An industrial Internet of Things data collection method based on edge-cloud collaboration

Abstract

The present invention provides an industrial Internet of Things data collection method based on edge-cloud collaboration, which belongs to the field of data analysis technology, including: S1, deploying multiple edge nodes at the industrial site to collect the original data of industrial equipment and perform real-time preprocessing; S2, the edge node adopts a hybrid algorithm combined with a deep learning model to further process the preprocessed data; S3, during the data processing process, the edge node monitors the operating status of the system through a real-time feedback mechanism, and dynamically adjusts the data collection frequency, data compression rate and processing strategy through a preset strategy; S4, the cloud performs a deep analysis of the received preprocessed data and constructs a multi-dimensional analysis model to generate analysis results and optimization strategies; S5, the cloud feeds back the analysis results and optimization strategies to the edge node, and guides the edge node to adjust the data collection and processing strategies. The present invention can improve data collection efficiency, ensure data security, and reduce cloud pressure.

Description

Technical Field

[0001] The present invention relates to the field of data analysis technology, and in particular to an industrial Internet of Things data collection method based on edge-cloud collaboration. Background Art

[0002] The Industrial Internet of Things (IIoT) has been widely used. A large number of devices and sensors in factory workshops are connected to the Internet, forming a huge network of machine interconnections. These smart devices and sensors continuously generate massive amounts of industrial data, including production data, status data, environmental data, etc. How to efficiently, safely and intelligently collect and process these industrial big data is crucial to realizing intelligent manufacturing and optimizing production processes. However, traditional industrial data collection methods have some significant defects:

[0003] 1) Inefficiency. Complete reliance on centralized cloud data collection and limited network bandwidth will lead to data congestion and increased latency;

[0004] 2) Security risks. A large amount of sensitive data can be easily attacked and stolen during network transmission;

[0005] 3) High cost. The need to deploy a large amount of cloud computing resources or deploy large-scale edge computing nodes on site will keep operating costs high.

[0006] To solve the above-mentioned defects, the present invention proposes an industrial Internet of Things data collection method based on edge-cloud collaboration. Summary of the invention

[0007] The purpose of the present invention is to provide an industrial Internet of Things data collection method based on edge-cloud collaboration, which greatly improves the intelligence and adaptability of data collection and processing. It is suitable for complex and changeable industrial Internet of Things environments, improves data collection efficiency, ensures data security, and reduces cloud pressure.

[0008] To achieve the above object, the present invention provides an industrial Internet of Things data collection method based on edge-cloud collaboration, comprising the following steps:

[0009] S1. Deploy multiple edge nodes at industrial sites to collect raw data from industrial equipment and perform real-time preprocessing of the collected raw data through adaptive Kalman filters;

[0010] S2, the edge node uses a hybrid algorithm combined with a deep learning model to further process the preprocessed data;

[0011] S3. During data processing, the edge node monitors the system's operating status through a real-time feedback mechanism and dynamically adjusts the data collection frequency, data compression rate, and processing strategy through preset strategies;

[0012] S4. The cloud performs in-depth analysis on the received pre-processed data and builds a multi-dimensional analysis model to generate analysis results and optimization strategies;

[0013] S5. The cloud feeds back the analysis results and optimization strategies to the edge nodes, guiding them to adjust data collection and processing strategies.

[0014] Preferably, in step S1, the original data includes: temperature sensor data, pressure sensor data, flow sensor data, vibration sensor data, current data, voltage data, rotation speed data, product detection data, production count data, alarm status data and control instruction data.

[0015] The adaptive Kalman filter improves the quality of data by dynamically adjusting the filter parameters by real-time monitoring of the statistical characteristics of the input data (such as mean, variance, etc.). Preprocessing includes: noise suppression, data interpolation, dynamic adjustment of filter parameters, outlier detection and correction, feature extraction, signal enhancement and data compression.

[0016] Preferably, in step S1, the adaptive Kalman filter is used to perform real-time estimation and noise suppression on the original data. It is a recursive algorithm including two steps of prediction and update:

[0017] First, the state prediction equation is described as:

[0018]

[0019] Among them, the state prediction equation is based on the current state and control input u k-1 , predict the state of the next moment k

[0020] The state prediction process depends on the state transfer matrix A and the control matrix B. At the same time, the prediction covariance P k|k-1 is also calculated, which represents the uncertainty of the prediction. The prediction formula for covariance is as follows:

[0021] P k|k-1 =AP k-1|k-1 A T +Q h (2)

[0022] Among them, Q h is the process noise covariance; P k-1|k-1 is the predicted covariance at the current time k; P k|k-1 is the predicted covariance at the next time step k;

[0023] When the new observation z k When it arrives, the Kalman filter updates the state estimate. First, the Kalman gain K is calculated.k , which determines the weight between the predicted value and the new observation value. The calculation formula of Kalman gain is:

[0024] K k =P k|k-1 H T (HP k|k-1 H T +R) -1 (3)

[0025] Where H is the observation matrix; R is the observation noise covariance;

[0026] The Kalman gain is then used to update the state estimate, combining the prediction with the new observation to get the corrected state:

[0027]

[0028] Among them, z k is the measured value; is the predicted measurement; It is the state obtained by correction at the next moment.

[0029] Preferably, in step S2, the processing process of the hybrid algorithm combined with the deep learning model is as follows:

[0030] The preprocessed data were input into convolutional neural networks and long short-term memory networks for feature extraction and time series analysis, respectively;

[0031] Among them, convolutional neural networks are used for feature extraction of image data and time series data:

[0032]

[0033] Among them, y i,j,l is the convolution output of the lth filter; i, j are image positions or coordinates; x i+m,j+n Represents the pixel value or feature map value of the input image at position (i+m,j+n). ,ω m,n,l is the weight value of the lth convolution kernel at position (m,n); b l is the bias term of the lth convolution kernel; M, N are the sizes of the convolution kernel; m, n represent the offset of the convolution kernel relative to the current center position, which is used to traverse all elements of the convolution kernel;

[0034] LSTM networks are able to capture dependencies in time series data:

[0035] f t =σ(W f ·[h t-1 ,x t ]+b f )(6)

[0036] Where: f t is the output of the forget gate; W f is the weight matrix of the forget gate; h t-1 is the hidden state at the previous moment; x t is the current input; b f is the bias term; σ is the Sigmoid activation function;

[0037] h t =o t tanh(C t )(7)

[0038] o t =σ(W o ·[h t-1 ,x t ]+b o )(8)

[0039] Among them, h t is the hidden state at the current moment; o t is the activation value of the output gate; W o is the weight matrix of the output gate; b o is the bias term; C t is the state of the memory unit at the current moment.

[0040] The data processed by the hybrid algorithm can be directly used for local decision-making at the edge node, such as detecting anomalies, triggering alarms, etc. At the same time, these processing results can also be further transmitted to the cloud.

[0041] Preferably, in step S3, the real-time feedback mechanism includes: the edge node continuously monitors the operating status of the system during data processing, including CPU utilization, memory usage, network bandwidth status, and data processing delay;

[0042] The preset strategies include:

[0043] 1) Data collection frequency: Based on real-time feedback, if the network bandwidth is limited or the processor is overloaded, the edge node can dynamically reduce the data collection frequency, thereby reducing the system load.

[0044] 2) Data compression rate: Before transmission, the data compression rate can be adjusted according to the network conditions. For example, when the network is congested, a higher compression rate can be selected to reduce the amount of data and ensure the smooth transmission of key data.

[0045] 3) Processing strategy: Edge nodes can select different processing strategies based on the system operation status, such as selecting a lighter model for inference when resources are limited, or selecting a lower resolution image for processing.

[0046] Q-learning is the core tool in step S3. The edge node continuously explores and learns the effects of different processing strategies in different environments through reinforcement learning algorithms. Through this learning, the edge node can make better decisions in similar situations in the future, improving the overall efficiency and robustness of the system.

[0047] Preferably, in step S3, the edge node dynamically adjusts the data collection frequency, data compression rate and processing strategy through Q-Learning. The update formula of Q-Learning is as follows:

[0048]

[0049] Among them, Q(s t ,a t ) is in state s t Next, perform action a t Q value; α is the learning rate; r t+1 Is to perform action a t The immediate reward obtained after γ is the discount factor; is the next state s t+1 The maximum Q value.

[0050] Preferably, in step S4, the multi-dimensional analysis model includes an equipment health status analysis model, a production quality analysis model, a process optimization model and a predictive maintenance model;

[0051] Among them, the equipment health status analysis model uses a random forest classifier:

[0052]

[0053] Where: f(x) is the classification result; N h is the number of decision trees; h i (x) is the classification result of the i-th decision tree;

[0054] The production quality analysis model uses linear regression:

[0055] y=β0+β1x1+β2x2+···+β n x n (11)

[0056] Among them, y is the predicted value of product quality; x1,x2,···,x n are process parameters; β0,β1,···,β n is the regression coefficient;

[0057] The process optimization model uses the fitness function of the genetic algorithm:

[0058]

[0059] Among them, F(x) is the fitness of individual x; f i (x) is the objective function; ω i is the weight; N d is the number of individuals;

[0060] The predictive maintenance model uses the remaining useful life prediction formula:

[0061] RUL(t)=Tt(13)

[0062] Where: RUL(t) is the remaining useful life of the equipment at time t; T is the expected life of the equipment (which can be obtained through historical data or statistical models); t is the current time.

[0063] Therefore, the present invention adopts the above-mentioned industrial Internet of Things data collection method based on edge-cloud collaboration, and the beneficial technical effects are as follows:

[0064] (1) Improve the efficiency of industrial data collection: Make full use of the local computing power of edge nodes to implement pre-processing such as data filtering and compression at the source of data, which greatly reduces the pressure on network transmission and the pressure on the cloud. Adopt a hierarchical data processing strategy, and delay-sensitive process data can be processed at the edge without long-distance transmission, thus reducing delay.

[0065] (2) Reduce the cost of industrial data processing: Adopt a flexible "edge + cloud" architecture, without the need to invest a lot of money to deploy large-scale edge nodes or cloud computing resources; efficient division of labor and collaboration between edge nodes, higher resource utilization, saving computing and storage costs; the cloud is responsible for high-level tasks, and edge nodes handle lightweight tasks, which makes task scheduling more reasonable and reduces overall operating costs;

[0066] (3) Realize intelligent industrial data processing: The edge nodes perform intelligent pre-processing on the raw data, extract data features, and optimize data quality. The cloud performs in-depth analysis and modeling on the data to discover the valuable knowledge and patterns hidden therein. Through a closed-loop feedback control mechanism, the edge nodes can dynamically adjust their strategies and continuously optimize the entire system.

[0067] (4) Good scalability and adaptability: The edge adopts a distributed and scalable architecture, which allows for flexible increase or decrease of the number of edge nodes according to demand; it is seamlessly integrated with the cloud and can adjust the cloud resource ratio as needed, with good scalability; this method is generally applicable to various industrial scenarios.

[0068] In summary, the method of the present invention comprehensively solves the key challenges of efficiency, security, cost, etc. faced by traditional industrial big data collection by innovatively integrating the synergistic advantages of edge computing and cloud computing, realizes intelligent and efficient industrial data collection and processing, and provides strong technical support for the intelligent transformation of industry. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 This is a flow chart of an industrial Internet of Things data collection method based on edge-cloud collaboration of the present invention. DETAILED DESCRIPTION

[0070] The technical solution of the present invention is further described below through the accompanying drawings and embodiments.

[0071] Unless otherwise defined, technical or scientific terms used in the present invention shall have the common meanings understood by one having ordinary skills in the field to which the present invention belongs.

[0072] Embodiment 1

[0073] like Figure 1 As shown, the present invention provides an industrial Internet of Things data collection method based on edge-cloud collaboration, which specifically includes the following steps:

[0074] S1. Deploy multiple edge nodes at industrial sites and collect raw data from industrial equipment.

[0075] The original data includes:

[0076] Temperature sensor data (temperature readings of a device or environment);

[0077] Pressure sensor data (pressure readings of liquids or gases);

[0078] Flow sensor data (readings of liquid or gas flow);

[0079] Vibration sensor data (vibration readings while the machine is running, used to monitor equipment health);

[0080] Current data / voltage data (current and voltage readings of motors or other electrical equipment);

[0081] Speed ​​data (speed readings of rotating equipment such as fans, bearings, etc.);

[0082] Product inspection data (obtaining product dimensions, defects, and other information through vision or other sensors);

[0083] Production count data (counting the number of products produced);

[0084] Alarm status data and control instruction data (alarm information of equipment failure or abnormal status);

[0085] Control instruction data (control instructions issued to actuators such as valves, motors, etc.).

[0086] The collected raw data is preprocessed in real time through adaptive filters, which can include the following aspects:

[0087] Noise Suppression:

[0088] Low-pass filtering: used to remove high-frequency noise, suitable for smoothing sensor data such as temperature and pressure, and reducing random fluctuations in the data.

[0089] Adaptive median filtering: It is particularly suitable for processing outliers introduced by transient faults. It dynamically adjusts the filter window size and eliminates abnormal data points based on the median of neighboring data.

[0090] Data interpolation:

[0091] Time series interpolation: During data collection, data may be lost due to brief sensor failures or communication delays. Adaptive filters can interpolate based on contextual data to fill in the missing data.

[0092] Spatial interpolation: In a sensor array, if a sensor fails, the missing data can be inferred and interpolated by using data from neighboring sensors.

[0093] Dynamically adjust filter parameters:

[0094] Real-time monitoring data characteristics: According to the real-time monitoring data characteristics (such as mean, variance, frequency distribution, etc.), the adaptive filter can dynamically adjust its filtering parameters. For example, in the case of increased signal noise, the filter will automatically increase its filtering strength and increase the smoothing effect of the filter.

[0095] Adaptive frequency filtering: When noise in different frequency bands is detected, the adaptive filter can selectively filter out unwanted frequency components and retain useful signals. For example, in vibration data processing, the adaptive filter can automatically adjust the filtering frequency range for different types of mechanical vibration signals.

[0096] Outlier detection and correction:

[0097] Anomaly Detection: By analyzing the statistical characteristics of the data (such as sudden jumps, unusual high / low values), adaptive filters can automatically identify and flag potential outliers.

[0098] Data correction: For detected outliers, the adaptive filter can make reasonable corrections based on the trends of adjacent data to make the data more continuous and reliable.

[0099] Feature extraction:

[0100] Adaptive extraction of signal features: In some scenarios, such as vibration data or acoustic signals, adaptive filters can be combined with signal processing algorithms to extract important frequency components or patterns in real time. This can provide higher quality input data for subsequent deep learning models.

[0101] Signal Enhancement:

[0102] Adaptive signal enhancement: For data that may have weak signals, adaptive filters can enhance the key characteristics of the signal, making the data clearer and easier to process and analyze later, for example, by amplifying specific frequency components to highlight important signals.

[0103] Data Compression:

[0104] Real-time compression: While performing filtering, the adaptive filter can compress data in real time, reducing the amount of data and reducing the burden on network transmission and storage.

[0105] Through these preprocessing steps, the adaptive filter can ensure that the raw data has been optimized and purified before entering the subsequent processing link, thereby improving the data processing efficiency and accuracy of the overall system.

[0106] The adaptive Kalman filter is used to estimate and suppress noise of raw data in real time. It is a recursive algorithm that includes two steps: prediction and update:

[0107] First, the state prediction equation is described as:

[0108]

[0109] Among them, the state prediction equation is based on the current state and control input u k-1 , predict the state of the next moment k

[0110] The state prediction process depends on the state transfer matrix A and the control matrix B. At the same time, the prediction covariance P k|k-1 is also calculated, which represents the uncertainty of the prediction. The prediction formula for covariance is as follows:

[0111] P k|k-1 =AP k-1|k-1 A T +Q h (2)

[0112] Among them, Q h is the process noise covariance; P k-1|k-1 is the predicted covariance at the current time k; P k|k-1 is the predicted covariance at the next time step k;

[0113] When the new observation z k When it arrives, the Kalman filter updates the state estimate. First, the Kalman gain K is calculated. k , which determines the weight between the predicted value and the new observation value. The calculation formula of Kalman gain is:

[0114] K k =P k|k-1 H T (HP k|k-1 H T +R) -1 (3)

[0115] Where H is the observation matrix; R is the observation noise covariance;

[0116] The Kalman gain is then used to update the state estimate, combining the prediction with the new observation to get the corrected state:

[0117]

[0118] Among them, z k is the measured value; is the predicted measurement; It is the state obtained by correction at the next moment.

[0119] S2, the edge node uses a hybrid algorithm combined with a deep learning model to further process the preprocessed data;

[0120] The preprocessed data is passed to the convolutional neural network (CNN) for feature extraction. For image data or time series data, CNN can automatically learn and extract useful features, which can be the pattern of equipment vibration, flaws in product defect images, etc. At the same time, the data is passed to the long short-term memory network (LSTM) for time series analysis. LSTM can capture the time dependency in the data, perform trend prediction, fault detection, etc., and is particularly suitable for processing sensor data with long-term dependencies. The data processed by the hybrid algorithm can be directly used for local decision-making at the edge node, such as detecting anomalies, triggering alarms, etc. At the same time, these processing results can also be further transmitted to the cloud.

[0121] Specifically, convolutional neural networks are used for feature extraction of image data and time series data:

[0122]

[0123] Among them, y i,j,l is the convolution output of the lth filter; i, j are image positions; x i+m,j+nRepresents the pixel value or feature map value of the input image at position (i+m,j+n); ω m,n,l is the weight value of the lth convolution kernel at position (m,n); b l is the bias term of the lth convolution kernel; M, N are the sizes of the convolution kernel; m, n represent the offset of the convolution kernel relative to the current center position, which is used to traverse all elements of the convolution kernel;

[0124] LSTM networks are able to capture dependencies in time series data:

[0125] f t =σ(W f ·[h t-1 ,x t ]+b f ) (6)

[0126] Where: f t is the output of the forget gate; W f is the weight matrix of the forget gate; h t-1 is the hidden state at the previous moment; x t is the current input; b f is the bias term; σ is the Sigmoid activation function;

[0127] h t =o t tanh(C t ) (7)

[0128] o t =σ(W o ·[h t-1 ,x t ]+b o ) (8)

[0129] Among them, h t is the hidden state at the current moment; o t is the activation value of the output gate; W o is the weight matrix of the output gate; b o is the bias term; C t is the state of the memory unit at the current moment.

[0130] S3. During data processing, the edge node monitors the system's operating status through a real-time feedback mechanism and dynamically adjusts the data collection frequency, data compression rate, and processing strategy through preset strategies;

[0131] The preset strategies include:

[0132] 1) Data collection frequency: Based on real-time feedback, when the network bandwidth is limited or the processor is overloaded, the edge node dynamically reduces the data collection frequency, thereby reducing the system load;

[0133] 2) Data compression rate: Before transmission, adjust the data compression rate according to the network conditions;

[0134] 3) Processing strategy: The edge node selects different processing strategies according to the system operation status.

[0135] The edge nodes dynamically adjust the data collection frequency, data compression rate and processing strategy through the Q-Learning algorithm;

[0136] The update formula of the Q-Learning algorithm is as follows:

[0137]

[0138] Among them, Q(s t ,a t ) is in state s t Next, perform action a t Q value; α is the learning rate; r t+1 Is to perform action a t The immediate reward obtained after γ is the discount factor; m a axQ(s t+1 ,a t ) is the next state s t+1 The maximum Q value.

[0139] S4. The cloud performs in-depth analysis on the received pre-processed data and builds a multi-dimensional analysis model to generate analysis results and optimization strategies;

[0140] The multi-dimensional analysis model includes equipment health status analysis model, production quality analysis model, process optimization model and predictive maintenance model.

[0141] Among them, the equipment health status analysis model uses a random forest classifier:

[0142]

[0143] Where: f(x) is the classification result; N h is the number of decision trees; h i (x) is the classification result of the i-th decision tree;

[0144] The production quality analysis model uses linear regression:

[0145] y=β0+β1x1+β2x2+···+β n x n (11)

[0146] Among them, y is the predicted value of product quality; x1,x2,···,x n are process parameters; β0,β1,···,βn is the regression coefficient;

[0147] The process optimization model uses the fitness function of the genetic algorithm:

[0148]

[0149] Among them, F(x) is the fitness of individual x; f i (x) is the objective function; ω i is the weight; N d is the number of individuals;

[0150] The predictive maintenance model uses the remaining useful life prediction formula:

[0151] RUL(t)=Tt(13)

[0152] Where: RUL(t) is the remaining useful life of the equipment at time t; T is the expected life of the equipment; and t is the current time.

[0153] It is worth noting that the contents not elaborated in detail in the present invention are all prior art and are well known to those skilled in the art.

[0154] Therefore, the present invention adopts the above-mentioned industrial Internet of Things data collection method based on edge-cloud collaboration, which greatly improves the intelligence and adaptability of data collection and processing. It is suitable for complex and changeable industrial Internet of Things environments, improves data collection efficiency, ensures data security, and reduces cloud pressure.

[0155] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. An industrial Internet of Things data collection method based on edge-cloud collaboration, characterized in that: The following steps are involved: S1. Deploy multiple edge nodes at industrial sites to collect raw data from industrial equipment and perform real-time preprocessing of the collected raw data through adaptive Kalman filters; S2, the edge node uses a hybrid algorithm combined with a deep learning model to further process the preprocessed data; S3. During data processing, the edge node monitors the system's operating status through a real-time feedback mechanism and dynamically adjusts the data collection frequency, data compression rate, and processing strategy through preset strategies; S4. The cloud performs in-depth analysis on the received pre-processed data and builds a multi-dimensional analysis model to generate analysis results and optimization strategies; S5. The cloud feeds back the analysis results and optimization strategies to the edge nodes, guiding them to adjust data collection and processing strategies. In step S1, the adaptive Kalman filter includes two steps: prediction and update: First, the state prediction equation is described as: Among them, the state prediction equation is based on the current state and control input u k-1 , predict the state of the next moment k The state prediction process depends on the state transfer matrix A and the control matrix B. At the same time, the prediction covariance P k|k-1 It will also be calculated, and the prediction formula of covariance is as follows: P k|k-1 =AP k-1|k-1 From T +Q h (2) Among them, Q h is the process noise covariance; P k-1|k-1 is the predicted covariance at the current time k; P k|k-1 is the predicted covariance at the next time step k; When the new observation z k When it arrives, the Kalman filter updates the state estimate. First, the Kalman gain K is calculated. k , the calculation formula is: K k =P k|k-1 H T (HP k|k-1 H T +R) -1 (3) Where H is the observation matrix; R is the observation noise covariance; The Kalman gain is then used to update the state estimate, combining the prediction with the new observation to get the corrected state: Among them, z k is the measured value; is the predicted measurement; It is the state obtained by correction at the next moment.

2. According to the method of industrial Internet of Things data collection based on edge-cloud collaboration according to claim 1, it is characterized in that: In step S1, the original data includes: temperature sensor data, pressure sensor data, flow sensor data, vibration sensor data, current data, voltage data, rotation speed data, product detection data, production count data, alarm status data and control instruction data.

3. According to the method of industrial Internet of Things data collection based on edge-cloud collaboration according to claim 2, it is characterized in that: In step S2, the processing process of the hybrid algorithm combined with the deep learning model is as follows: The preprocessed data were input into convolutional neural networks and long short-term memory networks for feature extraction and time series analysis, respectively; Among them, convolutional neural networks are used for feature extraction of image data and time series data: Among them, y i,j,l is the convolution output of the lth filter; i, j are image positions; x i+m,j+n Represents the pixel value or feature map value of the input image at position (i+m,j+n); ω m,n,l is the weight value of the lth convolution kernel at position (m,n); b l is the bias term of the lth convolution kernel; M, N are the sizes of the convolution kernel; m, n represent the offset of the convolution kernel relative to the current center position, which is used to traverse all elements of the convolution kernel; LSTM networks are able to capture dependencies in time series data: f t =σ(W f ·[h t-1 ,x t ]+b f ) (6) Where: f t is the output of the forget gate; W f is the weight matrix of the forget gate; h t-1 is the hidden state at the previous moment; x t is the current input; b f is the bias term; σ is the Sigmoid activation function; h t =o t ·tanh(C t ) (7) the t =σ(W o ·[h t-1 ,x t ]+b o ) (8) Among them, h t is the hidden state at the current moment; o t is the activation value of the output gate; W o is the weight matrix of the output gate; b o is the bias term; C t is the state of the memory unit at the current moment.

4. According to the method of industrial Internet of Things data collection based on edge-cloud collaboration according to claim 3, it is characterized in that: In step S3, the preset strategy includes: 1) Data collection frequency: Based on real-time feedback, when the network bandwidth is limited or the processor is overloaded, the edge node dynamically reduces the data collection frequency, thereby reducing the system load; 2) Data compression rate: Before transmission, adjust the data compression rate according to the network conditions; 3) Processing strategy: The edge node selects different processing strategies according to the system operation status.

5. According to the method of industrial Internet of Things data collection based on edge-cloud collaboration according to claim 4, it is characterized in that: In step S3, the edge node dynamically adjusts the data collection frequency, data compression rate and processing strategy through the Q-Learning algorithm; The update formula of the Q-Learning algorithm is as follows: Among them, Q(s t ,a t ) is in state s t Next, perform action a t Q value; α is the learning rate; r t+1 Is to perform action a t The immediate reward obtained after γ is the discount factor; is the next state s t+1 The maximum Q value.

6. According to the method of industrial Internet of Things data collection based on edge-cloud collaboration according to claim 5, it is characterized in that: In step S4, the multi-dimensional analysis model includes an equipment health status analysis model, a production quality analysis model, a process optimization model, and a predictive maintenance model; Among them, the equipment health status analysis model uses a random forest classifier: Where: f(x) is the classification result; N h is the number of decision trees; h i (x) is the classification result of the i-th decision tree; The production quality analysis model uses linear regression: y=β0+β1x1+β2x2+···+β n x n (11) Among them, y is the predicted value of product quality; x1,x2,···,x n are process parameters; β0,β1,···,β n is the regression coefficient; The process optimization model uses the fitness function of the genetic algorithm: Among them, F(x) is the fitness of individual x; f i (x) is the objective function; ω i is the weight; N d is the number of individuals; The predictive maintenance model uses the remaining useful life prediction formula: RUL(t)=Tt (13) Where: RUL(t) is the remaining useful life of the equipment at time t; T is the expected life of the equipment; and t is the current time.

Images