Dynamic data updating method and system based on internet of things platform

Abstract

The application relates to the technical field of Internet of Things communication, in particular to a dynamic data updating method and system based on an Internet of Things platform. The dynamic data updating method based on the Internet of Things platform comprises the following steps: S1: constructing a state evolution perception unit in an edge gateway, maintaining a three-dimensional state evolution model for each device, and collecting device state data in real time, wherein an algorithm is used to construct a state evolution curve according to the device state data, and the state evolution curve comprises a real-time curve, a confirmation curve and an evolution curve; S2: obtaining a local curvature change rate and a prediction error bandwidth based on the state evolution curve, and identifying a key state node according to the local curvature change rate and the prediction error bandwidth. Through the three-dimensional state evolution model and the lightweight time sequence prediction constructed on the edge side from the state snapshot to the evolution track, the offline state is changed from unknowable to predictable and traceable, and the continuity and reliability of the state data are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of Internet of Things (IoT) communication technology, and in particular to a dynamic data update method and system based on an IoT platform. Background Technology

[0002] In IoT applications, device status data is typically aggregated and synchronized to the cloud via edge gateways, providing a unified status view for upper-layer applications through digital twins or device shadowing mechanisms. However, existing IoT shadowing mechanisms heavily rely on a continuous and stable network connection. When devices are in a weak network, experience disconnections, or experience fluctuations in API service quality, cloud status updates are forced to stop, causing the shadow status to remain at the snapshot taken just before the disconnection. During this period, the application layer can only access outdated and static device status, unable to perceive the actual operational trends of the device while it is offline, making it difficult to support the high real-time and high-reliability business requirements of real-time monitoring, predictive maintenance, and remote control.

[0003] Traditional solutions often employ a passive compensation mechanism combining local caching and delayed synchronization. This involves uploading accumulated offline state data in batches to the cloud for overwriting updates after network recovery. This approach only provides post-event data replenishment and cannot provide timely and reliable state information to the application layer during offline periods. More critically, existing solutions generally lack online modeling capabilities for device state evolution trends. They fail to proactively predict future device states during the disconnection window and lack a dynamic evaluation and feedback mechanism for the reliability of prediction results. This results in a triple dilemma of "unknowable, unreliable, and unusable" state data during offline periods, severely hindering the application of IoT systems in fields with extremely high requirements for state continuity and data reliability, such as industrial control, connected vehicles, and smart healthcare.

[0004] Therefore, there is an urgent need to develop a dynamic data update method and system based on the Internet of Things platform, which can maintain the ability to perceive, predict and trust the evolution of device status even when the connection is lost, and realize a paradigm upgrade from passive supplementation to active inference, from single snapshot to dynamic trajectory, and from indiscriminate synchronization to differentiated integration. Summary of the Invention

[0005] To overcome the shortcomings of existing IoT shadow mechanisms, such as interrupted cloud state updates and the application layer only being able to obtain outdated snapshots when the network is weak or disconnected, which prevent the perception of the actual evolution trend of devices during offline periods, and the lack of continuous state modeling and dynamic feedback on prediction reliability in traditional passive completion schemes, resulting in data being unknowable, unreliable, and unusable during offline periods, this invention provides a dynamic data update method and system based on an IoT platform.

[0006] The technical solution is as follows: A dynamic data update method based on an IoT platform, including the following steps:

[0007] S1: Construct a state evolution perception unit in the edge gateway, maintain a three-dimensional state evolution model for each device, and collect device state data in real time. Use an algorithm to construct a state evolution curve based on the device state data. The state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve.

[0008] S2: Based on the state evolution curve, obtain the local curvature change rate and prediction error bandwidth, and identify key state nodes according to the local curvature change rate and prediction error bandwidth;

[0009] S3: When a decline in API connection quality is detected or a device enters a weak network area, the state evolution extrapolation model is activated. A lightweight time series prediction model is used to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories. The multi-step state extrapolation trajectories are then compared with the latest evolution curve in the cloud to generate dynamic confidence labels.

[0010] S4: Generate a prediction summary based on the multi-step state extrapolation trajectory and confidence level identifier, perform a multi-channel attempt synchronization strategy, synchronize the prediction summary to the cloud, and perform trajectory fusion with the cloud evolution curve to obtain a fused evolution curve;

[0011] S5: The cloud continuously sends requests to the API to detect the connection status according to the set time threshold. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded and trajectory alignment and difference fusion are performed.

[0012] Preferably, the step of constructing a state evolution perception unit in the edge gateway maintains a three-dimensional state evolution model for each device and collects device state data in real time. An algorithm is used to construct a state evolution curve based on the device state data. The state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve. Specifically: the three-dimensional state evolution model is a model that maintains the real-time curve, confirmation curve, and evolution curve; the real-time curve is a smoothed sequence of instantaneous state values ​​after filtering and denoising; the confirmation curve is a sequence of state snapshots submitted and confirmed in two stages via the cloud; the evolution curve is based on a long short-term memory network prediction model, using historical window data as input, and outputting predicted state values ​​and their confidence intervals for multiple future time steps to form a prediction trajectory; the device state data time series is collected through a sliding window, and a trend extraction algorithm is used to fit the state change pattern; the device state data includes instantaneous state values, confirmed state values, evolution data, and historical window data.

[0013] Preferably, the step of obtaining the local curvature change rate and prediction error band width based on the state evolution curve, and identifying key state nodes based on the local curvature change rate and prediction error band width, includes: calculating the first and second derivatives of each point based on three adjacent points on the evolution curve using the central difference method, and substituting them into the curvature formula. Obtain local curvature values The ratio of the difference between adjacent curvature values ​​to the corresponding time interval is calculated to obtain the local curvature change rate. Based on the probability distribution output by the time series prediction model, the upper and lower limits of the confidence interval for each prediction time are extracted, and the prediction error band width is the difference between the upper and lower limits of the confidence interval. Key state nodes are obtained based on the obtained local curvature change rate and prediction error band width. The key state nodes include trend inflection points, prediction divergence nodes, and highly sensitive state nodes.

[0014] Preferably, the step of obtaining key state nodes based on the obtained local curvature change rate and prediction error band width, wherein the key state nodes include trend inflection points, prediction divergence nodes, and highly sensitive state nodes, includes: if the local curvature change rate exceeds a preset first threshold, then the corresponding state point is determined to be a trend inflection point; if the prediction error band width exceeds a preset second threshold, then the corresponding state point is determined to be a prediction divergence node; if the trend inflection point, prediction divergence node, or instantaneous state value exceeds the corresponding preset safe operating range, then the corresponding state point is determined to be a highly sensitive state node.

[0015] Preferably, when a decrease in API connection quality or a device entering a weak network area is detected, the state evolution extrapolation model is activated. A lightweight time-series prediction model is used to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories. The multi-step state extrapolation trajectories are then compared with the latest evolution curve in the cloud to generate dynamic confidence indicators. This includes: for trend inflection points, extracting evolution data within the shortest effective window after the inflection point, using adaptive exponential smoothing to quickly fit the new trend direction, and performing only short-step extrapolation to avoid error accumulation; for prediction divergence nodes, when the prediction error band width exceeds the extrapolation threshold, the model... The extrapolation step size is automatically reduced to half of the default value, and a conservative prediction mode is switched. At the same time, a Naive Bayes model is enabled for parallel extrapolation, and the result with a narrow confidence interval is selected as the final output. For highly sensitive state nodes, a multi-model ensemble extrapolation strategy is enabled, and a long short-term memory network, a Holt-Winters model, and a simple moving average model are run simultaneously. The extrapolation trajectory is dynamically weighted and fused based on the inverse error of each model in the previous prediction window. Multi-step state extrapolation is performed on key state nodes on the local evolution curve to obtain multi-step state extrapolation trajectory. The confidence level is obtained by comparing the multi-step state extrapolation trajectory with the latest evolution curve in the cloud.

[0016] Preferably, the step of obtaining a confidence level identifier by comparing the multi-step state extrapolation trajectory with the latest evolution curve in the cloud includes: the confidence level identifier includes a confidence score and a confidence level; the confidence level includes a first confidence level, a second confidence level, and a third confidence level; when the confidence score is higher than a first confidence threshold, it is rated as the first confidence level; when the confidence score is higher than a second confidence threshold but lower than the first confidence threshold, it is rated as the second confidence level; and when the confidence score is lower than the second confidence threshold, it is rated as the third confidence level; the standard deviation of the historical prediction error in the cloud is defined as... The confidence score is obtained through the confidence scoring formula, which is:

[0017] ;

[0018] in Score the confidence level; The trajectory vector for the time period corresponding to the multi-step state extrapolation trajectory; This is the trajectory vector corresponding to the latest evolution curve in the cloud for the corresponding time period; Dynamic time-warped distance is used as the distance metric for trajectory similarity. As a regulating factor; To prevent extremely small constants with a denominator of zero.

[0019] Preferably, the step of generating a prediction summary based on the multi-step state extrapolation trajectory and confidence level identifier, performing a multi-channel attempt synchronization strategy, synchronizing the prediction summary to the cloud, and fusing it with the cloud evolution curve to obtain a fused evolution curve includes: compressing the multi-step state extrapolation trajectory, confidence score, and confidence level to obtain the prediction summary, and then synchronizing and updating it to the cloud database using a multi-channel attempt synchronization strategy; the multi-channel attempt synchronization strategy includes a first-priority low-bandwidth backup channel strategy. If the first-priority attempt succeeds, the prediction summary is sent; if it fails, it enters a second-priority device direct connection relay strategy. In this case, an adjacent online device is found as a relay to send the prediction summary through short-distance communication, and the online device forwards it to the cloud database. If it fails, it enters a third-priority storage and waiting timed window strategy. In this case, the next prediction time window is defined and added to the prediction summary, and the prediction summary is stored in the edge gateway persistent queue, waiting for the next prediction time window to arrive to try the first-priority low-bandwidth backup channel strategy; when the prediction summary arrives at the cloud, the fused evolution curve is obtained through the cloud trajectory fusion model.

[0020] Preferably, the step of obtaining the fusion evolution curve through the cloud trajectory fusion model after the predicted summary reaches the cloud includes: the cloud trajectory fusion model takes the multi-step state extrapolation trajectory in the predicted summary and the cloud evolution curve as input, uses the Kalman filter algorithm to perform optimal state estimation, dynamically maps the confidence score in the predicted summary to the weight factor of the observation noise covariance matrix, the higher the confidence score, the more the fusion result is biased towards the multi-step state extrapolation trajectory, and the fusion state value at each time moment is generated by recursively calculating by minimizing the prediction error covariance, forming a continuous fusion evolution curve.

[0021] Preferably, the cloud continuously sends requests to the API to detect the connection status according to a set time threshold. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level identifier. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded and trajectory alignment and difference fusion are performed, including: when the connection status is abnormal, if the confidence level is the first confidence level, the interface displays the fused status value; if the confidence level is the second confidence level, the interface simultaneously displays the fused status value and its prediction error band width; if the confidence level is the third confidence level, the interface prioritizes displaying the last confirmed status value. The system overlays and displays the multi-step state extrapolation trajectory, prediction error band width, and prediction divergence warning indicators. When the connection is restored, the edge gateway continuously records the state evolution curve during the offline period and uploads the complete trajectory segment after the network is restored. The cloud uses a timestamp alignment algorithm to match the multi-step state extrapolation trajectory with the cloud evolution curve point by point within the overlapping time window, calculates the average absolute deviation of the state value at each moment as the state deviation of the overlapping area, and automatically updates the cloud evolution curve to the edge trajectory when the state deviation is less than the difference tolerance threshold. When the state deviation exceeds the difference tolerance threshold, a manual confirmation process is triggered.

[0022] Preferably, the dynamic data update system based on the Internet of Things platform further includes:

[0023] The state evolution perception module constructs a state evolution perception unit in the edge gateway, maintains a three-dimensional state evolution model for each device, collects device state data in real time, and uses an algorithm to construct a state evolution curve based on the device state data. The state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve.

[0024] The key node identification module obtains the local curvature change rate and prediction error bandwidth based on the state evolution curve, and identifies key state nodes according to the local curvature change rate and prediction error bandwidth.

[0025] The evolution extrapolation module activates the state evolution extrapolation model when it detects a decline in API connection quality or when the device enters a weak network area. It uses a lightweight time series prediction model to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories.

[0026] The confidence assessment module compares the trajectory of the multi-step state extrapolation with the latest evolution curve in the cloud to generate dynamic confidence labels.

[0027] The cloud trajectory fusion module generates a prediction summary based on the multi-step state extrapolation trajectory and confidence identifier, performs a multi-channel attempt synchronization strategy, synchronizes the prediction summary to the cloud, and performs trajectory fusion with the cloud evolution curve to obtain a fusion evolution curve.

[0028] The visualization and interaction module continuously sends requests to the API in the cloud according to a set time threshold to detect the connection status. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded and trajectory alignment and difference fusion are performed.

[0029] Compared with existing technologies, the present invention has the following advantages: By providing a dynamic data update method and system based on an IoT platform, the present invention achieves the following beneficial effects compared with existing IoT shadow mechanisms: First, from state snapshots to evolutionary trajectories, by constructing a three-dimensional state evolution model and lightweight time-series prediction at the edge, the device state is upgraded from discrete, lagging snapshots to continuous, forward-looking evolutionary trajectories. Even during disconnection, multi-step extrapolation trajectories with confidence indicators can still be provided, making offline states predictable and traceable, significantly improving the continuity and reliability of state data; Second, from passive triggering to active sensing, based on the local curvature change rate and prediction error band width, trend inflection points, divergence nodes, and highly sensitive state nodes are automatically identified, and evolutionary features drive the gate... The differentiated extrapolation and synchronization scheduling of key nodes transforms batch retransmission after network recovery into priority processing of critical states, significantly reducing redundant data transmission and cloud fusion overhead. Third, from fixed priority to adaptive optimization, the multi-channel synchronization strategy has environmental awareness capabilities, dynamically selecting low-bandwidth direct connection, device direct connection relay, or timed window waiting based on network quality, and dynamically adjusting cloud fusion weights in conjunction with confidence scores to achieve optimal matching between synchronization resources and prediction reliability. Fourth, from flag feedback to visual closed loop, the application query interface presents fusion status values, prediction error bands, and final confirmation values ​​differentiated according to confidence levels, visually feeding back the state evolution process, prediction reliability boundaries, and offline risks to the user in a closed loop, greatly enhancing the interpretability of system behavior and human-computer interaction decision-making capabilities. In summary, this invention achieves a comprehensive upgrade in device status under weak network and disconnection scenarios, from passive supplementation to active inference, from single snapshots to dynamic trajectories, and from indiscriminate synchronization to adaptive fusion. It effectively solves the technical bottlenecks of data being unknowable, unreliable, and unusable during offline periods, and provides solid support for high real-time and high-reliability applications in industrial IoT, vehicle-road collaboration, and telemedicine. Attached Figure Description

[0030] Figure 1 This is a flowchart of a dynamic data update method based on an Internet of Things (IoT) platform.

[0031] Figure 2 This is a schematic diagram of a dynamic data update system based on an Internet of Things (IoT) platform. Detailed Implementation

[0032] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0033] Example 1: A dynamic data update method based on an IoT platform, such as... Figure 1 As shown, it includes the following steps:

[0034] S1: Construct a state evolution perception unit in the edge gateway, maintain a three-dimensional state evolution model for each device, and collect device state data in real time. Use an algorithm to construct a state evolution curve based on the device state data. The state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve.

[0035] The three-dimensional state evolution model is a model that maintains real-time curves, confirmation curves, and evolution curves; the real-time curve is a smoothed sequence of instantaneous state values ​​after filtering and denoising; the confirmation curve is a sequence of state snapshots submitted and confirmed in two stages via the cloud; the evolution curve is based on a long short-term memory network prediction model, using historical window data as input, and outputting state prediction values ​​and their confidence intervals for multiple future time steps to form a prediction trajectory; the device state data time series is collected through a sliding window, and a trend extraction algorithm is used to fit the state change pattern; the device state data includes instantaneous state values, confirmed state values, evolution data, and historical window data.

[0036] It should also be noted that the three-dimensional state evolution model does not refer to three-dimensional coordinates in physical space, but rather to three parallel curves that characterize the state of the same device from three different temporal semantic dimensions. The real-time curve is continuously refreshed at a high sampling frequency. After the original signal is filtered by a digital filter to remove high-frequency noise and outliers, a smooth and instantly resolvable state waveform is formed, providing operators with a low-latency operational monitoring view. The confirmation curve represents a state snapshot verified by a distributed consensus protocol. It is generated based on a cloud-based two-phase commit mechanism to ensure that each state update meets atomicity and durability constraints during cross-node replication, thereby providing a conflict-free and auditable copy of the device state at the logical level. The evolution curve is generated by a time-series prediction engine driven by a long short-term memory network. Based on the multi-dimensional state sequence within a historical window, it extrapolates the expected trajectory of the device over several time steps in the future and simultaneously outputs the prediction uncertainty quantified in the form of confidence intervals, providing a probabilistic reference for forward-looking operation and maintenance decisions.

[0037] S2: Based on the state evolution curve, obtain the local curvature change rate and prediction error bandwidth, and identify key state nodes according to the local curvature change rate and prediction error bandwidth;

[0038] Based on three adjacent points on the evolution curve, the first and second derivatives of each point are calculated using the central difference method, and then substituted into the curvature formula. Obtain local curvature values The ratio of the difference between adjacent curvature values ​​to the corresponding time interval is calculated to obtain the local curvature change rate. Based on the probability distribution output by the time series prediction model, the upper and lower limits of the confidence interval for each prediction time are extracted, and the prediction error band width is the difference between the upper and lower limits of the confidence interval. Key state nodes are obtained based on the obtained local curvature change rate and prediction error band width. The key state nodes include trend inflection points, prediction divergence nodes, and highly sensitive state nodes.

[0039] If the rate of change of local curvature exceeds a preset first threshold, the corresponding state point is determined to be a trend inflection point; if the width of the prediction error band exceeds a preset second threshold, the corresponding state point is determined to be a prediction divergence node; if the trend inflection point, prediction divergence node, or instantaneous state value exceeds the corresponding preset safe operating range, the corresponding state point is determined to be a highly sensitive state node.

[0040] It should also be noted that if the local curvature change rate exceeds the preset first threshold, the corresponding state point will be marked as a trend inflection point. The trend inflection point is used to trigger the reset of the evolution extrapolation model and the correction of the prediction direction. The first threshold is extracted from the curvature change rate samples that have been confirmed as trend inflection points from the historical normal operation data of the equipment, and the 90th percentile is taken as the default threshold; it can also be manually adjusted by the user according to the sensitivity of the working conditions. This threshold is used to determine whether the state point is a trend inflection point.

[0041] If the prediction error bandwidth exceeds a preset second threshold, the corresponding state point is marked as a prediction divergence node. This prediction divergence node is used to trigger prediction mechanism degradation and uncertainty control. The second threshold is set by statistically analyzing the mean and standard deviation of historical prediction error bandwidths, and is also dynamically determined based on the maximum tolerance boundary for prediction divergence in the business context. This threshold is used to identify prediction divergence nodes.

[0042] If any of the above two types of nodes, or any instantaneous state value, deviates from the predefined safe operating range, the corresponding state point will be marked as a highly sensitive state node. The highly sensitive state node is used to drive differentiated resource scheduling and synchronization assurance. The safe operating range is directly defined based on the equipment's factory design specifications and the extreme value range of the state under historical normal operating conditions, or by using statistical process control methods to extract the upper and lower control limits from stable historical data as the range boundary. This range is used to determine the highly sensitive state node.

[0043] S3: When a decline in API connection quality is detected or a device enters a weak network area, the state evolution extrapolation model is activated. A lightweight time series prediction model is used to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories. The multi-step state extrapolation trajectories are then compared with the latest evolution curve in the cloud to generate dynamic confidence labels.

[0044] For trend inflection points, the evolution data within the shortest effective window after the inflection point is extracted, and an adaptive exponential smoothing method is used to quickly fit the new trend direction, performing only short-step extrapolation to avoid error accumulation. For prediction divergence nodes, when the prediction error band width exceeds the extrapolation threshold, the model automatically reduces the extrapolation step size to half of the default value and switches to a conservative prediction mode. At the same time, a Naive Bayes model is used for parallel extrapolation, and the result with a narrow confidence interval is selected as the final output. For highly sensitive state nodes, a multi-model ensemble extrapolation strategy is used, simultaneously running a Long Short-Term Memory network, a Holt-Winters model, and a simple moving average model, and dynamically weighting and fusing the extrapolation trajectory based on the inverse of the error of each model in the previous prediction window. Multi-step state extrapolation is performed on key state nodes on the local evolution curve to obtain multi-step state extrapolation trajectories. The confidence level is obtained by comparing the multi-step state extrapolation trajectories with the latest evolution curve in the cloud.

[0045] The confidence level identifier includes a confidence score and a confidence level; the confidence level includes a first confidence level, a second confidence level, and a third confidence level. A confidence score higher than the first confidence threshold is assigned to the first confidence level; a confidence score higher than the second confidence threshold but lower than the first confidence threshold is assigned to the second confidence level; and a confidence score lower than the second confidence threshold is assigned to the third confidence level. The standard deviation of the historical prediction error in the cloud is defined as... The confidence score is obtained through the confidence scoring formula, which is:

[0046] ;

[0047] in Score the confidence level; The trajectory vector for the time period corresponding to the multi-step state extrapolation trajectory; This is the trajectory vector corresponding to the latest evolution curve in the cloud for the corresponding time period; Dynamic time-warped distance is used as the distance metric for trajectory similarity. As a regulating factor; To prevent extremely small constants with a denominator of zero.

[0048] It should also be noted that the short-step extrapolation refers to predicting only a very short time window in the future when the state trend changes. Its purpose is to quickly fit the new trend before the trend direction is fully stable and to avoid the error avalanche caused by long-term extrapolation. The extrapolation threshold is determined based on the statistical distribution of the historical prediction error band width. Usually, the mean plus three standard deviations is taken as the default benchmark to identify that the degree of prediction divergence has exceeded the controllable range. A fixed threshold may also be set directly according to the maximum tolerance boundary of state uncertainty in business. The parallel extrapolation refers to using two or more heterogeneous prediction models to perform independent extrapolation at the same time when the prediction is highly divergent and the confidence of a single model is insufficient. The result with the lowest uncertainty and the narrowest interval is selected as the final trajectory.

[0049] Definition of the first The root mean square error of the model in the previous prediction window is The root mean square error is obtained based on the model predictions and confirmation curves at each time point within the previous prediction window. First, the prediction deviation is calculated point-by-point and squared. Then, the average of the squared deviations at all times within the window is taken. Finally, the square root of this average is taken to obtain the root mean square error for that window. The fused trajectory is obtained by dynamically weighting and extrapolating the fused trajectories based on the inverse errors of each model in the previous prediction window. , , For each model, for future moments extrapolated values; To integrate weights, .

[0050] When the confidence score is higher than the first confidence threshold, it is rated as the first confidence level. The first confidence threshold is based on the sample distribution of historical confidence scores that has been confirmed as reliable extrapolation, and its 90th percentile is selected as the initial threshold to ensure that only predictions with sufficiently high scores are assigned the highest confidence level. When the confidence score is higher than the second confidence threshold but lower than the first confidence threshold, it is rated as the second confidence level. The second confidence threshold is based on the sample distribution of historical confidence scores that has been confirmed as unreliable extrapolation, and its 10th percentile is selected as the initial threshold. Predictions lower than this value will be judged as low confidence. The standard deviation of the historical prediction error in the cloud is obtained based on the recent extrapolation-confirmation error sequence stored in the cloud. Each error term is the difference between the multi-step extrapolation trajectory reported by the edge gateway in the past and the corresponding state value of the confirmation curve received by the cloud in the subsequent time.

[0051] In the confidence score formula, the trajectory vector corresponding to the time period of the multi-step state extrapolation trajectory. For the local edge gateway to generate future values ​​through multi-step extrapolation during weak network conditions A sequence of state prediction values ​​at each time step, in the form of: The time step is fixed at 1 second; the trajectory vector of the latest evolution curve in the cloud corresponds to the time period. A sequence of expected state values ​​that completely correspond to a time period, in the form of... If some time-based data is missing in the cloud due to disconnection, only the overlapping valid points are calculated; both trajectories are normalized to zero mean to eliminate the influence of dimensions; the adjustment factor is dynamically adjusted based on the historical error distribution of the prediction model and the current network fluctuation level to balance the sensitivity of the confidence score. Combining the different requirements of business scenarios for data real-time performance and accuracy, the factor value is adaptively set to enhance the discrimination in harsh scenarios and smooth the score in fault-tolerant scenarios. Through offline training and online feedback, continuous calibration is performed to ensure that the adjustment factor makes the confidence score truly reflect the reliability of the prediction data.

[0052] S4: Generate a prediction summary based on the multi-step state extrapolation trajectory and confidence level identifier, perform a multi-channel attempt synchronization strategy, synchronize the prediction summary to the cloud, and perform trajectory fusion with the cloud evolution curve to obtain a fused evolution curve;

[0053] A prediction summary is obtained by compressing the multi-step state extrapolation trajectory, confidence score, and confidence level. This summary is then synchronously updated to the cloud database using a multi-channel synchronization strategy. The multi-channel synchronization strategy includes a first-priority low-bandwidth backup channel strategy. If the first-priority attempt succeeds, the prediction summary is sent. If it fails, a second-priority direct device relay strategy is implemented, where adjacent online devices are found as relays to send the prediction summary via short-range communication, and the online devices forward it to the cloud database. If this fails, a third-priority storage and waiting time window strategy is implemented. A next prediction time window is defined and added to the prediction summary. The prediction summary is stored in the edge gateway's persistent queue, waiting for the next prediction time window to arrive before attempting the first-priority low-bandwidth backup channel strategy again. Once the prediction summary reaches the cloud, a fusion evolution curve is obtained using a cloud trajectory fusion model.

[0054] The cloud trajectory fusion model takes the multi-step state extrapolation trajectory in the prediction summary and the cloud evolution curve as input, and uses the Kalman filter algorithm to perform optimal state estimation. The confidence score in the prediction summary is dynamically mapped to the weight factor of the observation noise covariance matrix. The higher the confidence score, the more the fusion result is biased towards the multi-step state extrapolation trajectory. The fusion state value at each time step is generated by recursively calculating by minimizing the prediction error covariance, forming a continuous fusion evolution curve.

[0055] It should also be noted that, to ensure the integrity of the prediction digest during transmission and optimize bandwidth utilization, an optimized lossless coding scheme is used to compress the prediction digest. In the multi-channel synchronization mechanism, a backup transmission path is specifically designed for low-bandwidth conditions. This path employs a lightweight communication protocol with high fault tolerance, thus maintaining basic data reachability even when network quality fluctuates.

[0056] When the direct device relay scheme is enabled, a Bluetooth Mesh network device discovery protocol is used. The edge gateway broadcasts a device discovery request, including its own device ID and status digest. Upon receiving the request, nearby online devices reply with a response packet containing their own device ID, signal strength, and current load. Based on the signal strength (RSSI) and load, the edge gateway selects the optimal device as the relay node, establishing an encrypted point-to-point connection (such as DTLS) and transmitting the predicted digest data. The relay device forwards the received data to the cloud database through its main communication channel. If direct relay cannot be established, it immediately switches to a temporary waiting mode, during which the predicted digest is encapsulated and temporarily stored locally. The definition of the next prediction time window is based on the duration of the current network anomaly and the confidence level of the predicted data. The start time and interval of the next window are dynamically determined, and the optimal time window is selected by combining historical availability statistics of each backup communication channel to improve the synchronization success rate. By estimating the expected data change rate within the window, the window length is adaptively adjusted to balance status timeliness and communication load.

[0057] All summary data awaiting synchronization is sent to a highly reliable persistent queue deployed on the edge gateway. This queue uses storage media with power-loss protection to prevent data loss or corruption during the waiting period. When the preset next synchronization time arrives, transmission will be attempted first through the aforementioned low-bandwidth backup channel according to a predetermined strategy priority. The entire synchronization process is designed as a recursive closed-loop process until the data is successfully submitted to the cloud storage system, thereby ensuring the eventual consistency of the data synchronization task in an intermittent connection environment.

[0058] S5: The cloud continuously sends requests to the API to detect the connection status according to the set time threshold. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded and trajectory alignment and difference fusion are performed.

[0059] When the connection status is abnormal, the interface displays the fused status value when the confidence level is first; when the confidence level is second, the interface displays both the fused status value and its prediction error band width; when the confidence level is third, the interface prioritizes displaying the last confirmed status value, and overlays the multi-step status extrapolation trajectory, prediction error band width, and prediction divergence warning indicator. When the connection status is restored, the edge gateway continuously records the status evolution curve during offline periods and uploads the complete trajectory segment after network recovery. The cloud uses a timestamp alignment algorithm to match the multi-step status extrapolation trajectory with the cloud evolution curve point by point within the overlapping time window, calculates the average absolute deviation of the status value at each moment as the status deviation of the overlapping area, and automatically updates the cloud evolution curve to the edge trajectory when the status deviation is less than the difference tolerance threshold. When the status deviation exceeds the difference tolerance threshold, a manual confirmation process is triggered.

[0060] It should also be noted that when the IoT platform detects a device connection interruption or abnormal status, it will dynamically adjust the information presentation strategy of the front-end interface based on the confidence level of the currently cached data to balance the tension between data timeliness and reliability. If the confidence level is the first level, the interface will directly present the state estimate after multi-model weighted fusion, aiming to provide operators with the simplest and most direct operational view. If the confidence level is the second level, the interface will display the fused state value along with the current prediction error band width, which is an extended piece of information. Presented in a visual interval format, it helps users intuitively understand the uncertainty range of the prediction results, thereby preserving the necessary prudent space in monitoring and decision-making. If the confidence level is the third confidence level, the default extrapolation result is no longer of sufficient reference value. The interface will prioritize displaying the last confirmed status value before the device goes offline, and overlay and draw multi-step status extrapolation trajectory, prediction error band at the corresponding time, and prominent prediction divergence warning mark on this benchmark. This view design aims to fully restore the system's extrapolation process of device status during offline period, assisting operation and maintenance personnel in quickly identifying potential risks and locating abnormal nodes.

[0061] After the connection is restored, the complete offline evolution trajectory uploaded by the edge gateway and the cloud evolution curve within the overlapping time window are used as input. A timestamp alignment algorithm is used for point-by-point matching, and the absolute value of the difference between the two trajectory state values ​​at each moment is calculated. Then, the average difference of all moments within the window is taken to obtain the average absolute deviation as the state deviation. When the state deviation is less than the difference tolerance threshold, the cloud evolution curve is automatically updated to the edge trajectory. The difference tolerance threshold is determined based on the statistical distribution of historical trajectory deviations under normal device operation, and is taken as two to three times the average of historical state deviations. A fixed tolerance limit is also directly set according to the device measurement accuracy and the maximum allowable deviation range of the service for state consistency. When the state deviation exceeds the difference tolerance threshold, not only will a confirmation process requiring manual intervention be triggered, but a detailed difference analysis report will also be automatically generated. This report will list the data points with deviations item by item and preliminarily analyze their causes, thereby assisting operation and maintenance personnel in quickly locating the root cause of the problem.

[0062] Example 2: Based on Example 1, a dynamic data update system based on an IoT platform, such as... Figure 2 As shown, it also includes:

[0063] The state evolution perception module constructs a state evolution perception unit in the edge gateway, maintains a three-dimensional state evolution model for each device, collects device state data in real time, and uses an algorithm to construct a state evolution curve based on the device state data. The state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve.

[0064] The key node identification module obtains the local curvature change rate and prediction error bandwidth based on the state evolution curve, and identifies key state nodes according to the local curvature change rate and prediction error bandwidth.

[0065] The evolution extrapolation module activates the state evolution extrapolation model when it detects a decline in API connection quality or when the device enters a weak network area. It uses a lightweight time series prediction model to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories.

[0066] The confidence assessment module compares the trajectory of the multi-step state extrapolation with the latest evolution curve in the cloud to generate dynamic confidence labels.

[0067] The cloud trajectory fusion module generates a prediction summary based on the multi-step state extrapolation trajectory and confidence identifier, performs a multi-channel attempt synchronization strategy, synchronizes the prediction summary to the cloud, and performs trajectory fusion with the cloud evolution curve to obtain a fusion evolution curve.

[0068] The visualization and interaction module continuously sends requests to the API in the cloud according to a set time threshold to detect the connection status. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded and trajectory alignment and difference fusion are performed.

[0069] The above description is merely an embodiment of the present invention and does not limit the patent scope of the present invention. Any equivalent structural or procedural transformations made based on the content of the present invention specification, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of the present invention.

Claims

1. A dynamic data update method based on an Internet of Things (IoT) platform, characterized in that: Includes the following steps: S1: Construct a state evolution perception unit in the edge gateway, maintain a three-dimensional state evolution model for each device, and collect device state data in real time. Use an algorithm to construct a state evolution curve based on the device state data. The state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve. S2: Based on the state evolution curve, obtain the local curvature change rate and prediction error bandwidth, and identify key state nodes according to the local curvature change rate and prediction error bandwidth; S3: When a decline in API connection quality is detected or a device enters a weak network area, the state evolution extrapolation model is activated. A lightweight time series prediction model is used to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories. The multi-step state extrapolation trajectories are then compared with the latest evolution curve in the cloud to generate dynamic confidence labels. S4: Generate a prediction summary based on the multi-step state extrapolation trajectory and confidence level identifier, perform a multi-channel attempt synchronization strategy, synchronize the prediction summary to the cloud, and perform trajectory fusion with the cloud evolution curve to obtain a fused evolution curve; S5: The cloud continuously sends requests to the API to detect the connection status according to the set time threshold. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded and trajectory alignment and difference fusion are performed.

2. The dynamic data update method based on an IoT platform according to claim 1, characterized in that, The process involves constructing a state evolution perception unit in the edge gateway, maintaining a three-dimensional state evolution model for each device, and collecting device state data in real time. An algorithm is used to construct a state evolution curve based on the device state data. This state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve. Specifically: the three-dimensional state evolution model is a model for maintaining the real-time curve, confirmation curve, and evolution curve; the real-time curve is a smoothed sequence of instantaneous state values ​​after filtering and denoising; the confirmation curve is a sequence of state snapshots submitted and confirmed in two stages via the cloud; the evolution curve is based on a long short-term memory network prediction model, using historical window data as input, and outputting predicted state values ​​and their confidence intervals for multiple future time steps to form a prediction trajectory; the process involves collecting time series of device state data through a sliding window and using a trend extraction algorithm to fit the state change pattern; the device state data includes instantaneous state values, confirmed state values, evolution data, and historical window data.

3. The dynamic data update method based on an IoT platform according to claim 1, characterized in that, The process of obtaining the local curvature change rate and prediction error band width based on the state evolution curve, and identifying key state nodes based on the local curvature change rate and prediction error band width, includes: calculating the first and second derivatives of each point based on three adjacent points on the evolution curve using the central difference method, and substituting them into the curvature formula. Obtain local curvature values The ratio of the difference between adjacent curvature values ​​to the corresponding time interval is calculated to obtain the local curvature change rate. Based on the probability distribution output by the time series prediction model, the upper and lower limits of the confidence interval for each prediction time are extracted, and the prediction error band width is the difference between the upper and lower limits of the confidence interval. Key state nodes are obtained based on the obtained local curvature change rate and prediction error band width. The key state nodes include trend inflection points, prediction divergence nodes, and highly sensitive state nodes.

4. The dynamic data update method based on an IoT platform according to claim 3, characterized in that, The process of obtaining key state nodes based on the obtained local curvature change rate and prediction error band width, wherein the key state nodes include trend inflection points, prediction divergence nodes, and highly sensitive state nodes, includes: if the local curvature change rate exceeds a preset first threshold, then the corresponding state point is determined to be a trend inflection point; if the prediction error band width exceeds a preset second threshold, then the corresponding state point is determined to be a prediction divergence node; if the trend inflection point, prediction divergence node, or instantaneous state value exceeds the corresponding preset safe operating range, then the corresponding state point is determined to be a highly sensitive state node.

5. The dynamic data update method based on an IoT platform according to claim 1, characterized in that, When a decline in API connection quality or a device entering a weak network area is detected, the state evolution extrapolation model is activated. A lightweight time-series prediction model is used to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories. The multi-step state extrapolation trajectories are then compared with the latest evolution curve in the cloud to generate dynamic confidence labels. This includes: for trend inflection points, extracting evolution data within the shortest effective window after the inflection point, using adaptive exponential smoothing to quickly fit the new trend direction, and performing only short-step extrapolation to avoid error accumulation; for prediction divergence nodes, when the prediction error band width exceeds the extrapolation threshold, the model automatically... The extrapolation step size is reduced to half of the default value, and a conservative prediction mode is switched. At the same time, a Naive Bayes model is enabled for parallel extrapolation, and the result with a narrow confidence interval is selected as the final output. For highly sensitive state nodes, a multi-model ensemble extrapolation strategy is enabled, and a Long Short-Term Memory network, a Holt-Winters model, and a simple moving average model are run simultaneously. The extrapolation trajectory is dynamically weighted and fused based on the inverse error of each model in the previous prediction window. Multi-step state extrapolation is performed on key state nodes on the local evolution curve to obtain multi-step state extrapolation trajectories. The confidence level is obtained by comparing the multi-step state extrapolation trajectory with the latest evolution curve in the cloud.

6. The dynamic data update method based on an IoT platform according to claim 5, characterized in that, The step of obtaining a confidence level identifier by comparing the multi-step state extrapolation trajectory with the latest evolution curve in the cloud includes: the confidence level identifier comprising a confidence score and a confidence level; the confidence level comprising a first confidence level, a second confidence level, and a third confidence level; a first confidence level when the confidence score is higher than a first confidence threshold; a second confidence level when the confidence score is higher than a second confidence threshold but lower than a first confidence threshold; and a third confidence level when the confidence score is lower than a second confidence threshold; the standard deviation of the historical prediction error in the cloud is defined as... The confidence score is obtained through the confidence scoring formula, which is: ； in Score the confidence level; The trajectory vector for the time period corresponding to the multi-step state extrapolation trajectory; This is the trajectory vector corresponding to the latest evolution curve in the cloud for the corresponding time period; Dynamic time-warped distance is used as the distance metric for trajectory similarity. As a regulating factor; To prevent extremely small constants with a denominator of zero.

7. The dynamic data update method based on an IoT platform according to claim 1, characterized in that, The process of generating a predicted summary based on the multi-step state extrapolation trajectory and confidence level, implementing a multi-channel attempt synchronization strategy, synchronizing the predicted summary to the cloud, and fusing it with the cloud evolution curve to obtain a fused evolution curve includes: compressing the multi-step state extrapolation trajectory, confidence score, and confidence level to obtain the predicted summary, and then updating it to the cloud database using a multi-channel attempt synchronization strategy; the multi-channel attempt synchronization strategy includes a first-priority low-bandwidth backup channel strategy. If the first-priority attempt succeeds, the predicted summary is sent; if it fails, it enters a second-priority device direct connection relay strategy, where an adjacent online device is found as a relay to send the predicted summary via short-distance communication, and the online device forwards it to the cloud database; if it fails again, it enters a third-priority storage and waiting timed window strategy, where the next prediction time window is defined and added to the predicted summary, the predicted summary is stored in the edge gateway's persistent queue, and the first-priority low-bandwidth backup channel strategy is attempted when the next prediction time window arrives; after the predicted summary arrives at the cloud, the fused evolution curve is obtained through a cloud trajectory fusion model.

8. The dynamic data update method based on an IoT platform according to claim 7, characterized in that, Once the predicted summary reaches the cloud, the fusion evolution curve is obtained through the cloud trajectory fusion model. This includes: the cloud trajectory fusion model takes the multi-step state extrapolation trajectory in the predicted summary and the cloud evolution curve as input, uses the Kalman filter algorithm to perform optimal state estimation, dynamically maps the confidence score in the predicted summary to the weighting factor of the observation noise covariance matrix. The higher the confidence score, the more the fusion result is biased towards the multi-step state extrapolation trajectory. The fusion state value at each moment is generated by recursively calculating by minimizing the prediction error covariance, forming a continuous fusion evolution curve.

9. The dynamic data update method based on an Internet of Things platform according to claim 1, characterized in that, The cloud continuously sends requests to the API to monitor connection status according to a set time threshold. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded, and trajectory alignment and difference fusion are performed, including: when the connection status is abnormal, if the confidence level is the first confidence level, the interface displays the fused status value; if the confidence level is the second confidence level, the interface simultaneously displays the fused status value and its prediction error band width; if the confidence level is the third confidence level, the interface prioritizes displaying the last confirmed status value, and... The system overlays and displays the multi-step state extrapolation trajectory, prediction error band width, and prediction divergence warning indicators. When the connection is restored, the edge gateway continuously records the state evolution curve during the offline period and uploads the complete trajectory segment after the network is restored. The cloud uses a timestamp alignment algorithm to match the multi-step state extrapolation trajectory with the cloud evolution curve point by point within the overlapping time window. The average absolute deviation of the state value at each moment is calculated as the state deviation of the overlapping area. When the state deviation is less than the difference tolerance threshold, the cloud evolution curve is automatically updated to the edge trajectory. When the state deviation exceeds the difference tolerance threshold, a manual confirmation process is triggered.

10. A dynamic data update system based on an Internet of Things (IoT) platform, used to implement the dynamic data update method based on an IoT platform as described in any one of claims 1-9, characterized in that, Also includes: The state evolution perception module constructs a state evolution perception unit in the edge gateway, maintains a three-dimensional state evolution model for each device, collects device state data in real time, and uses an algorithm to construct a state evolution curve based on the device state data. The state evolution curve includes a real-time curve, a confirmation curve, and an evolution curve. The key node identification module obtains the local curvature change rate and prediction error bandwidth based on the state evolution curve, and identifies key state nodes according to the local curvature change rate and prediction error bandwidth. The evolution extrapolation module activates the state evolution extrapolation model when it detects a decline in API connection quality or when the device enters a weak network area. It uses a lightweight time series prediction model to perform multi-step state extrapolation on key state nodes to obtain multi-step state extrapolation trajectories. The confidence assessment module compares the trajectory of the multi-step state extrapolation with the latest evolution curve in the cloud to generate dynamic confidence labels. The cloud trajectory fusion module generates a prediction summary based on the multi-step state extrapolation trajectory and confidence identifier, performs a multi-channel attempt synchronization strategy, synchronizes the prediction summary to the cloud, and performs trajectory fusion with the cloud evolution curve to obtain a fusion evolution curve. The visualization and interaction module continuously sends requests to the API in the cloud according to a set time threshold to detect the connection status. When the connection status is abnormal, the application query will provide a differentiated interface display based on the confidence level. When the connection status is restored, the complete evolution trajectory during the offline period is uploaded and trajectory alignment and difference fusion are performed.

Images