Wireless operation and maintenance method and system based on Internet of Things

Abstract

The invention relates to the technical field of the Internet of Things, in particular to a wireless operation and maintenance method and system based on the Internet of Things, and the method comprises the steps: S1, collecting original operation data through Internet of Things equipment, and converting the original operation data into a current equipment state vector; s2, obtaining an abnormal score and a state tension item of the current equipment in combination with the state vector of the current equipment and the vector of the neighbor equipment; s3, constructing a response strength scoring function based on the abnormal score and the state tension item, and outputting a response strength value through the response strength scoring function; s4, inputting the response intensity value into a structure mapper, and outputting a control instruction through the structure mapper; s5, the control instruction is issued to the equipment through the communication channel; and S6, the Internet of Things equipment executes operation according to the control instruction, and feeds back an execution result to the system.

Description

Technical Field

[0001] The present invention relates to the technical field of the Internet of Things, and particularly to a wireless operation and maintenance method and system based on the Internet of Things. Background Art

[0002] With the wide application of Internet of Things technology, a large number of wireless terminal devices are deployed in complex scenarios such as industrial sites, urban infrastructure, and agricultural environments, undertaking key tasks such as data collection, status monitoring, and control execution. These devices are usually connected to a cloud platform or an edge gateway through wireless communication to achieve remote operation and maintenance functions, such as device status update, fault repair, system upgrade, etc. The essence of wireless operation and maintenance is to reliably and timely manage and maintain devices without interrupting the normal operation of the devices. However, with the continuous expansion of the scale of the Internet of Things system, the types of devices tend to be diversified, and the deployment environment is more complex, and the existing wireless operation and maintenance technologies are facing serious challenges.

[0003] First of all, the wireless communication environment where Internet of Things devices are located generally has problems such as channel congestion, spectrum interference, and multipath fading, resulting in unstable transmission of operation and maintenance instructions and a high packet loss rate, especially more serious in low-power wide-area communication protocols such as LoRa and Zigbee. Secondly, the device status perception mechanism generally adopts a timing reporting or simple threshold alarm strategy. Once a device loses contact due to communication anomalies or enters an abnormal state, the system often cannot accurately identify its operating state and lacks a mechanism to infer its state from the surrounding devices. In addition, since most devices are deployed in resource-constrained places (such as underground, far suburbs, and the wild) and are usually powered by batteries, their communication, computing, and sensing capabilities are limited. Traditional cloud-centered operation and maintenance strategies (such as centralized scheduling and unified control) are not feasible in terms of timeliness and energy consumption. Although existing research has tried to introduce methods such as edge computing, multi-channel polling, and fault-tolerant broadcasting to improve operation and maintenance reliability, there are often problems such as overweight models, static strategies, and lack of feedback, and it is impossible to truly achieve high-robust and high-efficiency wireless operation and maintenance in complex scenarios.

[0004] To sum up, the current wireless operation and maintenance technologies have significant bottlenecks in three aspects: (1) Unmeasurable status: There is no reliable judgment mechanism when a device loses contact or has communication anomalies; (2) Inaccessible instructions: Spectrum congestion and interference lead to control instruction failures; (3) Unclosed control: It is difficult to verify the execution status of the device, and the system cannot form an operation and maintenance closed loop of "recognition - decision - execution - verification". Therefore, there is an urgent need for a new operation and maintenance method and system with state cognitive intelligence, wireless interference perception ability, and adaptive control scheduling mechanism, which can effectively adapt to the heterogeneity, instability, and resource limitations of the Internet of Things environment and improve the stability, self-healing ability, and intelligent level of the overall system. Summary of the Invention

[0005] To solve the problems in the above-mentioned background art, the present invention provides a wireless operation and maintenance method and system based on the Internet of Things.

[0006] To achieve the above object, the technical solution adopted by the present invention is:

[0007] A wireless operation and maintenance method based on the Internet of Things, including:

[0008] S1: Collect the original operation data through Internet of Things devices, and convert the original operation data into the current device state vector;

[0009] S2: Combine the current device state vector and the neighbor device vector to obtain the anomaly score and the state tension term of the current device;

[0010] S3: Construct a response intensity scoring function based on the anomaly score and the state tension term, and output the response intensity value through the response intensity scoring function;

[0011] S4: Input the response intensity value into the structure mapper, and output the control instruction through the structure mapper;

[0012] S5: Send the control instruction to the device through the communication channel;

[0013] S6: The Internet of Things device executes the operation according to the control instruction and feeds back the execution result to the system.

[0014] Preferably, the expression of the anomaly score in S2 is:

[0015]

[0016] where s i is the anomaly score, z i is the current device state vector, z j is the neighbor device state vector, w ij is the state influence weight of neighbor device j on current device i, λ is the regularization term coefficient; Ω(z i ) is the regularization term, is the set of neighbor devices.

[0017] Preferably, the expression of the state tension term in S2 is:

[0018]

[0019] where Ω(z i ) is the state tension term.

[0020] Preferably, S1 specifically includes:

[0021] S1.1: Obtain the original operation data, and send the original operation data to the edge node periodically through the wireless protocol;

[0022] S1.2: Standardize the original operation data through edge nodes;

[0023] S1.3: Perform data augmentation on the standardized original operation data to generate positive sample pairs;

[0024] S1.4: Input the positive sample pairs into the encoder to generate the current device status vector through the encoder.

[0025] Preferably, the communication channel generation in S5 includes:

[0026] Regularly scan available channels through the communication module of the Internet of Things device and record the historical communication status records of each channel;

[0027] Based on the historical communication status records, predict multiple channels in a moving average manner;

[0028] Input the prediction results into the comprehensive scheduling scoring function and output the scoring results of multiple channels through the comprehensive scheduling scoring function;

[0029] Select the channel with the smallest score value as the communication channel.

[0030] Preferably, the expression of the response intensity scoring function in S3 is:

[0031]

[0032] where ψ i is the behavior trust factor, is the average value of the device status in the past T time windows, β1, β2, β3 are empirical weight coefficients, and σ(·) is the Sigmoid function.

[0033] Preferably, S6 specifically includes:

[0034] S6.1: After the Internet of Things device receives the control instruction, call the local control module to execute the operation through the preset control mapping table;

[0035] S6.2: After the operation is completed, the Internet of Things device constructs a feedback data packet through the same communication channel and returns the feedback data packet to the system;

[0036] S6.3: The system receives the feedback data packet, extracts the execution result flag bit of the feedback data packet, and generates a feedback scalar.

[0037] Preferably, the feedback data packet in S6.2 includes the instruction ID, device ID, operation type, and execution result.

[0038] Preferably, the raw data in S1 includes voltage, device chip temperature, wireless reception strength, network response latency, and device load.

[0039] A wireless operation and maintenance system based on the Internet of Things, which is applied to the above-mentioned wireless operation and maintenance method based on the Internet of Things, and includes:

[0040] Internet of Things devices for collecting raw operation data;

[0041] An encoder for generating a state vector of the device;

[0042] A structure mapper for dividing the response strength value into different control levels according to intervals and generating control instructions

[0043] The beneficial effects of the present invention are as follows:

[0044] By constructing a collaborative cognitive mechanism between device states and a dynamic adaptation mechanism for communication links, intelligent and stable wireless remote operation and maintenance control are realized. Specifically, the present invention introduces a self-learning state modeling mechanism on the device side in device state judgment, so that in the case of lack of central control or loss of connection of individual devices, the system can still share state information through surrounding devices and achieve auxiliary judgment, thereby improving the accuracy and coverage of anomaly detection.

[0045] In terms of the execution of operation and maintenance instructions, the present invention designs a communication link self-sensing and dynamic scheduling mechanism, which can real-time sense the current wireless channel state and dynamically select the optimal frequency band or alternative channels based on the current interference situation, significantly improving the transmission success rate of instructions.

[0046] In terms of the overall control process, the present invention proposes a feedback closed-loop system driven by state judgment, realizing a continuous closed-loop control chain from state recognition, instruction triggering, path selection to result feedback, truly breaking through the boundary limitations of the traditional centralized operation and maintenance model, and adapting to the actual needs of wide device distribution, uncertain communication, and fast response control in the Internet of Things environment. The present invention has the characteristics of light structure, flexible deployment, control closed-loop, and strong robustness, and is particularly suitable for remote, complex, and resource-constrained Internet of Things wireless operation and maintenance scenarios. Description of the Drawings

[0047] Figure 1 It is a flowchart of the wireless operation and maintenance method based on the Internet of Things in a specific embodiment of the present invention;

[0048] Figure 2 It is a flowchart of generating a device state vector in a specific embodiment of the present invention;

[0049] Figure 3 It is a flowchart of feedback of execution results in a specific embodiment of the present invention;

[0050] Figure 4 This is the block diagram of the wireless operation and maintenance system based on the Internet of Things in a specific embodiment of the present invention. Specific embodiments

[0051] Please refer to Figures 1-4 As shown, the present invention relates to a wireless operation and maintenance method and system based on the Internet of Things, specifically including:

[0052] Step 1: Device status coding and modeling [[ID=1E]]

[0053] The purpose of this step is to convert the original operation data collected by Internet of Things devices into a state vector z i , which serves as the basis for anomaly recognition and control decision-making in subsequent steps. Considering the highly heterogeneous and scarce annotation of device data in the Internet of Things environment, we use the SimCLR framework in self-supervised learning methods to encode and model the original data. The SimCLR structure is simple, the training is stable, and it adapts to the requirements of edge deployment, and can effectively extract the low-dimensional semantic representation of the device operation state.

[0054] Input:

[0055] Device original data x i : Refers to the sensor readings uploaded by the device in each sampling period, including:

[0056] – Voltage (such as the real-time battery voltage of a 3.3V power supply device);

[0057] – Device chip temperature (such as read by the internal ADC);

[0058] – Wireless reception strength RSSI (such as sampled by Wi-Fi or LoRa modules);

[0059] – Network response delay (obtained from the time difference of sending ACK);

[0060] – Device load (such as CPU usage rate, memory occupancy), etc.

[0061] The data is periodically sent to the edge node through wireless protocols such as LoRa, Wi-Fi, Zigbee, etc. Each sampling of the data forms a set of x i , whose dimension is d, such as

[0062] First, the collected original data x i is standardized so that different types of sensor data can be learned under the same dimension. The standardization uses the Z-score method: subtract the mean of each dimension and divide by the standard deviation.

[0063] Subsequently, it enters the self-supervised encoding process. We use the SimCLR framework to generate sample pairs in the training stage: for the original data xi , two views are generated through two different data augmentation methods and as a positive sample pair. The augmentation methods include:

[0064] 1. Gaussian perturbation augmentation: Add Gaussian white noise with a mean of 0 and a standard deviation of σ =

[0065] 0.05 to each sensor dimension;

[0066] 2. Time window shearing: Crop sub - segments from the original time series to disrupt the sampling alignment.

[0067] The two augmented samples are respectively input into the encoder f θ to generate the corresponding state vectors:

[0068]

[0069] where:

[0070] f θ is a self - supervised encoder model with a structure of two - layer 1D convolution (the convolution kernel size of each layer is 3, and the number of channels is 64 / 128) + flattening + one - layer fully connected (128 - dimensional);

[0071] are state vectors generated under the same device, at the same moment but with different augmentation methods.

[0072] During the training process, the following contrast loss function is used to optimize the model to make the positive sample vectors close and the negative sample vectors far apart:

[0073]

[0074] where:

[0075] sim(·,·) represents the calculation of cosine similarity;

[0076] τ is the temperature coefficient, which controls the smoothness of the distribution;

[0077] z k is the state vector of other samples, used as negative samples.

[0078] After the training is completed, the encoder f θ is deployed to the edge node or the device side, and only the inference part is retained. During actual operation, each set of device data x i will be directly mapped to the state vector:

[0079] z i = f θ (x i ); (3)

[0080] Step output:

[0081] State vector z i : Represents the current operating state of device i, which is a dense embedding vector of dimension 128 and serves as the basis for subsequent anomaly recognition.

[0082] Step 2: Neighbor collaborative judgment and state anomaly recognition

[0083] In the Internet of Things wireless operation and maintenance system, devices are often deployed in complex environments such as industrial workshops, underground tunnels, and remote sites, where there are problems such as unstable communication links, device disconnection, and signal drift. In this case, relying on the device's own state upload for anomaly recognition is extremely fragile. Therefore, this step proposes a creative neighbor collaborative state judgment mechanism. By leveraging the state vectors of other communicable devices in the network and through state consensus analysis and similarity enhancement modeling, it determines whether the target device is in an abnormal state. Different from traditional centralized judgment or rule-based warning methods, this mechanism combines the geometric characteristics of the device state space and the network topology characteristics, and still has robust judgment ability under the conditions of unstable wireless links and missing device data.

[0084] Input:

[0085] Device state vector z i : Output from step one, generated by the encoder f θ and is the current state embedding vector of device i;

[0086] Set of neighbor device states Set of state vectors obtained through the broadcast of neighboring devices, and the adjacency relationship is dynamically determined by communicability (e.g., RSSI greater than the threshold, stable signal);

[0087] State propagation weight w ij : Represents the "consensus force" of neighbor j on the state of device i, determined based on signal strength, historical consistency, and an adaptive update mechanism.

[0088] Device i constructs its neighbor set by listening to the broadcast packets of surrounding devices during the current cycle Each neighbor device j broadcasts its state vector z j , and these vectors are from the trained state encoder f θ , and are transmitted after quantization and compression (e.g., through 8-bit fixed-point compression). After receiving them, device i constructs the neighbor state set {z j} and calculates the state consistency score between its own state and that of the neighbors.

[0089] We first define the anomaly score s of device i i as its state vector zi Weighted dissimilarity with neighbor device status:

[0090] Where:

[0091] z i is the current device status vector;

[0092] z j is the neighbor device status vector;

[0093] w ij is the status influence weight of neighbor j on device i, defined as:

[0094]

[0095] used to emphasize the greater influence of neighbors with stronger signals and higher stability on the judgment result;

[0096] λ is the regularization term coefficient, controlling the deviation penalty intensity;

[0097] Ω(z i ) is an innovative designed regularization term, used to capture the tension offset phenomenon in the state space, that is, whether there is a "pulling" trend of device i in the state space (the state deviates in multiple directions):

[0098]

[0099] It represents the overall pulling force of the device status deviating from the neighbors. If the neighbor status directions are highly dispersed, the value of Ω is large, which helps to determine "status drift" or "edge failure".

[0100] The creativity of the above design lies in:

[0101] introducing the "tension potential energy" of neighbor status as an abnormal regularization term, overcoming the limitation that the traditional similarity mean is insensitive to "boundary drift";

[0102] The status propagation weight w ij is very close to the actual Internet of Things scenario and is dynamically determined by RSSI, avoiding blind averaging.

[0103] Device i will compare the calculated anomaly score s i with the dynamic anomaly threshold δ i to determine whether it is abnormal currently. The threshold δ i no longer uses a static value, but introduces a local dynamic adaptive mechanism:

[0104] δ i = μ i + α·σ i ; (7)

[0105] Wherein:

[0106] μ i is the average of the anomaly scores in the past T cycles of i;

[0107] σ i is the standard deviation of the anomaly scores within this cycle;

[0108] α is a coefficient for adjusting the sensitivity (e.g., α = 2).

[0109] This design can dynamically adjust the threshold according to the historical state fluctuations of the device locally, enhancing the adaptability of the model to different operating modes and device types.

[0110] Step 3: Generation of operation and maintenance operation strategy

[0111] This step aims to further determine whether operation and maintenance operations are required based on the anomaly score s i calculated in the previous step and the state tension term Ω(z i ), and generate structured control instructions Different from the traditional strategy based on single threshold judgment, this step designs a hierarchical operation and maintenance strategy generation mechanism that combines the degree of anomaly, neighbor situation tension, and response cost perception, enabling the system to adaptively select the most appropriate control action under various abnormal situations (such as: instantaneous interference, continuous disconnection, state drift).

[0112] First, we define a comprehensive response intensity scoring function φ i , which is used to quantify the "urgency of control trigger" of the current state of the device. This function not only integrates the current anomaly score s i and the tension term Ω(z i ), but also adds a behavior trust adjustment term ψ i , which is used to capture the deviation degree between the historical behavior of the device and the current anomaly, forming a more stable strategy scoring basis. The scoring function is as follows:

[0113]

[0114] Wherein:

[0115] s i is the anomaly score;

[0116] Ω(z i ) is the state tension term;

[0117] ψ i is the behavior trust factor, and the calculation method is Where represents the average value of the device state in the past T time windows, reflecting the behavioral stability of the device;

[0118] β1, β2, and β3 are empirical weight coefficients that satisfy β1 + β2 + β3 = 1;

[0119] σ(·) is the Sigmoid function that maps the score to [0, 1], facilitating subsequent classification and policy mapping.

[0120] The creativity of this function lies in the introduction of the behavior stability feedback term ψ i , which prevents the device from being misjudged as abnormal during accidental fluctuations. Meanwhile, considering the state tension and the degree of abnormality, it is designed specifically for the actual scenario where "wireless devices may be occasionally disconnected but not necessarily failed".

[0121] Next, according to the value of φ i , the system will enter the policy mapping process. We designed a third-order response policy structure mapper that divides φ i into different control levels by intervals and introduced an action risk penalty term ρ(t i ), which is used to enhance the decision-making prudence in high-risk control scenarios.

[0122] The final policy objective function is:

[0123]

[0124] Where:

[0125] is the control instruction output, t i is the operation type (such as 0 = ignore, 1 = reconnect,

[0126] 2 = restart, 3 = enter the safe mode), p i is the additional parameter (such as waiting duration, affected module);

[0127] · is the system-predefined policy space;

[0128] is the target response level of each type of policy (such as the light policy θ1 = 0.3, the heavy policy θ3 = 0.8);

[0129] ρ(t i ) is the risk value corresponding to this operation (such as the restart operation brings a high risk of system interruption);

[0130] λ is the risk control coefficient, which prevents high-risk operations from being triggered when unnecessary;

[0131] Minimizing this function can obtain the most suitable operation policy that fits the current state.

[0132] The core innovation of this design lies in:

[0133] The response intensity score φi Instead of simple weighting, it considers the stability of historical states;

[0134] The strategy selection is a penalty-optimized process rather than a fixed-threshold trigger;

[0135] Introducing the action risk penalty as a constraint term for the first time in the patent reflects that "control is not necessarily the more intense the better", which is especially applicable to scenarios of embedded, low-power, and remote devices.

[0136] The system will select from the control strategy library such that the minimum combined output is the final instruction The instruction content includes the operation type t i and the parameter p i , which will serve as the core input for communication scheduling and instruction issuance in the next step.

[0137] Step output:

[0138] Control instruction represents the optimal operation and maintenance operation strategy and parameter combination of the current device for use in subsequent steps; this output is the central scheduling information in the system control chain and an important medium for realizing closed-loop control.

[0139] Step 4: Channel Prediction and Operation and Maintenance Path Scheduling

[0140] In the previous step, the system has generated the operation and maintenance control instruction for device i This control operation needs to be accurately issued to the target device through the wireless communication network. However, in the Internet of Things scenario, the communication link quality fluctuates greatly, especially in complex deployment environments such as industrial, field, and underground, where problems such as channel congestion, short-term interference, or the device being unable to receive are likely to occur. Therefore, the goal of this step is to dynamically select the best communication channel based on the type of control instruction and the network state prediction result, and issue the control instruction to device i to ensure the maximization of the instruction delivery success rate.

[0141] The communication module of device i (such as the SX1276 LoRa chip) will regularly scan the available channels and record the historical communication status of each channel. Taking LoRa as an example, the device can enable multiple frequency bands (such as 433 MHz, 470 MHz, etc.) and record whether the last K communications on each frequency band were successful.

[0142] For example, the failure record of channel f is where 1 represents a communication failure (such as not receiving an ACK), and 0 represents success.

[0143] The system predicts the channels using the moving average method:

[0144]

[0145] Wherein:

[0146] · is the availability prediction of device i for channel f in the current period;

[0147] · indicates whether the k-th recent communication has failed (1 means failure);

[0148] · The moving average ensures that the prediction is only based on recent situations and is not affected by long-term redundancy.

[0149] Considering the different urgencies of control operation types t i (such as t i = 2 indicates system restart with high failure cost), we introduce the control operation risk sensitivity ρ(t i ). Its value is preset by the system: for example, when the light control operation t i = 1, ρ(1) = 0.2, while when the heavy control operation t i = 2, ρ(2) = 0.8.

[0150] We define the following comprehensive scheduling scoring function:

[0151]

[0152] Wherein:

[0153] · indicates whether the most recent communication on this channel has failed;

[0154] · λ is the risk modulation factor, and it is recommended to set λ = 1.0 or adjust according to the system importance;

[0155] · The smaller the overall score, the more reliable the channel and the more suitable for issuing the current control instruction.

[0156] The system selects the channel with the smallest score:

[0157]

[0158] Finally, the control instruction will be sent to device i through channel by the communication module (such as the LoRa driver stack) to perform actions such as packing, sending, and waiting for ACK. If this channel fails, the system will continue to try in the sub-optimal channels for up to M times until successful or giving up beyond the threshold.

[0159] For example:

[0161] Suppose the control instruction Indicates that the current operation type is "system reboot"; parameter p i ={module: 3, delay: 5}, indicating that it acts on module 3 and is executed with a delay of 5 seconds; the currently available channels are The moving average of the failure rates of their last 5 communications is as follows

[0162]

[0163] Whether the last communication failed

[0164]

[0165] The operation risk coefficient ρ(2)=0.8 (because t i =2 is a reboot, which belongs to a high-risk operation);

[0166] The risk adjustment factor is set to λ = 1.0.

[0167] Next, score each channel:

[0168] Score of f1:

[0169]

[0170] Score of f2:

[0171]

[0172] Score of f3:

[0173]

[0174] Therefore, the lowest score is f1, that is:

[0175] The final system selects f1 as the sending channel of the control instruction and sends the instruction to the target device i.

[0176] Step 5: Instruction execution and status feedback update

[0177] In the previous step, the system has successfully sent the operation and maintenance control instruction through the channel to the target device i. This step is responsible for driving the device to actually execute the control operation and wirelessly feedback the execution result (success or failure) to the upper-level system, thus closing the entire operation and maintenance closed-loop.

[0178] After receiving the control instruction by device i, it will act according to the control type t in iti and parameter p i , and call the local control module for operation through a preset control mapping table.

[0179] For example:

[0180] If t i = 1 (communication reconnection), the device will re-initialize the communication stack (such as restarting the LoRa module);

[0181] If t i = 2 (module restart), then reset the specified module p i .module, and after the execution is completed, re-enter the main loop;

[0182] If t i = 3 (security mode switching), then write to the register flag bit to enter the low-power mode.

[0183] After the operation is completed, the device will build a feedback data packet through the same channel and return it to the edge node or the master control server. The feedback content includes instruction ID, device ID, operation type, execution result, etc. The system extracts the execution result flag bit in it and generates a feedback scalar r i :

[0184]

[0185] This feedback mechanism is implemented through ACK receipt or application layer heartbeat judgment at the communication protocol layer. For important control types (such as t i = 2), the system supports retransmitting N times within T seconds when no feedback is received until r i ≠ 0 or the retry threshold is exceeded.

[0186] The feedback mechanism on the device side is completed through a state machine and driver callbacks in software implementation, for example, by using the control thread and communication thread of FreeRTOS in cooperation. All feedback information will be sent to the central state database to drive the next round of judgment, control closed-loop or manual intervention reminder.

[0187] The above embodiments are only descriptions of the preferred embodiments of the present invention, and do not limit the scope of the present invention. Without departing from the design spirit of the present invention, various deformations and improvements made by those of ordinary skill in the art to the technical solution of the present invention shall fall within the protection scope determined by the claims of the present invention.

Claims

1. A wireless operation and maintenance method based on the Internet of Things, characterized in that Including: S1: Collect the original operation data through the Internet of Things device, and convert the original operation data into the current device state vector; S2: Combine the current device state vector and the neighbor device vector to obtain the anomaly score and the state tension term of the current device; S3: Construct a response intensity scoring function based on the anomaly score and the state tension term, and output the response intensity value through the response intensity scoring function; S4: Input the response intensity value into the structure mapper, and output the control instruction through the structure mapper; S5: Send the control instruction to the device through the communication channel; S6: The Internet of Things device executes the operation according to the control instruction and feeds back the execution result to the system.

2. The wireless operation and maintenance method based on the Internet of Things according to claim 1, characterized in that The expression of the anomaly score in S2 is: where s i is the anomaly score, z i is the current device state vector, z j is the neighbor device state vector, w ij is the state influence weight of neighbor device j on the current device i, λ is the regularization term coefficient; Ω(z i ) is the regularization term, is the set of neighbor devices.

3. The wireless operation and maintenance method based on the Internet of Things according to claim 1, wherein The expression of the state tension term in S2 is: where Ω(z i ) is the state tension term.

4. The wireless operation and maintenance method based on the Internet of Things according to claim 1, wherein The specific content of S1 includes: S1.1: Obtain the original operation data, and send the original operation data to the edge node periodically through the wireless protocol; S1.2: Standardize the original operation data through the edge node; S1.3: Perform data augmentation on the standardized original operation data to generate positive sample pairs; S1.4: Input the positive sample pairs into the encoder, and generate the current device state vector through the encoder.

5. The wireless operation and maintenance method based on the Internet of Things according to claim 1, characterized in that, The generation of the communication channel in S5 includes: Regularly scan the available channels through the communication module of the Internet of Things device, and record the historical communication status records of each channel; Based on the historical communication status records, predict multiple channels in a moving average manner; Input the prediction result into the comprehensive scheduling scoring function, and output the scoring results of multiple channels through the comprehensive scheduling scoring function; Select the channel with the smallest scoring value as the communication channel.

6. The wireless operation and maintenance method based on the Internet of Things according to claim 1, wherein The expression of the response intensity scoring function in S3 is: where ψ i is the behavior trust factor, is the average value of the device status in the past T time windows, and β1, β2, β3 are empirical weight coefficients, and σ(·) is the Sigmoid function.

7. The wireless operation and maintenance method based on the Internet of Things according to claim 1, characterized in that The specific content of S6 includes: S6.1: After receiving the control instruction, the Internet of Things device calls the local control module to execute the operation through the preset control mapping table; S6.2: After the operation is completed, the Internet of Things device constructs a feedback data packet through the same communication channel and returns the feedback data packet to the system; S6.3: The system receives the feedback data packet, extracts the execution result flag bit of the feedback data packet, and generates a feedback scalar.

8. The wireless operation and maintenance method based on the Internet of Things according to claim 7, characterized in that The feedback data packet in S6.2 includes the instruction ID, device ID, operation type, and execution result.

9. The wireless operation and maintenance method based on the Internet of Things according to claim 7, characterized in that, The original data in S1 includes voltage, device chip temperature, wireless reception intensity, network response delay, and device load.

10. A wireless operation and maintenance system based on the Internet of Things, the wireless operation and maintenance system being applied to the wireless operation and maintenance method based on the Internet of Things according to any one of claims 1-9, characterized in that, Including: Internet of Things device, used to collect the original operation data; Encoder, used to generate the state vector of the device; Structure mapper, used to divide the response intensity value into different control levels according to the interval and generate control instructions.

Images