Internet-of-things data transmission method and system based on Internet-of-things middle station

Abstract

The invention provides an internet of things data transmission method and system based on an internet of things middle station, and relates to the technical field of communication, and the method comprises the steps: collecting multivariate perception data based on an internet of things perception layer which comprises a sensor network composed of a plurality of sensors; according to a coding time sequence matrix corresponding to the multivariate perception data, a data transmission period characteristic constructed by an internet-of-things middle station and a bidirectional long-short-term memory network model, obtaining an adaptive transmission period; and dynamically adjusting the data transmission time interval of the sensing layer of the Internet of Things according to the adaptive transmission period. According to the invention, the real-time monitoring capability and the response speed of the Internet of Things platform can be improved.

Description

Technical Field

[0001] The present invention relates to the field of communication technology, and in particular to an Internet of Things data transmission method and system based on an Internet of Things middle station. Background Art

[0002] With the rapid development of Internet of Things (IoT) technology, the widespread application of various smart devices has gradually penetrated into people's daily life and work. IoT devices are connected through wireless networks to achieve data exchange and remote control between devices, greatly improving production efficiency and convenience of life.

[0003] The Chinese patent with publication number CN118869742A discloses a method and a middle platform for urban Internet of Things data processing, which includes: collecting Internet of Things data from different devices and systems, and standardizing the Internet of Things data; the standardization processing includes data cleaning, conversion, mapping and integration; storing the Internet of Things data after standardization processing; based on the Internet of Things data after standardization processing, providing data services for various applications, and the data services include data query, data sharing and data visualization. However, the solution provided by the above application relies on centralized supply chain management, and the number of sensor node devices is huge and the types are diverse. Each sensor node device may contain different sensors, actuators, data processing units and communication modules. In such a complex distributed environment, it is easy to cause data transmission delays, network congestion and single point failures. Therefore, it is very necessary to provide an Internet of Things data transmission method and system based on the Internet of Things middle platform to improve storage query efficiency. Summary of the invention

[0004] In view of this, the present invention proposes an IoT data transmission method and system based on an IoT middle station. By utilizing a bidirectional long short-term memory network model combined with a coding time series matrix and transmission cycle characteristics, the appropriate adaptive transmission cycle can be accurately predicted, and the transmission interval can be dynamically adjusted so that the sensor network can quickly respond to environmental changes, which helps to improve the real-time monitoring capability and response speed of the IoT middle station.

[0005] The present invention provides an Internet of Things data transmission method based on an Internet of Things middle station, the method comprising:

[0006] Collecting multi-sensory data based on the Internet of Things perception layer, wherein the Internet of Things perception layer includes a sensor network composed of multiple sensors;

[0007] Obtain an adaptive transmission cycle according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the IoT platform, and the bidirectional long short-term memory network model;

[0008] The data transmission time interval of the Internet of Things perception layer is dynamically adjusted according to the adaptive transmission period.

[0009] On the basis of the above technical solution, preferably, the adaptive transmission cycle is obtained according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the IoT platform, and the bidirectional long short-term memory network model, specifically including:

[0010] Obtaining a coding time series matrix, inputting the coding time series matrix into a bidirectional long short-term memory network model, and obtaining multivariate time series features;

[0011] Based on the time period of two adjacent data transmission processes collected by the IoT center, a data transmission cycle feature is constructed, and the data transmission cycle feature and the multivariate time series feature are fused and input into the bidirectional long short-term memory network model to obtain the optimal adaptive transmission period.

[0012] On the basis of the above technical solution, preferably, the obtaining of the encoding time series matrix specifically includes:

[0013] Extracting a set of homogeneous IoT sensors in the IoT perception layer, and determining a sequence start position and an initial subsequence length corresponding to the homogeneous IoT sensor set, and constructing an initial coding time series matrix according to the sequence start position, the initial subsequence length and the perception time series matrix;

[0014] Iterating the initial encoded time series matrix, and updating the homogeneous IoT sensor set and removing processed sensor nodes in each iteration;

[0015] According to the positions of the remaining sensor nodes in the homogeneous IoT sensor set, multiple time-warped distance matrices corresponding to the sensor nodes are calculated, and a time-warped distance set is constructed based on the distance metric values ​​of the time-warped distance matrices;

[0016] Selecting a time series index having a minimum distance metric value from the time-warping distance set, and merging a time series segment corresponding to the time series index into the initial encoded time series matrix;

[0017] When the homogeneous IoT sensor set is an empty set, the iteration process ends and the encoded time series matrix is ​​output.

[0018] More preferably, after obtaining the encoding time series matrix, the method further includes:

[0019] The encoded time series matrix is ​​input into a multi-head convolutional neural network, a sequence feature vector corresponding to the sequence length of the convolution filter in the multi-head convolutional neural network is extracted, and the sequence feature vector is feature fused with the hidden state output vector in the bidirectional long short-term memory network model.

[0020] More preferably, the expression of the convolution head output function in the multi-head convolutional neural network is:

[0021]

[0022] Among them, c h,i represents the output of the i-th feature map of the h-th convolutional head, represents the output function of the i-th feature map of the h-th convolution head at the sequence position j, φ represents the moving step of the convolution operation, n1 represents the input time series length covered by the sliding window during each convolution operation, and n2 represents the time series length of the input convolution head. represents the attention weight of the i-th feature map of the h-th convolution head at the sequence position j, softmax() represents the normalization function, and v T represents the transpose of the attention score weight matrix, tanh() represents the hyperbolic tangent activation function, W represents the input transformation matrix, represents the feature dimension vector contained in each time step in the local time series segment starting from sequence position j and with a length of n1, U represents the state transformation matrix, and h j-1 represents the hidden state vector at sequence position j, b c represents the bias term corresponding to the attention weight, represents the gate value of the i-th feature map of the h-th convolution head at the sequence position j, σ represents the activation function, [;] represents the vector concatenation operation, W g represents the weight matrix corresponding to the gating layer, b g represents the bias term corresponding to the gating layer, It represents the residual term of the i-th feature map of the h-th convolution head at the sequence position j, and ⊙ represents the element-by-element multiplication operation.

[0023] More preferably, the output characteristic function expression of the bidirectional long short-term memory network is:

[0024]

[0025] E t =LN(S t +Res(x t ))

[0026] O t =MLP([E t ; F t ])

[0027] Among them, h t represents the hidden state output function of the bidirectional long short-term memory network at time t, [;] represents the vector concatenation operation, represents the hidden state output vector of the forward long short-term memory network at time t, represents the hidden state output vector of the backward long short-term memory network at time t, F t represents the joint feature function, fusion() represents the operator merging function, H represents the total number of convolution heads, I represents the total number of feature maps, S t represents the characteristic function after sensor information sharing, γ represents the information sharing weight parameter, ⊙ represents the element-by-element multiplication operation, represents the output function of the i-th feature map of the h-th convolutional head at sequence position j, E t represents the feature function after feature enhancement, LN() represents the layer normalization function, Res() represents the residual connection function, O t represents the output feature function of the bidirectional long short-term memory network, and MLP() represents a multi-layer perceptron function.

[0028] More preferably, the collecting of multi-sensory data based on the IoT perception layer further includes:

[0029] Dynamically collect multivariate perception data using the IoT perception layer, obtain the data upload time of each sensor in the IoT perception layer, receive the multivariate perception data through an edge server, and perform deduplication and time series optimization on the multivariate perception data to obtain a perception time series matrix;

[0030] Event recognition and classification are performed on the perception time series matrix, data aggregation and compression are performed according to the classification result, and the aggregated and compressed perception time series matrix is ​​uploaded to the cloud service layer for decompression and storage.

[0031] More preferably, dynamically adjusting the data transmission time interval of the Internet of Things perception layer according to the adaptive transmission period specifically includes:

[0032] An update instruction is sent to each sensor node through the Internet of Things middle station, and the update instruction includes the adaptive transmission cycle, so that each sensor node dynamically adjusts the data collection interval and transmission time interval according to the adaptive transmission cycle.

[0033] In a second aspect of the present application, an Internet of Things data transmission system based on an Internet of Things middle station is provided, wherein the Internet of Things data transmission system comprises a data acquisition module, a cycle acquisition module and a cycle adjustment module connected in sequence, wherein:

[0034] The data acquisition module is used to collect multi-sensory data based on the Internet of Things perception layer, wherein the Internet of Things perception layer includes a sensor network composed of multiple sensors;

[0035] The cycle acquisition module is used to obtain an adaptive transmission cycle according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the Internet of Things platform, and the bidirectional long short-term memory network model;

[0036] The period adjustment module is used to dynamically adjust the data transmission time interval of the Internet of Things perception layer according to the adaptive transmission period.

[0037] In a third aspect of the present application, an electronic device is provided, comprising a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory.

[0038] The IoT data transmission method and system based on the IoT middle platform provided by the present invention have the following beneficial effects compared with the prior art:

[0039] (1) Through the sensor network composed of various sensors, it is possible to comprehensively monitor and collect multi-dimensional perception data in the environment. The multi-dimensional data collected by the perception layer is constructed through the construction of the coded time series matrix, which realizes the orderly integration and standardized processing of different sensor data. The bidirectional long short-term memory network model combined with the coded time series matrix and the transmission cycle characteristics can accurately predict the appropriate adaptive transmission cycle, and dynamically adjust the data transmission time interval according to the adaptive transmission cycle, which can significantly reduce unnecessary data transmission, reduce network bandwidth occupancy, and save energy consumption of sensor nodes. At the same time, the dynamic adjustment of the transmission interval enables the sensor network to respond quickly to environmental changes, ensuring that high-frequency data is obtained at critical moments, which helps to improve the real-time monitoring capability and response speed of the IoT middle station.

[0040] (2) By extracting a set of homogeneous IoT sensors and determining the corresponding sequence starting position and initial subsequence length, the initial coded time series matrix is ​​constructed, which ensures the orderly integration of different sensor data, helps to reduce data redundancy, avoid data conflicts, and ensure the accuracy and consistency of the coding matrix. In the iterative process, by calculating the time-warped distance matrix corresponding to multiple sensor nodes and constructing a time-warped distance set based on the distance metric, the similarity between different time series can be measured more accurately, thereby improving the quality of the coding matrix. At the same time, the time series index with the smallest distance metric value is selected from the time-warped distance set, and the corresponding time series fragment is merged into the initial coded time series matrix, ensuring that the most similar and representative data fragments are integrated first, thereby improving the efficiency and effect of data fusion. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0042] Figure 1 A schematic diagram of a flow chart of an Internet of Things data transmission method based on an Internet of Things middle station provided by the present invention;

[0043] Figure 2 A schematic diagram of the framework of the sensor node, the sending node and the aggregation node provided by the present invention;

[0044] Figure 3 A schematic diagram of the framework of the Internet of Things data transmission system provided by the present invention;

[0045] Figure 4 This is a schematic structural diagram of an electronic device provided by the present invention.

[0046] Explanation of the accompanying drawings: 1. Internet of Things data transmission system; 11. Data acquisition module; 12. Period acquisition module; 13. Period adjustment module; 2. Electronic device; 21. Processor; 22. Communication bus; 23. User interface; 24. Network interface; 25. Memory. DETAILED DESCRIPTION

[0047] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0048] The present invention discloses an Internet of Things data transmission method based on an Internet of Things middle station, referring to Figure 1 The method includes steps S1 to S3.

[0049] Step S1, collecting multi-sensory data based on the Internet of Things perception layer, wherein the Internet of Things perception layer includes a sensor network composed of multiple sensors.

[0050] In this embodiment, the IoT perception layer includes various types of sensors, such as temperature sensors, humidity sensors, light sensors, pressure sensors, gas sensors, motion sensors, etc. These sensors are deployed at different locations according to the requirements of specific application scenarios to comprehensively monitor environmental parameters. Sensor nodes are distributed in the monitoring area according to a predetermined network topology (such as star, mesh, tree, etc.) to ensure the coverage and redundancy of data collection and improve the robustness and reliability of the system.

[0051] Furthermore, each sensor sets a reasonable sampling frequency according to its monitoring object and application requirements. For example, temperature and humidity sensors can be sampled once a minute, while motion sensors may require a higher sampling frequency to capture rapidly changing motion data. The data collected by each sensor includes continuous analog signals and discrete event data. The data format is unified as a digital signal, and a standardized data format (such as JSON, XML, Protobuf, etc.) is used to facilitate subsequent processing and transmission. The collected raw data is initially filtered and denoised to ensure the accuracy and reliability of the data. For example, a moving average filter is used to eliminate the interference of environmental noise on the temperature sensor data. The sensor is calibrated to correct the systematic error and random error of the sensor to ensure the accuracy of the data. For example, a gas sensor is calibrated regularly to maintain its measurement accuracy.

[0052] Ensure the time synchronization of all sensor nodes through the Network Time Protocol (NTP) or a dedicated time synchronization protocol (such as IEEE 1588 Precision Time Protocol) to avoid data inconsistency caused by clock deviation. Add an accurate timestamp to each collected data to record the data collection time, supporting subsequent data association and time series analysis. For example, each temperature data has a timestamp accurate to milliseconds so that it can be accurately located in the time series matrix.

[0053] Low-power wireless communication protocols (such as LoRaWAN, Zigbee, Bluetooth Low Energy, etc.) are used to achieve wireless data transmission between sensor nodes and edge servers to ensure the stability and low energy consumption of data transmission. In some scenarios with high bandwidth or high reliability requirements, wired communication protocols (such as Ethernet, RS-485, etc.) can be used to connect sensor nodes and gateway devices. Sensor nodes regularly transmit the collected data in batches to edge servers to reduce the number of transmissions and save bandwidth. Critical event data (such as abnormally high temperature, dangerous gas leakage, etc.) is immediately sent to the edge server through a real-time transmission mechanism to ensure timely response.

[0054] This step also includes S11 to S12.

[0055] Step S11, use the Internet of Things perception layer to dynamically collect multi-sensory data, and obtain the upload data time of each sensor in the Internet of Things perception layer, receive the multi-sensory data through the edge server, and deduplicate and optimize the time series of the multi-sensory data to obtain the perception time series matrix.

[0056] In this step, a message queue system (such as MQTT Broker, Kafka, etc.) is used to receive and buffer data from sensors, and the received data is temporarily stored in the cache of the edge server (such as memory cache, SSD cache). Use the timestamp information of the data to detect and delete duplicate or nearly duplicate data. For example, if a sensor generates multiple identical data at the same timestamp, keep one and delete the remaining duplicate data. Identify and eliminate duplicate data with exactly the same or highly similar content to ensure the uniqueness and accuracy of the data set.

[0057] Furthermore, the deduplicated multivariate sensory data is organized in chronological order to form structured time series data. The data of each sensor is used as an independent time series dimension and aligned by time. Interpolation algorithms (such as linear interpolation, spline interpolation, etc.) or prediction models (such as moving average, ARIMA, etc.) are used to fill in the missing data in the time series. The data of different sensors are aligned to a unified time base according to the timestamp to form a unified time interval. For example, the data of all sensors are aligned to a time interval of once per minute. The time series data is normalized (such as Min-Max normalization) or standardized (such as Z-score standardization) to eliminate the dimensional differences of different sensor data. Event recognition and classification are performed on the perception time series matrix, data aggregation and compression are performed according to the classification results, and the aggregated and compressed perception time series matrix is ​​uploaded to the cloud service layer for decompression and storage. The optimized time series data is organized into a two-dimensional matrix, where the rows represent time steps and the columns represent the data of different sensors.

[0058] In one example, take the smart agriculture system as an example: the sensor types include soil moisture sensor, temperature sensor, light sensor and gas sensor. The sampling frequency is once a minute for soil moisture and temperature, once every 10 seconds for light sensor, and real-time sampling for gas sensor when abnormal gas concentration is detected. All sensors synchronize time with edge server through NTP, and sensor data is in JSON format. Each data contains sensor ID, data type, value and timestamp. The edge server runs MQTT Broker, and all sensors publish data to the specified topic through MQTT protocol, and the edge server subscribes and receives data. The received data is first stored in Redis cache as temporary storage to improve data processing speed. Using Redis's deduplication function, based on sensor ID and timestamp, ensure that there is only one unique data for each time step. Align soil moisture and temperature data to once a minute, and light data to once every 10 seconds, and fill in missing data through interpolation method. And perform Min-Max normalization on the data of each sensor to ensure that the data is in the same range, which improves the subsequent model processing effect. The soil moisture, temperature, light and gas concentration data collected once a minute are organized into a 4-column time series matrix, where each row represents the data of a time step. The constructed perception time series matrix is ​​stored in the SSD of the edge server and is ready to be uploaded to the cloud service layer for further processing.

[0059] Step S2, obtain the adaptive transmission cycle according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the Internet of Things center, and the bidirectional long short-term memory network model.

[0060] This step also includes S21 to S22.

[0061] Step S21, obtaining a coding time series matrix, inputting the coding time series matrix into a bidirectional long short-term memory network model, and obtaining multivariate time series features.

[0062] Furthermore, step S21 also includes S211 to S215.

[0063] Step S211, extracting a set of homogeneous IoT sensors in the IoT perception layer, and determining the sequence starting position and initial subsequence length corresponding to the homogeneous IoT sensor set, and constructing an initial coding time series matrix according to the sequence starting position, the initial subsequence length and the perception time series matrix.

[0064] In this step, the sensor features used to determine the isomorphism are determined, and the above sensor features include at least sensor type (such as temperature, humidity, pressure), sampling rate, data accuracy, and communication protocol. A feature vector is generated for each sensor, and the similarity between sensors is calculated using an appropriate similarity or distance measurement method (such as cosine similarity, Euclidean distance). A clustering algorithm (such as K-means, hierarchical clustering) or a threshold-based classification method is applied to divide sensors with high similarity into the same isomorphic set to obtain multiple isomorphic sensor sets, and the sensors in each set have high similarity in the defined feature dimension.

[0065] Select the time point with the earliest timestamp among all homogeneous sensor data as the starting position of the sequence to ensure data synchronization. If there is data missing or anomaly, select the first time point with complete data record as the starting position. According to the periodic characteristics of sensor data (such as daily cycle, weekly cycle), select a subsequence length that can cover a complete cycle to capture data trends and patterns. In the absence of obvious periodicity, a fixed sliding window length can be set, such as 30 minutes, 1 hour, etc. The specific length is determined by application requirements and data characteristics.

[0066] According to the determined sequence starting position and initial subsequence length, the corresponding data segment is extracted from the sensing time series matrix. This involves intercepting the data of the specified time period from the data stream of each homogeneous sensor.

[0067] Process missing values ​​and outliers for the above data to ensure data integrity and consistency. Ensure that the data of all sensors are aligned on the time axis to avoid data inconsistency caused by time offset. Arrange the extracted and preprocessed data according to the sensor set and time series to form a new matrix. The rows of the matrix usually represent different homogeneous sensors, and the columns represent the time steps. Depending on specific needs, the data may need to be encoded, such as normalization, standardization, or feature transformation.

[0068] In one example, when storing the signals collected by the sensor, not only the values ​​of non-zero elements but also the corresponding location information of the sensor should be stored. In the local sampling stage, each sensor node samples the signal and determines whether it is a sending node or a resting node in this period with a certain probability. In the data upload stage, the remaining nodes do not send data. The sending nodes send the sampled data to the collection point in turn. The collection point collects all the data of the sending nodes, amplifies them and forwards them to the cloud service layer. The cloud service layer reconstructs the signal transmission process in the observation area through the optimization algorithm.

[0069] like Figure 2As shown in Figure 1, assume that there are M nodes in the IoT perception layer, and select m nodes from the IoT perception layer. Whenever a new node is set, the new node is connected to the selected m nodes. The number of IoT perception layer nodes generated after the tth moment is t+M, and the number of edges in the IoT perception layer is mt.

[0070] The clustering coefficient of the above IoT perception layer can be expressed as:

[0071]

[0072] Step S212, iterating the initial coded time series matrix, and updating the homogeneous IoT sensor set and removing the processed sensor nodes in each iteration.

[0073] In this step, the initial coding time series matrix constructed in step S211 is used as the starting point of iteration. The homogeneous sensor set initially includes all sensor nodes that are identified as homogeneous. The processed sensor node list is empty at the beginning and is used to record the sensor nodes that have been processed during the iteration process.

[0074] Select a sensor node from the current homogeneous sensor set. The selection strategy can be based on specific requirements, such as polling selection, high priority sensors first, etc. Integrate the processed data segments into the initial encoded time series matrix. This step ensures that the matrix structure remains consistent and reflects the latest processing results. Remove the currently processed sensor node from the homogeneous sensor set and add it to the list of processed sensor nodes. This step avoids the same sensor being processed repeatedly in subsequent iterations. Re-evaluate and update the homogeneous sensor set based on the latest processing status, including re-matching features, recalculating sensor similarity, or adjusting homogeneous relationships to ensure the accuracy and relevance of the homogeneous sensor set.

[0075] Step S213, according to the positions of the remaining sensor nodes in the homogeneous IoT sensor set, calculate multiple time-warped distance matrices corresponding to the sensor nodes, and construct a time-warped distance set based on the distance metric values ​​of the time-warped distance matrices.

[0076] In this step, the remaining sensor nodes and their corresponding positions are extracted from the homogeneous sensor set, and the time series data of each sensor node in the perception time series matrix is ​​obtained to determine the comparison sensor pairs. According to the requirements, the sensor pairs that need to calculate the distance are selected. There are usually two ways (pairwise comparison to calculate the distance between each pair of sensors in the set or central sensor comparison, that is, selecting one or more central sensors and calculating the distance with other sensors one by one).

[0077] For each pair of selected sensor nodes, the time series data of the two sensor nodes are obtained, and the DTW distance between the two time series is calculated, allowing nonlinear alignment of the time axis to capture similar patterns, and the calculated DTW distance is stored in the corresponding position matrix. For each pair of sensor nodes, a distance value is generated, and finally a distance matrix is ​​formed. Each element (p, q) in the matrix represents the DTW distance between sensor node p and sensor node q.

[0078] Extract distance metrics from the generated multiple time-warped distance matrices and aggregate the distance values ​​in all matrices into a unified set. Calculate the statistical characteristics of the distance values, such as average distance, minimum distance, maximum distance, etc. According to the specific application requirements, select meaningful distance metrics to enter the time-warped distance set, such as the minimum distance is used to identify the most similar sensor pairs; threshold screening is to select sensor pairs with distance values ​​below a certain threshold, indicating that they are highly similar; ranking selection is to select the top N minimum distance values. The filtered key distance values ​​form the final time-warped distance set.

[0079] Step S214, selecting a time series index with a minimum distance metric value from the time warping distance set, and merging the time series segment corresponding to the time series index into the initial encoded time series matrix.

[0080] In this step, all distance metrics stored in the time-warped distance set are accessed. Each distance metric corresponds to a specific time series index, indicating the similarity between the sequence and the current encoding matrix. The time series index corresponding to the minimum distance metric is recorded, the most similar time series segment is identified, and the corresponding time series segment is extracted from the perception time series matrix according to the selected time series index. The extracted time series segment is added to the current initial encoding time series matrix. If there is relevant data in the encoding matrix, ensure that the merging of the new segment does not introduce data conflicts or redundancy. The processed sensor nodes are removed from the homogeneous IoT sensor set to avoid repeated processing in subsequent iterations. Prepare for the next round of iterations and continue to select the time series segment with the next minimum distance metric value for merging.

[0081] Step S215, when the set of homogeneous IoT sensors is an empty set, the iteration process ends and the encoded time series matrix is ​​output.

[0082] In this step, by extracting the homogeneous IoT sensor set and determining the corresponding sequence starting position and initial subsequence length, the initial coding time series matrix is ​​constructed to ensure the orderly integration of different sensor data. This helps to reduce data redundancy, avoid data conflicts, and ensure the accuracy and consistency of the coding matrix. During the iteration process, the homogeneous IoT sensor set is continuously updated and the processed sensor nodes are removed to ensure that each iteration can efficiently process the remaining sensor data. This dynamic update mechanism improves the operation efficiency of the algorithm, reduces unnecessary computational overhead, and improves the overall processing speed. By calculating the time-warped distance matrix corresponding to multiple sensor nodes and constructing a time-warped distance set based on the distance metric, the similarity between different time series can be measured more accurately. This precise similarity analysis helps the effective fusion of subsequent data and improves the quality of the coding matrix. The time series index with the smallest distance metric value is selected from the time-warped distance set, and the corresponding time series segment is merged into the initial coding time series matrix, ensuring that the most similar and representative data segments are integrated first. This process improves the efficiency and effect of data fusion and makes the coding matrix more representative and practical. When the set of homogeneous IoT sensors is an empty set, the iteration process ends automatically and the coded time series matrix is ​​output, which avoids unnecessary additional calculations and improves the automation and operation efficiency of the method. At the same time, the output coded time series matrix is ​​efficient and complete, which is suitable for subsequent data analysis and application.

[0083] Step S22, based on the time period of two adjacent data transmission processes collected by the IoT center, construct the data transmission cycle characteristics, fuse the data transmission cycle characteristics and the multivariate time series characteristics, and input them into the bidirectional long short-term memory network model to obtain the optimal adaptive transmission period.

[0084] In this step, the timestamp of each transmission is extracted from the data transmission record, and the time difference between two adjacent transmissions is calculated as the transmission cycle feature. The constructed data transmission cycle feature is merged with the extracted multivariate time series feature to form a comprehensive feature vector. The fusion method can use simple feature concatenation or more complex feature fusion methods, such as weighted fusion, principal component analysis (PCA) dimensionality reduction, etc.

[0085] The data set is divided into training set, validation set and test set according to a certain ratio. Usually the ratio is 70% training set, 15% validation set and 15% test set. And according to the continuity of the time series, the sliding window method is used to slice the long sequence into multiple subsequences to meet the input requirements of the bidirectional long short-term memory network model.

[0086] The bidirectional long short-term memory network model includes an input layer, a bidirectional LSTM layer, a fully connected layer and an output layer. The input layer is used to receive the fused feature vector with a shape of (time step, feature dimension). The bidirectional LSTM layer includes two parts, the forward LSTM and the backward LSTM, which are used to capture the context information of the sequence. The fully connected layer maps the output of the LSTM layer to the target variable space, that is, the predicted transmission period. The output layer is used to output the predicted transmission period value.

[0087] Furthermore, the encoded time series matrix is ​​input into a multi-head convolutional neural network, the sequence feature vector corresponding to the sequence length of the convolution filter in the multi-head convolutional neural network is extracted, and the sequence feature vector is feature fused with the hidden state output vector in the bidirectional long short-term memory network model.

[0088] In this embodiment, since multi-head convolution allows the model to use different convolution filters on different "heads", features are extracted from multiple angles to enhance the expressiveness of the model. The encoded time series matrix is ​​received, and the shape is (number of samples, time step, number of features, number of channels). Usually the number of channels is 1, representing a single time series feature. The convolution filter has multiple convolution kernels (filters), each filter has a different sequence length (convolution kernel size), and the output feature map of the multi-head convolution is flattened into a one-dimensional vector as a subsequent feature vector.

[0089] In this step, the expression of the convolution head output function in the multi-head convolutional neural network is:

[0090]

[0091] Among them, c h,i represents the output of the i-th feature map of the h-th convolutional head, represents the output function of the i-th feature map of the h-th convolution head at the sequence position j, φ represents the moving step of the convolution operation, n1 represents the input time series length covered by the sliding window during each convolution operation, and n2 represents the time series length of the input convolution head. represents the attention weight of the i-th feature map of the h-th convolution head at the sequence position j, softmax() represents the normalization function, and v T represents the transpose of the attention score weight matrix, tanh() represents the hyperbolic tangent activation function, W represents the input transformation matrix, represents the feature dimension vector contained in each time step in the local time series segment starting from sequence position j and with a length of n1, U represents the state transformation matrix, and h j-1 represents the hidden state vector at sequence position j, b c represents the bias term corresponding to the attention weight, represents the gate value of the i-th feature map of the h-th convolution head at the sequence position j, σ represents the activation function, [;] represents the vector concatenation operation, W g represents the weight matrix corresponding to the gating layer, b g represents the bias term corresponding to the gating layer, It represents the residual term of the i-th feature map of the h-th convolution head at the sequence position j, and ⊙ represents the element-by-element multiplication operation.

[0092] Furthermore, the feature extraction function of the bidirectional long short-term memory network is expressed as:

[0093]

[0094] in, Represents the hidden state output vector of the forward long short-term memory network at time t, LSTM f represents the forward LSTM unit, Represents the hidden state output vector of the backward long short-term memory network at time t, LSTM f Represents a backward LSTM unit.

[0095]

[0096] E t =LN(S t +Res(x t ))

[0097] The output characteristic function expression of the bidirectional long short-term memory network is:

[0098] O t =MLP([E t ; F t ])

[0099] Among them, h t represents the hidden state output function of the bidirectional long short-term memory network at time t, [;] represents the vector concatenation operation, represents the hidden state output vector of the forward long short-term memory network at time t, represents the hidden state output vector of the backward long short-term memory network at time t, F t represents the joint feature function, fusion() represents the operator merging function, H represents the total number of convolution heads, I represents the total number of feature maps, S t represents the characteristic function after sensor information sharing, γ represents the information sharing weight parameter, ⊙ represents the element-by-element multiplication operation, represents the output function of the i-th feature map of the h-th convolutional head at sequence position j, E t represents the feature function after feature enhancement, LN() represents the layer normalization function, Res() represents the residual connection function, Ot It represents the output feature function of the bidirectional long short-term memory network, and MLP() represents the multi-layer perceptron function.

[0100] Step S3, dynamically adjust the data transmission time interval of the Internet of Things perception layer according to the adaptive transmission cycle.

[0101] In this embodiment, an update instruction is sent to each sensor node through the IoT middle station, and the update instruction includes an adaptive transmission cycle, so that each sensor node dynamically adjusts the data collection interval and transmission time interval according to the adaptive transmission cycle.

[0102] In one example, a smart agriculture system uses Internet of Things (IoT) technology to monitor and manage farmland environments. The smart agriculture system includes an IoT middle station and multiple sensor nodes distributed in different farmland areas. These sensors are used to monitor key parameters such as soil moisture, temperature, and light intensity in real time. According to environmental changes and data analysis results, the IoT middle station needs to dynamically adjust the data collection and transmission intervals of each sensor node to optimize resource usage and improve the real-time nature of data.

[0103] The instructions use standardized data formats, such as JSON or XML, to ensure compatibility and parsability between the middle station and the sensor nodes. The core content of the instructions includes the adaptive transmission cycle, that is, the time interval between new data collection and transmission. In addition, other configuration parameters such as data collection accuracy and transmission priority can be included as needed. For example, an update instruction may contain a command type (such as "UPDATE_TRANSMISSION_INTERVAL") and a payload part, in which the new transmission interval time is specified (for example, 600 seconds, i.e. 10 minutes).

[0104] Furthermore, MQTT (Message Queuing Telemetry Transport Protocol) is selected as the communication protocol because its lightweight and publish / subscribe mode are very suitable for resource-constrained sensor nodes. MQTT supports topic filtering, which facilitates the selective sending of instructions. In addition, other optional protocols include CoAP (Constrained Application Protocol) and HTTP / HTTPS, but MQTT is usually a better choice depending on the resources and network environment of the sensor node.

[0105] The IoT middle platform generates new adaptive transmission cycle parameters based on the collected data and the preset adaptive algorithm (such as prediction using a bidirectional long short-term memory network model). The middle platform encapsulates these parameters in a predefined instruction format and sends them to the target sensor node through the selected communication protocol (such as MQTT). For example, if the system detects that the soil moisture in a certain area does not change much, the middle platform may decide to adjust the transmission interval from every 5 minutes to every 10 minutes to reduce the frequency of data transmission and save bandwidth and power consumption.

[0106] Authentication and authorization mechanisms (such as username / password or TLS certificate) are used to ensure that only authorized middle stations can send instructions to sensor nodes. At the same time, data encryption protocols (such as TLS / SSL) are used to protect data in transit and prevent instructions from being intercepted or tampered with by unauthorized parties. Sensor nodes have the ability to receive and process update instructions. Specifically, the sensor node will continue to listen to the instruction topic from the middle station. Once an update instruction is received, it will parse the instruction content, extract the new transmission interval parameters, and update the internal configuration to adjust the time interval for data collection and transmission. For example, when a sensor node receives an instruction to update the transmission interval to 600 seconds, it will immediately adjust its internal timer to reduce the frequency of data collection and transmission to once every 10 minutes.

[0107] After successfully receiving and applying the update command, the sensor node will send a confirmation message to the middle station, indicating that the command has been processed correctly. If the command processing fails, the node will also send a failure confirmation message with the corresponding error information. For example, after successfully updating the transmission interval, the sensor node will send a confirmation message containing the "success" status and the new transmission interval time to the middle station. If an error occurs during the update process, the node will send a confirmation message containing the "failure" status and error details.

[0108] Furthermore, the IoT middle station will record the time, target node, and confirmation status of each instruction sent. If a transmission failure or no confirmation is received, the middle station can resend the instruction or mark the problem node for subsequent processing. Sensor nodes should also record logs of receiving and executing instructions, including instruction content, reception time, and execution status. Based on the feedback from sensor nodes and system operation, the IoT middle station can dynamically adjust the transmission cycle strategy. For example, if the middle station receives frequent feedback from sensor nodes in a certain area that transmission failures occur, it may adjust the transmission interval or optimize the network configuration. In addition, the middle station can regularly evaluate the effectiveness of the adaptive transmission cycle, and continuously optimize the transmission strategy and algorithm model based on actual data and application requirements to achieve more efficient resource utilization and more real-time data monitoring.

[0109] By dynamically adjusting the data transmission interval according to actual needs, unnecessary frequent transmission can be avoided, the network bandwidth occupation can be effectively reduced, and the overall network efficiency can be improved. The adaptive transmission cycle can be adjusted in real time according to the network status to prevent data congestion during the peak period of network load and ensure the smoothness and stability of data transmission. Dynamic adjustment of the transmission interval means that the sensor device can enter a low-power state when it does not need to transmit data frequently, significantly reducing battery consumption and extending the service life of the device. Through synchronization control and priority mechanisms, the timely transmission of important data is ensured, data loss caused by data conflicts or network fluctuations is reduced, and the reliability of data transmission is improved. The time series matrix of multivariate perception data is optimized to improve the structured degree of data, facilitate subsequent event identification and classification, and improve the accuracy of data processing. Using the bidirectional long short-term memory network (Bi-LSTM) model, the optimal transmission cycle is predicted based on historical data and real-time network conditions, so that the system can adapt autonomously to environmental changes and demand changes. The support of multiple sensor types and the adaptive adjustment mechanism make the system highly flexible and able to adapt to changes in different application scenarios and needs.

[0110] Based on the above method, an IoT data transmission system based on IoT middle platform is proposed. Figure 3 The Internet of Things data transmission system 1 includes a data acquisition module 11, a cycle acquisition module 12 and a cycle adjustment module 13, wherein:

[0111] The data acquisition module 11 is used to collect multi-sensory data based on the Internet of Things perception layer, wherein the Internet of Things perception layer includes a sensor network composed of multiple sensors;

[0112] The cycle acquisition module 12 is used to obtain the adaptive transmission cycle according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the Internet of Things middle station, and the bidirectional long short-term memory network model;

[0113] The cycle adjustment module 13 is used to dynamically adjust the data transmission time interval of the Internet of Things perception layer according to the adaptive transmission cycle.

[0114] In one example, the period acquisition module 12 is used to obtain the encoding time series matrix, input the encoding time series matrix into the bidirectional long short-term memory network model, and obtain multivariate time series features; based on the time periods of two adjacent data transmission processes collected by the Internet of Things center, the data transmission period features are constructed, and the data transmission period features and the multivariate time series features are fused and input into the bidirectional long short-term memory network model to obtain the optimal adaptive transmission period.

[0115] In one example, the period acquisition module 12 is used to extract a set of homogeneous IoT sensors in the IoT perception layer, and determine the sequence starting position and initial subsequence length corresponding to the homogeneous IoT sensor set, and construct an initial coding time series matrix according to the sequence starting position, the initial subsequence length and the perception time series matrix; iterate the initial coding time series matrix, and update the homogeneous IoT sensor set and remove the processed sensor nodes in each iteration; calculate multiple time-warped distance matrices corresponding to the sensor nodes according to the positions of the remaining sensor nodes in the homogeneous IoT sensor set, and construct a time-warped distance set based on the distance metric value of the time-warped distance matrix; select the time series index with the smallest distance metric value from the time-warped distance set, and merge the time series segment corresponding to the time series index into the initial coding time series matrix; when the homogeneous IoT sensor set is an empty set, end the iteration process and output the coding time series matrix.

[0116] In one example, the period acquisition module 12 is used to input the encoded time series matrix into a multi-head convolutional neural network, extract the sequence feature vector corresponding to the length of the convolution filter sequence in the multi-head convolutional neural network, and perform feature fusion of the sequence feature vector with the hidden state output vector in the bidirectional long short-term memory network model.

[0117] In one example, the expression of the convolution head output function in a multi-head convolutional neural network is:

[0118]

[0119] Among them, c h,i represents the output of the i-th feature map of the h-th convolutional head, represents the output function of the i-th feature map of the h-th convolution head at the sequence position j, φ represents the moving step of the convolution operation, n1 represents the input time series length covered by the sliding window during each convolution operation, and n2 represents the time series length of the input convolution head. represents the attention weight of the i-th feature map of the h-th convolution head at the sequence position j, softmax() represents the normalization function, and v T represents the transpose of the attention score weight matrix, tanh() represents the hyperbolic tangent activation function, W represents the input transformation matrix, represents the feature dimension vector contained in each time step in the local time series segment starting from sequence position j and with a length of n1, U represents the state transformation matrix, and h j-1 represents the hidden state vector at sequence position j, b c represents the bias term corresponding to the attention weight, represents the gate value of the i-th feature map of the h-th convolution head at the sequence position j, σ represents the activation function, [;] represents the vector concatenation operation, W g represents the weight matrix corresponding to the gating layer, b g represents the bias term corresponding to the gating layer, It represents the residual term of the i-th feature map of the h-th convolution head at the sequence position j, and ⊙ represents the element-by-element multiplication operation.

[0120] In an example, the output feature function expression of the bidirectional long short-term memory network is:

[0121]

[0122] E t =LN(S t +Res(x t ))

[0123] O t =MLP([E t ; F t ])

[0124] Among them, h t represents the hidden state output function of the bidirectional long short-term memory network at time t, [;] represents the vector concatenation operation, represents the hidden state output vector of the forward long short-term memory network at time t, represents the hidden state output vector of the backward long short-term memory network at time t, F t represents the joint feature function, fusion() represents the operator merging function, H represents the total number of convolution heads, I represents the total number of feature maps, S t represents the characteristic function after sensor information sharing, γ represents the information sharing weight parameter, ⊙ represents the element-by-element multiplication operation, represents the output function of the i-th feature map of the h-th convolutional head at sequence position j, E t represents the feature function after feature enhancement, LN() represents the layer normalization function, Res() represents the residual connection function, O t It represents the output feature function of the bidirectional long short-term memory network, and MLP() represents the multi-layer perceptron function.

[0125] In one example, the data acquisition module 11 is used to dynamically collect multi-sensory data using the Internet of Things perception layer, and obtain the data upload time of each sensor in the Internet of Things perception layer, receive the multi-sensory data through the edge server, and deduplicate and optimize the time series of the multi-sensory data to obtain a perception time series matrix; perform event recognition and classification on the perception time series matrix, perform data aggregation and compression according to the classification results, and upload the aggregated and compressed perception time series matrix to the cloud service layer for decompression and storage.

[0126] In one example, the cycle adjustment module 13 is used to send an update instruction to each sensor node through the IoT platform, and the update instruction includes an adaptive transmission cycle, so that each sensor node dynamically adjusts the data collection interval and transmission time interval according to the adaptive transmission cycle.

[0127] See also Figure 4 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 4 As shown, the electronic device 2 may include: at least one processor 21 , at least one network interface 24 , a user interface 23 , a memory 25 , and at least one communication bus 22 .

[0128] The communication bus 22 is used to realize the connection and communication between these components.

[0129] The user interface 23 may include a display screen (Display) and a camera (Camera), and the optional user interface 23 may also include a standard wired interface and a wireless interface.

[0130] The network interface 24 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0131] Among them, the processor 21 may include one or more processing cores. The processor 21 uses various interfaces and lines to connect various parts in the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 25, and calling data stored in the memory 25. Optionally, the processor 21 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field-Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 21 can integrate one or more combinations of a central processing unit (Central Processing Unit, CPU), a graphics processor (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 21, and it can be implemented separately through a chip.

[0132] Among them, the memory 25 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read-Only Memory). Optionally, the memory 25 includes a non-transitory computer-readable storage medium. The memory 25 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 25 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned method embodiments, etc.; the data storage area may store data involved in the above-mentioned method embodiments, etc. The memory 25 may also be optionally at least one storage device located away from the aforementioned processor 21. As Figure 4 As shown, the memory 25 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program of an Internet of Things data transmission method based on an Internet of Things middle station.

[0133] exist Figure 4In the electronic device 2 shown, the user interface 23 is mainly used to provide an input interface for the user and obtain data input by the user; and the processor 21 can be used to call the application program of the Internet of Things data transmission method based on the Internet of Things middle station stored in the memory 25. When executed by one or more processors, the electronic device executes one or more methods in the above-mentioned embodiments.

[0134] A computer-readable storage medium stores instructions, which, when executed by one or more processors, enable the computer to execute one or more methods in the above-mentioned embodiments.

[0135] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application.

[0136] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0137] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0138] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0139] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0140] 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 memory. Based on this understanding, the technical solution of the present application, 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, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: various media that can store program codes, such as USB flash drives, mobile hard drives, magnetic disks or optical disks.

[0141] The above are only exemplary embodiments of the present disclosure, and the scope of the present disclosure cannot be limited thereto. That is, any equivalent changes and modifications made according to the teachings of the present disclosure are still within the scope of the present disclosure. After considering the disclosure of the specification and the truth of practice, it will be easy for those skilled in the art to think of other embodiments of the present disclosure. This application is intended to cover any variation, use or adaptive change of the present disclosure, which follows the general principles of the present disclosure and includes common knowledge or customary techniques in the technical field that are not recorded in the present disclosure.

[0142] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for transmitting data in the Internet of Things based on an Internet of Things middle station, characterized in that: The method comprises: Collecting multi-sensory data based on the Internet of Things perception layer, wherein the Internet of Things perception layer includes a sensor network composed of multiple sensors; Obtain an adaptive transmission cycle according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the IoT platform, and the bidirectional long short-term memory network model; The data transmission time interval of the Internet of Things perception layer is dynamically adjusted according to the adaptive transmission period.

2. The method for transmitting data of the Internet of Things based on the Internet of Things mid-station according to claim 1, characterized in that: The step of obtaining the adaptive transmission cycle according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the IoT platform, and the bidirectional long short-term memory network model specifically includes: Obtaining a coding time series matrix, inputting the coding time series matrix into a bidirectional long short-term memory network model, and obtaining multivariate time series features; Based on the time period of two adjacent data transmission processes collected by the IoT center, a data transmission cycle feature is constructed, and the data transmission cycle feature and the multivariate time series feature are fused and input into the bidirectional long short-term memory network model to obtain the optimal adaptive transmission period.

3. The method for transmitting data of the Internet of Things based on the Internet of Things middle station as claimed in claim 2, characterized in that: The obtaining of the encoding time series matrix specifically includes: Extracting a set of homogeneous IoT sensors in the IoT perception layer, and determining a sequence start position and an initial subsequence length corresponding to the homogeneous IoT sensor set, and constructing an initial coding time series matrix according to the sequence start position, the initial subsequence length and the perception time series matrix; Iterating the initial encoded time series matrix, and updating the homogeneous IoT sensor set and removing processed sensor nodes in each iteration; According to the positions of the remaining sensor nodes in the homogeneous IoT sensor set, multiple time-warped distance matrices corresponding to the sensor nodes are calculated, and a time-warped distance set is constructed based on the distance metric values ​​of the time-warped distance matrices; Selecting a time series index having a minimum distance metric value from the time-warping distance set, and merging a time series segment corresponding to the time series index into the initial encoded time series matrix; When the homogeneous IoT sensor set is an empty set, the iteration process ends and the encoded time series matrix is ​​output.

4. The method for transmitting data of the Internet of Things based on the Internet of Things middle station as claimed in claim 3, characterized in that: After obtaining the encoding time series matrix, the method further includes: The encoded time series matrix is ​​input into a multi-head convolutional neural network, a sequence feature vector corresponding to the sequence length of the convolution filter in the multi-head convolutional neural network is extracted, and the sequence feature vector is feature fused with the hidden state output vector in the bidirectional long short-term memory network model.

5. The method for transmitting data of the Internet of Things based on the Internet of Things middle station as claimed in claim 4, characterized in that: The expression of the convolution head output function in the multi-head convolutional neural network is: Among them, c h,i represents the output of the i-th feature map of the h-th convolutional head, represents the output function of the i-th feature map of the h-th convolution head at the sequence position j, φ represents the moving step of the convolution operation, n1 represents the input time series length covered by the sliding window during each convolution operation, and n2 represents the time series length of the input convolution head. represents the attention weight of the i-th feature map of the h-th convolution head at the sequence position j, softmax() represents the normalization function, and v T represents the transpose of the attention score weight matrix, tanh() represents the hyperbolic tangent activation function, W represents the input transformation matrix, represents the feature dimension vector contained in each time step in the local time series segment starting from sequence position j and with a length of n1, U represents the state transformation matrix, and h j-1 represents the hidden state vector at sequence position j, b c represents the bias term corresponding to the attention weight, represents the gate value of the i-th feature map of the h-th convolution head at the sequence position j, σ represents the activation function, [;] represents the vector concatenation operation, W g represents the weight matrix corresponding to the gating layer, b g represents the bias term corresponding to the gating layer, It represents the residual term of the i-th feature map of the h-th convolution head at the sequence position j, and ⊙ represents the element-by-element multiplication operation.

6. The method for transmitting data of the Internet of Things based on the Internet of Things mid-station according to claim 1, characterized in that: The output characteristic function expression of the bidirectional long short-term memory network is: E t =LN(S t +Res(x t )) O t =MLP([E t ;F t ]) Among them, h t represents the hidden state output function of the bidirectional long short-term memory network at time t, [;] represents the vector concatenation operation, represents the hidden state output vector of the forward long short-term memory network at time t, represents the hidden state output vector of the backward long short-term memory network at time t, F t represents the joint feature function, fusion() represents the operator merging function, H represents the total number of convolution heads, I represents the total number of feature maps, S t represents the characteristic function after sensor information sharing, γ represents the information sharing weight parameter, ⊙ represents the element-by-element multiplication operation, represents the output function of the i-th feature map of the h-th convolutional head at sequence position j, E t represents the feature function after feature enhancement, LN() represents the layer normalization function, Res() represents the residual connection function, O t represents the output feature function of the bidirectional long short-term memory network, and MLP() represents a multi-layer perceptron function.

7. The method for transmitting data of the Internet of Things based on the Internet of Things mid-station as claimed in claim 3, characterized in that: The collecting of multi-sensory data based on the Internet of Things perception layer also includes: Dynamically collect multivariate perception data using the IoT perception layer, obtain the data upload time of each sensor in the IoT perception layer, receive the multivariate perception data through an edge server, and perform deduplication and time series optimization on the multivariate perception data to obtain a perception time series matrix; Event recognition and classification are performed on the perception time series matrix, data aggregation and compression are performed according to the classification result, and the aggregated and compressed perception time series matrix is ​​uploaded to the cloud service layer for decompression and storage.

8. The method for transmitting data of the Internet of Things based on the Internet of Things mid-station according to claim 7, characterized in that: The dynamically adjusting the data transmission time interval of the Internet of Things perception layer according to the adaptive transmission cycle specifically includes: An update instruction is sent to each sensor node through the Internet of Things middle station, and the update instruction includes the adaptive transmission cycle, so that each sensor node dynamically adjusts the data collection interval and transmission time interval according to the adaptive transmission cycle.

9. An Internet of Things data transmission system based on an Internet of Things middle station, characterized in that: The Internet of Things data transmission system (1) comprises a data acquisition module (11), a cycle acquisition module (12) and a cycle adjustment module (13) which are connected in sequence, wherein: The data acquisition module (11) is used to collect multi-sensory data based on the Internet of Things perception layer, wherein the Internet of Things perception layer includes a sensor network composed of multiple sensors; The cycle acquisition module (12) is used to acquire an adaptive transmission cycle according to the encoding time series matrix corresponding to the multivariate perception data, the data transmission cycle characteristics constructed by the Internet of Things platform, and the bidirectional long short-term memory network model; The cycle adjustment module (13) is used to dynamically adjust the data transmission time interval of the Internet of Things perception layer according to the adaptive transmission cycle.

10. An electronic device, characterized in that: The electronic device (2) comprises a processor (21), a memory (25), a user interface (23) and a network interface (24), wherein the memory (25) is used to store instructions, the user interface (23) and the network interface (24) are used to communicate with other devices, and the processor (21) is used to execute the instructions stored in the memory (25) so that the electronic device (2) executes the method according to any one of claims 1 to 8.

Images