Method and system for processing reporting peak shifting of Internet of Things equipment

Abstract

The invention discloses an internet of things equipment reporting peak shifting processing method and system, and relates to the technical field of internet of things communication and intelligent water affairs, and the method comprises the steps: predicting the handshake time delay of the next period through a heterogeneous parameter time delay alignment model based on the historical communication characteristics of a water meter; according to the server capability, constructing dynamic slot pressure-bearing characteristics representing the concurrent margin and priority of each slot; fusing the local pressure bearing feature corresponding to the initial slot position and the water meter priority to generate an end cloud heterogeneous state collaborative sensing tensor; when the concurrency is insufficient, inputting the tensor into a scheduling model to determine a target idle slot position; and finally, calculating new wakeup time according to the target slot position and the predicted time delay, asynchronously issuing a control message and feeding back parameters in a closed loop by relying on the shadow cache pool. The problems of concurrent scheduling deadlock and bottom layer space-time dislocation of massive equipment are effectively solved, and the delivery success rate of downlink messages is increased.

Description

Technical Field

[0001] This invention relates to the fields of Internet of Things (IoT) communication and smart water management technology, specifically to a method and system for handling peak reporting issues from IoT devices. Background Technology

[0002] With the deepening development of smart water management, a massive number of low-power wide-area network devices at the sensing layer need to frequently send data reports to a unified cloud platform. During specific business cycles or sudden events (such as peak water usage in the morning and evening, or pipeline leaks), a large number of devices often wake up and attempt to establish network connections simultaneously within the same time window. This sudden, large-scale concurrent reporting can instantly generate a huge communication signaling storm, which can easily lead to network congestion and exhaustion of radio frequency resources on the base station side, and can also directly break through the maximum concurrent processing capacity of the application server, causing large-scale data packet loss and system crashes. How to efficiently and accurately schedule the reporting sequence of massive heterogeneous IoT devices has become a technical bottleneck in the current process of refined management and large-scale evolution of smart water management platforms.

[0003] To address the aforementioned problem of massive concurrent device congestion, existing data staggered reporting technologies typically rely on two conventional solutions: one is a static discretization method based on terminal attributes, which uses the physical unique identifier of the terminal device (such as IMEI) and algorithms such as modulo to evenly and statically distribute the planned reporting time points of the devices, thereby reducing the concurrency conflict rate; the other is a dynamic adjustment method based on cloud load, which combines the average data processing time of the server with the current connection status, and the software platform issues delay commands to the devices according to the macro load balancing strategy to roughly adjust the next reporting time of the devices.

[0004] However, the aforementioned existing technologies still have the following significant drawbacks when dealing with complex and ever-changing real-world water management conditions: First, existing solutions typically rely on fixed historical average connection times to estimate the timing of command issuance. Under environmental disturbances such as increased battery resistance due to severe cold or decreased signal-to-noise ratio during peak communication periods, the actual network handshake time of the device exhibits drastic non-linear fluctuations, easily deviating from the extremely short RF listening window of low-power devices, leading to downlink control command delivery failures. Second, existing dynamic adjustment mechanisms generally adopt a rigid unidirectional traversal rule of delaying when a shortage occurs, and do not deeply couple server concurrency capacity with business priority. When a sudden pipe burst causes a surge in high-priority alarm data that exceeds the concurrency limit, later-arriving critical alarm data will be blindly postponed to a remote idle period, losing the real-time value of sudden warnings. Finally, existing technologies lack an asynchronous command alignment mechanism for deeply dormant devices. Blindly sending proactive paging signals on the server side is extremely wasteful of signaling resources and makes it difficult to drive a peak-shifting scheduling system to form an adaptive optimization closed loop.

[0005] Therefore, the present invention provides a method and system for handling reporting peak errors on Internet of Things (IoT) devices. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for handling reporting peaks on Internet of Things (IoT) devices, so as to solve the existing problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution: a method for handling peak reporting issues on Internet of Things (IoT) devices, comprising the following steps: S1. Based on the historical communication feature sequence of the target water meter, the predicted handshake delay of the target water meter in the next communication cycle is obtained by constructing a pre-trained heterogeneous parameter delay alignment model. S2. Based on the server's maximum concurrent capacity and the average processing time for a single data item, construct a multi-dimensional dynamic slot carrying capacity characteristic. The slot carrying capacity characteristic represents the concurrent capacity of each time slot and the service priority of the resident device. S3. Determine the initial slot index based on the identification code of the target water meter, the original wake-up time and the predicted handshake delay, and fuse the local bearing pressure characteristics corresponding to the initial slot index with the service priority of the target water meter to generate the edge-cloud heterogeneous state collaborative perception tensor. S4. When the concurrent margin at the initial slot index does not meet the access conditions, the edge-cloud heterogeneous state collaborative perception tensor is input into the intelligent game scheduling model to establish a target idle slot index that meets the carrying requirements. S5. Calculate the new wake-up time of the target water meter based on the target idle slot index and the predicted handshake delay, and asynchronously send a control message containing the new wake-up time to the target water meter through the shadow device cache pool, and provide parameter feedback to the heterogeneous parameter delay alignment model and the intelligent game scheduling model based on the sending result.

[0008] A further improvement of this invention is that the heterogeneous parametric delay alignment model is configured to obtain the predicted handshake delay of the target water meter in the next communication cycle, including: extracting the measured handshake time, battery voltage attenuation characteristics, and base station received signal reference power of the target water meter from N previous communication cycles within a preset sliding time window, performing time-series splicing, and constructing a multi-dimensional time-series input vector; the heterogeneous parametric delay alignment model inputs the multi-dimensional time-series input vector into the delay twin prediction model, and outputs the predicted handshake delay and dynamic handshake delay characteristics of the target water meter in the next communication cycle.

[0009] A further improvement of this invention lies in that the heterogeneous parametric delay alignment model is constructed based on a long short-term memory network, and its internal processing logic includes: using the forget gate mechanism within the model to filter out the random delay jitter feature weights caused by fast channel fading in the multi-dimensional time-series input vector; extracting the nonlinear delay growth trend caused by the battery voltage decay feature through the input gate and state update mechanism within the model; and transforming the nonlinear delay growth trend into the predicted handshake delay via a fully connected layer mapping. .

[0010] A further improvement of this invention lies in the fact that the edge-cloud heterogeneous state collaborative sensing tensor The acquisition process includes: calculating the initial expected mapping time based on the target water meter's identification code, original wake-up time, allowed deviation time, and predicted handshake delay, and mapping it to the initial slot index; extracting a local matrix fragment of the slot bearing pressure feature around the initial slot index as the local bearing pressure feature; and inputting the local bearing pressure feature, the target water meter's service priority, and the dynamic handshake delay feature into a state fusion function to generate the edge-cloud heterogeneous state collaborative perception tensor.

[0011] A further improvement of this invention is that the scheduling model is an intelligent game-theoretic scheduling model; the endpoint-cloud heterogeneous state collaborative perception tensor is input into the scheduling model to establish a target idle slot index that meets the carrying requirements, including: inputting the endpoint-cloud heterogeneous state collaborative perception tensor into the intelligent game-theoretic scheduling model, and the intelligent game-theoretic scheduling model outputting a target action instruction containing either a preemptive takeover or a neighboring zone shift; when the service priority of the target water meter is the highest-level alarm service, and the concurrency margin at the initial slot index is less than or equal to zero, the target action instruction is a preemptive takeover instruction to expel low-priority conventional devices in that slot; when the target water meter is a conventional service priority and encounters a concurrency limit, the target action instruction is a neighboring zone shift instruction, configured to traverse and find neighboring available idle slots with a value greater than zero as the target idle slot index.

[0012] A further improvement of this invention is that, after establishing the target available slot index that meets the carrying capacity requirements, it also includes a step of providing parameter feedback to the intelligent game-theoretic scheduling model, including: calculating the comprehensive off-peak scheduling adaptation index after executing the action command. Its calculation formula is expressed as: ;in, For the initial slot index, For the final determined target free slot index, The length of the dynamically concurrent capacity array. The number of devices to be evicted during a preemptive takeover. Initial slot index The total number of devices already stationed at the location. and The preset weighting coefficients are used to determine the comprehensive peak-shifting scheduling adaptation index. The reward feedback value is converted into a reinforcement learning value, and the intelligent game scheduling model is iterated, configured to maintain the action instructions output by the intelligent game scheduling model at a preset index. Within a monotonically increasing health range.

[0013] A further improvement of the present invention is that the target water meter delivery process includes: the server silently pushes a control message containing the new wake-up time into the shadow device cache pool of the core network; monitoring the status of the target water meter, and when the target water meter wakes up according to the original wake-up time and actively reports data within the communication receiving window, the cached control message and acknowledgment packet are merged and delivered to the target water meter.

[0014] A further improvement of this invention is that the allowable deviation time is dynamically set based on the service priority of the target water meter. When the service priority is the highest level alarm service, the allowable deviation time is less than or equal to a preset first duration threshold; when it is a regular meter reading service, the allowable deviation time is equal to a preset second duration threshold, and the first duration threshold is less than the second duration threshold.

[0015] On the other hand, the present invention provides an IoT device reporting peak handling system, comprising: The latency twin prediction module is configured to obtain the predicted handshake latency of the target water meter in the next communication cycle based on the historical communication feature sequence of the target water meter and through a heterogeneous parametric latency alignment model. The pressure-bearing characteristic construction module is configured to construct multi-dimensional dynamic slot pressure-bearing characteristics based on the server's maximum concurrency and the average processing time of a single data item. The pressure-bearing characteristics represent the concurrency margin of each time slot and the service priority of the resident device. The state tensor coupling module is configured to perform initial discretization based on the device identification code and original wake-up time of the target water meter, combined with the predicted handshake delay, to determine the initial slot index, and to integrate the local bearing pressure characteristics corresponding to the initial slot index with the service priority of the target water meter to generate a physical delay and service concurrency end-cloud heterogeneous state collaborative perception tensor. The intelligent game scheduling module is configured to input the edge-cloud heterogeneous state collaborative perception tensor into the intelligent game scheduling model when the concurrent margin at the initial slot index does not meet the access conditions. The intelligent game scheduling model then outputs a target action instruction containing either over-level preemption or adjacent-band translation to establish a target free slot index that meets the carrying requirements. The shadow instruction execution module is configured to calculate the new wake-up time of the target water meter based on the target free slot index and the predicted handshake delay, and asynchronously send a control message containing the new wake-up time to the target water meter through the shadow device cache pool, and provide parameter feedback to the heterogeneous parameter delay alignment model and the intelligent game scheduling model based on the sending result.

[0016] A further improvement of the present invention is that the time-delay twin prediction module includes a multi-dimensional time-series input unit and a prediction computing power unit; The multi-dimensional timing input unit is configured to concatenate the extracted measured handshake time, battery voltage attenuation features, and base station received signal reference power into a multi-dimensional timing input vector within a preset sliding time window. The heterogeneous parameter delay alignment model is deployed within the prediction computing unit. The heterogeneous parameter delay alignment model filters out the weights of accidental delay jitter features through a forget gate mechanism and extracts the nonlinear delay growth trend caused by the battery voltage decay feature through a state update mechanism.

[0017] Compared with the prior art, the beneficial effects of the present invention are: 1. This invention first constructs a heterogeneous parameter delay alignment model, extracts the historical handshake time of the water meter, battery attenuation, and signal power for multi-dimensional time-series extrapolation, and obtains accurate predicted handshake delay; it solves the problem of physical spatiotemporal misalignment of the calculation time due to drastic fluctuations in hardware and network latency caused by the sudden drop in temperature or communication peaks in the existing fixed-means scheme; it achieves precise alignment between downlink commands and the water meter's extremely short communication survival window, greatly improving the message delivery success rate and avoiding invalid reconnection and additional power consumption.

[0018] 2. By constructing a heterogeneous state collaborative perception tensor for edge and cloud environments and inputting it into an intelligent game-theoretic scheduling model for preemption or shifting decisions, the scheduling conflict problem caused by the rigid traversal rule of defaulting to follow when encountering a shortage, which leads to the blind postponement of high-priority alarm data and loss of real-time warning value, is solved under high-concurrency disturbances such as sudden pipe bursts. This achieves priority protection and flexible allocation of high-priority alarm data under extreme server pressure, ensuring that critical warnings are prioritized for access in a very short time and improving the system's emergency response capability.

[0019] 3. By using the shadow command asynchronous delivery mechanism, control messages are silently pushed into the core network cache database and then merged and sent when the water meter wakes up and actively reports at the original time. This solves the problem of huge signaling waste caused by the traditional server actively paging water meters in deep sleep, as well as the problem of uplink and downlink communication synchronization failure caused by the water meter's radio frequency shutdown. It realizes low-power off-peak command closed-loop delivery and seamless reset of water meter wake-up parameters, ensuring the stable operation of the off-peak allocation system. Attached Figure Description

[0020] Figure 1 This is a flowchart of a method for handling peak reporting errors on an IoT device according to the present invention; Figure 2 This is a framework diagram of an IoT device reporting peak handling system according to the present invention. Detailed Implementation

[0021] The technical solution of the present invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments of the present invention and the specific features in the embodiments are detailed descriptions of the technical solution of the present invention, rather than limitations thereof. In the absence of conflict, the embodiments of the present invention and the technical features in the embodiments can be combined with each other.

[0022] The term "and / or" is merely a description of the relationship between related objects, indicating that there can be three relationships. For example, A and / or B can represent three cases: A exists alone, A and B exist simultaneously, and B exists alone.

[0023] Example 1 Figure 1 This embodiment illustrates a flowchart of a method for handling peak reporting issues on IoT devices, with the following steps: In smart water management scenarios, a large number of low-power wide-area network (LPWAN) devices need to send staggered data reports to a unified server. Conventional staggered data reporting solutions often use a fixed historical average connection time to estimate the device wake-up time. In this conventional approach, the time taken for an IoT device to wake up, connect to the base station, and then establish a connection with the server is recorded in a collection. And calculate the average of the set. .

[0024] However, conventional averaging schemes have serious limitations. When the external temperature drops sharply, increasing the battery's internal resistance, or when mobile communication peak hours cause a decrease in the signal-to-noise ratio, these underlying hardware and environmental disturbances can cause the water meter wake-up and network handshake times to fluctuate dramatically and non-linearly. Using a fixed average latency will cause the communication point calculated by the server to deviate from the water meter's actual extremely short survival window, resulting in downlink control message delivery failure. To solve this underlying spatiotemporal misalignment problem, this embodiment introduces a heterogeneous parameter delay alignment model for accurate delay estimation.

[0025] Before applying the heterogeneous parameter delay alignment model online, an offline training phase is required. Specifically, the server pre-constructs a training set containing a massive amount of historical samples. The input feature matrix of each training sample strictly corresponds to the measured handshake latency sequence, battery voltage attenuation parameter, and base station received signal reference power sequence within the historical sliding time window. The true label for model training is set as the actual handshake latency of the device in the corresponding next real communication cycle. This scheme constructs a loss function based on mean squared error to supervise model learning. The mathematical expression of the loss function is: Where M is the batch training sample size. The predicted handshake delay is the output of the forward propagation of the heterogeneous parametric delay alignment model. To obtain the true label values, the backpropagation algorithm combined with an optimizer is used to iteratively update the gradients of the forget gate, input gate, and fully connected layer weights within the model until the loss function converges to below a preset accuracy threshold. This yields the pre-trained heterogeneous parameter delay-aligned model, ensuring a highly robust mathematical and physical foundation for the online inference process.

[0026] Specifically, within a preset sliding time window, the historical data of the target water meter is obtained. This is a sequence of communication parameters. This sequence includes the historical measured handshake times of the target water meter. to Battery voltage decay parameter and the base station received signal reference power The measured handshake time, battery voltage attenuation parameters, and base station received signal reference power are concatenated in time sequence to construct a multi-dimensional time sequence input vector. Its formula can be expressed as .

[0027] Multidimensional time-series input vector The input is fed into a heterogeneous parametric time-delay alignment model built on a long short-term memory network. The model's internal processing logic involves using a forgetting gate mechanism to filter out multi-dimensional temporal input vectors. The weight of the random delay jitter parameter caused by fast channel fading is used to bring the coefficients of the retention rate matrix of the identified discontinuous high-frequency disturbance features close to zero through the Sigmoid activation function, thereby suppressing the interference of the random delay jitter parameter on the long-term memory mainline state during cell state updates.

[0028] By utilizing the model's internal input gates and state update mechanism, parameters derived from battery voltage decay are extracted. This leads to a nonlinear latency growth trend. Through mapping at the fully connected layer, this nonlinear latency growth trend is transformed into the predicted handshake latency of the target water meter in the next communication cycle. and dynamic handshake latency attributes The preset sliding time window is preferably the most recent 30 days, with a historical sampling count of [number missing]. The optimal setting is fifty times, which effectively filters out short-term network disturbances and fully preserves long-term hardware aging trends.

[0029] In some possible embodiments, when smart water management equipment is located in outdoor pipe networks during the harsh winters of northern regions, the battery discharge capacity drops sharply due to the low temperature. Conventional fixed-means algorithms cannot detect this change at all, while the heterogeneous parametric delay alignment model of this invention can keenly capture the changes in the previous reports. The slight decrease and the increase in time allowed for an accurate deduction of the extended period due to the severe cold. This ensures that the basic parameters for subsequent scheduling operations do not deviate.

[0030] However, while the heterogeneous parameter delay alignment model addresses the communication alignment issue for individual water meters, a new system-level scheduling conflict arises. Due to the highly accurate predicted handshake delay, in the face of sudden application-layer pulse disturbances such as pipe bursts, a large number of alarm devices carrying the highest service priority will connect to the server within the same extremely narrow real-world time slice. This highly accurate concentration can instantly exceed the server's maximum concurrent processing capacity within the corresponding time period. Existing peak-shifting mechanisms typically employ a one-way traversal rule of delaying when encountering missing data when facing excessive concurrent capacity. This rigid blind callback allocation leads to high-priority alarm data arriving later being blindly postponed to remote idle time, resulting in the loss of the real-time value of sudden early warnings. To address this high-concurrency conflict caused by accurate prediction, this embodiment further introduces a multi-dimensional state dimensionality upgrade and tensor coupling mechanism.

[0031] First, the server collects the processing time of each piece of data from the device on the server and stores it in a collection. The unit is milliseconds, and the set is calculated. The average value is the average processing time. To facilitate subsequent time slot calculations, if If it is less than 1000 milliseconds, then Approximately 1 second; if If it is less than 2000 milliseconds, then Approximately 2 seconds, and so on. Then, based on the server's maximum concurrency... Average time for processing a single data entry To construct multi-dimensional dynamic slot pressure parameters, the specific steps are as follows: Initialize a dynamic concurrent capacity array of length L. The formula for calculating the array length L is: ,in This represents the allowed deviation time for the business logic itself, and its unit is set to minutes. This is a dynamic concurrency capacity array. It records the status of the sets of devices that went online within each time period, and its actual values ​​are expressed as follows: The data is presented in the form of numbers, where positive numbers indicate the number of devices still allowed to come online within that time period, and negative numbers indicate that the maximum number of concurrent devices has been exceeded. The number of devices within a given time period is counted using a separate counting thread on the server side. And calculate the concurrency margin for each time slot. The corresponding calculation formula is: Simultaneously, the highest service priority of the devices residing in each time slot is extracted. Concurrency margin With the highest business priority Perform overlay mapping to generate dynamic slot pressure parameters that include the two-dimensional states of all time slots. Subsequently, based on the target water meter's unique communication identifier IMEI and original wake-up time... Allowable deviation time and the predicted handshake delay obtained from the aforementioned calculations The initial expected mapping time is obtained through calculation, and the formula for calculating the initial expected mapping time is as follows: .

[0032] For example, if a device's original wake-up time is 12:00, it is allowed to report an offset time. Given a communication time of 5 minutes, a unique IMEI of 865042565189128, and a server connection time of 20 seconds, the initially allocated off-peak reporting time is calculated to be 12:02:40, meaning the device will begin self-wake-up at 12:02:40. The calculated initial expected mapping time is then directly converted into a dynamic concurrent capacity array. Initial slot index in Extract dynamic slot pressure parameters. In the initial slot index The surrounding local matrix fragments are compared with the target water meter's own business priority. and the aforementioned dynamic handshake latency attribute Common input state fusion function This generates a collaborative sensing tensor for heterogeneous state between the edge and cloud. .

[0033] To ensure the computability and dimensionality consistency of the generated edge-cloud heterogeneous state collaborative sensing tensor, the state fusion function $\Psi$ in this embodiment is specifically configured as a nonlinear mapping network containing a multidimensional tensor splicing operator and a multilayer perceptron. The specific generation process is as follows: First, the extracted local bearing pressure feature matrix fragment and the target water meter's own business priority constant are spliced ​​together using the multidimensional tensor splicing operator. and dynamic handshake latency attribute vector The features are concatenated along the feature dimension to form an initial one-dimensional joint feature vector. This initial one-dimensional joint feature vector is then input into a multilayer perceptron containing at least two hidden layers. A nonlinear activation function (such as ReLU) is used for high-dimensional feature extraction and dimensionality reduction mapping, ultimately outputting a heterogeneous state collaborative perception tensor with a fixed dimension. This nonlinear fusion structure effectively eliminates the dimensional differences between the physical layer's time parameters and the business layer's capacity parameters, preventing feature contamination caused by simple numerical addition.

[0034] At this point, when the system determines the initial slot index Concurrency margin at the location When the value is less than or equal to zero, or when the slot is locked by a high-priority service, the conventional postponement method will fail. This embodiment resolves this deadlock situation of rate limiting and preemption by introducing an intelligent game-theoretic scheduling model. The generated edge-cloud heterogeneous state collaborative sensing tensor... As a multidimensional state representation of the current system environment, it is directly input into the high-dimensional state perception layer of the intelligent game scheduling model. The intelligent game scheduling model preferably employs a deep reinforcement learning network, which is based on the business priority of the target water meter. Output the target action command within the action space. This is done when determining the service priority of the target water meter. This is the highest level alarm service, and the initial slot index is... When the concurrency margin is insufficient, the model outputs a preemption command to forcibly evict low-priority regular meter reading devices in that slot; when the target water meter is a regular business priority and also encounters the concurrency limit, the model outputs an adjacent band shift command to traverse and find nearby available free slots with a value greater than zero.

[0035] For the aforementioned intelligent game scheduling model, the underlying network architecture is preferably a Deep Q-Network (DQN) or a Proximal Policy Optimization Network (PPO). In this deep reinforcement learning framework, the model's input state space is the edge-cloud heterogeneous state collaborative perception tensor, while the model's action space is discretized and defined as a finite instruction set.

[0036] Specifically, when the model outputs a preemptive action command, the system's underlying command parser translates this abstract command into a forced sleep delay downlink control signaling for the low-priority device in the initial slot, unlocks its physical MAC address lock on the current time slot, and simultaneously writes the communication identifier of the high-priority alarm device into the slot's access whitelist; when outputting an adjacent band translation action command, the sliding step generator inside the scheduling model outputs the specific offset step distance, i.e. The absolute value is used to accurately locate and lock the target free slot index. By establishing clear network architecture constraints and mapping software and hardware action signaling, the physical executability of AI scheduling strategies at the underlying IoT communication protocol layer is fully guaranteed.

[0037] After executing the above action instructions, the system will establish a target available slot index that meets the load-bearing requirements. At this point, the system will adjust the dynamic concurrent capacity array. The value is adjusted synchronously, that is, the execution is performed. as well as The matrix update operation. To ensure the long-term rationality of scheduling and avoid falling into local scheduling failures or causing low-priority equipment to be continuously evicted, resulting in reporting stagnation, this scheme constructs a comprehensive peak-shifting scheduling adaptation index E, the calculation formula of which is expressed as follows: In this formula, This represents the aforementioned initial slot index. This represents the index of the final identified target free slot, where L is the length of the dynamic concurrent capacity array. This refers to the number of low-priority devices that were actually removed during a preemptive takeover attempt. Initial slot index The total number of devices already stationed at the location. and To preset the weight coefficients, the preferred values ​​are set to 0.6 and 0.4 respectively. This optimal setting is based on prioritizing the control of the time offset step size to ensure the arrival rate, while also considering the system disturbance rate caused by preemptive actions. The system converts the comprehensive off-peak scheduling adaptation index E into a reward feedback value for reinforcement learning, and continuously updates the weights of the intelligent game scheduling model through a gradient ascent strategy, ensuring that the action commands output by the model always remain within the monotonically increasing optimization interval of the comprehensive off-peak scheduling adaptation index E. If, after traversing and exploring the entire domain of the scheduling model, the dynamic concurrent capacity array is found... If all slot data is less than zero, it means that the single server's capacity limit has been reached. At this point, the system will adopt a load balancing strategy and deploy an additional server to manage device communication.

[0038] In some possible implementations, for example, if a main pipe bursts in an older residential area, hundreds of the highest-priority alarm data messages flood in instantly, exhausting the server's concurrent pool for the corresponding time period. Traditional queuing mechanisms can only keep these alarm data messages, which are crucial to the safety of the pipeline network, waiting in the queue for tens of minutes. However, the intelligent game-theoretic scheduling model of this invention can instantly activate a preemptive mechanism in this scenario, removing all regular monthly water meter readings from the current slot and properly relocating them to a distant slot, ensuring that hundreds of pipe burst alarms are processed within the allowable deviation time of just a few seconds. All data has been prioritized for access and reporting, greatly ensuring the timeliness of urban water monitoring data flow.

[0039] The target free slot index is determined through an intelligent game scheduling model. The next crucial step is to seamlessly synchronize the new scheduling results to the target water meter. Since water meters generally operate in low-power mode and are in a deep sleep state for extended periods, conventional server-side active wake-up mechanisms are not only extremely resource-intensive but also likely to fail to synchronize because the water meter is not in radio frequency monitoring mode. Therefore, this embodiment constructs a shadow command asynchronous delivery mechanism to completely solve this uplink-downlink synchronization problem.

[0040] First, based on the established target free slot index Predicting handshake latency Calculate the new wake-up time of the target water meter. The calculation formula is expressed as follows: .

[0041] For example, if the original self-wake-up time of the target water meter is 12:00, the reported offset time is allowed. The time is 5 minutes, and the connection time to the server is 20 seconds, when the target slot index is established by the scheduler. When the value is 3 and the server processing time is 2 seconds, the new self-wake-up time is calculated as 12:00 minus 5 minutes minus 20 seconds plus 2 seconds multiplied by 3, resulting in a final established time of 11:54:46. The system records this new self-wake-up reporting time, which contains a unique identifier, into the set. Persistent storage is performed within this process. The new wake-up time is then calculated. After that, the server will never initiate an active paging process, but will instead send a message containing the new wake-up time. The control messages are silently pushed into the shadow device cache pool of the core network for temporary storage. The server continuously checks whether the device is online; if it is offline, it sends a delay command and continuously monitors the status of the target water meter. When the target water meter wakes up according to the original wake-up time... When the device wakes up and proactively reports its business data within a very short paging listening window, the server immediately merges and sends the control message temporarily stored in the shadow device's cache pool with the network confirmation packet, delivering the command to reset the self-wake-up time to the target water meter. Upon receiving the merged packet, the target water meter automatically resets its internal clock wake-up parameters. Simultaneously, the system extracts the actual latency deviation during this command delivery process as a feedback penalty, feeding it back to the heterogeneous parameter latency alignment model and the intelligent game scheduling model, completing the closed-loop iteration of model parameters across the entire business process. This process, through the extremely short receiving window inherent in the low-power wide-area network protocol, not only achieves command delivery with low additional power consumption but also eliminates dependence on static parameters by relying on a real feedback loop, ensuring continuous optimization and stable operation of subsequent massive water meter allocation.

[0042] The threshold and weight settings involved in this embodiment can be set by default according to the present invention, or can be set by those skilled in the art.

[0043] Example 2 Figure 2 This invention illustrates a framework diagram of an error reporting and peak handling system for IoT devices. Based on the same inventive concept as Embodiment 1, this invention provides an error reporting and peak handling system for IoT devices, comprising: The delay twin prediction module is used to obtain the predicted handshake delay of the target water meter in the next communication cycle based on the historical communication feature sequence of the target water meter and through a heterogeneous parametric delay alignment model. The time-delay twin prediction module includes a multi-dimensional time-series input unit and a prediction computing power unit; The multi-dimensional timing input unit is configured to concatenate the extracted measured handshake time, battery voltage attenuation features, and base station received signal reference power into a multi-dimensional timing input vector within a preset sliding time window. The heterogeneous parameter delay alignment model is deployed within the prediction computing unit. The heterogeneous parameter delay alignment model filters out the weights of accidental delay jitter features through a forget gate mechanism and extracts the nonlinear delay growth trend caused by the battery voltage decay feature through a state update mechanism.

[0044] The pressure-bearing characteristic construction module is configured to construct multi-dimensional dynamic slot-bearing pressure characteristics based on the server's maximum concurrency and the average processing time for a single data entry. These pressure-bearing characteristics characterize the concurrency margin of each time slot and the service priority of the residing device; including: The address space initialization unit is configured to allocate a dynamic concurrent capacity array of length L in memory based on the allowable deviation time of the service and the average processing time of a single data item.

[0045] The concurrent capacity statistics unit is configured to call an independent counting thread to count the total number of devices allocated in each time slot in real time. And combined with the maximum number of concurrent users Calculate the concurrency margin .

[0046] Priority feature appending unit is configured to extract the highest service priority of the device residing in each time slot. Combine it with the corresponding concurrency margin Integration and Upgraded Output of Dynamic Slot Pressure Parameters .

[0047] The state tensor coupling module is configured to perform initial discretization based on the device identifier code and original wake-up time of the target water meter, combined with the predicted handshake delay, to determine the initial slot index. It then integrates the local load-bearing pressure characteristics corresponding to the initial slot index with the service priority of the target water meter to generate a physical delay and service concurrency end-cloud heterogeneous state collaborative perception tensor; including: The baseline discrete mapping unit is configured to obtain the initial expected mapping time and is positioned as a dynamic concurrent capacity array. Initial slot index in .

[0048] The local feature extraction unit is configured to use the initial slot index. Centered on the dynamic pressure parameters of the slot, extract the parameters. A local matrix fragment.

[0049] State tensor fusion unit with built-in state fusion function This is used to correlate the local matrix fragment with the target water meter's own service priority. and dynamic handshake delay parameters The data is then spliced ​​together to generate a tensor for collaborative perception of hardware latency and business concurrency between the cloud and the edge.

[0050] The intelligent game scheduling module is configured to input the edge-cloud heterogeneous state collaborative perception tensor into the intelligent game scheduling model when the concurrent reserve at the initial slot index does not meet the access conditions. The intelligent game scheduling model then outputs a target action command containing either preemption or adjacent-band shift to establish a target free slot index that meets the carrying requirements. This includes: The high-dimensional state perception unit is configured to receive the edge-cloud heterogeneous state collaborative perception tensor output by the environment tensor coupling module and transform it into the current environment state space input that can be recognized by the reinforcement learning network.

[0051] The strategic game decision-making unit has built-in action space mapping logic. When the target water meter is an alarm service and the initial slot availability is insufficient, it forcibly outputs a preemptive command; when it is a regular service and the availability is insufficient, it outputs an adjacent band shift command, thereby establishing the target free slot index. .

[0052] The scheduling adaptation evolution unit has a built-in index evaluation function to calculate the comprehensive peak-shifting scheduling adaptation index and generate reward feedback values. It maintains the optimization direction of the decision layer through gradient updates.

[0053] The shadow instruction execution module is configured to calculate the new wake-up time of the target water meter based on the target free slot index and the predicted handshake delay, and asynchronously send a control message containing the new wake-up time to the target water meter through the shadow device cache pool. Based on the sending result, it provides parameter feedback to the heterogeneous parameter delay alignment model and the intelligent game scheduling model. This includes: The wake-up time reset unit is configured to be based on the target free slot index established by the scheduler. By predicting the handshake delay, a new wake-up time is derived.

[0054] The shadow asynchronous buffer unit includes a control message buffer interface that interfaces with the core network. It is configured to silently store control messages during the water meter's dormant period and merge the message with the network acknowledgment packet and send it out within a very short radio frequency reception window when the water meter actively reports data.

[0055] The closed-loop feedback routing unit is configured to extract the actual time delay deviation data as a penalty parameter after the instruction is issued, and then route it back to the time delay twin prediction module and the intelligent game scheduling module for network weight iteration.

[0056] It is important to note that the heterogeneous parametric delay alignment model and the intelligent game scheduling model constructed in this invention are not two isolated data processing modules, but rather constitute an inseparable cross-level causal control chain and a deeply coupled linkage mechanism. In conventional peak-shifting architectures, time deviation and concurrent scheduling typically operate as independent branch logics; however, in this invention, it is precisely because of the underlying physical fields, such as extreme low temperatures causing battery voltage decay, or rapid channel fading deep within the pipeline causing severe fluctuations in RSRP, that the heterogeneous parametric delay alignment model can accurately extract and output the dynamic handshake delay characteristics representing these fluctuations.

[0057] This dynamic handshake latency feature, as the core driving dimension, is deeply integrated into the edge-cloud heterogeneous state collaborative perception tensor. In other words, it is precisely because of the anomalies in the underlying physical field that the expected latency shift occurs. This forced misalignment in space and time directly causes the originally assigned slot of the target water meter to drift, forcibly pushing it into the congested time slot where the concurrent capacity is exhausted, thus triggering... The extreme conditions not only force dynamic changes in the mapping relationship of the upper-level concurrency matrix, but also, as the core physical precondition, directly trigger the execution of the preemptive or adjacent-band shift mechanism in the intelligent game scheduling model.

[0058] This invention, through the aforementioned mechanism, implements a logic for triggering high-dimensional game-theoretic scheduling and preemption based on the abrupt changes in the state tensor characteristics of the cloud at the bottom layer, caused by changes in underlying physical parameters and handshake delays. The latency derivation results at the bottom layer not only form the state foundation for the top-level scheduling, but the action feedback from the top-level scheduling further corrects the loop through a shadow buffer pool. This deep coupling and linkage between the two in terms of function and data flow overcomes the discrepancy between traditional underlying physical communication protocols and upper-layer software load balancing.

[0059] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0060] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0061] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0062] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0063] The embodiments of the present invention have been described above with reference to the accompanying drawings. However, the present invention is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of the present invention without departing from the spirit and scope of the claims. All of these forms are within the protection scope of the present invention.

Claims

1. A method for handling peak reporting issues on Internet of Things (IoT) devices, characterized in that: Includes the following steps: S1. Based on the historical communication feature sequence of the target water meter, the predicted handshake delay of the target water meter in the next communication cycle is obtained by constructing a pre-trained heterogeneous parameter delay alignment model. S2. Based on the server's maximum concurrent capacity and the average processing time for a single data item, construct a multi-dimensional dynamic slot carrying capacity characteristic. The slot carrying capacity characteristic represents the concurrent capacity of each time slot and the service priority of the resident device. S3. Determine the initial slot index based on the identification code of the target water meter, the original wake-up time and the predicted handshake delay, and fuse the local bearing pressure characteristics corresponding to the initial slot index with the service priority of the target water meter to generate the edge-cloud heterogeneous state collaborative perception tensor. S4. When the concurrent margin at the initial slot index does not meet the access conditions, the edge-cloud heterogeneous state collaborative perception tensor is input into the intelligent game scheduling model to establish a target idle slot index that meets the carrying requirements. S5. Calculate the new wake-up time of the target water meter based on the target idle slot index and the predicted handshake delay, and asynchronously send a control message containing the new wake-up time to the target water meter through the shadow device cache pool, and provide parameter feedback to the heterogeneous parameter delay alignment model and the intelligent game scheduling model based on the sending result.

2. The method for handling peak reporting issues on IoT devices according to claim 1, characterized in that: The heterogeneous parametric delay alignment model is configured to obtain the predicted handshake delay of the target water meter in the next communication cycle, including: extracting the measured handshake time, battery voltage attenuation characteristics, and base station received signal reference power of the target water meter in the past N times within a preset sliding time window, performing time-series splicing, and constructing a multi-dimensional time-series input vector; the heterogeneous parametric delay alignment model inputs the multi-dimensional time-series input vector into the delay twin prediction model, and outputs the predicted handshake delay and dynamic handshake delay characteristics of the target water meter in the next communication cycle.

3. The method for handling peak reporting issues on IoT devices according to claim 2, characterized in that: The heterogeneous parametric delay alignment model is built on a long short-term memory network. Its internal processing logic includes: using the forget gate mechanism within the model to filter out the random delay jitter feature weights caused by fast channel fading in the multi-dimensional time-series input vector; extracting the nonlinear delay growth trend caused by the battery voltage decay feature through the input gate and state update mechanism within the model; and transforming the nonlinear delay growth trend into the predicted handshake delay via a fully connected layer mapping. .

4. The method for handling peak reporting errors on IoT devices according to claim 2, characterized in that: The edge-cloud heterogeneous state collaborative sensing tensor The acquisition process includes: calculating the initial expected mapping time based on the target water meter's identification code, original wake-up time, allowed deviation time, and predicted handshake delay, and mapping it to the initial slot index; extracting a local matrix fragment of the slot bearing pressure feature around the initial slot index as the local bearing pressure feature; and inputting the local bearing pressure feature, the target water meter's service priority, and the dynamic handshake delay feature into a state fusion function to generate the edge-cloud heterogeneous state collaborative perception tensor.

5. The method for handling peak reporting issues on IoT devices according to claim 1, characterized in that: The scheduling model is an intelligent game-theoretic scheduling model. The process involves inputting the edge-cloud heterogeneous state collaborative perception tensor into the scheduling model to establish a target idle slot index that meets the carrying requirements. This includes: inputting the edge-cloud heterogeneous state collaborative perception tensor into the intelligent game-theoretic scheduling model, and having the intelligent game-theoretic scheduling model output a target action command containing either a preemptive takeover or a neighboring zone shift. When the target water meter's service priority is the highest-level alarm service, and the concurrent reserve at the initial slot index is less than or equal to zero, the target action command is a preemptive takeover command to expel low-priority conventional devices in that slot. When the target water meter has a conventional service priority and encounters a concurrency limit, the target action command is a neighboring zone shift command, configured to traverse and find adjacent available idle slots with a value greater than zero as the target idle slot index.

6. A method for handling peak reporting issues on IoT devices according to claim 5, characterized in that: After establishing the target available slot index that meets the carrying capacity requirements, the process also includes a step of providing parameter feedback to the intelligent game scheduling model, including: calculating the comprehensive off-peak scheduling adaptation index after executing the action command. Its calculation formula is expressed as: ;in, For the initial slot index, For the final determined target free slot index, The length of the dynamically concurrent capacity array. The number of devices to be evicted during a preemptive takeover. Initial slot index The total number of devices already stationed at the location. and The preset weighting coefficients are used to determine the comprehensive peak-shifting scheduling adaptation index. The reward feedback value is converted into a reinforcement learning value, and the intelligent game scheduling model is iterated, configured to maintain the action instructions output by the intelligent game scheduling model at a preset index. Within a monotonically increasing health range.

7. The method for handling peak reporting issues on IoT devices according to claim 1, characterized in that: The target water meter delivery process includes: the server silently pushing a control message containing the new wake-up time into the shadow device cache pool of the core network; monitoring the status of the target water meter; and when the target water meter wakes up according to the original wake-up time and actively reports data within the communication receiving window, merging the cached control message and confirmation packet and sending them to the target water meter.

8. A method for handling peak reporting issues on IoT devices according to claim 4, characterized in that: The allowable deviation time is dynamically set based on the service priority of the target water meter. When the service priority is the highest level alarm service, the allowable deviation time is less than or equal to a preset first duration threshold. When it is a regular meter reading service, the allowable deviation time is equal to a preset second duration threshold, and the first duration threshold is less than the second duration threshold.

9. A peak reporting processing system for Internet of Things (IoT) devices, used to execute a peak reporting processing method for IoT devices as described in any one of claims 1-8, characterized in that: include: The latency twin prediction module is configured to obtain the predicted handshake latency of the target water meter in the next communication cycle based on the historical communication feature sequence of the target water meter and through a heterogeneous parametric latency alignment model. The pressure-bearing characteristic construction module is configured to construct multi-dimensional dynamic slot pressure-bearing characteristics based on the server's maximum concurrency and the average processing time of a single data item. The pressure-bearing characteristics represent the concurrency margin of each time slot and the service priority of the resident device. The state tensor coupling module is configured to perform initial discretization based on the device identification code and original wake-up time of the target water meter, combined with the predicted handshake delay, to determine the initial slot index, and to integrate the local bearing pressure characteristics corresponding to the initial slot index with the service priority of the target water meter to generate a physical delay and service concurrency end-cloud heterogeneous state collaborative perception tensor. The intelligent game scheduling module is configured to input the edge-cloud heterogeneous state collaborative perception tensor into the intelligent game scheduling model when the concurrent margin at the initial slot index does not meet the access conditions. The intelligent game scheduling model then outputs a target action instruction containing either over-level preemption or adjacent-band translation to establish a target free slot index that meets the carrying requirements. The shadow instruction execution module is configured to calculate the new wake-up time of the target water meter based on the target free slot index and the predicted handshake delay, and asynchronously send a control message containing the new wake-up time to the target water meter through the shadow device cache pool, and provide parameter feedback to the heterogeneous parameter delay alignment model and the intelligent game scheduling model based on the sending result.

10. The IoT device reporting peak-shifting processing system according to claim 9, characterized in that: The time-delay twin prediction module includes a multi-dimensional time-series input unit and a prediction computing power unit; The multi-dimensional timing input unit is configured to concatenate the extracted measured handshake time, battery voltage attenuation features, and base station received signal reference power into a multi-dimensional timing input vector within a preset sliding time window. The heterogeneous parameter delay alignment model is deployed within the prediction computing unit. The heterogeneous parameter delay alignment model filters out the weights of accidental delay jitter features through a forget gate mechanism and extracts the nonlinear delay growth trend caused by the battery voltage decay feature through a state update mechanism.

Images