pressure data: The pressure of the equipment or environment.

[0087] Optical data: light intensity, infrared data.

[0088] Audio data: ambient or device sound.

[0089] Image / video data: Visual information collected by the camera.

[0090] Current / voltage data: device electrical status.

[0091] GPS data: Geographic location information.

[0092] Collected in real time through different sensors, forming multimodal data input.

[0093] For example, when multimodal data is collected in real time through sensors installed on terminal devices, the temperature sensor records the current room temperature once a minute; the humidity sensor records the humidity level once a minute; the camera captures the video stream in real time for monitoring; and the microphone captures the ambient sound in real time. When a request is generated, the user sends a data processing request through the smart home APP, requiring abnormal detection of temperature and humidity data, and sending the abnormal data to the mobile phone notification. After receiving the request, it is added to the request queue.

[0094] Step S20: allocate computing resources according to the priority of the multimodal data processing request, and dynamically adjust the request queue order.

[0095] In this step, see Figure 2 As shown, when allocating computing resources according to the priority of multimodal data processing requests and dynamically adjusting the request queue order, the following steps are included:

[0096] Step S201: receiving and parsing a multimodal data processing request, and performing weighted round-robin scheduling according to the priority and urgency of the request;

[0097] Step S202: dynamically adjust the priority queue order in real time according to the priority of the multimodal data processing request, and dynamically allocate computing resources;

[0098] Step S203: assign a weight value to each request according to the priority and urgency of the request, process the request according to the weight value in each round of scheduling, and set the processing times according to the weight;

[0099] Step S204: poll the requests in the queue each time the scheduling is performed, and process the weighted number of each request;

[0100] Step S205: dynamically adjust the order and weight of requests in the request queue according to the processing of requests and the addition of new requests.

[0101] Exemplarily, when parsing received requests and identifying the priority of processing tasks, anomaly detection takes precedence over normal data analysis. When requests are assigned weight values, the weight of anomaly detection requests is set high (e.g., 100) and the weight of normal data analysis requests is set low (e.g., 30). High-weight requests (anomaly detection) are processed first, and more processing power and memory are allocated to high-priority anomaly detection requests.

[0102] Step S30: parse the multimodal data processing request, discretize the multimodal data obtained by the analysis, convert the continuous data into a discrete format, and interpolate the missing data to obtain discrete approximate data.

[0103] In this step, when obtaining discrete approximation data, it also includes processing the discretized data through a normalization formula and fitting the normalized data using a piecewise approximation; wherein, see Figure 3 As shown in Figure 1, the following steps are included when fitting the normalized data using piecewise approximation:

[0104] Step S301, segmenting the normalized discrete data according to the change points;

[0105] Step S302: Apply cubic spline to each discrete data segment for piecewise approximation fitting, and use the least squares method to determine the parameters of the fitting function;

[0106] Step S303: synthesize the fitting results of all segments, make the function values ​​and derivatives of the connection points smooth, and generate complete discrete approximation data.

[0107] Step S40: extract and fuse features through discrete approximation data, standardize the fused feature data, and generate a reconstructed data set.

[0108] In this step, see Figure 4 As shown, generating a reconstructed data set includes the following steps:

[0109] Step S401, parsing and acquiring continuous time series data collected from the sensor;

[0110] Step S402: sampling the continuous data at a predetermined time interval and converting it into discrete data points;

[0111] Step S403: Map the data value to an integer level range of discrete numerical values ​​to obtain discrete data;

[0112] Step S404: Detect missing points in discrete data and fill in the missing data by linear interpolation;

[0113] Step S405: extract and fuse features, perform standardization on the fused feature data, and generate a reconstructed data set.

[0114] The reconstructed data set is a complete data set including all original data points, interpolated and filled data points, and feature vectors.

[0115] Taking the intelligent city traffic monitoring system as an example, the sensors installed on traffic lights and surveillance cameras collect traffic flow data, vehicle speed information, and video surveillance data. When processing data discretization and fitting, the continuous time series data (vehicle speed data) obtained from the sensor is first discretized and converted into data points in a discrete format. For example, the vehicle speed data per minute is recorded as a discrete value; in the discretized data, if the vehicle speed data is missing at certain time points, the vehicle speed at the missing point is estimated by linear interpolation, and the discretized data is normalized to standardize the data value to the range of 0 to 1, and then piecewise fitting is performed. The normalized discrete data is segmented according to the change points. According to the different change patterns of the vehicle speed data during the morning peak and evening peak, the segmented processing is performed in these two time periods, and the cubic spline is applied to the data of each segment for fitting, and the fitting results of all segments are synthesized to make the function value and derivative of the connection point smooth, and generate complete discrete approximation data.

[0116] Then feature extraction and data reconstruction are carried out. The continuous time series data (vehicle speed, traffic flow) collected from the sensor is parsed and prepared for sampling. The continuous data is sampled according to the predetermined time interval (every minute) and converted into discrete data points. The missing points in the discrete data are detected and filled in these data points through linear interpolation. The key features of peak vehicle speed and average flow are extracted from the discrete data, and the extracted features are fused to generate feature vectors. The fused feature data are standardized, and a complete reconstructed data set is generated by combining all the original data points, interpolated and filled data points, and extracted feature vectors. This effectively improves the accuracy of data processing and the responsiveness of the system, providing strong support for intelligent traffic management.

[0117] Step S50: perform anomaly detection on the extracted feature data based on the anomaly detection model, mark the anomaly points and the features of the anomaly points, classify the anomaly points through the classification model, and identify the anomaly type of the anomaly points.

[0118] In this step, see Figure 5 As shown, before performing anomaly detection on the extracted feature data based on the anomaly detection model, anomaly detection model training is also included, and the steps are as follows:

[0119] Step S501, mapping the feature vector in the reconstructed data set to the mean and variance of the latent space through the encoder network, and using the mean and variance to generate samples of the latent space;

[0120] Step S502: Decoding the latent space samples into sample data through a decoder network;

[0121] Step S503: construct a loss function for VAE model training, calculate the reconstruction error as the basic indicator for anomaly detection based on the generated sample data, record the reconstruction error of each sample data, and set the anomaly threshold based on the reconstruction error distribution of the training data;

[0122] Step S504: When the reconstruction error of the sample data exceeds the abnormal threshold, it is marked as an abnormal point.

[0123] When training the anomaly detection model, the latent space mapping is first performed, and the feature vectors in the reconstructed data set are mapped to the mean and variance of the latent space through the encoder network. The mean and variance are used to generate latent space samples; then sample data is generated, and the latent space samples are decoded into sample data using the decoder network. When the loss function is constructed, the training loss function of the variational autoencoder (VAE) model is constructed. The reconstruction error of the sample data is calculated, and the reconstruction error of each sample is recorded. The anomaly threshold is set according to the reconstruction error distribution of the training data. Data whose reconstruction error exceeds the threshold is marked as an anomaly.

[0124] In this embodiment, see Figure 6 As shown, the classification model training includes:

[0125] Step S511, extracting outlier features from an anomaly detection model by reconstructing the error;

[0126] Step S512: manually mark the types of outliers, and divide the data with the marked outlier types into a training set and a test set;

[0127] Step S513: taking the extracted outlier features as input and the category of the outlier as output label, using the training set to train the classification model, adjusting the model parameters through cross-validation, and obtaining a trained classification model;

[0128] Step S514: Use the trained classification model to classify the abnormal points and identify the abnormal types of the abnormal points.

[0129] When training the classification model, first perform feature extraction and labeling to extract outlier features from the anomaly detection model. Manually label the outlier types and divide the data into training and test sets. Then perform training and optimization, using the outlier features as input and the outlier types as output labels, and use the training set to train the classification model. Adjust the model parameters through cross-validation, and finally use the trained classification model to classify the outliers and identify their anomaly types.

[0130] Step S60: Based on historical data and abnormal points detected in real time, the anomaly detection model is automatically retrained, and the multimodal data collected in real time is detected based on the updated anomaly detection model. The prediction result is compared with the preset warning threshold. If the threshold is exceeded, a warning signal is issued and a fault report is generated.

[0131] By implementing the above steps, the system can efficiently process multimodal data, accurately detect anomalies, and automatically optimize the detection model based on real-time data to ensure the accuracy and timely response of traffic monitoring.

[0132] In some embodiments, in the digital terminal data processing method of the present invention, when multimodal data is collected in real time based on sensors installed on the terminal device, the data is processed based on the collaboration mechanism of the edge computing node and the cloud platform; wherein an edge computing node for receiving and processing locally collected multimodal data is deployed on the terminal device;

[0133] Run anomaly detection models on edge nodes to perform real-time anomaly detection on preprocessed data and mark abnormal data;

[0134] Compress and filter the data after anomaly detection, and send the compressed anomaly events and filtered data to the cloud platform;

[0135] The classification model deployed on the cloud platform classifies the abnormal points of abnormal events and identifies the abnormal types of the abnormal points.

[0136] Through the above steps, the edge computing nodes deployed by the present invention are responsible for real-time processing and preliminary detection, while the cloud platform is responsible for more in-depth analysis and classification, thereby improving the overall data processing efficiency and accuracy.

[0137] The digital terminal data processing method of the present invention uses real-time multimodal data and dynamically allocates computing resources based on the execution priority of multimodal data processing requests and adjusts the processing order of request queues, which significantly reduces the delay in data transmission to the cloud, improves the system's processing efficiency for multimodal data, realizes dynamic allocation of resources, ensures that key requests are processed in a timely manner, and optimizes the system's response time and resource utilization.

[0138] The present invention also converts continuous data into discrete format and interpolates missing data, reducing the impact of data loss and improving data integrity and accuracy. Through feature extraction and fusion, the generated reconstructed data set has higher representativeness and processing accuracy, which helps to improve the analysis and prediction accuracy of subsequent models. The present invention combines anomaly detection models with classification models, which can timely identify anomalies in data and accurately classify anomaly types, thereby improving the sensitivity and accuracy of fault detection. Model retraining is automatically triggered based on real-time and historical data to ensure that the accuracy of the anomaly detection model continues to improve with changes in data, enhancing the system's adaptability to new anomaly patterns.

[0139] By comparing the prediction results with the preset warning threshold, the present invention can promptly issue a warning signal when an abnormality occurs, helping users to identify potential problems in advance; the generated fault report provides detailed information on the abnormality, helping users to quickly locate and solve the problem, thereby improving the system maintenance efficiency and user experience.

[0140] It should be understood that, although described in a certain order, these steps are not necessarily performed in sequence in the above order. Unless there is clear explanation in this article, the execution of these steps does not have strict order restriction, and these steps can be performed in other orders. Moreover, a part of the steps of the present embodiment may include a plurality of steps or a plurality of stages, and these steps or stages are not necessarily performed at the same time, but can be performed at different times, and the execution order of these steps or stages is not necessarily performed in sequence, but can be performed in turn or alternately with at least a part of the steps or stages in other steps or other steps.

[0141] In one embodiment, the present invention provides a digital terminal data processing system for executing the digital terminal data processing method, the system comprising:

[0142] Multimodal data acquisition module: configured on the terminal device, used to receive multimodal data collected by the sensor in real time and receive modal data processing requests sent from the terminal device;

[0143] Computing resource allocation module: used to allocate computing resources according to the priority of multimodal data processing requests and dynamically adjust the request queue order;

[0144] Data processing module: used to parse multimodal data processing requests and obtain reconstructed data sets after processing multimodal data;

[0145] Anomaly detection module: used to perform anomaly detection on the extracted feature data through anomaly detection model, marking anomalies and anomaly features;

[0146] Classification and recognition module: used to classify abnormal points through classification models and identify the abnormal types of abnormal points;

[0147] Model update retraining module, which is used to automatically trigger the retraining of the anomaly detection model and update the model parameters based on historical data and anomalies detected in real time;

[0148] Early warning monitoring module: It is used to detect multimodal data collected in real time based on the updated anomaly detection model, compare the prediction results with the preset early warning threshold, and if the threshold is exceeded, a warning signal is issued and a fault report is generated.

[0149] In this embodiment, the data processing module includes:

[0150] The data parsing submodule is used to parse multimodal data processing requests, perform discretization processing, convert continuous data into discrete format, and interpolate missing data to generate discrete approximate data;

[0151] The extraction and fusion submodule is used to perform feature extraction, feature fusion and standardization on discrete approximation data to generate a reconstructed data set.

[0152] In this embodiment, the digital terminal data processing system adopts the steps of a digital terminal data processing method as described above during execution. Therefore, the operation process of the digital terminal data processing system will not be described in detail in this embodiment.

[0153] To sum up, the digital terminal data processing method and system of the present invention can significantly improve the efficiency and accuracy of data processing, enhance the capabilities of anomaly detection and fault management, and provide users with more intelligent, efficient and reliable data processing solutions.

[0154] In one embodiment, a computer device is also provided in an embodiment of the present invention, comprising at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor performs the steps of the digital terminal data processing method.

[0155] In one embodiment, the present invention further provides a computer-readable storage medium, which stores computer instructions, and the computer instructions are used to enable the computer to execute the steps of the digital terminal data processing method.

[0156] A person skilled in the art can understand that all or part of the processes in the above-mentioned embodiments can be realized by instructing the relevant hardware through a computer program represented by computer instructions, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory.

[0157] Non-volatile memory may include read-only memory, magnetic tape, floppy disk, flash memory or optical storage, etc. Volatile memory may include random access memory or external cache memory. As an illustration and not limitation, RAM may be in various forms, such as static random access memory or dynamic random access memory, etc.

[0158] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for processing digital terminal data, characterized in that: The method comprises the following steps: Collect multimodal data in real time based on sensors installed on terminal devices, and receive multimodal data processing requests sent from terminal devices; Allocate computing resources according to the priority of multimodal data processing requests and dynamically adjust the request queue order; Parse the multimodal data processing request, discretize the multimodal data obtained by the analysis, convert the continuous data into discrete format, and interpolate the missing data to obtain discrete approximate data; Extract and fuse features through discrete approximation data, standardize the fused feature data, and generate a reconstructed data set; Based on the anomaly detection model, the extracted feature data is subjected to anomaly detection, anomalies and anomaly feature marks are marked, anomalies are classified through the classification model, and the anomaly type of the anomaly is identified; Based on historical data and abnormal points detected in real time, the anomaly detection model is automatically retrained. The multimodal data collected in real time is detected based on the updated anomaly detection model, and the prediction results are compared with the preset warning threshold. If the threshold is exceeded, a warning signal is issued and a fault report is generated. Wherein, when multimodal data is collected in real time based on sensors installed on terminal devices, the data is processed based on the collaboration mechanism between edge computing nodes and cloud platforms; wherein, edge computing nodes for receiving and processing locally collected multimodal data are deployed on the terminal devices; Run anomaly detection models on edge nodes to perform real-time anomaly detection on preprocessed data and mark abnormal data; Compress and filter the data after anomaly detection, and send the compressed anomaly events and filtered data to the cloud platform; Classify the abnormal points of abnormal events based on the classification model deployed on the cloud platform and identify the abnormal types of the abnormal points; Among them, when allocating computing resources according to the priority of the multimodal data processing request and dynamically adjusting the request queue order, the following steps are included: Receive and parse multimodal data processing requests, and perform weighted round-robin scheduling based on the priority and urgency of the requests; Dynamically adjust the priority queue sorting in real time according to the priority of multimodal data processing requests, and dynamically allocate computing resources; Assign a weight value to each request based on its priority and urgency, process the request according to the weight value in each round of scheduling, and set the processing times based on the weight; During each scheduling, poll the requests in the queue and process the weighted number of each request; Dynamically adjust the order and weight of requests in the request queue based on request processing and the addition of new requests; Wherein, when obtaining discrete approximation data, it also includes processing the discretized data through a normalization formula, and fitting the normalized data using piecewise approximation; wherein: when fitting the normalized data using piecewise approximation, it includes the following steps: Segment the normalized discrete data according to the change points; Apply cubic spline to each discrete data segment for piecewise approximation fitting, and use the least square method to determine the parameters of the fitting function; The fitting results of all segments are synthesized to smooth the function values ​​and derivatives at the connection points and generate complete discrete approximation data; Before performing anomaly detection on the extracted feature data based on the anomaly detection model, anomaly detection model training is also included, and the steps are as follows: Map the feature vectors in the reconstructed dataset to the mean and variance of the latent space through the encoder network, and use the mean and variance to generate samples of the latent space; Decode the latent space samples into sample data through the decoder network; Construct the loss function of VAE model training, calculate the reconstruction error as the basic indicator of anomaly detection based on the generated sample data, record the reconstruction error of each sample data, and set the anomaly threshold based on the reconstruction error distribution of the training data; When the reconstruction error of the sample data exceeds the abnormal threshold, it is marked as an abnormal point.

2. The method for processing digital terminal data according to claim 1, characterized in that: Generating a reconstructed dataset includes the following steps: Parse and obtain continuous time series data collected from sensors; Continuous data is sampled at predetermined time intervals and converted into discrete data points; Map the data values ​​to the integer level range of discrete numerical values ​​to obtain discrete data; Detect missing points in discrete data and fill in the missing data by linear interpolation; Feature extraction and fusion are performed, and the fused feature data is standardized to generate a reconstructed data set.

3. The method for processing digital terminal data according to claim 2, characterized in that: The reconstructed data set is a complete data set including all original data points, interpolated data points and feature vectors.

4. The method for processing digital terminal data according to claim 3, characterized in that: When training a classification model, it includes: Extract outlier features from anomaly detection models through reconstruction errors; Manually label the types of outliers and divide the labeled data into training sets and test sets; The extracted outlier features are used as input, the category of the outlier is used as the output label, the classification model is trained using the training set, and the model parameters are adjusted through cross-validation to obtain a trained classification model; Use the trained classification model to classify the outliers and identify the anomaly types of the outliers.

5. A digital terminal data processing system, characterized in that: For executing the digital intelligent terminal data processing method according to any one of claims 1 to 4, the digital intelligent terminal data processing system comprises: Multimodal data acquisition module: configured on the terminal device, used to receive multimodal data collected by the sensor in real time and receive modal data processing requests sent by the terminal device; Computing resource allocation module: used to allocate computing resources according to the priority of multimodal data processing requests and dynamically adjust the request queue order; Data processing module: used to parse multimodal data processing requests and obtain reconstructed data sets after processing multimodal data; Anomaly detection module: used to perform anomaly detection on the extracted feature data through anomaly detection model, marking abnormal points and abnormal point features; Classification and recognition module: used to classify abnormal points through classification models and identify the abnormal types of abnormal points; Early warning monitoring module: It is used to detect multimodal data collected in real time based on the updated anomaly detection model, compare the prediction results with the preset early warning threshold, and if the threshold is exceeded, a warning signal is issued and a fault report is generated.

6. The digital terminal data processing system according to claim 5, characterized in that: The data processing module comprises: The data parsing submodule is used to parse multimodal data processing requests, perform discretization processing, convert continuous data into discrete format, and interpolate missing data to generate discrete approximate data; The extraction and fusion submodule is used to perform feature extraction, feature fusion and standardization on discrete approximation data to generate a reconstructed data set.

Images