Separated transmission equipment data processing method and system and storage medium

Abstract

The embodiment of the invention provides a data processing method and system for separated transmission equipment and a storage medium, and belongs to the technical field of data processing. The method comprises the steps of obtaining current task data and historical task data; according to the current task data and the historical task data, at least one selected feature template is selected from a preset feature template library according to a preset selection rule, target feature templates are obtained, and each target feature template forms a target instruction set. And sending the target instruction set to the edge device, and receiving a target feature set acquired and processed by the edge device according to the target instruction set. And performing interpolation fusion on the target feature set and the historical task data according to a preset interpolation fusion reconstruction model to obtain a unified tensor. And inputting the unified tensor into a preset fault prediction model to obtain prediction data. Unnecessary information uploading is avoided, and the problems of resource waste and data missing are solved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a data processing method, system and storage medium for a separate transmission device. Background Art

[0002] In modern intelligent systems, an increasing number of devices are deployed in edge environments to undertake data collection and preliminary processing tasks, such as smart security cameras, smart meters, and in-vehicle terminals. These edge devices connect to central servers and upload the collected data to the central servers for further analysis and decision-making. In this type of architecture, edge devices and central servers are decoupled in terms of physical deployment and computing functions. In related technologies, edge devices all use "static" processing logic, such as fixed compression algorithms or feature extraction rules. These algorithms lack the ability to perceive task semantics or central analysis needs, resulting in uploaded data with low relevance to the task, wasting bandwidth and affecting the accuracy of the central server's predictions on the uploaded data. Therefore, how to improve the accuracy of central server predictions under resource constraints has become a technical problem that needs to be solved urgently. Summary of the Invention

[0003] The main purpose of the embodiments of the present application is to propose a data processing method, system and storage medium for a separate transmission device, aiming to improve the accuracy of central server prediction under resource constraints.

[0004] To achieve the above-mentioned objectives, a first aspect of an embodiment of the present application provides a method for processing data of a separate transmission device, which is applied to a central server. The method includes:

[0005] Get current task data and historical task data;

[0006] Selecting at least one selected feature template from a preset feature template library according to the current task data and the historical task data according to a preset selection rule to obtain a target feature template; wherein each of the target feature templates constitutes a target instruction set;

[0007] Sending the target instruction set to the edge device, and receiving the target feature set collected and processed by the edge device according to the target instruction set;

[0008] Interpolating and fusing the target feature set and the historical task data according to a preset interpolation fusion reconstruction model to obtain a unified tensor;

[0009] The unified tensor is input into a preset fault prediction model to obtain prediction data.

[0010] In some embodiments, selecting at least one selected feature template from a preset feature template library according to a preset selection rule based on the current task data and the historical task data to obtain a target feature template includes:

[0011] Constructing a task attention vector based on a preset feature template library and the current task data; wherein the task attention vector represents the degree of attention of the current task to each of the selected feature templates;

[0012] Obtaining a comprehensive attention score for each of the selected feature templates based on the task attention vector and the historical task data;

[0013] A preset number of selected feature templates with the highest scores are selected according to the comprehensive attention score to obtain the target feature template.

[0014] In some embodiments, the feature template library includes at least one selected feature template, and constructing the task attention vector based on the preset feature template library and the current task data includes:

[0015] Obtaining a semantic embedding vector of the current task data;

[0016] Obtaining a description embedding vector of each of the selected feature templates;

[0017] The task attention vector is obtained according to the semantic embedding vector and the description embedding vector; wherein the task attention vector includes a task attention score for each selected feature template.

[0018] To achieve the above objectives, a second aspect of an embodiment of the present application provides a data processing method for a separate transmission device, which is applied to an edge device. The method includes:

[0019] Receive a target instruction set; wherein the target instruction set is generated by a central server based on current task data and historical task data;

[0020] Selecting from the initial feature processing module set according to the target instruction set and a preset resource scheduling loss function to obtain at least one target feature processing module;

[0021] Perform feature extraction on the collected data according to the target instruction set and the target feature processing module to obtain a target feature set;

[0022] The target feature set is sent to the central server.

[0023] In some embodiments, after extracting features from the collected data according to the target instruction set and the target feature processing module to obtain a target feature set, the method further includes:

[0024] Obtaining a stability estimation value of the target feature set according to a preset modal stability estimation function;

[0025] If the stability estimate is greater than or equal to a preset threshold, the target feature set is not updated;

[0026] If the stability estimation value is less than a preset threshold, the target feature set is subjected to dimensionality downsampling processing to obtain an updated target feature set.

[0027] In some embodiments, the target feature set includes initial sub-features, and after extracting features from the collected data according to the target instruction set and the target feature processing module to obtain the target feature set, the method further includes:

[0028] Obtaining an abnormal value of the initial sub-feature according to a preset dual-sensing abnormality trigger function;

[0029] Obtaining a confidence value of the initial sub-feature according to a preset confidence dynamic suppression function;

[0030] Acquire the initial sub-feature whose abnormal value is greater than a preset abnormal threshold and whose confidence value is greater than a preset confidence threshold, and obtain at least one target sub-feature;

[0031] The target feature set is updated according to the target sub-feature to obtain the updated target feature set.

[0032] In some embodiments, the target instruction set includes multiple target instructions, each of which includes a module number. The target processing module is obtained by selecting from the initial feature processing module set according to the target instruction set and a preset resource scheduling loss function, including:

[0033] Selecting from the initial feature processing module set according to the module number to obtain a first feature processing module set;

[0034] At least one target feature processing module is selected from the first feature processing module set according to the resource scheduling loss function.

[0035] To achieve the above objectives, a third aspect of an embodiment of the present application provides a separate transmission device data processing system, which is applied to a central server. The system includes:

[0036] Acquisition module, used to obtain current task data and historical task data;

[0037] A first selection module is configured to select at least one selected feature template from a preset feature template library according to the current task data and the historical task data according to a preset selection rule to obtain a target feature template; wherein each target feature template constitutes a target instruction set;

[0038] A first sending module is configured to send the target instruction set to an edge device and receive a target feature set collected and processed by the edge device according to the target instruction set;

[0039] A fusion module, configured to interpolate and fuse the target feature set and the historical task data according to a preset interpolation fusion reconstruction model to obtain a unified tensor;

[0040] The prediction module is used to input the unified tensor into a preset fault prediction model to obtain prediction data.

[0041] To achieve the above objectives, a fourth aspect of an embodiment of the present application provides a separate transmission device data processing system, which is applied to an edge device. The system includes:

[0042] A receiving module, configured to receive a target instruction set; wherein the target instruction set is generated by a central server based on current task data and historical task data;

[0043] a second selection module, configured to select from the initial feature processing module set according to the target instruction set and a preset resource scheduling loss function to obtain at least one target feature processing module;

[0044] The extraction module is used to extract features from the collected data according to the target instruction set and the target feature processing module to obtain a target feature set.

[0045] The second sending module is used to send the target feature set to the central server.

[0046] To achieve the above-mentioned purpose, the fifth aspect of the embodiment of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method described in the first aspect and the method described in the second aspect.

[0047] The present application proposes a method, system and storage medium for processing data of a separate transmission device, which obtains current task data and historical task data. Based on the current task data and historical task data, at least one selected feature template is selected from a preset feature template library according to preset selection rules to obtain a target feature template, wherein each target feature template constitutes a target instruction set. The target instruction set is sent to the edge device, and the target feature set collected and processed by the edge device according to the target instruction set is received. The target feature set and the historical task data are interpolated and fused according to a preset interpolation fusion reconstruction model to obtain a unified tensor. The unified tensor is input into a preset fault prediction model to obtain predicted data. The target instruction set is related to the current task, which enables the edge device to upload only data features that are highly relevant to the current task, avoiding unnecessary information upload. The context-aware reconstruction model restores and completes the received sparse target feature set to overcome the data missing problem caused by edge preprocessing and bandwidth limitations. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] Figure 1 This is a flow chart of a data processing method for a separate transmission device provided in an embodiment of the present application;

[0049] Figure 2 yes Figure 1 Flowchart of step S102 in FIG.

[0050] Figure 3 is a flow chart of a data processing method for a separate transmission device provided by another embodiment of the present application;

[0051] Figure 4 yes Figure 3 Flowchart of step S302 in FIG.

[0052] Figure 5 is a flow chart of a data processing method for a separate transmission device provided in the third embodiment of the present application;

[0053] Figure 6 is a flow chart of a data processing method for a separate transmission device provided in a fourth embodiment of the present application;

[0054] Figure 7 It is a structural diagram of a separate transmission device data processing system provided in an embodiment of the present application;

[0055] Figure 8 It is a structural diagram of a separate transmission device data processing system provided by another embodiment of the present application;

[0056] Figure 9 This is a schematic diagram of the hardware structure of the electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of this application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0058] It should be noted that although the system diagrams illustrate functional module divisions and the flowcharts illustrate logical sequences, in certain circumstances, the steps shown or described may be performed in a sequence that differs from the module divisions in the system or the sequence in the flowcharts. The terms "first," "second," and so on, in the specification, claims, and drawings, are used to distinguish similar items and are not necessarily used to describe a specific sequence or precedence.

[0059] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this application pertains. The terms used herein are for the purpose of describing the embodiments of this application only and are not intended to limit this application.

[0060] In modern intelligent systems, an increasing number of devices are deployed in edge environments to undertake data collection and preliminary processing tasks, such as smart security cameras, smart meters, and in-vehicle terminals. These edge devices connect to central servers and upload the collected data to the central servers for further analysis and decision-making. In this type of architecture, edge devices and central servers are decoupled in terms of physical deployment and computing functions. In related technologies, edge devices all use "static" processing logic, such as fixed compression algorithms or feature extraction rules. These algorithms lack the ability to perceive task semantics or central analysis requirements, resulting in a low correlation between the uploaded data and the task, which not only wastes bandwidth but also affects the accuracy of the central server's predictions on the uploaded data.

[0061] Based on this, the embodiments of the present application provide a data processing method, system and storage medium for a separated transmission device, which aims to select a target feature template from a preset feature template library according to preset selection rules based on current task data and historical task data. Each target feature template constitutes a target instruction set, so that the target instruction set has a high correlation with the current task. The edge device can upload the corresponding target feature set to the central server according to the target instruction set. Since the edge device only uploads necessary features, the problem of bandwidth waste is solved. The central server will also interpolate and fuse the target feature set based on historical task data, realizing a context-aware data reconstruction mechanism, so that the central server can be more accurate when predicting the data uploaded by the edge device.

[0062] The data processing method, system and storage medium of the separated transmission device provided in the embodiments of the present application are specifically explained through the following embodiments. First, the data processing method of the separated transmission device in the embodiments of the present application is described.

[0063] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Artificial Intelligence (AI) is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use knowledge to achieve optimal results.

[0064] Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.

[0065] The data processing method for a separated transmission device provided in the embodiments of the present application can be applied to a terminal, can be applied to a server side, and can also be software running in a terminal or a server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or as a server cluster or distributed system composed of multiple physical servers, or as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the data processing method for a separated transmission device, etc., but is not limited to the above forms.

[0066] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments in which tasks are performed by remote processing devices connected via a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0067] Please refer to Figure 1 , Figure 1This is a flowchart of a separate transmission device data processing method provided in an embodiment of the present application, and the separate transmission device data processing method is applied to a central server. Figure 1 The method may include but is not limited to steps S101 to S105.

[0068] Step S101, obtaining current task data and historical task data;

[0069] Step S102, selecting at least one selected feature template from a preset feature template library according to the current task data and the historical task data according to a preset selection rule to obtain a target feature template; wherein each target feature template constitutes a target instruction set;

[0070] Step S103: Send the target instruction set to the edge device, and receive the target feature set collected and processed by the edge device according to the target instruction set;

[0071] Step S104: interpolating and fusing the target feature set and the historical task data according to a preset interpolation fusion reconstruction model to obtain a unified tensor;

[0072] Step S105 , inputting the unified tensor into a preset fault prediction model to obtain prediction data.

[0073] In step S101 of some embodiments, the current task data is generated by the system scheduler or manual configuration, and includes task type, related tags, and attention modalities. Task types include equipment overheating prediction and anomaly detection, and related tags include equipment overheating, motor anomalies, or structural looseness. The attention modalities are the types of data required to identify related tags. Such as temperature, vibration, image, and voltage. In one example, if the task type is bearing wear prediction, the related tag is wear, and the attention modalities are vibration and sound signals. Historical task data is stored in the context cache, which includes historically uploaded feature sets and corresponding model output results.

[0074] In one example, T represents current task data, H represents historical task data, then T = (task_type, target_label, modality_focus), task_type represents the task type, target_label represents the relevant label, and modality_focus represents the focus modality. represents the feature subset uploaded by the edge device for the tith time, y t-i Indicates the model output result after the ti-th fault prediction model predicts the feature subset uploaded for the ti-th time.

[0075] See also Figure 2In some embodiments, step S102 may include steps S201 to S203:

[0076] Step S201: constructing a task attention vector based on a preset feature template library and current task data; wherein the task attention vector represents the degree of attention of the current task to each selected feature template;

[0077] Step S202: obtaining a comprehensive attention score for each selected feature template based on the task attention vector and historical task data;

[0078] Step S203 , selecting a preset number of feature templates with the highest scores according to the comprehensive attention scores to obtain a target feature template.

[0079] In step S201 of some embodiments, a feature template library is predefined. Each selected feature template f j It is a structured feature processing module, such as f1 is the signal sliding mean and kurtosis calculation, f2 is the image patch local gradient statistics, f3 is the volatility in the temperature time series window and the frequency band energy concentration in the Fourier transform. j All contain the mode M j , processing type, parameter range, historical performance and other structural fields.

[0080] Get the semantic embedding vector E of the current task data T , and the selected feature template f j The description embedding vector E j After comparing the similarities, a task attention vector r is constructed to screen the selected feature templates. The task attention vector represents the degree of attention of the current task to each selected feature template, as shown in the following formula (1):

[0081]

[0082] Among them, r j represents the task attention score of the j-th selected feature template, E T The semantic embedding vector representing the current task data can be generated by averaging keyword vectors or semantic models. For example, E T =Encoder("task_type: overheat prediction; target_label: overheat; modality_focus: temperature, vibration"). E j Represents the selected feature template f j The description embedding vector, E k Represents the selected feature template f kThe description embedding vector. ||·||2 represents the Euclidean distance, which is normalized by softmax to ensure ∑ j r j =1, task attention vector r = {r1, r2, ..., r m}, that is, the task attention vector is composed of all task attention scores.

[0083] In some embodiments, in steps S202 to S203, the task attention vector and historical task data are combined to calculate the comprehensive attention score of each selected feature template, and a residual-guided regularization term is introduced to strengthen features with good historical performance. Specifically, as shown in the following formula (2):

[0084]

[0085] Among them, a j represents the comprehensive attention score of the j-th selected feature template, r j represents the task attention score of the jth selected feature template, λ represents the residual term adjustment factor, which is used to control the influence weight of historical task data, Is an indicator function, if f j Appears in feature subset In the case of y, it is 1. t-i represents the model output result after the ti-th fault prediction model predicts the feature subset uploaded for the ti-th time, y true,t-i is the ti-th true label, |y t-i -y true,t-i | represents the prediction error. (1-error) indicates the contribution of this feature subset to improving accuracy.

[0086] It should be noted that when the preset number is k, according to the comprehensive attention score a j , from the feature template library The top k selected feature templates with the highest scores are selected as target feature templates, and each target feature template constitutes a target instruction set. The specific formula (3) is as follows:

[0087]

[0088] Where I represents the target instruction set, which consists of k structured instructions. Represents the feature template library, a1,...,a m Represents all comprehensive attention scores, and k represents the number of predictions. Each target instruction i j Including template_id, params, modality, priority_score and other fields. Target instruction structure example:

[0089] i1={"id":3,"type":"gradient","modality":"temperature","params":{"window":5}}.

[0090] i2={"id":8,"type":"fft","modality":"vibration","params":{"band":[10,50]}}.

[0091] In steps S201 to S203 shown in the embodiment of the present application, a task attention vector is constructed through a feature template library and current task data. Based on the task attention vector and historical task data, a comprehensive attention score is obtained for each selected feature template. A preset number of selected feature templates with the highest scores are selected based on the comprehensive attention scores to obtain target feature templates and construct a target instruction set. By generating a target instruction set related to the current task, the edge device uploads the target feature set according to the target instruction set, and can only upload data features that are highly relevant to the current task. Avoid unnecessary information uploading and improve transmission efficiency and data value density from the source.

[0092] In step S103 of some embodiments, the central server generates a target instruction set related to the current task and sends the target instruction set to the edge device. The edge device sets relevant parameters according to the target instruction set, collects relevant data, obtains a target feature set based on the collected data, and sends the target feature set to the central server.

[0093] In step S104 of some embodiments, since the edge device only uploads the filtered target feature set F * , resulting in the lack of complete context in the central server. To ensure that the fault prediction model can obtain continuous and uniform dimensional input, it can be based on the historical task data H and the target feature set F * , perform time completion and structure fusion to complete the unified tensor 's construction.

[0094] Specifically, in order to achieve structure completion and modality consistency, a temporal interpolation-context fusion reconstruction model is constructed, namely the interpolation fusion reconstruction model Its goal is to transform the target feature set F * Fusion with historical task data H generates a unified tensor The overall process is: Constructing the modal-time matrix Where d is the feature dimension set, L is the length of the sliding time window (such as 20 steps); if a certain moment or mode is in F * If it is missing, it is marked as a missing point and used as the interpolation target to be estimated; the interpolation fusion is used to reconstruct the model Perform missing completion, as shown in the following formula (4):

[0095]

[0096] in, Represents the reconstructed unified tensor, M represents the modality-time matrix containing some missing parts, directly from F * Construct. Mask represents the missing position Boolean matrix, HistPattern(H) represents the "modal temporal trend" extracted from the historical task data H, used for interpolation. Fill(·) is the context-guided interpolation algorithm defined by the central end (such as mode-preserving interpolation and trend-guided estimation).

[0097] In one example, to ensure that the interpolation structure is consistent with the actual change trend, a change trend preservation term Δ is introduced to fit the first-order change direction in each historical mode, and this direction is used to guide the estimation of the missing segment during completion, as shown in the following formula (5):

[0098]

[0099] Among them, Δ j represents the main change direction of mode j, Represents the average eigenvalue of mode j at time t. This direction is used to determine whether the current missing value should be filled upward or downward. For example, when the historical value increases continuously, the missing value adopts the "interpolation upward" strategy. This filling strategy is different from the conventional mean filling or linear interpolation, and is more consistent with the task semantics. After all valuations are completed, the output

[0100] In step S105 of some embodiments, the unified tensor After input into the fault prediction model, prediction data is output, for example, the prediction data includes failure probability, classification label and predicted value.

[0101] It should be noted that in order to support the subsequent system feedback module, the prediction error E of each round is recorded to determine whether the fault prediction model is stable or whether further instruction optimization is required. This is shown in the following formula (6):

[0102]

[0103] Among them, E represents the prediction error, which is written to the system log. If it is a regression task, then Can be |yy true |, if it is a classification task, then Can be the cross entropy loss, y is the output result of the fault prediction model, y true is the true label (from expert system or sensor annotation).

[0104] To better support subsequent feedback analysis, the fault prediction model also outputs a set of characteristic response vectors Δ′ = {Δ′ j}, each Δ′ j express The influence of the jth dimension on the output y is usually achieved through internal attention or derivative estimation of the model. The response value structure is shown in the following formula (7):

[0105]

[0106] Among them, Δ′ j represents the feature response of the jth dimension, y represents the output result of the fault prediction model, Represents a unified tensor of the jth dimension. At the same time, the current feature subset and model output results need to be stored in the historical task data.

[0107] Steps S101 to S105 shown in the embodiment of the present application are performed by obtaining current task data and historical task data. According to the current task data and historical task data, at least one selected feature template is selected from the preset feature template library according to the preset selection rules to obtain a target feature template, wherein each target feature template constitutes a target instruction set. The target instruction set is sent to the edge device, and the target feature set collected and processed by the edge device according to the target instruction set is received. The target feature set and the historical task data are interpolated and fused according to the preset interpolation fusion reconstruction model to obtain a unified tensor. The unified tensor is input into the preset fault prediction model to obtain predicted data. The target instruction set is related to the current task, and enables the edge device to upload only data features that are highly relevant to the current task, avoiding unnecessary information uploading. The context-aware reconstruction model restores and completes the received sparse target feature set to overcome the data missing problem caused by edge preprocessing and bandwidth limitations.

[0108] Please refer to Figure 3 , Figure 3 This is a flowchart of a separate transmission device data processing method provided by another embodiment of the present application, and the separate transmission device data processing method is applied to an edge device. Figure 3 The method may include but is not limited to steps S301 to S304.

[0109] Step S301, receiving a target instruction set; wherein the target instruction set is generated by a central server based on current task data and historical task data;

[0110] Step S302: selecting from the initial feature processing module set according to the target instruction set and the preset resource scheduling loss function to obtain at least one target feature processing module;

[0111] Step S303, performing feature extraction on the collected data according to the target instruction set and the target feature processing module to obtain a target feature set;

[0112] Step S304: Send the target feature set to the central server.

[0113] In step S301 of some embodiments, the central server generates a target instruction set highly relevant to the current task based on the current task data and historical task data, and then sends the target instruction set to the edge device. This enables the edge device to extract only the data features highly relevant to the current task based on the target instruction set, thus avoiding unnecessary information upload.

[0114] See also Figure 4 In some embodiments, the target instruction set includes multiple target instructions, the target instructions include module numbers, and step S302 may include steps S401 to S402:

[0115] Step S401, selecting from the initial feature processing module set according to the module number to obtain a first feature processing module set;

[0116] Step S402 : selecting from the first feature processing module set according to the resource scheduling loss function to obtain at least one target feature processing module.

[0117] In step S401 of some embodiments, each target instruction corresponds to an initial feature processing module deployed locally on the edge device, and all initial feature processing modules on an edge device constitute an initial feature processing module set. j Contains the template_id, modality, type, params, and priority_score fields. template_id is the module ID, modality is the mode of action, such as "temperature," "vibration," and "image." type is the processing method, such as gradient, spectral_energy, and local_stats. params are configuration parameters (window size, frequency band range, etc.). priority_score is the overall attention score. The target instruction set determines the initial feature processing module and parameter configuration that the edge module should execute.

[0118] Specifically, a selection is made from the initial feature processing module set based on all module numbers contained in the target instruction set. The selected initial feature processing module is called the first feature processing module, and all first feature processing modules constitute the first feature processing module set. The first feature processing module is the module required to complete the target instruction. For example, the "equipment overheating prediction" task usually requires modules such as the sliding mean, volatility, and temperature rise rate under the temperature mode. The "abnormal vibration identification" task prefers spectrum analysis, peak statistics, and short-time energy analysis modules under the vibration mode.

[0119] In step S402 of some embodiments, to address the issues of limited computing resources and modality imbalance in edge devices, the execution granularity of the first feature processing module to be run can be dynamically adjusted through a resource scheduling loss function. The resource scheduling loss function is shown in the following formula (8):

[0120]

[0121] in, represents the resource scheduling loss function, c j Indicates the resource cost (such as computing power and memory) expected to be consumed by the jth instruction. R represents the current remaining resource budget of the edge device, p j =t j .priority_score, where priority_score represents the comprehensive attention score of the selected feature template, t j .priority_score represents the composition of i j The sum of the comprehensive attention scores of all selected feature templates of the instruction, ∈ is a small constant to prevent division by zero. The system minimizes As a target, a first feature processing module with high task relevance and low overhead is selected and retained to obtain at least one target feature processing module.

[0122] It's important to note that by retaining only the target feature processing modules with high task relevance and low overhead, and executing the current task through these modules, resource overhead can be further reduced while completing the current task, enabling stable execution of critical tasks on edge devices with limited computing power. For example, when running "Fault Warning" on a low-power device, activating only the vibration peak and frequency energy modules can avoid CPU overload, increased power consumption, or excessive latency. This results in a more stable system and more responsive response. For example, in the "Device Overheat Prediction" task, retaining the "Temperature Mean" and "Temperature Fluctuation" modules, which have low computational complexity but high task weight, still provides sufficient discrimination without the need for resource-intensive image feature extraction. This preserves the core capabilities of the task while saving resources.

[0123] In some embodiments, in steps S403 to S404, after selecting the target feature processing module, the edge device sets parameters of the target feature processing module according to the target instruction set, performs feature extraction on the data collected by the target feature processing module, obtains a target feature set, and sends the target feature set to the central server.

[0124] In steps S401 to S404 illustrated in this embodiment, a target instruction set is received, wherein the target instruction set is generated by a central server based on current task data and historical task data. At least one target feature processing module is selected from the initial feature processing module set according to the target instruction set and the preset resource scheduling loss function. Feature extraction is performed on the collected data according to the target instruction set and the target feature processing module to obtain a target feature set, and the target feature set is sent to the central server. Since the target instruction set is highly relevant to the current task, the target instruction set can be used to guide the edge device to extract data features that are highly relevant to the current task, thereby avoiding unnecessary information uploads that cause waste of resources. Further, the resource scheduling loss function is used to retain target feature processing modules with high relevance to the execution task and low overhead. When the current task is completed, the resource overhead is further reduced to prevent insufficient computing power of the edge device.

[0125] See also Figure 5 In some embodiments, after step S403, the data processing method of the separate transmission device may further include but is not limited to steps S501 to S503:

[0126] Step S501, obtaining a stability estimation value of a target feature set according to a preset modal stability estimation function;

[0127] Step S502: If the stability estimate is greater than or equal to the preset threshold, the target feature set is not updated;

[0128] Step S503: If the stability estimation value is less than a preset threshold, the target feature set is subjected to dimensionality downsampling processing to obtain an updated target feature set.

[0129] In step S501 of some embodiments, when bandwidth is limited, it can be determined whether the target feature set can be automatically downsampled or combined for representation by using the stability estimation value of the target feature set. The stability estimation value of the target feature set is obtained by using the modal stability estimation function, as shown in the following formula (9):

[0130]

[0131] S (m) represents the stability estimate, Var(X (m) ) represents the variance of the target feature set within the time window, μ(X(m) ) represents the mean, and δ is a stabilization factor to prevent division by zero. The smaller S(m), the weaker the fluctuation of the target feature set, the higher the information redundancy, and the more suitable the downsampling.

[0132] In steps S502 to S503 of some embodiments, τ s Indicates setting the threshold. When S (m) ≥τ s , there is no need to downsample the target feature set, so the target feature set is not updated. S (m) <τ s When the target feature set is updated, it is necessary to perform dimensionality downsampling on the target feature set to obtain an updated target feature set. Dimensionality downsampling can be performed using average pooling or principal component retention. This dimensionality downsampling process can save data transmission costs.

[0133] In steps S501 to S503 of this embodiment, the target feature set is automatically downsampled, thereby further saving data transmission costs and reducing bandwidth.

[0134] See also Figure 6 In some embodiments, after step S403, the data processing method of the separate transmission device may further include but is not limited to steps S601 to S604:

[0135] Step S601, obtaining an abnormal value of the initial sub-feature according to a preset dual-sensing abnormality trigger function;

[0136] Step S602, obtaining the confidence value of the initial sub-feature according to a preset confidence dynamic suppression function;

[0137] Step S603: obtaining an initial sub-feature whose abnormal value is greater than a preset abnormal threshold and whose confidence value is greater than a preset confidence threshold, and obtaining at least one target sub-feature;

[0138] Step S604: update the target feature set according to the target sub-feature to obtain an updated target feature set.

[0139] In step S601 of some embodiments, the target feature set includes initial sub-features, and the target feature set F = {f'1, f'2, ..., f' p}, f′ j Represents the initial sub-feature.

[0140] In order to further reduce the data volume of the target feature set and save bandwidth, the feature subset F that is "truly worth uploading" can be identified from the target feature set F. * Therefore, the task-modality dual-sensing abnormal trigger function is used for identification. The dual-sensing abnormal trigger function is shown in the following formula (10):

[0141] g j =α·|f′ j -μ j |+β·TE j +λ·a j , (10)

[0142] Among them, g j represents the outlier, α, β, and λ are balance factors, which are preset when the edge device is deployed. j Represents the current initial sub-feature, μ j Indicates the current historical feature mean, calculated from historical task data. TE j Represents the selected feature template f j The mode M j The time entropy of the feature in the last n periods is used to determine whether the mode is in the "disturbance period". j is the comprehensive attention score (task-specific regularization term) of the jth selected feature template, which represents the task weight coefficient corresponding to the feature and reflects the importance of the feature template in the current task. g Indicates the preset abnormal threshold. When g j >θ g When f′ j Determined to be an "abnormal candidate upload feature".

[0143] In step S602 of some embodiments, in order to avoid frequent uploading due to short-term disturbances, the current initial sub-feature f′ is measured by a confidence dynamic suppression function. j Whether it has continuously brought valuable model prediction optimization over the past period of time. This value is updated by the central server training result feedback round by round and executed by the edge device. The confidence dynamic suppression function is shown in the following formula (11):

[0144]

[0145] in, Represents the confidence update value of the tth round, which is continuously updated through formula (11) to obtain the confidence value Γ of the initial sub-feature j η represents the momentum coefficient, y t-1 represents the output of the fault prediction model in round t-1, y true,t-1 represents the true label of the t-1th round. represents the modal prediction variance, and δ is a stabilization factor to prevent division by zero. If the error is significantly reduced after a feature history is uploaded, then Γ j Will continue to rise.

[0146] In some embodiments, steps S603 and S604 obtain an initial sub-feature whose outlier value is greater than a preset outlier threshold and whose confidence value is greater than a preset confidence threshold, thereby obtaining at least one target sub-feature. The target feature set is then updated based on the target sub-feature to obtain an updated target feature set. That is, the updated target feature set only includes the target sub-feature. This reduces the amount of transmitted data and further saves data transmission costs.

[0147] The judgment method of whether the abnormal value is greater than the preset abnormal threshold and the confidence value is greater than the preset confidence threshold is shown in the following formula (12):

[0148] F * ={f′ j ∈F|f j >θ g ∧Γ j >θ Γ}, (12)

[0149] Among them, F * represents the updated target feature set, f' j represents the initial sub-feature, F represents the original target feature set, g j represents the abnormal value of the initial sub-feature, θ g Indicates the preset abnormal threshold. Γ j represents the confidence value of the initial sub-feature, θ Γ Indicates the preset trust threshold. When g j >θ g and Γ j >θ Γ , the initial sub-feature is used as the target sub-feature, and the updated target feature set only includes the target sub-feature.

[0150] In steps S601 to S604 of this embodiment, an abnormal value greater than a preset abnormality threshold indicates a significant structural change. A confidence value greater than the preset confidence threshold indicates that the historical upload is valid. Upload to the central server is only performed when both significant structural change and valid historical upload are met, further saving data transmission costs and bandwidth.

[0151] See also Figure 7 The present application also provides a separate transmission device data processing system, which is applied to a central server and can implement the separate transmission device data processing method applied to the central server. The system includes:

[0152] Acquisition module 701, used to acquire current task data and historical task data;

[0153] The first selection module 702 is configured to select at least one selected feature template from a preset feature template library according to a preset selection rule based on the current task data and the historical task data to obtain a target feature template; wherein each target feature template constitutes a target instruction set;

[0154] The first sending module 703 is configured to send the target instruction set to the edge device and receive the target feature set collected and processed by the edge device according to the target instruction set;

[0155] A fusion module 704 is configured to interpolate and fuse the target feature set and the historical task data according to a preset interpolation fusion reconstruction model to obtain a unified tensor;

[0156] The prediction module 705 is used to input the unified tensor into a preset fault prediction model to obtain prediction data.

[0157] The specific implementation of the separate transmission device data processing system is substantially the same as the specific embodiment of the separate transmission device data processing method applied to the central server, and will not be described in detail here.

[0158] See also Figure 8 The present application also provides a separate transmission device data processing system, which is applied to an edge device and can implement the separate transmission device data processing method applied to the edge device. The system includes:

[0159] Receiving module 801, for receiving a target instruction set; wherein the target instruction set is generated by a central server based on current task data and historical task data;

[0160] A second selection module 802 is configured to select from the initial feature processing module set according to the target instruction set and a preset resource scheduling loss function to obtain at least one target feature processing module;

[0161] The extraction module 803 is used to extract features from the collected data according to the target instruction set and the target feature processing module to obtain a target feature set.

[0162] The second sending module 804 is configured to send the target feature set to the central server.

[0163] The specific implementation of the separate transmission device data processing system is basically the same as the specific embodiment of the separate transmission device data processing method applied to the edge device mentioned above, and will not be repeated here.

[0164] An embodiment of the present application further provides an electronic device comprising a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the aforementioned data processing method for a separate transmission device applied to a central server and the data processing method for a separate transmission device applied to an edge device. The electronic device can be any smart terminal, including a tablet computer and an in-vehicle computer.

[0165] See also Figure 9 , Figure 9 The hardware structure of an electronic device according to another embodiment is shown. The electronic device includes:

[0166] The processor 901 can be implemented as a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0167] The memory 902 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 902 can store an operating system and other application programs. When the technical solutions provided in the embodiments of this specification are implemented through software or firmware, the relevant program code is stored in the memory 902 and is called by the processor 901 to execute the data processing method for a separate transmission device applied to a central server and the data processing method for a separate transmission device applied to an edge device of this application;

[0168] Input / output interface 903, used to implement information input and output;

[0169] Communication interface 904, used to implement communication interaction between this device and other devices, which can be achieved through wired means (such as USB, network cable, etc.) or wireless means (such as mobile network, WiFi, Bluetooth, etc.);

[0170] Bus 905 , which transmits information between various components of the device (e.g., processor 301 , memory 302 , input / output interface 903 , and communication interface 904 );

[0171] The processor 901 , the memory 902 , the input / output interface 903 and the communication interface 904 are connected to each other in communication within the device via a bus 905 .

[0172] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the above-mentioned data processing method of the separated transmission device applied to the central server and the data processing method of the separated transmission device applied to the edge device.

[0173] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0174] The embodiments described in the embodiments of this application are intended to more clearly illustrate the technical solutions of the embodiments of this application and do not constitute a limitation on the technical solutions provided by the embodiments of this application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0175] Those skilled in the art will understand that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0176] The system embodiment described above is merely illustrative. The units described as separate components may or may not be physically separate, i.e., they may be located in one place or distributed across multiple network units. Some or all of the modules may be selected based on actual needs to achieve the objectives of this embodiment.

[0177] Those skilled in the art will appreciate that all or some of the steps in the methods, systems, and functional modules / units in the devices disclosed above may be implemented as software, firmware, hardware, or appropriate combinations thereof.

[0178] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0179] It should be understood that in this application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0180] In the several embodiments provided in this application, it should be understood that the disclosed systems and methods can be implemented in other ways. For example, the system embodiments described above are merely illustrative. For example, the division of the above-mentioned units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of systems or units, which can be electrical, mechanical or other forms.

[0181] The units described above as separate components may or may not be physically separate, and 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 these units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0182] In addition, the functional units in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0183] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes multiple instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store programs.

[0184] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but are not intended to limit the scope of the present invention. Any modifications, equivalent substitutions, and improvements made by those skilled in the art without departing from the scope and essence of the present invention should be within the scope of the present invention.

Claims

1. A data processing method for a separate transmission device, characterized in that: Applied to a central server, the method includes: Get current task data and historical task data; Selecting at least one selected feature template from a preset feature template library according to the current task data and the historical task data according to a preset selection rule to obtain a target feature template; wherein each of the target feature templates constitutes a target instruction set; Sending the target instruction set to the edge device, and receiving the target feature set collected and processed by the edge device according to the target instruction set; Interpolating and fusing the target feature set and the historical task data according to a preset interpolation fusion reconstruction model to obtain a unified tensor; The unified tensor is input into a preset fault prediction model to obtain prediction data.

2. The method according to claim 1, characterized in that The step of selecting at least one selected feature template from a preset feature template library according to the current task data and the historical task data according to a preset selection rule to obtain a target feature template includes: Constructing a task attention vector based on a preset feature template library and the current task data; wherein the task attention vector represents the degree of attention of the current task to each of the selected feature templates; Obtaining a comprehensive attention score for each of the selected feature templates based on the task attention vector and the historical task data; A preset number of selected feature templates with the highest scores are selected according to the comprehensive attention score to obtain the target feature template.

3. The method according to claim 1, characterized in that The feature template library includes at least one selected feature template, and constructing a task attention vector according to the preset feature template library and the current task data includes: Obtaining a semantic embedding vector of the current task data; Obtaining a description embedding vector of each of the selected feature templates; The task attention vector is obtained according to the semantic embedding vector and the description embedding vector; wherein the task attention vector includes a task attention score for each selected feature template.

4. A data processing method for a separate transmission device, characterized in that: Applied to an edge device, the method includes: Receive a target instruction set; wherein the target instruction set is generated by a central server based on current task data and historical task data; Selecting from the initial feature processing module set according to the target instruction set and a preset resource scheduling loss function to obtain at least one target feature processing module; Perform feature extraction on the collected data according to the target instruction set and the target feature processing module to obtain a target feature set; The target feature set is sent to the central server.

5. The method according to claim 4, characterized in that After extracting features from the collected data according to the target instruction set and the target feature processing module to obtain a target feature set, the method further includes: Obtaining a stability estimation value of the target feature set according to a preset modal stability estimation function; If the stability estimate is greater than or equal to a preset threshold, the target feature set is not updated; If the stability estimation value is less than a preset threshold, the target feature set is subjected to dimensionality downsampling processing to obtain an updated target feature set.

6. The method according to claim 4, characterized in that The target feature set includes initial sub-features, and after extracting features from the collected data according to the target instruction set and the target feature processing module to obtain the target feature set, the method further includes: Obtaining an abnormal value of the initial sub-feature according to a preset dual-sensing abnormality trigger function; Obtaining a confidence value of the initial sub-feature according to a preset confidence dynamic suppression function; Acquire the initial sub-feature whose abnormal value is greater than a preset abnormal threshold and whose confidence value is greater than a preset confidence threshold, and obtain at least one target sub-feature; The target feature set is updated according to the target sub-feature to obtain the updated target feature set.

7. The method according to claim 4, characterized in that The target instruction set includes multiple target instructions, each of which includes a module number. The target processing module is obtained by selecting from the initial feature processing module set according to the target instruction set and a preset resource scheduling loss function, including: Selecting from the initial feature processing module set according to the module number to obtain a first feature processing module set; At least one target feature processing module is selected from the first feature processing module set according to the resource scheduling loss function.

8. A separate transmission equipment data processing system, characterized in that: Applied to a central server, the system includes: Acquisition module, used to obtain current task data and historical task data; A first selection module is configured to select at least one selected feature template from a preset feature template library according to the current task data and the historical task data according to a preset selection rule to obtain a target feature template; wherein each target feature template constitutes a target instruction set; A first sending module is configured to send the target instruction set to an edge device and receive a target feature set collected and processed by the edge device according to the target instruction set; A fusion module, configured to interpolate and fuse the target feature set and the historical task data according to a preset interpolation fusion reconstruction model to obtain a unified tensor; The prediction module is used to input the unified tensor into a preset fault prediction model to obtain prediction data.

9. A separate transmission equipment data processing system, characterized in that: Applied to edge devices, the system includes: A receiving module, configured to receive a target instruction set; wherein the target instruction set is generated by a central server based on current task data and historical task data; a second selection module, configured to select from the initial feature processing module set according to the target instruction set and a preset resource scheduling loss function to obtain at least one target feature processing module; The extraction module is used to extract features from the collected data according to the target instruction set and the target feature processing module to obtain a target feature set. The second sending module is used to send the target feature set to the central server.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the data processing method of a separate transmission device according to any one of claims 1 to 3 and the data processing method of a separate transmission device according to any one of claims 4 to 7 are implemented.

Images