A Low-Power Data Transmission Method and System for IoT Terminals Based on Edge Collaboration

Abstract

This invention discloses a low-power data transmission method and system for IoT terminals based on edge collaboration, relating to the field of IoT wireless communication technology. The method includes: an event-driven acquisition step that uses a change detection algorithm to determine data changes, triggering transmission only when the change exceeds an adaptive threshold; otherwise, entering a deep sleep mode; a transmission scheduling step that uses deep reinforcement learning to optimize transmit power, modulation and coding, and transmission time slots; an edge collaborative processing step that performs semantic analysis and redundancy fusion on multi-terminal data before batch forwarding; a group collaborative transmission step that achieves channel state sharing and energy-balanced proxy forwarding within a group; and a feedback optimization step that feeds back the transmission effect to the terminal to update parameters. This invention, through an end-edge-cloud collaborative architecture and a closed-loop optimization mechanism, reduces terminal standby power consumption to the microampere level, improves energy efficiency, and extends network lifetime.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) wireless communication technology, and in particular to a low-power data transmission method and system for IoT terminals based on edge collaboration. Background Technology

[0002] With the rapid development and large-scale deployment of IoT technology, massive numbers of IoT terminal devices are widely used in smart cities, industrial monitoring, environmental sensing, smart agriculture, and other fields. These terminal devices are typically battery-powered, and due to size and cost constraints, battery capacity is limited. Battery life has become a key bottleneck restricting the long-term stable operation of IoT systems. How to maximize terminal battery life while ensuring data transmission timeliness and reliability has become a core challenge facing the development of IoT technology.

[0003] Chinese invention CN119854832A discloses a wireless communication method and system for IoT data transmission. The method includes real-time monitoring of the IoT environment and collection of environmental data, dynamic selection of the optimal frequency band and implementation of adaptive frequency hopping based on the environmental data, dynamic allocation of available channels based on the current network status and device requirements, preprocessing and fusing data from different IoT terminals to form data packets, performing edge computing to determine the optimal transmission path based on the current network status after the data packets are generated, and transmitting the data packets to the destination device. During the data transmission process, a feedback adjustment mechanism is formed by monitoring the transmission status and environmental changes to continuously optimize network performance.

[0004] However, the aforementioned existing technologies have the following shortcomings: First, the scheme mainly focuses on frequency band selection and channel allocation optimization at the network layer, lacking consideration for low-power design on the terminal side. The terminal needs to continuously monitor environmental data, resulting in high standby power consumption, which is difficult to meet the long lifespan requirements of battery-powered terminals. Second, the transmission scheduling of this scheme uses a genetic algorithm for channel optimization, but does not consider the impact of the terminal's battery status on the transmission strategy. It cannot dynamically adjust the transmission power and transmission frequency according to the terminal's energy status, and energy efficiency needs to be improved. Third, the edge computing function of this scheme is mainly used for transmission path selection, and does not achieve semantic-level compression and redundancy removal of data reported by multiple terminals. When multiple terminals report similar data, there is a large amount of redundant transmission, which increases the transmission overhead of the wide area network. Fourth, this scheme does not involve a cooperative transmission mechanism between neighboring terminals. Each terminal transmits data independently, and it is impossible to achieve energy balance through intra-group cooperation, causing some terminals to run out of power prematurely, affecting the overall lifespan of the network. Fifth, the feedback adjustment mechanism of this scheme is coarse-grained, mainly adjusting network layer parameters, and does not form a fine-grained closed-loop optimization of terminal acquisition thresholds and transmission strategies, making it difficult to achieve the optimal balance between terminal power consumption and transmission performance.

[0005] Therefore, there is an urgent need for an IoT data transmission method that can balance low power consumption of the terminal, high transmission efficiency, and data timeliness. Summary of the Invention

[0006] To address the technical problems of high terminal power consumption, low transmission energy efficiency, and lack of terminal collaboration mechanisms in existing technologies, this invention provides a low-power data transmission method and system for IoT terminals based on edge collaboration.

[0007] The technical solution adopted in this invention is as follows:

[0008] A low-power data transmission method for IoT terminals based on edge collaboration includes: an event-driven acquisition step, where the IoT terminal uses a lightweight change detection algorithm to determine the change magnitude of raw data collected by sensors. Data transmission is triggered when the data change magnitude exceeds an adaptive change threshold; otherwise, the terminal enters a deep sleep mode. The adaptive change threshold is dynamically adjusted based on historical data change patterns and current battery power. A transmission scheduling step involves a transmission scheduling module constructing a state encoding vector, which includes the current battery power status, channel quality indicators, data buffer size, and latency constraints. This state encoding vector is input into a deep reinforcement learning decision network, which outputs transmit power configuration, modulation and coding scheme, and transmission time slot selection. The reward function of the deep reinforcement learning decision network comprehensively optimizes energy efficiency and transmission reliability. Edge collaboration is also included. The processing steps include: an edge computing node receiving data reported by multiple IoT terminals, performing semantic analysis on the data to identify redundant information, deduplicating and fusing the identified redundant information to generate fused compressed data, and forwarding the fused compressed data in batches to a cloud server; a group collaborative transmission step where neighboring IoT terminals form a transmission collaboration group, and terminals within the group share channel state information through short-range low-power communication, selecting energy-sufficient nodes as proxy nodes based on the remaining power of each terminal in the group, and having the proxy nodes forward the data to be transmitted by other nodes; and a feedback optimization step where the edge computing node statistically analyzes the transmission success rate, average latency, and terminal energy consumption indicators, and sends these indicators as feedback information to the IoT terminals, which then update the adaptive change threshold and the policy parameters of the deep reinforcement learning decision network based on the feedback information.

[0009] Preferably, the sliding window has a window length of 8 to 32 sampling points. In the deep sleep mode, the terminal shuts down the wireless transceiver and the main processor, leaving only the low-power timer and sensor interface in standby mode.

[0010] Preferably, the first power threshold is 20% to 30% of the battery capacity, and the second power threshold is 60% to 80% of the battery capacity.

[0011] Preferably, the state encoding vector has a dimension of 12 to 24.

[0012] Preferably, the preset similarity threshold ranges from 0.85 to 0.95, and the compression ratio is 40% to 60% of the original data volume.

[0013] Preferably, the preset proxy threshold ranges from 1.2 to 1.5.

[0014] This invention also provides a low-power data transmission system for IoT terminals based on edge collaboration, comprising: an event-driven acquisition module, used to determine the change magnitude of raw data collected by sensors using a lightweight change detection algorithm; triggering data transmission when the data change magnitude exceeds an adaptive change threshold; and causing the terminal to enter a deep sleep mode when the data change magnitude does not exceed the adaptive change threshold; a transmission scheduling module, used to construct a state encoding vector containing battery power status, channel quality indicators, data buffer size, and latency constraint information, and output transmit power configuration, modulation and coding scheme, and transmission time slot selection based on a deep reinforcement learning decision network; an edge collaborative processing module, deployed on edge computing nodes, used to perform semantic analysis and redundancy removal on data reported by multiple IoT terminals, generate fused compressed data, and forward it in batches to a cloud server; a group collaborative transmission module, used to organize neighboring terminals into transmission collaboration groups to achieve channel state information sharing, and select energy-sufficient nodes as proxy nodes for data proxy forwarding; and a feedback optimization module, used to statistically analyze transmission performance indicators and feed them back to the event-driven acquisition module and the transmission scheduling module to update the corresponding parameters.

[0015] The beneficial effects of this invention are as follows: By employing an event-driven data acquisition strategy and a deep sleep mechanism, the standby power consumption of the terminal is reduced to the microamp level, achieving a power consumption reduction of over 95% compared to continuous monitoring mode, significantly extending the terminal's battery life; by comprehensively optimizing transmission power, modulation coding, and transmission time slots through deep reinforcement learning algorithms, energy efficiency is improved by 30% to 50% while ensuring transmission reliability; by semantic compression and redundancy removal at edge nodes, the amount of data transmitted over the wide area network is reduced by 40% to 60%, lowering network transmission overhead; by employing a group-based collaborative transmission and energy balancing proxy mechanism, the energy consumption of terminals within a group tends to be balanced, extending the overall network lifetime by 20% to 35%; and by employing a closed-loop feedback optimization mechanism, adaptive adjustments to the acquisition threshold and transmission strategy are achieved, enabling the system to adapt to dynamically changing environmental conditions. Attached Figure Description

[0016] Figure 1This is a flowchart illustrating a low-power data transmission method for IoT terminals based on edge collaboration, provided in an embodiment of the present invention.

[0017] Figure 2 This is a schematic diagram of the architecture of a low-power data transmission system for IoT terminals based on edge collaboration, provided in one embodiment of the present invention. Detailed Implementation

[0018] Please refer to the attached document. Figures 1-2 To make the objectives, technical solutions, and advantages of the present invention clearer, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort should fall within the scope of protection of the present invention.

[0019] Reference Figure 1 This invention provides a low-power data transmission method for IoT terminals based on edge collaboration. The method constructs a three-layer collaborative architecture of end-edge-cloud, deploys edge computing nodes between IoT terminals and cloud servers, and achieves the optimal balance between low power consumption and high timeliness through deep coupling and collaboration of five core steps: event-driven acquisition, intelligent transmission scheduling, edge collaborative processing, group collaborative transmission, and closed-loop feedback optimization.

[0020] Step S1: Event-driven data acquisition.

[0021] The event-driven acquisition step is the core of this invention to achieve low power consumption of the terminal. This step uses a lightweight change detection algorithm to determine whether the sensor data has changed significantly, and triggers data transmission only when necessary, while the terminal enters a deep sleep mode the rest of the time.

[0022] In one embodiment of the present invention, the IoT terminal periodically wakes up to perform data sampling, with the sampling period set to vary from 100 milliseconds to 10 seconds depending on the application scenario. After collecting raw data from the sensor, the terminal uses a lightweight change detection algorithm to determine the magnitude of data changes. The lightweight change detection algorithm proposed in this invention employs a sliding window mean comparison method, which has low computational complexity and is suitable for running on resource-constrained IoT terminals.

[0023] Specifically, let the current sampling time be... The current sample value is The length of the sliding window is The formula for calculating the historical average within the sliding window is:

[0024] ,

[0025] in, It is the arithmetic mean of the historical sampled values ​​within the sliding window. For the first The sampled values ​​at each sampling time. The value represents the length of the sliding window, ranging from 8 to 32 sampling points.

[0026] The formula for calculating the change in the current sample value compared to the historical mean is:

[0027] ,

[0028] in, The magnitude of data change at the current moment. This is the current sampled value. This represents the historical average within the sliding window.

[0029] This invention further proposes a dynamic adjustment mechanism for an adaptive changing threshold. Adaptive changing threshold The system dynamically adjusts based on historical data patterns and current battery level; the calculation formula is as follows:

[0030] ,

[0031] in, For adaptive changing thresholds, As the baseline threshold, This is the battery power adjustment function. The historical fluctuation adjustment function is used. This represents the current remaining battery percentage. This represents the standard deviation of historical data.

[0032] The specific form of the battery power regulation function is as follows:

[0033] ,

[0034] in, The first power threshold is set at 20% to 30% of the battery capacity. The second power threshold is set to 60% to 80% of the battery capacity. When the battery power is lower than the first power threshold, the system increases the threshold to reduce the transmission frequency and extend the terminal's operating time; when the battery power is higher than the second power threshold, the system decreases the threshold to improve the timeliness of data acquisition and transmission.

[0035] The specific form of the historical fluctuation adjustment function is as follows:

[0036] ,

[0037] in, This is an adjustment coefficient, with a value ranging from 0.1 to 0.3. Let N be the standard deviation of the data over the most recent N sampling periods. This is the average of the data over the most recent N sampling periods. This function increases the threshold accordingly in scenarios with large data fluctuations, avoiding frequent triggering of unnecessary transmissions.

[0038] When the magnitude of data change Exceeding the adaptive change threshold When the data changes, the terminal triggers a data transmission process, packaging the current data and sending it to the transmission scheduling module. When the data change amplitude does not exceed the adaptive change threshold, the terminal enters a deep sleep mode. In deep sleep mode, the terminal shuts down the wireless transceiver, main processor, and most peripherals, leaving only the low-power timer and sensor interface in standby mode, reducing standby power consumption to below 10 microamps. In a preferred embodiment of the invention, a microcontroller with an ultra-low-power sleep mode is used, and the measured deep sleep current is 3 to 8 microamps, which is more than three orders of magnitude lower than the 15 to 30 milliamps of normal operation.

[0039] It should be noted that the output of the event-driven acquisition step includes two parts: when transmission is triggered, the data packet to be transmitted (data stream D03) and the change detection result (data stream D02) are output to the transmission scheduling module; when transmission is not triggered, the historical data in the sliding window is updated and the system enters sleep mode to wait for the next sampling cycle.

[0040] Step S2, transmission scheduling steps.

[0041] The transmission scheduling step is the core of the energy efficiency optimization in this invention. This step uses a deep reinforcement learning algorithm to dynamically optimize the transmit power, modulation and coding scheme and transmission time slot selection, so as to maximize energy efficiency while ensuring transmission reliability.

[0042] In one embodiment of the present invention, the transmission scheduling module first constructs a state encoding vector. The state encoding vector is the input to the deep reinforcement learning decision network and contains multi-dimensional features describing the current terminal state and the environment state. The multi-factor state encoding method proposed in this invention encodes the key factors affecting transmission decisions into a unified vector representation.

[0043] State encoding vector The composition is as follows:

[0044] ,

[0045] in, for The state encoding vector at time step, This is a normalized percentage representation of the current battery state, with a value ranging from 0 to 1. This is the normalized representation of the channel signal-to-noise ratio. This is the estimated bit error rate for the current channel. This represents the proportion of cached data to the total cache capacity. This is the normalized value of the maximum tolerable latency for the current data. For the one-hot encoded vector of data priority category, Other auxiliary features include historical transmission success rate and channel state change trends. In a preferred embodiment of the invention, the state coding vector has a dimension of 12 to 24.

[0046] After the state encoding vector is constructed, it is input into the deep reinforcement learning decision network. The deep reinforcement learning decision network used in this invention is based on a deep Q-network architecture, containing 3 to 5 fully connected hidden layers, with 64 to 256 neurons in each layer. The network input is the state encoding vector, and the output is the Q-value estimate of each action in the action space.

[0047] The action space is defined as follows:

[0048] ,

[0049] in, For the action space, For the first Level transmit power configuration, It is a set of selectable transmit power levels, comprising 4 to 8 discrete power levels. For the first A type of modulation and coding scheme, It is a set of optional modulation and coding schemes, including BPSK, QPSK, 16QAM, etc. For the first One transmission time slot, This is the set of available transmission time slots.

[0050] One of the core innovations of this invention is the proposal of a composite reward function that comprehensively optimizes energy efficiency and transmission reliability. Reward Function The calculation formula is:

[0051] ,

[0052] in, for The overall reward value at any given moment. , , The weighting coefficients are satisfied. , For energy efficiency rewards, As a reward item for transmission reliability, This is a delay penalty.

[0053] The formula for calculating the energy efficiency bonus is:

[0054] ,

[0055] in, The amount of data successfully transmitted, in bits. The energy consumed in this transmission, measured in joules. This is a normalization coefficient to ensure that the energy efficiency bonus value falls within a reasonable range.

[0056] The formula for calculating the transmission reliability bonus is as follows:

[0057] ,

[0058] in, The current transmission success rate, A preset reliability target is typically set between 0.95 and 0.99. A full score is awarded when the transmission success rate reaches the target; otherwise, a partial score is awarded.

[0059] The formula for calculating the delay penalty is:

[0060] ,

[0061] in, This is the delay penalty coefficient. This refers to the actual transmission completion delay. This represents the maximum tolerable delay. A negative penalty is incurred when the actual delay exceeds the maximum tolerable delay.

[0062] In a preferred embodiment of the present invention, the typical value of the weighting coefficient is... , , This configuration prioritizes energy efficiency optimization while also considering transmission reliability and latency constraints. The weighting can be adjusted based on actual needs in different application scenarios; for example, it can be increased for latency-sensitive applications. The value of .

[0063] The training of the deep reinforcement learning decision network employs an experience replay and target network separation technique. The experience replay buffer stores the historical state-action-reward-new state quadruples, with a batch size set between 32 and 128. The target network synchronizes with the main network every 100 to 500 training steps. The learning rate is set between 0.0001 and 0.001, and the Adam optimizer is used for parameter updates.

[0064] The transmission scheduling module configures physical layer transmission parameters based on the output of the deep reinforcement learning decision network, including setting the gain of the transmit power amplifier, configuring the modulation scheme and coding rate of the modem, and selecting the transmission time slot in the time division multiple access system. After configuration, the terminal transmits data packets in the specified time slot with the specified power and modulation scheme.

[0065] Step S3, Edge Collaboration Processing Step.

[0066] The edge collaborative processing step, executed on edge computing nodes, is the core component of this invention for optimizing transmission overhead. As the middle layer of the end-edge-cloud architecture, the edge computing node undertakes functions such as data aggregation, semantic analysis, redundancy removal, and batch forwarding, effectively reducing the amount of data transmitted over a wide area network.

[0067] In one embodiment of the present invention, an edge computing node is deployed on the access gateway or base station side, and receives data reported by multiple IoT terminals within its coverage area through a wireless access network. The edge node first decodes and temporarily stores the received data, waiting for all terminal data to arrive within the convergence window time.

[0068] The core of edge collaborative processing is semantic analysis and redundancy identification. The semantic redundancy calculation algorithm proposed in this invention is based on feature vector similarity measurement and can identify semantic-level redundancy information between data reported by different terminals.

[0069] Specifically, for each terminal's reported data packets First, extract its feature vector. :

[0070] ,

[0071] in, The feature extraction function can be implemented using a lightweight convolutional network or a statistical feature extraction method. In a preferred embodiment of the present invention, considering the characteristics of IoT sensor data, the feature vector includes statistical features such as data mean, variance, maximum value, minimum value, and slope of the trend, and the feature vector dimension is 8 to 32 dimensions.

[0072] For any two terminals and Calculate the cosine similarity between the feature vectors of the reported data:

[0073] ,

[0074] in, For the terminal and terminal semantic similarity of data The inner product of two eigenvectors. and These are the magnitudes of the two eigenvectors, respectively.

[0075] When the semantic similarity exceeds a preset similarity threshold, redundant information is determined to exist between the two data packets. (Preset similarity threshold) The value ranges from 0.85 to 0.95, and this threshold can be adjusted according to the redundancy tolerance of the specific application scenario. After redundancy identification is completed, the edge nodes perform fusion processing on the redundant data.

[0076] The redundant data fusion strategy adopted in this invention is as follows: For data groups identified as mutually redundant, a complete set of data is retained as a baseline, and other data only retain the differences from the baseline data. The structure of the fused compressed data packet is as follows:

[0077] ,

[0078] in, To merge compressed data packets, As benchmark data, the data with the largest volume or the highest information entropy is usually selected as the benchmark. For the first The differences between redundant data and baseline data, This represents the number of data packets in the redundant data group. Metadata includes auxiliary information such as terminal identifiers, timestamps, and raw data indexes.

[0079] Through the above-described fusion processing, the compression ratio can reach 40% to 60% of the original data volume. In the experimental verification of this invention, for a scenario with 100 terminals deployed, the average redundancy between terminal data is about 35%, and after edge fusion processing, the amount of data transmitted over the wide area network is reduced by about 45%.

[0080] Edge nodes employ a batch forwarding strategy to send the fused and compressed data to the cloud server. The triggering conditions for batch forwarding include: the amount of fused data reaching a preset data volume threshold, or the waiting time reaching a preset time threshold. In a preferred embodiment of the invention, the data volume threshold is set to 80% of the buffer capacity, and the time threshold is set to 1 to 10 seconds, which can be configured according to the latency requirements of the actual application.

[0081] In addition, the edge collaborative processing module also undertakes protocol conversion functions. IoT terminals typically use low-power wide area network or personal area network protocols such as LoRa, NB-IoT, and ZigBee, while edge nodes connect to the cloud via Ethernet or 4G / 5G cellular networks. Edge nodes complete the conversion between different network protocols, shielding the heterogeneity of the underlying network.

[0082] Step S4, Group Collaborative Transmission Step.

[0083] The group cooperative transmission step is the core of the network lifetime optimization in this invention. This step achieves energy balance within the group through cooperation between neighboring terminals, preventing some terminals from running out of power prematurely.

[0084] In one embodiment of the invention, geographically proximate IoT terminals are organized into transmission cooperation groups. The grouping strategy can be based on geographical distance, channel correlation, or application logic. In a preferred embodiment of the invention, a geographical distance-based grouping strategy is used, grouping terminals with a physical distance less than the short-range communication coverage area into the same cooperation group, with each group containing 3 to 10 terminals.

[0085] Terminals within the collaborative group share channel state information (CSS) with each other using short-range low-power communication technologies (such as Bluetooth Low Energy (BLE) or IEEE 802.15.4). CSS includes parameters such as signal-to-noise ratio (SNR), bit error rate (BER), and available frequency points measured by each terminal. Through this shared CSS, each terminal can learn about the channel conditions of other terminals within the group, providing a basis for collaborative transmission decisions.

[0086] One of the core innovations of this invention is the proposed energy-balanced proxy selection algorithm. This algorithm dynamically selects proxy nodes based on the remaining power of each terminal in the group, allowing nodes with sufficient power to forward data from other nodes, thereby achieving a balanced distribution of energy consumption within the group.

[0087] Specifically, for each terminal within the collaboration group Calculate its energy balance factor:

[0088] ,

[0089] in, For the terminal Energy balance factor, For the terminal The current remaining battery power, This is the average remaining battery level of all terminals within the group.

[0090] The selection criteria for proxy nodes are that the energy balance factor is the largest and exceeds the preset proxy threshold. :

[0091] ,

[0092] in, For the selected proxy node, A preset proxy threshold is set, ranging from 1.2 to 1.5. This threshold ensures that only nodes with relatively abundant energy are selected as proxy nodes, avoiding low-energy nodes bearing additional forwarding burdens.

[0093] After a proxy node is selected, other terminals within the group send their data to the proxy node via short-range communication. Short-range communication consumes significantly less power than wide-area network (WAN) transmission, thus having a minimal impact on the energy consumption of the proxied terminals. The proxy node collects the group's data and forwards it to the edge computing nodes via a unified WAN interface.

[0094] The structure of the proxy forwarding data packets is as follows:

[0095] ,

[0096] in, Forward data packets for the proxy, For proxy node identification, and The proxied terminals are respectively Identification and data, For timestamps.

[0097] Agent roles are dynamically rotated within the collaboration group. When the energy balance factor of the current agent node drops below a preset agent threshold, an agent switching process is triggered, and a new agent node is selected. In a preferred embodiment of the invention, the evaluation period for agent switching is set to 5 to 15 minutes to balance agent switching overhead and energy balance effect.

[0098] Another advantage of the group-based collaborative transmission step is that it provides transmission redundancy and reliability guarantees. When the channel conditions between a terminal and an edge node are poor, data can be forwarded using a proxy node within the group with good channel conditions, thereby improving the overall transmission success rate.

[0099] Step S5: Feedback and optimization steps.

[0100] The feedback optimization step is the core of this invention to achieve closed-loop adaptive optimization. This step feeds back the transmission effect to the front-end module, driving the dynamic adjustment of the acquisition threshold and transmission strategy to form a complete closed-loop control.

[0101] In one embodiment of the present invention, the edge computing node continuously collects statistics on the transmission performance indicators of each terminal. The core indicators collected include transmission success rate, average transmission latency, and estimated terminal power consumption.

[0102] Transmission success rate The calculation formula is:

[0103] ,

[0104] in, The transmission success rate within the statistical period. The number of data packets successfully transmitted. This represents the total number of data packets transmitted.

[0105] Average transmission delay The calculation formula is:

[0106] ,

[0107] in, For average transmission delay, For the first The reception time of each data packet, For the first The time it takes for each data packet to be sent.

[0108] Terminal energy consumption estimates are calculated based on transmission power and transmission duration:

[0109] ,

[0110] in, For the estimated terminal energy consumption, For the first The transmission power of the next transmission. For the first The duration of each transmission.

[0111] Edge nodes use the aforementioned statistical metrics as feedback information and send them to the corresponding IoT terminals via downlink channels. The transmission cycle for feedback information can be set from 1 minute to 10 minutes, ensuring timeliness while avoiding excessive downlink communication overhead.

[0112] The closed-loop parameter update algorithm proposed in this invention dynamically adjusts the terminal's acquisition threshold and transmission strategy based on feedback information. The specific rules for parameter updating are as follows:

[0113] When the transmission success rate is lower than the preset success rate threshold At that time, the following adjustments will be made:

[0114] ,

[0115] ,

[0116] in, For the updated adaptive change threshold, The threshold adjustment step size ranges from 0.05 to 0.15. For the updated energy efficiency reward weights, Adjust the step size for weighting. Lowering the change threshold encourages terminals to transmit data more frequently, while reducing the energy efficiency weight makes the system more inclined to choose a high-reliability transmission configuration.

[0117] When the average latency exceeds the preset latency threshold At that time, the slot selection strategy of the deep reinforcement learning decision network is adjusted to prioritize low-competition slots, while the delay penalty weights are adjusted. Increase appropriately.

[0118] When the terminal's energy consumption exceeds the preset energy threshold At that time, the following adjustments will be made:

[0119] ,

[0120] in, The threshold increment is set to 0.1 to 0.2 when energy consumption is too high. Increasing the threshold reduces the transmission frequency, while the deep reinforcement learning decision network tends to choose a low-power transmission configuration.

[0121] The closed-loop control relationship formed by feedback optimization is as follows: the transmission trigger generated by the event-driven acquisition step affects the transmission frequency, the transmission frequency affects the terminal energy state through the energy consumption accumulation of the transmission scheduling step, and the energy state adjusts the change threshold of the event-driven acquisition step in reverse through the parameter update of the feedback optimization step; similarly, the transmission configuration generated by the transmission scheduling step affects the transmission success rate and latency, and these indicators affect the policy update of the deep reinforcement learning decision network through the feedback optimization step.

[0122] In the experimental verification of this invention, the system employing the aforementioned closed-loop feedback optimization mechanism can adaptively adjust parameters to cope with environmental changes. In scenarios with deteriorating channel conditions, the system automatically increases the transmit power and reduces the modulation order to ensure transmission reliability; in scenarios with insufficient battery power, the system automatically reduces the transmission frequency and selects a low-power configuration to extend operating time.

[0123] The low-power data transmission method for IoT terminals based on edge collaboration provided by this invention has achieved significant technical results after system testing and verification.

[0124] The test environment was configured as follows: 100 IoT terminal nodes were deployed, each equipped with a 2000mAh lithium battery. The terminal processors were ARM Cortex-M4 series low-power microcontrollers, and the edge computing nodes were edge servers using Intel Xeon processors. The test scenario was industrial environment monitoring, with sensor data including parameters such as temperature, humidity, and vibration, and a sampling period of 1 second.

[0125] Compared with the control group of the CN119854832A technical solution, the method of the present invention achieves improvements in the following indicators: the average standby power consumption of the terminal is reduced from 25 mA in continuous monitoring mode to 8 μA in event-driven mode, a power consumption reduction of more than 99%; the energy efficiency (data transmission volume per unit energy consumption) is increased from 1.2 Mbit / J to 2.1 Mbit / J, an improvement of 75%; the amount of data transmitted over the wide area network is reduced by 47%, effectively reducing network load; the overall network lifespan (defined as the time when the first terminal battery is depleted) is extended by 32%, from 45 days to 59 days; and the transmission success rate remains stable at over 97% in scenarios with dynamically changing channel conditions.

[0126] Reference Figure 2 The present invention also provides a low-power data transmission system for IoT terminals based on edge collaboration. The system's modules correspond to the steps in the above method embodiments, including an event-driven acquisition module 1, a transmission scheduling module 2, an edge collaboration processing module 3, a group collaboration transmission module 4, and a feedback optimization module 5.

[0127] Event-driven acquisition module 1 is deployed on the IoT terminal side and is used to determine the magnitude of change in the raw data collected by the sensor using a lightweight change detection algorithm. As described in the method embodiment, event-driven acquisition module 1 uses a sliding window mean comparison algorithm to calculate the magnitude of data change. When the magnitude of change exceeds an adaptive change threshold, data transmission is triggered; otherwise, the terminal enters a deep sleep mode. The hardware implementation of event-driven acquisition module 1 includes a low-power sensor interface circuit, a sliding window data buffer, and a threshold comparator. In a preferred embodiment of the present invention, the threshold comparison function can be implemented through a hardware comparator to further reduce the processor wake-up frequency.

[0128] The transmission scheduling module 2 is also deployed on the IoT terminal side, used to construct state encoding vectors and output transmission parameter configurations based on the deep reinforcement learning decision network. As described in the method embodiment, the state encoding vector of the transmission scheduling module 2 includes battery power status, channel quality indicators, data buffer size, and latency constraint information. The deep reinforcement learning decision network outputs transmit power configuration, modulation and coding scheme, and transmission time slot selection. The interface between the transmission scheduling module 2 and the event-driven acquisition module 1 includes transmission trigger signals and the transmission of data packets to be transmitted. Considering the limited computing resources of the IoT terminal, the deep reinforcement learning decision network adopts a lightweight design, with the number of hidden layer neurons controlled between 64 and 128, and the model parameter size not exceeding 10KB.

[0129] Edge collaborative processing module 3 is deployed on an edge computing node to perform semantic analysis and redundancy removal on data reported by multiple IoT terminals. As described in the method embodiment, edge collaborative processing module 3 identifies redundant information through feature extraction and cosine similarity calculation, and then forwards the redundant data in batches to the cloud server after fusion and compression. The hardware carrier of edge collaborative processing module 3 can be an embedded computing platform deployed on the access gateway side or an edge server on the base station side. Edge collaborative processing module 3 is connected to the terminal-side module via a wireless access network and to the cloud server via a wide area network.

[0130] The group cooperative transmission module 4 is distributed and deployed on the IoT terminal side to organize neighboring terminals into transmission cooperative groups and realize proxy forwarding. As described in the method embodiment, the group cooperative transmission module 4 shares channel state information through short-range low-power communication and selects proxy nodes based on energy balancing factors. The implementation of the group cooperative transmission module 4 depends on the short-range communication interface (such as Bluetooth Low Energy BLE module) and intra-group coordination protocol of the terminal.

[0131] The statistical function of the feedback optimization module 5 is deployed on the edge computing node, while the parameter update function is deployed on the IoT terminal side. The edge-side feedback optimization module 5 statistically analyzes transmission success rate, average latency, and terminal energy consumption indicators, sending the statistical results as feedback information to the terminal. The terminal-side feedback optimization module 5 performs parameter updates based on the received feedback information, adjusting the adaptive threshold of the event-driven acquisition module 1 and the deep reinforcement learning decision network strategy parameters of the transmission scheduling module 2.

[0132] The five modules described above form a deeply coupled closed-loop collaborative architecture: the output of the event-driven acquisition module 1 triggers the action of the transmission scheduling module 2; the transmission parameters configured by the transmission scheduling module 2 affect the quality of data received by the edge collaborative processing module 3; the proxy forwarding of the group collaborative transmission module 4 provides supplementary data sources for the edge collaborative processing module 3; and the parameter updates of the feedback optimization module 5 inversely adjust the behavior strategies of the event-driven acquisition module 1 and the transmission scheduling module 2. This closed-loop collaborative architecture enables the system to continuously and adaptively optimize, maintaining optimal performance under dynamically changing environmental conditions.

[0133] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. A low-power data transmission method for IoT terminals based on edge collaboration, characterized in that, include: In the event-driven acquisition step, the IoT terminal uses a lightweight change detection algorithm to determine the change magnitude of the raw data collected by the sensor. When the data change magnitude exceeds the adaptive change threshold, data transmission is triggered. When the data change magnitude does not exceed the adaptive change threshold, the terminal enters a deep sleep mode. The adaptive change threshold is dynamically adjusted according to the historical data change pattern and the current battery level. In the transmission scheduling step, the transmission scheduling module constructs a state encoding vector, which includes the current battery power status, channel quality indicators, data buffer size, and latency constraint information. The state encoding vector is input into a deep reinforcement learning decision network, which outputs transmit power configuration, modulation and coding scheme, and transmission time slot selection. The reward function of the deep reinforcement learning decision network comprehensively optimizes energy efficiency and transmission reliability. In the edge collaborative processing step, the edge computing node receives data reported by multiple IoT terminals, performs semantic analysis on the data reported by the multiple IoT terminals to identify redundant information, performs deduplication and fusion processing on the identified redundant information to generate fused compressed data, and forwards the fused compressed data to the cloud server in batches. In the group-based collaborative transmission step, neighboring IoT terminals form a transmission collaboration group. Terminals within the group share channel status information through short-range low-power communication. Based on the remaining power of each terminal in the group, a node with sufficient energy is selected as a proxy node, which forwards the data to be transmitted from other nodes. In the feedback optimization step, the edge computing node statistically analyzes the transmission success rate, average latency, and terminal energy consumption indicators, and sends these indicators as feedback information to the IoT terminal. The IoT terminal updates the adaptive change threshold and the policy parameters of the deep reinforcement learning decision network based on the feedback information.

2. The method according to claim 1, characterized in that, The lightweight change detection algorithm includes: calculating the absolute value of the difference between the current sampled value and the historical average within the sliding window as the change amplitude. The window length of the sliding window is 8 to 32 sampling points. In the deep sleep mode, the terminal shuts down the wireless transceiver and the main processor, and only retains the low-power timer and sensor interface in standby mode.

3. The method according to claim 1, characterized in that, The dynamic adjustment of the adaptive change threshold includes: increasing the adaptive change threshold to reduce the transmission frequency when the battery power is lower than a first power threshold, and decreasing the adaptive change threshold to improve data timeliness when the battery power is higher than a second power threshold. The first power threshold is 20% to 30% of the battery capacity, and the second power threshold is 60% to 80% of the battery capacity.

4. The method according to claim 1, characterized in that, The state coding vector has a dimension of 12 to 24, the current battery power status is represented by a normalized percentage, the channel quality indicators include signal-to-noise ratio and bit error rate, and the delay constraint information includes data timeliness level and maximum tolerable delay.

5. The method according to claim 1, characterized in that, The reward function includes an energy efficiency reward and a transmission reliability reward. The energy efficiency reward is the ratio of the amount of data successfully transmitted to the energy consumed, and the transmission reliability reward is the difference between the transmission success rate and the preset reliability target. The reward function combines the energy efficiency reward and the transmission reliability reward by weighted summation.

6. The method according to claim 1, characterized in that, The semantic analysis includes: extracting feature vectors of data reported by each terminal, calculating the cosine similarity between feature vectors of data from different terminals, and determining redundant data when the cosine similarity exceeds a preset similarity threshold. The preset similarity threshold ranges from 0.85 to 0.

95.

7. The method according to claim 1, characterized in that, The fusion process includes: semantically compressing redundant data while preserving the differences, with a compression ratio of 40% to 60% of the original data volume; the triggering condition for the batch forwarding is that the fused data volume reaches a preset data volume threshold or the waiting time reaches a preset time threshold.

8. The method according to claim 1, characterized in that, The step of selecting a node with sufficient energy as a proxy node includes: calculating the energy balance factor of each terminal in the group, wherein the energy balance factor is the ratio of the remaining power of the terminal to the average remaining power of the group, and selecting the terminal with the largest energy balance factor that is greater than a preset proxy threshold as the proxy node, wherein the preset proxy threshold ranges from 1.2 to 1.

5.

9. The method according to claim 1, characterized in that, The parameter update based on feedback information includes: reducing the adaptive change threshold and increasing the weight of the transmit power configuration when the transmission success rate is lower than the preset success rate threshold; prioritizing the selection of low-latency transmission slots when the average latency exceeds the preset latency threshold; and increasing the adaptive change threshold and reducing the transmit power when the terminal energy consumption exceeds the preset energy consumption threshold.

10. A low-power data transmission system for IoT terminals based on edge collaboration, used to implement the method according to any one of claims 1-9, characterized in that, include: The event-driven acquisition module is used to determine the magnitude of change in the raw data acquired by the sensor through a lightweight change detection algorithm. When the magnitude of the data change exceeds the adaptive change threshold, data transmission is triggered. When the magnitude of the data change does not exceed the adaptive change threshold, the terminal enters a deep sleep mode. The transmission scheduling module is used to construct a state coding vector containing information such as battery power status, channel quality indicators, data buffer size, and latency constraints, and outputs transmit power configuration, modulation and coding scheme, and transmission time slot selection based on a deep reinforcement learning decision network. The edge collaborative processing module, deployed on edge computing nodes, is used to perform semantic analysis and redundancy removal on data reported by multiple IoT terminals, generate fused and compressed data, and forward it to the cloud server in batches. The group cooperative transmission module is used to organize neighboring terminals to form a transmission cooperative group, realize the sharing of channel state information, and select energy-sufficient nodes as proxy nodes for data proxy forwarding. The feedback optimization module is used to collect transmission performance metrics and feed them back to the event-driven acquisition module and the transmission scheduling module to update the corresponding parameters.

Images