Redcap terminal power consumption anomaly detection method and device based on deep learning model, medium and terminal

Abstract

The application provides a RedCap terminal power consumption anomaly detection method and device based on a deep learning model, a medium and a terminal. The method comprises: collecting power consumption time sequence data and service log data of the RedCap terminal in each power consumption scene, and aligning the time sequences of the collected power consumption time sequence data and service log data based on a preset time alignment mechanism; preprocessing the power consumption time sequence data and service log data after time sequence alignment, and extracting features from the preprocessed power consumption time sequence data and service log data to obtain power consumption features and service features; inputting the extracted power consumption features and service features into a deep learning model for training to construct a power consumption anomaly detection model; and deploying the constructed power consumption anomaly detection model to determine the anomaly of the current RedCap terminal power consumption to be detected and locate the anomaly root cause. The application can realize high-precision, high-efficiency and full-scene RedCap terminal power consumption detection.

Description

Technical Field

[0001] This application relates to the field of wireless communication technology, and in particular to a method, apparatus, medium, and terminal for detecting abnormal power consumption in RedCap terminals based on a deep learning model. Background Technology

[0002] With the large-scale deployment of fifth-generation mobile communication technology (5G), 5G RedCap (Reduced Capability), as a core technology for lightweight 5G IoT, is widely used in IoT scenarios such as industrial sensing, smart homes, and remote meter reading due to its advantages of low cost, low power consumption, and medium-to-low speed transmission. To meet low power consumption requirements, RedCap terminals typically employ complex power consumption mechanisms such as deep sleep mode (PSM), periodic sleep mode (eDRX), and dynamic air interface scheduling. In actual operation, the terminal's power consumption state dynamically switches according to changes in communication behavior, working scenarios, and configuration parameters, exhibiting characteristics of fragmented power consumption features, transient state changes, and coupled abnormal scenarios. This places extremely high demands on the fine-grainedness, accuracy, and scenario coverage of power consumption anomaly detection.

[0003] Currently, the mainstream RedCap terminal power consumption testing in the industry mainly includes the following technical solutions: (1) Automated detection method based on fixed threshold: The power consumption time series data of the terminal is collected by a high-precision power meter and compared with the pre-set static threshold to determine whether it is abnormal. If the power consumption data collected in real time exceeds the static threshold, it is determined to be power consumption abnormal; if it is within the static threshold range, it is determined to be power consumption normal; (2) Manual anomaly detection method based on engineer experience: Senior test engineers manually reproduce the test scenario, decompose the power consumption curve data, and compare the air interface signaling process with the terminal operation log to manually analyze and determine whether there is a hidden power consumption abnormality fault and preliminarily screen the anomaly type; (3) Detection method based on single-modal machine learning: Its core logic is to extract only the statistical features and time domain features of the power consumption time series data, and rely on traditional machine learning models such as SVM and random forest to complete the anomaly classification and judgment.

[0004] However, the aforementioned RedCap terminal power consumption testing solutions still mainly rely on traditional manual testing and automated static threshold testing, with a small amount of single-modal machine learning testing methods introduced. These methods cannot meet the requirements for high-precision, high-efficiency, and all-scenario power consumption testing, thus seriously affecting the battery life and operational stability of terminal products.

[0005] Therefore, it is necessary to provide a RedCap terminal power consumption anomaly detection method, device, medium, and terminal based on a deep learning model to solve the above-mentioned problems in the prior art. Summary of the Invention

[0006] In view of the shortcomings of the prior art described above, the purpose of this application is to provide a RedCap terminal power consumption anomaly detection method, device, medium and terminal based on a deep learning model, to solve the technical problem that the prior art cannot meet the power consumption detection requirements of high precision, high efficiency and all scenarios.

[0007] To achieve the above and other related objectives, the first aspect of this application provides a RedCap terminal power consumption anomaly detection method based on a deep learning model, comprising: collecting power consumption time-series data and service log data of the RedCap terminal under various power consumption scenarios, and aligning the collected power consumption time-series data and service log data according to a preset time alignment mechanism; preprocessing the time-aligned power consumption time-series data and service log data, and extracting features from the preprocessed power consumption time-series data and service log data to obtain power consumption features and service features; inputting the extracted power consumption features and service features into a deep learning model for training to construct a power consumption anomaly detection model; and deploying the constructed power consumption anomaly detection model to determine anomalies and locate the root causes of anomalies in the power consumption of the RedCap terminal to be detected.

[0008] In some embodiments of the first aspect of this application, the timing alignment of the collected power consumption timing data and service log data based on a preset time alignment mechanism specifically includes: outputting a pulse signal based on the general input / output interface of the RedCap terminal, and triggering the synchronous acquisition of power consumption timing data and service log data according to the output pulse signal; outputting a heartbeat event at preset intervals based on the firmware of the RedCap terminal, calculating the clock drift of the power consumption timing data and service log data acquisition based on the output heartbeat event, and correcting the calculated clock drift based on a linear regression model to obtain the timing-corrected power consumption timing data and service log data; and dynamically matching the timing-corrected power consumption timing data and service log data based on dynamic time warping technology to obtain the timing-aligned power consumption timing data and service log data.

[0009] In some embodiments of the first aspect of this application, the preprocessing of time-aligned power consumption time-series data and service log data, and the feature extraction of the preprocessed power consumption time-series data and service log data to obtain power consumption features and service features, specifically includes: cleaning the time-aligned power consumption time-series data and service log data, and filtering out noise from the cleaned power consumption time-series data based on wavelet transform technology; segmenting the noise-filtered power consumption time-series data according to the service events in the cleaned service log data to obtain multiple power consumption segments, and labeling each power consumption segment with a corresponding service tag; extracting power consumption features including statistical features, temporal features, and morphological features from each power consumption segment with a service tag, and extracting service features from the cleaned service log data; and normalizing the extracted power consumption features and service features to convert them into a data format suitable for input to a deep learning model.

[0010] In some embodiments of the first aspect of this application, the training process of the power consumption anomaly detection model includes: the input layer of a deep learning model synchronously receiving the input normalized power consumption features and service features; inputting the input normalized power consumption features and service features into the feature encoding layer of the deep learning model for feature encoding to obtain a power consumption time-series feature vector and a service semantic feature vector; inputting the power consumption time-series feature vector and the service semantic feature vector into the multi-feature fusion layer of the deep learning model, wherein the multi-feature fusion layer performs feature fusion on the input power consumption time-series feature vector and the service semantic feature vector based on a cross-attention mechanism to obtain a fused feature vector; and the output layer of the deep learning model receiving the input fused feature vector and outputting whether an anomaly exists and the anomaly type when an anomaly occurs.

[0011] In some embodiments of the first aspect of this application, the feature encoding layer includes a power consumption feature encoding layer and a service semantic encoding layer; wherein, the power consumption feature encoding layer encodes the input power consumption features based on a one-dimensional convolutional neural network to obtain a power consumption time-series feature vector; and the service semantic encoding layer encodes the input service features based on an embedding layer to obtain a service semantic feature vector.

[0012] In some embodiments of the first aspect of this application, the method further includes: iteratively optimizing the constructed power consumption anomaly detection model based on a preset model iteration optimization mechanism to update the deployed power consumption anomaly detection model.

[0013] In some embodiments of the first aspect of this application, the iterative optimization of the constructed power consumption anomaly detection model based on a preset model iterative optimization mechanism to update the deployed power consumption anomaly detection model specifically includes: acquiring automated test data and filtering the acquired automated test data based on a preset filtering mechanism to select valid samples; inputting the selected valid samples into the power consumption anomaly detection model for forward inference, outputting inference results, and extracting the model inference confidence corresponding to each valid sample from the inference results; comparing the model inference confidence corresponding to each valid sample with a preset confidence threshold, if the model inference confidence is higher than the preset confidence threshold, the model inference confidence is lowered. If the model's inference confidence is greater than or equal to a preset confidence threshold, a pseudo-label is generated for the valid sample. If the model's inference confidence is less than the preset confidence threshold, the corresponding valid sample is placed in a manual review pool for manual verification. Valid samples with pseudo-labels are input into the incremental learning model for incremental training to update the parameters of the power consumption anomaly detection model. The performance of the updated power consumption anomaly detection model is evaluated. If the performance of the updated power consumption anomaly detection model improves, it is deployed. If the performance of the updated power consumption anomaly detection model deteriorates, it is rolled back to the stable version.

[0014] To achieve the above and other related objectives, a second aspect of this application provides a RedCap terminal power consumption anomaly detection device based on a deep learning model, comprising: a data acquisition module, used to acquire power consumption time-series data and service log data of the RedCap terminal under various power consumption scenarios, and to align the acquired power consumption time-series data and service log data according to a preset time alignment mechanism; a feature extraction module, used to preprocess the time-aligned power consumption time-series data and service log data, and to extract features from the preprocessed power consumption time-series data and service log data to obtain power consumption features and service features; a model building module, used to input the extracted power consumption features and service features into a deep learning model for training, to build a power consumption anomaly detection model; and a model deployment module, used to deploy the built power consumption anomaly detection model to determine anomalies and locate the root causes of anomalies in the power consumption of the RedCap terminal to be detected.

[0015] To achieve the above and other related objectives, a third aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method.

[0016] To achieve the above and other related objectives, a fourth aspect of this application provides an electronic terminal, including a memory, a processor, and a computer program stored in the memory; the processor executes the computer program to implement the method.

[0017] As described above, the RedCap terminal power consumption anomaly detection method, apparatus, medium, and terminal based on deep learning models of this application have the following beneficial effects:

[0018] By synchronously collecting power consumption time-series data and service log data from RedCap terminals under various power consumption scenarios, and aligning the collected power consumption time-series data and service log data according to a preset time alignment mechanism, the time-aligned power consumption time-series data and service log data are preprocessed and feature extracted to obtain power consumption features and service features. Then, the extracted power consumption features and service features are input into a deep learning model for training to build a power consumption anomaly detection model. Finally, the built power consumption anomaly detection model is deployed to determine the power consumption of the RedCap terminal to be detected and locate the root cause of the anomaly, realizing high-precision, high-efficiency and full-scenario RedCap terminal power consumption detection, thereby effectively accelerating the iteration speed of RedCap terminal chip development and comprehensively improving the power consumption performance and long-term operational reliability of terminal products. Attached Figure Description

[0019] Figure 1 The diagram shown is a flowchart of a RedCap terminal power consumption anomaly detection method based on a deep learning model in one embodiment of this application.

[0020] Figure 2 The diagram shown illustrates the working principle of the RedCap terminal power consumption anomaly detection method based on a deep learning model in one embodiment of this application.

[0021] Figure 3 The diagram shown is a flowchart illustrating the training process of a power consumption anomaly detection model in one embodiment of this application.

[0022] Figure 4 The diagram shows a flowchart of iterative optimization of the power consumption anomaly detection model in one embodiment of this application.

[0023] Figure 5 The diagram shown is a block diagram of a RedCap terminal power consumption anomaly detection device based on a deep learning model, according to one embodiment of this application.

[0024] Figure 6 The diagram shown is a structural schematic of an electronic terminal according to an embodiment of this application. Detailed Implementation

[0025] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. It should be noted that, unless otherwise specified, the following embodiments and features in the embodiments can be combined with each other.

[0026] In the embodiments of this application, the terms "first" and "second" are used to distinguish identical or similar items with essentially the same function and effect. Those skilled in the art will understand that the terms "first" and "second" do not limit the quantity or execution order, and that the terms "first" and "second" do not necessarily imply that they are different.

[0027] It should be noted that, in the embodiments of this application, the words "exemplary" or "for example" indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0028] In this application embodiment, "at least one" refers to one or more, and "more than one" refers to two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can represent: a, b, c, ab, ac, bc, or abc, where a, b, and c can be single or multiple.

[0029] Before providing a further detailed description of the present invention, the nouns and terms used in the embodiments of the present invention are explained, and the nouns and terms used in the embodiments of the present invention are subject to the following interpretations:

[0030] <1> Deep learning models are machine learning methods that mimic the structure of the human brain's neural networks, automatically learning complex features from large-scale data through multi-layered neural network models.

[0031] <2> General Purpose Input Output (GPIO): A programmable digital interface used for bidirectional data interaction between embedded systems or microcontrollers and external devices.

[0032] <3> Firmware refers to software programs embedded in hardware devices. It is the foundation and core of device operation and exists in various electronic devices such as routers, cameras, and smartphones. It is responsible for device initialization and low-level control, and it is usually stored in non-volatile memory such as EPROM or EEPROM, which allows firmware to be retained even after power failure.

[0033] <4> Linear Regression Model: A basic and widely used machine learning algorithm used to predict continuous values. It assumes a linear relationship between the target variable and the feature variables and attempts to find a best-fitting straight line to describe this relationship.

[0034] <5> Dynamic Time Warping (DTW) is an algorithm used to calculate the similarity between two time series. It is particularly suitable for series with different lengths or misaligned time steps. It uses a dynamic programming strategy to perform non-linear time-domain alignment of the time series, thereby finding the optimal path to minimize the distance between the two series.

[0035] <6> Data cleaning refers to the systematic organization and correction of raw data, identifying and correcting identifiable errors, including handling missing values, duplicate values, outliers, and standardizing data formats, to make the data more accurate, complete, and consistent.

[0036] <7> Wavelet Transform (WT) is a technique that decomposes a signal into a series of wavelet functions. It can simultaneously provide the time (or space) and frequency information of the signal and is suitable for processing non-stationary signals.

[0037] <8> A one-dimensional convolutional neural network (1D-CNN) is a deep learning model specifically designed for processing sequential data, such as time series, text, and speech signals. It is a variant of the convolutional neural network (CNN). Compared to a two-dimensional convolutional neural network (2D-CNN), the convolutional operations in 1D-CNN are performed only on one-dimensional sequences. Its basic structure includes an input layer, convolutional layers, activation functions, pooling layers, and fully connected layers. The input layer receives one-dimensional sequence data; the convolutional layers extract local features using one-dimensional convolutional kernels; the activation functions enhance non-linear expressive power; the pooling layers downsample to reduce data dimensionality; and the fully connected layers are used for global analysis and classification.

[0038] <9> Embedding layer: A hidden layer in a neural network that maps high-dimensional, discrete data into low-dimensional, continuous vector representations to capture the latent semantic information of the data and improve the efficiency of model processing.

[0039] <10> Cross-Attention is an extension of the attention mechanism. Its core idea is that the query comes from one sequence, while the key and value come from another sequence, thereby achieving information alignment and fusion between the two sequences.

[0040] <11> Incremental learning models, also known as continuous learning or life-long learning, are machine learning methods designed to enable models to continuously learn new knowledge in dynamic environments while retaining existing knowledge, without retraining the entire model.

[0041] <12> Elastic Weight Consolidation (EWC) is a regularization-based continuous learning method that preserves old knowledge when training new tasks by calculating the importance of parameters to old tasks and restricting the updates of key parameters.

[0042] <13> Learning without Forgetting (LWF) is a technique that uses knowledge distillation to retain knowledge from old tasks while training for new tasks, aiming to solve the catastrophic forgetting problem in neural networks.

[0043] <14> Incremental Classifier and Representation Learning (iCaRL) algorithm: This is a type of incremental learning method that aims to learn new concepts incrementally while dealing with limited training data. It achieves this through approximate average example classification, group-based example selection, knowledge distillation, and prototype training.

[0044] Currently, the main RedCap terminal power consumption testing solutions in the industry include automated detection methods based on fixed thresholds, manual anomaly detection methods based on engineer experience, and detection methods based on single-modal machine learning. However, these existing technical solutions all have significant technical defects. Specifically: (1) Automated detection methods based on fixed thresholds are currently the most widely used automated detection methods in the industry. However, they rely only on a single power consumption data dimension and static threshold judgment, and cannot effectively distinguish between normal business fluctuations and real power consumption anomalies. They can only achieve basic screening of regular and explicit power consumption anomalies, and are completely ineffective for fine-grained, intermittent, and scenario-coupled power consumption anomalies. At the same time, they require frequent manual adjustment of thresholds, resulting in extremely poor adaptability. (2) Manual anomaly detection methods based on engineer experience are currently the main supplementary means for fine-grained power consumption anomaly detection in the industry. Although they can detect implicit power consumption anomalies, they can still detect hidden power consumption anomalies. However, since the entire testing process relies on the personal experience of senior test engineers, the judgment standards of different personnel are different, resulting in poor consistency of test results. In addition, the manual analysis process is cumbersome, and the analysis of a single test case can take several hours, resulting in extremely low testing efficiency. It is also impossible to achieve 24-hour uninterrupted testing, making it difficult to cover a large number of complex test scenarios. At the same time, the labor cost is high and the scale of implementation is difficult. (3) The detection method based on single-modal machine learning is a traditional intelligent detection upgrade solution. Although this solution introduces intelligent detection capabilities, it still does not break through the technical limitations of single power consumption data. It does not associate with core business behaviors such as air interface scheduling and protocol interaction, and cannot explain the underlying reasons for power consumption anomalies. It cannot guide the optimization design of terminal chips. Furthermore, the model is a static and fixed model with no iterative learning capability, which cannot adapt to the power consumption characteristics of new scenarios and has weak generalization ability. It can be seen that these existing technical solutions are generally difficult to adapt to the industry testing needs of rapid chip iteration, continuous scenario updates, high precision, and high efficiency. Moreover, the power consumption hidden dangers are frequently missed, which seriously affects the battery life and operational stability of terminal products. To address this, this application provides a RedCap terminal power consumption anomaly detection method, device, medium, and terminal based on a deep learning model. This method achieves high-precision, high-efficiency, all-scenario, and model-iteratively optimized terminal power consumption detection. Furthermore, the power consumption anomaly detection model ultimately constructed can further identify the root cause of the anomaly when power consumption anomalies are identified, thereby effectively guiding terminal chip design and improving the battery life and operational stability of terminal products.

[0045] It should be noted that the RedCap terminal power consumption anomalies detected in this application include explicit power consumption anomalies and fine-grained implicit power consumption anomalies. Among them, fine-grained implicit power consumption anomalies include implicit power redundancy, abnormal micro-wake-up, microampere-level module leakage, and scene-coupled intermittent anomalies.

[0046] To facilitate understanding of the embodiments of this application, in conjunction with Figure 1 and Figure 2 Detailed explanation. Figure 1 The diagram illustrates a flowchart of a RedCap terminal power consumption anomaly detection method based on a deep learning model, according to an embodiment of the present invention. Figure 2 This illustration shows a schematic diagram illustrating the working principle of a RedCap terminal power consumption anomaly detection method based on a deep learning model, according to an embodiment of the present invention. The RedCap terminal power consumption anomaly detection method based on a deep learning model in this embodiment includes the following steps:

[0047] Step S11: Collect power consumption timing data and service log data of RedCap terminal under various power consumption scenarios, and align the collected power consumption timing data and service log data according to the preset time alignment mechanism.

[0048] By synchronously collecting two types of core data—power consumption time-series data and business logs—we achieve full-scenario, high-precision data coverage. Furthermore, a pre-designed time alignment mechanism is implemented to achieve microsecond-level precise data alignment, abandoning the existing single power consumption threshold judgment mode. This provides data support for subsequent multi-dimensional feature fusion and correlation modeling, solving the problems of single detection dimensions and time-series misalignment in traditional detection from a data perspective.

[0049] For example, power consumption timing data is collected using a high-precision power meter, covering all operating modes of the RedCap terminal, including deep sleep mode (PSM), periodic sleep mode (eDRX), DRX (Discontinuous Reception) scheduling, and data transmission and reception. The sampling rate is dynamically adjustable from 1MHz to 10MHz, accurately capturing instantaneous power consumption changes. Log service data is collected collaboratively with an air interface tester via the terminal chip debugging interface. The log service data collected by the air interface tester includes core service data such as air interface scheduling parameters, RRC (Radio Resource Control) signaling interactions, and MAC (Media Access Control) layer protocol flows. The log service data collected by the terminal chip debugging interface includes logs of hardware and software module start / stop, sleep / wake events, etc. All collected power consumption timing data and service log data carry high-precision timestamps, laying the foundation for subsequent data alignment and feature association.

[0050] In some embodiments of this application, the timing alignment of the collected power consumption timing data and service log data based on a preset time alignment mechanism specifically includes: outputting a pulse signal based on the general input / output interface of the RedCap terminal, and triggering the synchronous acquisition of power consumption timing data and service log data based on the output pulse signal; outputting a heartbeat event at preset intervals based on the firmware of the RedCap terminal, calculating the clock drift of the power consumption timing data and service log data acquisition based on the output heartbeat event, and correcting the calculated clock drift based on a linear regression model to obtain the timing-corrected power consumption timing data and service log data; and dynamically matching the timing-corrected power consumption timing data and service log data based on dynamic time warping technology to obtain the timing-aligned power consumption timing data and service log data.

[0051] Specifically, firstly, pulse signals are output from the general purpose input / output interface (GPIO) of the RedCap terminal. The pulse signals of key events such as wake-up and preamble transmission are used as a global time reference to calibrate the zero point of power consumption timing data and service log data. This ensures that the high-precision power meter, the terminal chip debugging interface, and the air interface tester all start sampling from the same time reference, thereby triggering the simultaneous acquisition of power consumption timing data and service log data. Secondly, the terminal firmware outputs a heartbeat event every preset time, such as 10ms. This heartbeat event is captured by each acquisition device, and the clock drift of each acquisition device is calculated. The clock drift of multiple acquisition devices is corrected by combining a linear regression model, thereby obtaining the timing-corrected power consumption timing data and service log data. Finally, dynamic time warping (DTW) technology is used to achieve dynamic matching and alignment between the power consumption timing curve and the service log event sequence, thereby realizing accurate timing binding of power consumption timing data and service log data. This involves employing a triple synchronization mechanism, which combines GPIO hardware triggering, software heartbeat calibration, and dynamic time warping, to solve the problem of timing misalignment in multi-source data. This precisely controls the data synchronization error to within 50μs, achieving microsecond-level timing alignment between power consumption timing data and business log data.

[0052] Step S12: Preprocess the timing-aligned power consumption timing data and service log data, and extract features from the preprocessed power consumption timing data and service log data to obtain power consumption features and service features.

[0053] In some embodiments of this application, the preprocessing of time-aligned power consumption time-series data and service log data, and the feature extraction of the preprocessed power consumption time-series data and service log data to obtain power consumption features and service features, specifically includes: cleaning the time-aligned power consumption time-series data and service log data, and filtering out noise from the cleaned power consumption time-series data based on wavelet transform technology; segmenting the noise-filtered power consumption time-series data according to the service events in the cleaned service log data to obtain multiple power consumption segments, and labeling each power consumption segment with a corresponding service tag; extracting power consumption features including statistical features, temporal features, and morphological features from each power consumption segment with a service tag, and extracting service features from the cleaned service log data; and normalizing the extracted power consumption features and service features to convert them into a data format suitable for input to a deep learning model.

[0054] Specifically, the multi-source data after time-series alignment is first cleaned to remove invalid data, thereby improving the accuracy of subsequent model training. Simultaneously, wavelet transform technology is used to remove noise from the power consumption time-series data. Furthermore, the power consumption time-series data is segmented according to business events in the business log data, resulting in multiple power consumption segments, each labeled with a corresponding business tag. Then, feature mining is performed on the time-series aligned multi-source data to obtain statistical and business features. Finally, the power consumption and business features extracted from different working modes are normalized to convert them into a standard data format suitable for deep learning model input. For example, the extracted power consumption features include statistical features, temporal features, and morphological features. Statistical features include basic statistical data such as time-series mean, variance, and peak value, describing the overall level and fluctuation of power consumption. Temporal features include rise time, duration, and duty cycle, extracting the morphological parameters of the power consumption waveform on the time axis. Morphological features are related data obtained by comparing similarity with a standard power consumption template, describing the degree of deviation between the current power consumption behavior and the standard morphology. The extracted business features transform semi-structured logs into structured events, and encode business semantic feature information such as air interface scheduling, protocol type, and module status, thereby generating standardized business semantic feature vectors.

[0055] Step S13: Input the extracted power consumption features and service features into a deep learning model for training to construct a power consumption anomaly detection model.

[0056] By deeply integrating power consumption characteristics with business characteristics, the system synchronously learns the power consumption change patterns and business behavior characteristics of the terminal, accurately distinguishes between normal power consumption fluctuations and actual power consumption anomalies, and reverse-analyzes the causes of anomalies based on attention weights. This enables interpretable detection of power consumption anomalies and root cause localization, solving the shortcomings of traditional intelligent detection solutions that lack traceability and cannot interpret detection results. As a result, a multi-dimensional power consumption anomaly detection model with intrinsic feature correlations is constructed, breaking through the limitations of single-dimensional analysis.

[0057] In some embodiments of this application, the training process of the power consumption anomaly detection model includes: the input layer of a deep learning model synchronously receiving the input normalized power consumption features and service features; inputting the input normalized power consumption features and service features into the feature encoding layer of the deep learning model for feature encoding to obtain a power consumption time-series feature vector and a service semantic feature vector; inputting the power consumption time-series feature vector and the service semantic feature vector into the multi-feature fusion layer of the deep learning model, wherein the multi-feature fusion layer performs feature fusion on the input power consumption time-series feature vector and the service semantic feature vector based on a cross-attention mechanism to obtain a fused feature vector; and the output layer of the deep learning model receiving the input fused feature vector and outputting whether an anomaly exists and the anomaly type when an anomaly occurs.

[0058] In some embodiments of this application, the feature encoding layer includes a power consumption feature encoding layer and a service semantic encoding layer; wherein, the power consumption feature encoding layer encodes the input power consumption features based on a one-dimensional convolutional neural network to obtain a power consumption time-series feature vector; and the service semantic encoding layer encodes the input service features based on an embedding layer to obtain a service semantic feature vector.

[0059] like Figure 3As shown, the deep learning model's network architecture includes an input layer, a feature encoding layer, a multi-feature fusion layer, and an output layer. This architecture enables deep correlation learning between power consumption features and business features. Specifically, the input layer simultaneously receives normalized power consumption features and business features; the feature encoding layer encodes power consumption features using a 1D-CNN to obtain a power consumption temporal feature vector, while the embedding layer encodes discrete business features to obtain a business semantic feature vector; the multi-feature fusion layer, based on a cross-attention mechanism, automatically calculates the correlation weights between the two types of features, achieving deep fusion of multi-dimensional features to obtain a fused feature vector; the output layer simultaneously performs binary classification of anomalies (normal / abnormal) and, in cases of anomalies, performs multi-class classification of anomalies (e.g., sleep leakage, false wake-up, etc.), thereby training a power consumption anomaly detection model. It should be noted that during the initial training phase, multiple types of typical anomaly scenarios, such as 20 types, are manually constructed, labeled samples are collected, and basic training is completed to quickly cover known power consumption defect patterns.

[0060] Step S14: Deploy the constructed power consumption anomaly detection model to determine anomalies and locate the root causes of anomalies in the power consumption of the RedCap terminal to be detected.

[0061] Specifically, during the RedCap terminal testing process, multi-dimensional data collected, aligned, and preprocessed in real time is input into the trained power consumption anomaly detection model. This replaces the traditional single-threshold judgment logic with multi-feature correlation analysis. Furthermore, by learning standard power consumption patterns and anomaly characteristics, the power consumption anomaly detection model automatically identifies transient fluctuations, intermittent triggers, and scenario-coupled power consumption anomalies. It accurately captures minute defects such as implicit power redundancy, abnormal micro-wake-up, and microampere-level module leakage, comprehensively covering various explicit, implicit, and refined power consumption anomaly scenarios, achieving fine-grained, full-scenario power consumption anomaly identification. Simultaneously, based on the cross-attention weights of the power consumption anomaly detection model, the core business events triggering the power consumption anomalies are located in reverse, enabling the explanation and root cause localization of power consumption anomalies, providing precise data support for terminal chip optimization.

[0062] In some embodiments of this application, the method further includes: iteratively optimizing the constructed power consumption anomaly detection model based on a preset model iteration optimization mechanism to update the deployed power consumption anomaly detection model.

[0063] Since existing power consumption detection methods are all fixed and static, lacking a data-driven closed-loop iterative mechanism, they cannot adapt to new power consumption anomaly conditions brought about by RedCap communication protocol upgrades, terminal hardware version iterations, and complex network scenario updates. Their versatility and adaptability under long-term detection are severely insufficient. Therefore, in this embodiment, a model iterative optimization mechanism is designed to iteratively optimize the power consumption anomaly detection model. Based on the already trained model, the model parameters are updated only using newly added measured samples, without the need to retrain the entire dataset. This achieves low-cost, continuous iterative upgrades of the model, thereby adapting to the power consumption conditions of new scenarios.

[0064] In some embodiments of this application, the iterative optimization of the constructed power consumption anomaly detection model based on a preset model iteration optimization mechanism to update the deployed power consumption anomaly detection model specifically includes: acquiring automated test data and filtering the acquired automated test data based on a preset filtering mechanism to select valid samples; inputting the selected valid samples into the power consumption anomaly detection model for forward inference, outputting inference results, and extracting the model inference confidence corresponding to each valid sample from the inference results; comparing the model inference confidence corresponding to each valid sample with a preset confidence threshold; if the model inference confidence is higher than the preset confidence threshold, the model inference confidence is lowered. If the model's inference confidence is greater than or equal to a preset confidence threshold, a pseudo-label is generated for the valid sample. If the model's inference confidence is less than the preset confidence threshold, the corresponding valid sample is placed in a manual review pool for manual verification. Valid samples with pseudo-labels are input into the incremental learning model for incremental learning training to update the parameters of the power consumption anomaly detection model. The performance of the updated power consumption anomaly detection model is evaluated. If the performance of the updated power consumption anomaly detection model improves, it is deployed. If the performance of the updated power consumption anomaly detection model deteriorates, it is rolled back to the stable version.

[0065] like Figure 4As shown, firstly, automated test data is acquired, and then high-value, valid samples are selected from the massive amount of automated test data through a four-layer screening mechanism of completeness, quality, novelty, and uncertainty. The purpose of completeness screening is to remove incomplete data on business events; the purpose of quality screening is to remove unclear, low-quality data; the purpose of novelty screening is to select new scenario samples that can expand the knowledge boundary of the model; and the purpose of uncertainty screening is to select samples that are close to the decision boundary. For samples that cannot be clearly classified in the final screening, they are placed in a manual review pool for manual verification. Then, the selected valid samples are input into the power consumption anomaly detection model for forward inference. Based on the output inference results, the model inference confidence score for each sample is extracted. This score is compared to a confidence threshold. If the confidence score is greater than or equal to the threshold, the sample is considered a high-confidence sample, and a pseudo-label is automatically generated. If the confidence score is less than the threshold, the sample is considered a low-confidence sample and is placed in a manual review pool. After manual verification, it is added to the sample library, continuously enriching the model's anomaly identification scenarios and enabling dynamic adaptation of the model to power consumption characteristics changes in new hardware, protocols, and networking scenarios. An incremental learning model is then used, employing high-confidence samples with pseudo-labels for incremental training of the power consumption anomaly detection model. The incremental learning algorithm updates the model parameters, eliminating the need for retraining with the entire dataset and reducing iteration costs. Finally, after each model iteration, a separate test set is used for performance evaluation testing. Only models with improved performance are automatically deployed; otherwise, they are rolled back to a stable version. This led to the establishment of a fully automated model iteration closed-loop mechanism, enabling autonomous upgrading of detection capabilities.

[0066] By designing a four-layer intelligent sample screening mechanism, automatic pseudo-label annotation, incremental learning online training mode, automatic evaluation and deployment mechanism, and a closed-loop self-evolution mechanism for accumulating unknown anomalies, this mechanism continuously iterates and optimizes model performance based on automated measured data. Without manual intervention or full data retraining, the model can autonomously adapt to new power consumption anomalies caused by terminal hardware upgrades, protocol iterations, and network scenario updates, thus achieving autonomous and continuous upgrades of detection capabilities.

[0067] For example, the incremental learning model employs a hybrid incremental learning algorithm combining Elastic Weight Consolidation (EWC), Forget-Free Learning (LWF), and Incremental Classification and Representation Learning (iCaRL). Specifically, EWC uses the Fisher Information Matrix (FIM) to evaluate the importance of each layer's parameters to historically known power consumption scenarios and anomalies. The incremental training loss function adds a regularization penalty term, imposing weight constraints on key historical parameters. This significantly limits large-scale updates to parameters related to old knowledge, preventing the model from forgetting typical historical anomalies such as PSM sleep leakage and false micro-wake-ups. Specifically, after initial model training, the FIM weights of all parameters in the 1D-CNN, cross-attention mechanism, and output layer are calculated. During incremental iterative training, the total loss function is constructed as the sum of the new sample classification loss and the EWC quadratic regularization loss, constraining only the backbone feature extraction layer and allowing moderate updates to the output classification head, balancing the preservation of old knowledge with adaptation to new anomalies. Forget-Free Learning (LWF) works by freezing the old model from the previous iteration as the teacher model, while the new model (student model) learns new labeled samples simultaneously. Furthermore, the teacher model's output soft labels for both old and new features are distilled. Distillation loss is used to constrain the student model's output distribution in historical scenarios to remain consistent with the old model. This approach preserves historical discrimination logic without relying on large amounts of historical sample storage. Specifically, before each incremental training iteration, the previous stable inference model is saved as the teacher model, and incremental samples and a small number of cached historical examples are input. Hard label classification loss and teacher soft output distillation loss are calculated simultaneously, thus solving the problems of missed detections of old anomalies and model drift caused by fine-tuning with only new samples. The Incremental Classification and Representation Learning (iCaRL) algorithm maintains a fixed-capacity historical example cache. In each incremental iteration, representative samples are selected from the high-value valid samples after four layers of screening and stored in the cache. During training, cached historical examples and newly added samples are jointly trained. By replaying a small number of historical samples, the old operating condition distribution is restored, compensating for the performance degradation of EWC and LwF when the data distribution drifts significantly. Specifically, the cache pool stores typical samples evenly according to the anomaly type (e.g., dormant leakage, module micro-wake-up, scene coupling anomalies, etc.), and high-confidence pseudo-label samples output from the four layers of screening are added to the cache first. The incremental batch mixes 80% new samples and 20% cached historical examples, and performs small-batch fine-tuning without retraining on the full original dataset. In other words, by using the Elastic Weight Consolidation (EWC) algorithm for parameter freezing, the Forget-Free Learning (LWF) technique for behavior imitation, and the Incremental Classification and Representation Learning (iCaRL) algorithm for data replay, the model achieves closed-loop self-evolutionary optimization to adapt to changes in power consumption characteristics of new hardware, new protocols, and new networking scenarios.

[0068] The RedCap terminal power consumption anomaly detection method based on a deep learning model provided in this application constructs a power consumption anomaly detection model based on the multi-feature correlation of power consumption characteristics and service characteristics. This solves the problems of high false negative and false positive rates, difficulty in identifying fine-grained anomalies, and inability to trace the source of anomalies in existing solutions. Furthermore, by building a data-driven closed-loop self-evolving iterative model architecture, it addresses the pain points of poor adaptability and inability to adapt to scenario iterations in existing static models. Ultimately, it achieves high-precision, automated, interpretable, and iterative detection of RedCap terminal power consumption anomalies, significantly reducing manual testing costs, improving testing efficiency, comprehensively optimizing the power consumption testing quality of terminal chips, and accelerating the R&D and iteration process of RedCap chip products.

[0069] The RedCap terminal power consumption anomaly detection method based on deep learning model provided in this application has the following beneficial effects: (1) The detection accuracy is greatly improved and the defect identification capability is significantly enhanced: the power consumption time series data and business log data are accurately bound through the microsecond-level triple time alignment mechanism, and multi-feature fusion modeling is performed based on cross-attention, which completely gets rid of the static threshold constraint and can significantly improve the defect identification capability; (2) It has interpretability and engineering guidance value and realizes the root cause of anomalies: it breaks through the limitation of a single data dimension and realizes the multi-feature fusion judgment logic of power consumption time series data and business log data; based on the cross-attention weight reverse analysis of the cause of anomalies, it accurately locates the underlying business root cause of power consumption anomalies, no longer simply outputs the abnormal results, and can directly guide the chip hardware design and software scheduling logic optimization, which has strong engineering implementation value; (3) Fully automated closed-loop self-evolution and continuous upgrade of model adaptation capability: a closed-loop self-evolution mechanism including four-layer intelligent sample screening, automatic pseudo-label labeling, incremental online training, automatic evaluation deployment, and unknown anomaly accumulation is built. Based on automated test data, the model performance is continuously iterated and optimized. No manual intervention or full data retraining is required. The model can autonomously adapt to new power consumption anomalies caused by terminal hardware upgrades, protocol iterations, and network scenario updates. The generalization ability and long-term availability are greatly improved. (4) Full scenario coverage and accurate identification of fine-grained hidden anomalies: Effectively identify fine-grained fluctuations, intermittent triggers, and scenario-coupled power consumption anomalies that cannot be detected by traditional threshold methods. Accurately identify hidden power consumption redundancy, abnormal micro-wake-up, micro-ampere level module leakage and other small power consumption defects. It fully covers various explicit, implicit, and refined power consumption anomaly scenarios. The scenario adaptability is far superior to traditional static rule detection schemes. (5) Test efficiency is increased exponentially and labor costs are greatly reduced: 24-hour uninterrupted fully automated data collection, analysis, and anomaly judgment are achieved. The analysis time of a single test case is shortened from several hours to several seconds. The overall test efficiency is greatly improved and can adapt to the mass production test requirements of large batches of chips. There is no need to rely on senior engineers to carry out complex analysis work. The standard difference of manual judgment is eliminated, and the test results are standardized and normalized.

[0070] Figure 5 This is a schematic block diagram of the RedCap terminal power consumption anomaly detection device based on a deep learning model provided in an embodiment of this application. Figure 5 As shown, the RedCap terminal power consumption anomaly detection device 500 based on a deep learning model includes:

[0071] The acquisition module 501 is used to acquire power consumption time-series data and service log data of RedCap terminal under various power consumption scenarios, and to align the acquired power consumption time-series data and service log data according to a preset time alignment mechanism.

[0072] The feature extraction module 502 is used to preprocess the time-aligned power consumption time-series data and service log data, and to extract features from the preprocessed power consumption time-series data and service log data to obtain power consumption features and service features.

[0073] The model building module 503 is used to input the extracted power consumption features and service features into a deep learning model for training, so as to build a power consumption anomaly detection model.

[0074] The model deployment module 504 is used to deploy the constructed power consumption anomaly detection model, and to determine the anomaly of the power consumption of the RedCap terminal to be detected and locate the root cause of the anomaly.

[0075] In existing technologies, RedCap's automated power consumption detection relies solely on a single power consumption timing threshold and static fixed rules, which cannot effectively distinguish between power consumption changes caused by normal business fluctuations and actual power consumption anomalies. It completely fails to detect fine-grained, intermittent, and scenario-coupled anomalies, resulting in high rates of missed and false detections and extremely poor scenario adaptability. Furthermore, existing refined power consumption detection methods heavily depend on the personal experience of senior engineers, requiring complex scenario reviews, power consumption data breakdown, and air interface signaling comparison analysis. This has a very high operational threshold, high labor costs, and long testing cycles. Different engineers have different judgment standards, making it impossible to achieve standardized, large-scale, and continuous 24 / 7 detection. Simultaneously, existing power consumption detection methods only focus on the superficial data of power consumption values, without relating it to core business behaviors such as air interface scheduling, protocol interaction, and module start / stop. They can only determine whether power consumption is abnormal, but cannot pinpoint the underlying business root cause of the anomaly, making it difficult to directly guide terminal chip hardware design and software scheduling logic optimization. To address this, this application provides a RedCap terminal power consumption anomaly detection device based on a deep learning model, which improves the accuracy and efficiency of terminal power consumption detection, covers all power consumption scenarios, and associates power consumption time-series data with business log data. In the case of an anomaly, it can locate the underlying business root cause of the anomaly, thereby directly guiding the terminal chip design and scheduling logic optimization.

[0076] It should be understood that the specific process of each module performing the above-mentioned steps has been described in detail in the above method embodiments, and will not be repeated here for the sake of brevity.

[0077] It should also be understood that the module division in the embodiments of this application is illustrative and only represents a logical functional division; in actual implementation, there may be other division methods. Furthermore, the functional modules in the various embodiments of this application can be integrated into a single processor, exist as separate physical entities, or be integrated into a single module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0078] Figure 6 This is a schematic block diagram of the electronic terminal provided in an embodiment of this application. Figure 6 As shown, the electronic terminal 600 includes at least one processor 601, a memory 602, at least one network interface 603, and a user interface 605. The various components in the electronic terminal 600 are coupled together via a bus system 604. It is understood that the bus system 604 is used to implement communication between these components. In addition to a data bus, the bus system 604 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in… Figure 6 The general will label all buses as bus systems.

[0079] The user interface 605 may include a monitor, keyboard, mouse, trackball, clicker, button, touchpad, or touch screen.

[0080] It is understood that memory 602 can be volatile memory or non-volatile memory, or both. Non-volatile memory can be read-only memory (ROM) or programmable read-only memory (PROM), which serves as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM) and synchronous static random access memory (SSRAM). The memories described in the embodiments of this invention are intended to include, but are not limited to, these and any other suitable categories of memory.

[0081] In this embodiment of the invention, the memory 602 is used to store various types of data to support the operation of the electronic terminal 600. Examples of this data include: any executable program for operation on the electronic terminal 600, such as the operating system 6021 and application programs 6022; the operating system 6021 contains various system programs, such as the framework layer, core library layer, driver layer, etc., for implementing various basic services and handling hardware-based tasks. The application program 6022 may contain various applications, such as a media player, browser, etc., for implementing various application services. The methods provided in this embodiment of the invention can be included in the application program 6022.

[0082] The methods disclosed in the above embodiments of the present invention can be applied to processor 601, or implemented by processor 601. Processor 601 may be an integrated circuit chip with signal processing capabilities. In the implementation process, each step of the above method can be completed by the integrated logic circuit of the hardware in processor 601 or by instructions in the form of software. The processor 601 may be a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 601 can implement or execute the methods, steps, and logic block diagrams disclosed in the embodiments of the present invention. General-purpose processor 601 may be a microprocessor or any conventional processor, etc. The steps of the accessory optimization method provided in the embodiments of the present invention can be directly reflected as being executed by a hardware decoding processor, or being executed by a combination of hardware and software modules in the decoding processor. The software module may be located in a storage medium, which is located in memory. The processor reads the information in the memory and combines it with its hardware to complete the steps of the aforementioned method.

[0083] In an exemplary embodiment, the electronic terminal 600 may be used by one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), or complex programmable logic devices (CPLDs) to perform the aforementioned method.

[0084] According to the method provided in the embodiments of this application, this application also provides a computer-readable storage medium storing program code, which, when executed on a computer, causes the computer to perform... Figures 1 to 4 The method of any of the embodiments shown.

[0085] As used in this specification, the terms "component," "module," "system," etc., are used to refer to computer-related entities, hardware, firmware, combinations of hardware and software, software, or software in execution. For example, a component can be, but is not limited to, a process running on a processor, a processor, an object, an executable file, an execution thread, a program, and / or a computer. As illustrated, applications running on computing devices and computing devices can both be components. One or more components may reside in a process and / or an execution thread, and components may be located on a single computer and / or distributed among two or more computers. Furthermore, these components can be executed from various computer-readable media on which various data structures are stored. Components can communicate, for example, via local and / or remote processes based on signals having one or more data packets (e.g., data from two components interacting with another component between a local system, a distributed system, and / or a network, such as the Internet interacting with other systems via signals).

[0086] Those skilled in the art will recognize that the various illustrative logical blocks and steps described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this application.

[0087] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0088] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0089] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0090] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0091] In the above embodiments, the functions of each functional unit can be implemented entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. A computer program product includes one or more computer instructions (programs). When the computer program instructions (programs) are loaded and executed on a computer, all or part of the flow or function according to the embodiments of this application is generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. Computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. Available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., high-density digital video discs (DVDs)), or semiconductor media (e.g., solid-state disks (SSDs)).

[0092] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0093] In summary, in response to the technical problem that existing technologies cannot meet the requirements of high precision, high efficiency, all-scenario, and iteratively updatable power consumption detection, this application provides a RedCap terminal power consumption anomaly detection method, device, medium, and terminal based on a deep learning model. It focuses on accuracy, efficiency, scenario adaptability, root cause localization, and iterative optimization capabilities, and has the following beneficial effects: (1) The detection accuracy is greatly improved, the ability to identify anomalies and defects is significantly enhanced, the anomaly coverage is comprehensively expanded, and the scenario adaptability is extremely strong: This application not only breaks through the constraints of static thresholds, reducing the false alarm rate from more than 30% to less than 2%, and the false alarm rate from 15% to less than 1%, accurately solving the problem of frequent false alarms and missed detections in traditional detection; it can also effectively identify the problems that existing solutions are completely ineffective in. The method detects fine-grained fluctuations, intermittent triggers, scene-coupled power consumption anomalies, as well as microampere-level leakage current, implicit power consumption redundancy and other minor defects; in addition, it does not require frequent manual debugging and modification of threshold parameters, and can achieve autonomous adaptation to different working modes, hardware and software configurations, and network environment power consumption characteristics of the terminal, with adaptability far exceeding that of existing detection schemes; (2) The test efficiency is increased exponentially, the labor cost and operation threshold are greatly reduced, and the test results are standardized and highly consistent: This application can realize 24-hour uninterrupted fully automated data collection, analysis and anomaly judgment, and the analysis time of a single test case is shortened from several hours to several seconds, and the overall test efficiency is increased by more than 10 times, which can adapt to the mass production test needs of large batches of chips; and, it does not require the reliance on senior engineers to carry out complex The complex scenario review, power consumption breakdown, and signaling comparison work have reduced the labor cost by 90%, completely solving the problems of high threshold and high cost of manual detection; at the same time, it eliminates the judgment bias caused by the difference in personal experience of different engineers, unifies the detection standard, and realizes the standardized and normalized output of power consumption anomaly detection; (3) Breaking through the limitations of a single data dimension, possessing interpretability and engineering guidance value, and having the ability to iterate and evolve independently: This application integrates power consumption data and business behavior data with multiple features to realize the judgment of anomalies, fundamentally solving the technical shortcomings of high false detection rate and single feature of single-modal intelligent detection; in addition, it can accurately locate the underlying business root cause corresponding to power consumption anomalies, and is no longer simply outputting abnormal results, so as to directly guide the terminal chip hardware design. The design and software scheduling logic are optimized, giving it strong engineering application value. At the same time, it abandons the traditional static and fixed model and achieves continuous iteration and upgrading of the model based on incremental learning. It can autonomously learn new abnormal scenarios and power consumption characteristics, greatly improving the model's generalization ability and long-term availability. It realizes the technological innovation from human experience-driven and static rule judgment to data intelligence-driven and dynamic iterative optimization. It has significantly improved in five dimensions: detection accuracy, testing efficiency, anomaly coverage, scenario adaptability, and engineering practicality. It effectively accelerates the R&D iteration speed of RedCap chips, comprehensively improves the power consumption performance and long-term operational reliability of terminal products, and is widely applicable to power quality management scenarios of lightweight 5G IoT terminal products such as industrial sensing, smart homes, and remote meter reading.Therefore, this application effectively overcomes the various shortcomings of the prior art and has high industrial application value.

[0094] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A method for detecting abnormal power consumption in RedCap terminals based on a deep learning model, characterized in that, include: Collect power consumption time-series data and service log data of RedCap terminals under various power consumption scenarios, and align the collected power consumption time-series data and service log data according to a preset time alignment mechanism; The power consumption time series data and service log data after time alignment are preprocessed, and features are extracted from the preprocessed power consumption time series data and service log data to obtain power consumption features and service features. The extracted power consumption features and service features are input into a deep learning model for training to construct a power consumption anomaly detection model. The constructed power consumption anomaly detection model is deployed to determine anomalies in the power consumption of the RedCap terminal to be detected and to locate the root cause of the anomalies.

2. The RedCap terminal power consumption anomaly detection method based on a deep learning model according to claim 1, characterized in that, The process of aligning the collected power consumption timing data and service log data according to the preset time alignment mechanism includes: The general input / output interface of the RedCap terminal outputs pulse signals, and triggers the synchronous acquisition of power consumption timing data and service log data based on the output pulse signals. The firmware of the RedCap terminal outputs heartbeat events at preset intervals. Based on the output heartbeat events, the clock drift of power consumption timing data and service log data acquisition is calculated. The calculated clock drift is then corrected based on a linear regression model to obtain the timing-corrected power consumption timing data and service log data. Dynamic timing warping technology is used to dynamically match the timing-corrected power consumption timing data and service log data to obtain timing-aligned power consumption timing data and service log data.

3. The RedCap terminal power consumption anomaly detection method based on a deep learning model according to claim 1, characterized in that, The process of preprocessing the time-aligned power consumption time-series data and service log data, and extracting features from the preprocessed power consumption time-series data and service log data to obtain power consumption features and service features, specifically includes: The power consumption time series data and business log data after time alignment are cleaned, and noise is filtered out from the cleaned power consumption time series data based on wavelet transform technology. Based on the business events in the cleaned business log data, the power consumption time series data after filtering out noise is segmented to obtain multiple power consumption segments, and each power consumption segment is labeled with a corresponding business tag. Power consumption features, including statistical features, time-domain features, and morphological features, are extracted from each power consumption segment with business tags, and business features are extracted from the cleaned business log data. The extracted power consumption and service features are normalized to convert them into a data format suitable for input to deep learning models.

4. The RedCap terminal power consumption anomaly detection method based on a deep learning model according to claim 3, characterized in that, The training process of the power consumption anomaly detection model includes: The input layer based on the deep learning model synchronously receives the normalized power consumption characteristics and service characteristics of the input. The normalized power consumption features and business features are input into the feature encoding layer of the deep learning model for feature encoding to obtain the power consumption time-series feature vector and the business semantic feature vector. The power consumption time-series feature vector and the business semantic feature vector are input into the multi-feature fusion layer of the deep learning model. The multi-feature fusion layer performs feature fusion on the input power consumption time-series feature vector and business semantic feature vector based on the cross-attention mechanism to obtain the fused feature vector. The output layer of the deep learning model receives the input fused feature vector and outputs whether it is abnormal and the type of abnormality if it is abnormal.

5. The RedCap terminal power consumption anomaly detection method based on a deep learning model according to claim 4, characterized in that, The feature coding layer includes a power consumption feature coding layer and a service semantic coding layer; The power consumption feature encoding layer encodes the input power consumption features based on a one-dimensional convolutional neural network to obtain a power consumption time-series feature vector; the service semantic encoding layer encodes the input service features based on an embedding layer to obtain a service semantic feature vector.

6. The RedCap terminal power consumption anomaly detection method based on a deep learning model according to claim 1, characterized in that, Also includes: The power consumption anomaly detection model is iteratively optimized based on a pre-defined model iteration optimization mechanism to update the deployed power consumption anomaly detection model.

7. The RedCap terminal power consumption anomaly detection method based on a deep learning model according to claim 6, characterized in that, The power consumption anomaly detection model is iteratively optimized based on a preset model iteration optimization mechanism to update the deployed power consumption anomaly detection model. The specific process includes: Acquire automated test data and filter the acquired automated test data based on a preset filtering mechanism to select valid samples; The selected valid samples are input into the power consumption anomaly detection model for forward inference, the inference results are output, and the model inference confidence corresponding to each valid sample is extracted from the inference results. The model inference confidence of each valid sample is compared with a preset confidence threshold. If the model inference confidence is greater than or equal to the preset confidence threshold, a pseudo label is generated for the valid sample. If the model inference confidence is less than the preset confidence threshold, the corresponding valid sample is placed into the manual review pool for manual verification. Valid samples with pseudo-labels are input into the incremental learning model for incremental learning training to update the parameters of the power consumption anomaly detection model. The performance of the power consumption anomaly detection model after the parameter update is evaluated. If the performance of the power consumption anomaly detection model after the parameter update is improved, the updated power consumption anomaly detection model is deployed; if the performance of the power consumption anomaly detection model after the parameter update is degraded, it is rolled back to the stable version.

8. A RedCap terminal power consumption anomaly detection device based on a deep learning model, characterized in that, include: The acquisition module is used to collect power consumption time-series data and service log data of RedCap terminals under various power consumption scenarios, and to align the collected power consumption time-series data and service log data according to a preset time alignment mechanism. The feature extraction module is used to preprocess the time-aligned power consumption time-series data and service log data, and to extract features from the preprocessed power consumption time-series data and service log data to obtain power consumption features and service features. The model building module is used to input the extracted power consumption features and service features into the deep learning model for training, so as to build a power consumption anomaly detection model. The model deployment module is used to deploy the constructed power consumption anomaly detection model, and to determine the anomaly and locate the root cause of the anomaly in the power consumption of the RedCap terminal to be detected.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1 to 7.

10. An electronic terminal, comprising a memory, a processor, and a computer program stored in the memory, characterized in that, The processor executes the computer program to implement the method as described in any one of claims 1 to 7.

Images