IoT-based smart cabinet status monitoring system

Abstract

This application relates to the field of IoT intelligent monitoring technology, and discloses an IoT-based intelligent cabinet status monitoring system, including an edge sensing gateway and a cloud decision platform. The edge sensing gateway collects multi-source sensor data within the cabinet, performs time-series fusion analysis, and generates consumable change events with confidence ratings. The cloud decision platform receives and aggregates events from the edge sensing gateway, runs an adaptive prediction model based on historical event sequences and confidence ratings, generates event discrimination parameters, and sends them back to the edge sensing gateway. The edge sensing gateway calibrates the subsequent time-series fusion analysis process based on the received parameters. This IoT-based intelligent cabinet status monitoring system can self-assess sensing reliability and dynamically optimize judgment logic, solving the problem of long-term monitoring reliability degradation caused by sensor performance drift or environmental interference in existing intelligent cabinet monitoring systems, and achieving highly accurate and adaptive intelligent status monitoring.

Description

Technical Field

[0001] This application relates to the field of IoT intelligent monitoring technology, specifically an IoT-based intelligent cabinet status monitoring system. Background Technology

[0002] With the widespread application of IoT technology in asset management, smart cabinets are widely used for the automated management of consumables, tools, and other items. Existing smart cabinet monitoring systems typically rely on RFID and weight sensors deployed within the cabinet to collect data, which is then uploaded to a cloud platform for centralized display and alerts. However, such systems face a key challenge in long-term operation: sensor performance drifts over time or is affected by environmental factors such as temperature and humidity, leading to systematic biases or increased random errors in the collected data. Because existing systems lack an autonomous assessment and compensation mechanism for performance degradation of the underlying sensing units or environmental interference, they can only perform simple processing and display of the uploaded raw data. This results in a gradual decline in the accuracy and reliability of the monitoring results, leading to frequent false alarms and missed alarms. This makes it difficult for managers to trust the system data, requiring frequent manual inventory checks and calibrations, thus failing to achieve truly intelligent and unmanned reliable monitoring. Summary of the Invention

[0003] Therefore, it is necessary to provide an IoT-based smart cabinet status monitoring system that can self-evaluate and autonomously optimize, thereby maintaining high-precision monitoring capabilities in changing environments, in order to address the aforementioned technical problems.

[0004] This application provides an IoT-based smart cabinet status monitoring system, comprising: An edge-aware gateway, deployed in a smart cabinet, is configured to collect multi-source sensor data within the smart cabinet, perform time-series fusion analysis on the multi-source sensor data, and generate consumable change events with confidence ratings. The cloud-based decision-making platform is communicatively connected to the edge sensing gateway and configured to receive and aggregate consumable change events from at least one of the edge sensing gateways, and run an adaptive prediction model based on historical event sequences and the confidence rating to generate event discrimination parameters and send the event discrimination parameters to the edge sensing gateway. The edge-aware gateway is further configured to calibrate the subsequent time-series fusion analysis process based on the received event discrimination parameters.

[0005] In one embodiment, the edge-aware gateway includes: The sensing unit is configured to respond to cabinet door status change events, synchronously trigger and collect RFID scanning data, weight sensing data and vibration sensing data; The fusion processing unit, connected to the sensing unit, is configured to perform time alignment and logical correlation analysis on the RFID scanning data, the weight sensing data, and the vibration sensing data to determine the validity of the consumable access operation and generate a consumable change event with the confidence rating. The communication unit is connected to both the fusion processing unit and the cloud decision platform, and is configured to report the consumable change event to the cloud decision platform and receive the event discrimination parameters from the cloud decision platform.

[0006] In one embodiment, the fusion processing unit is further configured to: Receive the time-aligned set of RFID tags, weight sampling sequence, and vibration spectrum; The RFID tag set of the smart cabinet is compared at two moments: when the door is opened and when it is closed, and the RFID tags that have changed are identified. When a changed RFID tag is identified, and the weight sampling sequence has a step change value within the corresponding time period, and the step change value and the nominal weight of the single consumable corresponding to the changed RFID tag satisfy a preset integer multiple relationship, and the vibration spectrum does not have an impact energy component exceeding the threshold within the time period, a valid consumable change operation is determined to have occurred. The confidence rating of the consumable change event is calculated based on the weight matching error and the energy distribution of the vibration spectrum, wherein the weight matching error is used to characterize the deviation between the step change value and the theoretical weight change value.

[0007] In one embodiment, the weight matching error is calculated and determined through the following process: The nominal weight of a single target consumable is determined based on the changes in the RFID tag set. and expected number of changes Calculate the theoretical weight change: ; Extract the actual weight change value from the weight sampling sequence. ; Calculate the weight matching error: ; Based on the weight matching error, a weight matching determination is performed; the weight matching determination includes: obtaining a dynamic adaptive threshold. The dynamic adaptive threshold is generated and distributed through the cloud decision platform based on the event discrimination parameters; the weight matching error is compared with the state adaptive threshold, if... If the weight match is successful, a positive weight factor is generated to improve the confidence rating; otherwise, a negative weight factor is generated to lower the confidence rating.

[0008] In one embodiment, the cloud-based decision platform includes a model management unit, which is configured to: For any selected edge sensing gateway, monitor the time series of confidence ratings for the target consumable reported by that edge sensing gateway; When the time series indicates that the confidence rating is consistently below the confidence threshold for a predetermined number of consecutive consumable change events, the root cause analysis process is triggered. The root cause analysis process includes performing root cause analysis based on multiple feature dimensions of recent consumable change events with low confidence ratings. The feature dimensions include at least the sequence of weight matching errors and the environmental data sequence at the time of the event. Based on the results of the root cause analysis, event discrimination parameters for optimizing the dynamic adaptive threshold are generated and sent through the communication unit.

[0009] In one embodiment, the root cause analysis process includes: Calculate the temporal correlation between the sequence of weight matching errors and at least one environmental parameter data in the environmental data sequence; If the correlation between the weight matching error and the environmental parameter data exceeds a preset correlation threshold, the low confidence problem is determined to be related to environmental interference, and the event discrimination parameters with an environmental compensation strategy are generated; otherwise, the event discrimination parameters without an environmental compensation strategy are generated.

[0010] In one embodiment, generating the event discrimination parameters with an environmental compensation strategy includes: Based on the correspondence between the weight matching error and environmental parameter data in historical data, a mapping relationship between the environment and the error is fitted through regression analysis, and this mapping relationship is incorporated into the event discrimination parameters.

[0011] In one embodiment, the fusion processing unit is further configured to: Receive the event discrimination parameters having the mapping relationship, and reconstruct the mapping relationship; Before each weight matching determination, the current environmental parameter vector is obtained in real time. ; Calculate the environmental compensation amount based on the reconstructed mapping relationship: ; The received benchmark dynamically adaptive threshold Combined with the environmental compensation amount, a final judgment threshold applicable to the current environment is obtained: The final determination threshold is then used to determine the weight matching.

[0012] In one embodiment, the edge-aware gateway includes a parameter execution management unit, which is configured to: Maintain a local parameter version queue, which is used to cache multiple consecutively received versions of the event discrimination parameters in the order of receipt; When performing the time-series fusion analysis, the event discrimination parameters of the latest version in the local parameter version queue are followed; When the connection between the edge-aware gateway and the cloud-based decision-making platform is interrupted, the update of the parameter version queue is stopped, and the latest version of parameters cached before the interruption is locked to continue the timing fusion analysis until communication is restored.

[0013] In one embodiment, the process of obtaining the confidence rating includes: The weight matching determination result is mapped to the first confidence component. The analysis results based on the spectral energy distribution of the vibration spectrum obtained after time alignment are mapped to a second confidence component. ; Receive the composite weight vector issued by the cloud decision-making platform as part of the event discrimination parameters. The final confidence rating is then synthesized using the following formula. ; in, and The composition weights are dynamically adjustable, and .

[0014] The aforementioned IoT-based smart cabinet status monitoring system, through the collaboration of an edge sensing gateway and a cloud-based decision-making platform, generates a confidence rating representing the credibility of the judgment for a consumable change event at the edge. The cloud-based decision-making platform aggregates and analyzes the confidence rating and its corresponding historical data, intelligently diagnoses the root cause of the problem through an adaptive predictive model, and generates targeted event discrimination parameters, which are then sent to the edge. The edge uses these parameters to perform real-time calibration of the subsequent data fusion and analysis process. This closed-loop mechanism proactively evaluates the quality of its own sensing results and continuously and accurately optimizes the core judgment logic, thereby improving the system's long-term monitoring accuracy, stability, and autonomy in complex and changing environments. This provides reliable technical support for achieving unmanned and trustworthy intelligent consumable storage management. Attached Figure Description

[0015] Figure 1This is a structural block diagram of an IoT-based smart cabinet status monitoring system provided in an embodiment of this application. Detailed Implementation

[0016] To facilitate understanding of the technical solutions provided in the embodiments of this application, the background technology involved in the embodiments of this application will be described below.

[0017] In medical, industrial, and laboratory fields, precise storage and inventory monitoring of high-value consumables, specialized tools, and reagents are crucial. Smart cabinet systems based on IoT, RFID, and intelligent identification technologies have become the mainstream solution for achieving this goal. These systems typically deploy RFID readers and weight sensors within the cabinet. When the cabinet door is opened and consumables are retrieved or returned, the system attempts to determine the specific operation (e.g., which box of consumables was taken, and how many boxes were taken) by reading changes in the RFID tags and combining this with weight changes. These operation events are then uploaded to a remote cloud platform, where administrators can view inventory status, record reports, and perform basic status monitoring.

[0018] However, after long-term practice and in-depth research, it has been found that the existing technical solutions mentioned above, in pursuing automated sensing and remote monitoring, have overlooked a core contradiction affecting the long-term reliable operation of the system: the mismatch between the physical reliability of the IoT sensing layer and the ever-increasing demand for data reliability from the business application layer. Specifically: First, the measurement accuracy of sensors is not constant. Taking the core weight sensor as an example, it inherently possesses characteristics such as zero-point drift and sensitivity variations; at the same time, its readings are easily affected by changes in ambient temperature and humidity. For instance, in warehouses with large day-night temperature differences or in cleanrooms with controlled temperature and humidity, the weight measurement value of the same item may show slight but critical deviations under different environments. RFID reading and writing may experience occasional missed or misreadings due to tag placement, metal interference within the cabinet, or minor changes in antenna performance.

[0019] Secondly, the existing unidirectional data flow architecture of sensing-uploading-display lacks the ability to assess and handle the aforementioned underlying physical uncertainties. The system merely mechanically summarizes the raw data reported by sensors (such as "RFID tag A disappeared, weight decreased by X grams") and directly converts it into a definite business event (such as "one box of consumables A has been taken"). When sensor data deviates due to the aforementioned reasons (for example, a box was actually taken, but due to weight drift, the measured value deviates significantly from the standard weight), the system cannot determine the credibility of the event judgment and can only record and display it as a definite fact. Over time, the system's inventory data will develop inexplicable discrepancies with the actual physical inventory, causing administrators to lose trust in the system.

[0020] Ultimately, this problem manifests itself in the following way: to ensure accurate management, administrators have to frequently interrupt the system's automated operation and resort to manual inventory checks to calibrate the data. This significantly diminishes the automation value of the smart cabinet, and remote monitoring becomes less meaningful due to unreliable data. The root cause is that the existing system only achieves the perception and presentation of status, but fails to assess and optimize its own perceived quality (i.e., reliability).

[0021] Therefore, the core technical problem that this embodiment aims to solve is: how to build an intelligent cabinet monitoring system that can self-assess perceived reliability and self-optimize accordingly, so as to ensure the accuracy and reliability of long-term monitoring data under the real conditions of sensor performance drift and environmental interference, and achieve highly reliable unmanned intelligent management. To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0022] The IoT-based smart cabinet status monitoring system provided in this embodiment is applied in scenarios such as medical, industrial, or laboratory settings for the intelligent management of consumables, tools, and other items. Figure 1 As shown, the system includes: An edge-aware gateway, deployed in a smart cabinet, is configured to collect multi-source sensor data within the smart cabinet, perform time-series fusion analysis on the multi-source sensor data, and generate consumable change events with confidence ratings. The cloud-based decision-making platform communicates with the edge-aware gateway and is configured to receive and aggregate consumable change events from at least one edge-aware gateway. It also runs an adaptive prediction model based on historical event sequences and confidence ratings to generate event discrimination parameters and send them to the edge-aware gateway. The edge-aware gateway is also configured to calibrate the subsequent time-series fusion analysis process based on the received event discrimination parameters.

[0023] Specifically, the edge sensing gateway is an integrated hardware module installed in the control compartment of the smart cabinet via a standard electrical interface. It integrates a microprocessor (such as an ARM Cortex-A series chip), storage unit, and multiple communication interfaces (such as RS485, CAN, and GPIO). Multi-source sensor data includes scan data from RFID readers connected through its interfaces, weight data output from weighing sensors, and vibration data output from vibration sensors. Time-series fusion analysis refers to a software algorithm running on the microprocessor that synchronizes, compares, and logically judges the time-correlated data from the aforementioned different sensors. The confidence rating is a numerical value calculated by this algorithm, such as a floating-point number between 0 and 1, used to quantitatively represent the credibility of the generated consumable change event (such as "removing item A").

[0024] The cloud-based decision-making platform is deployed on a cloud server cluster, providing a web service interface. It maintains communication with multiple edge-aware gateways via MQTT or HTTP protocols. The platform's database continuously stores all consumable change events reported by the gateways and their associated confidence ratings, forming a historical event sequence. The adaptive prediction model is a software model trained based on machine learning (such as gradient boosting trees or recurrent neural networks). This model takes features from the historical event sequence (such as consumable type, operation time, historical confidence trends, etc.) and the current confidence rating as input, performs inference calculations, and outputs a set of configuration parameters for optimizing edge awareness, i.e., event discrimination parameters. These parameters are distributed to the corresponding edge-aware gateways via an encrypted link.

[0025] Furthermore, the edge-aware gateway calibrates the subsequent time-series fusion analysis process, meaning its software has a dedicated parameter configuration module. When new event discrimination parameters are received via the communication interface, this module parses the parameters and uses them to update key thresholds, coefficients, or logical rules used within the fusion analysis algorithm. For example, it replaces the old threshold with a newly issued weight judgment threshold, thereby changing the strictness of subsequent weight matching judgments.

[0026] Based on the above, this system relies on the collaboration of an edge sensing gateway and a cloud-based decision-making platform. The edge side (edge ​​sensing gateway) is responsible for real-time sensing and preliminary judgment, producing results with confidence ratings; the cloud side (cloud-based decision-making platform) is responsible for macro-level monitoring and in-depth analysis, diagnosing events through models and providing event discrimination parameters; the edge side then performs a time-series fusion analysis process to optimize its own calibration fusion analysis. This closed loop enables the system to continuously perceive fluctuations in the quality of its output and proactively adjust its sensing strategy, thereby systematically and automatically maintaining high-precision monitoring even under component aging or environmental interference.

[0027] In one embodiment, to ensure that multi-source data can be effectively synchronized and processed, the edge-aware gateway includes a sensing unit, a fusion processing unit, and a communication unit.

[0028] The sensing unit is specifically configured to respond to cabinet door status change events, synchronously trigger and collect RFID scanning data, weight sensing data and vibration sensing data.

[0029] Specifically, the sensing unit consists of a signal conditioning circuit, an analog-to-digital converter, and digital logic circuits in hardware. A change in cabinet door status is indicated by a level transition signal generated by a magnetic reed switch sensor or photoelectric sensor mounted on the cabinet door. This signal serves as a trigger signal, simultaneously sent to the RFID reader controller, the weight sensor sampling circuit, and the vibration sensor sampling circuit. RFID scan data is generated by an UHF RFID reader (such as a reader module based on the Impinj R2000 chip), which, upon triggering, scans the antenna coverage area inside the cabinet and outputs a list of scanned electronic tag IDs. Weight sensing data is output from a high-precision strain gauge load cell (such as a sensor with a range of 30kg and an accuracy of ±1g), whose analog signal is sampled by a 24-bit analog-to-digital converter to obtain a digital weight sequence. Vibration sensing data is output from a triaxial MEMS accelerometer (such as Analog Devices' ADXL345), whose analog signal is filtered and converted from analog to digital to obtain a digital vibration acceleration sequence. This synchronous triggering mechanism ensures that different physical quantities are aligned at the same starting point in time.

[0030] The fusion processing unit is connected to the sensing unit. Its specific configuration is to perform time alignment and logical correlation analysis on RFID scanning data, weight sensing data and vibration sensing data to determine the validity of consumable access operations and generate consumable change events with confidence ratings.

[0031] Specifically, the fusion processing unit is physically the main microprocessor within the edge sensing gateway. Time alignment refers to the microprocessor assigning a unified, high-precision timestamp to the three types of data streams received and aligning the sequences according to the trigger time. Logical correlation analysis refers to a software algorithm running on the microprocessor that correlates RFID tag changes, weight curve characteristics, and vibration spectrum characteristics within a time window, and determines whether a complete and reliable consumable access operation is constituted according to preset rules.

[0032] The communication unit is connected to the fusion processing unit and the cloud decision platform respectively. Its specific configuration is to report consumable change events to the cloud decision platform and receive event discrimination parameters from the cloud decision platform.

[0033] Specifically, the communication unit consists of a 4G / 5G cellular network module (such as Quectel's EC20 series) or an NB-IoT module and its driving circuitry. It connects to the microprocessor via a UART or USB interface and is responsible for uploading the consumable change event data packets (in JSON or Protocol Buffers format) packaged by the microprocessor to the message queue or API interface specified by the cloud decision platform via the cellular network. It also listens for downlink channels to receive data packets sent from the cloud.

[0034] Based on the above, the synchronous triggering mechanism of the sensing unit is the foundation for subsequent accurate timing fusion, avoiding data misalignment caused by sampling delay. The fusion processing unit, as the core, executes the specific judgment logic. The communication unit ensures bidirectional smooth flow of data and instructions within the optimized closed loop. This modular design clearly defines functional boundaries, guaranteeing reliable system implementation and maintainability.

[0035] In one embodiment, to achieve highly reliable determination of consumable change events and to initially generate indicators for evaluating the credibility of such events, the fusion processing unit is further configured as follows: Receive the time-aligned RFID tag set, weight sampling sequence, and vibration spectrum; this refers to the fusion processing algorithm obtaining three sets of data from the buffer that have already undergone timestamp alignment: the RFID tag ID set at the door opening time and the door closing time (…). and ), a sequence of weight values ​​continuously sampled from the time the door opens to the time it closes And the vibration spectrum calculated by Fast Fourier Transform (FFT) within the same time period. .

[0036] The system compares the RFID tag sets of the smart cabinet at the moments of opening and closing, identifying RFID tags that have changed. Specifically, this is achieved by calculating the symmetric difference between the two sets. , The labels in the image are those of consumables that were suspected of being removed or placed during this operation.

[0037] When a changed RFID tag is identified, and the weight sampling sequence contains a step change value within the corresponding time period, and the step change value and the nominal weight of the consumable corresponding to the changed RFID tag satisfy a preset integer multiple relationship, and the vibration spectrum does not contain impact energy components exceeding a threshold within that time period, a valid consumable change operation is determined to have occurred. Specific steps include: 1) From the weight sequence... In this study, a stable weight step was identified using any existing change point detection algorithm (such as the mean-difference method based on a sliding window). 2) From Extract a consumable label from the database and query its nominal weight per item. 3) Calculate the ratio. and judge Is it close to a certain integer (i.e.) ,in For a small decimal, such as 0.1, Indicates to Rounding to the nearest integer is a standard mathematical function. 4) Analyze the vibration spectrum. Check whether the energy within a preset impact characteristic frequency band (such as the high-frequency band) exceeds the energy threshold statistically obtained under a static state. The algorithm determines a physical operation to be valid only if all of the above conditions are met simultaneously.

[0038] Based on the weight matching error and the energy distribution of the vibration spectrum, the confidence rating of the consumable change event is calculated, where the weight matching error characterizes the deviation between the step change value and the theoretical weight change value. Weight matching error The calculation is as follows ,in The energy distribution of the vibration spectrum can be scored using a normalized stationarity score. Indicate, for example ,in This refers to the actual energy in the impact frequency band. Confidence rating. The preliminary calculation can be considered as The function here It can be a weighted average or other monotonically decreasing function (about This ensures that the larger the error and the more violent the vibration, the lower the confidence level.

[0039] Based on the above, a triple correlation mechanism of RFID target identification, weight verification quantity, and vibration interference elimination reduces erroneous events caused by single sensor misjudgments (such as RFID missed reads or weight fluctuations). Simultaneously, key quantitative indicators (matching error, vibration energy) in the judgment process are used as inputs for calculating the confidence rating, giving the rating results clear physical meaning and a data foundation, providing high-quality, interpretable feedback signals for subsequent cloud-based optimization.

[0040] In one embodiment, to clarify the specific process of weight matching error and support the calculation of confidence level rating, the weight matching error is calculated and determined through the following process: The nominal weight of a single target consumable is determined based on changes in the RFID tag set. and expected number of changes Calculate the theoretical weight change: Expected number of changes Through analysis Confirm: If the label comes from If it disappears in the middle, then A positive value indicates that the item has been retrieved; if the label is in... If China and Singapore emerge, then A negative value indicates that something is being added. Read from the local consumables database.

[0041] Extract the actual weight change value from the weight sampling sequence. By positioning the weight-stabilized platform before the door opens (e.g., 5 seconds before) and after the door closes (e.g., 5 seconds after), the difference between their mean values ​​is obtained. .

[0042] Calculate the weight matching error: .

[0043] Weight matching is determined based on weight matching error; the weight matching determination includes: obtaining a dynamic adaptive threshold. The dynamic adaptive threshold is generated and distributed through a cloud-based decision-making platform based on event discrimination parameters; the weight matching error is compared with the dynamic adaptive threshold, and if... If the weight match is successful, a positive weighting factor is generated to improve the confidence rating. Otherwise, a negative weighting factor is generated to lower the confidence rating. Positive weighting factor It can be a fixed value (such as +1), or a variable. Negatively correlated function values. Negative weighting factors. It can be a fixed negative value (such as -1), or a value equal to... A positively correlated negative function. Dynamic adaptive threshold. It is one of the key parameters issued from the cloud, and its initial value can be preset by the system.

[0044] Based on the above, the error calculation method between theoretical and actual values ​​was clarified, and a dynamic threshold optimized by the cloud was used for judgment. This makes the judgment criteria no longer fixed and potentially unapplicable to all consumables or all states, but can be intelligently adjusted by the cloud based on historical performance. The judgment result is output in the form of weighting factors, providing a clear and structured input for the final confidence synthesis, and realizing effective integration between the edge judgment logic and the cloud optimization strategy.

[0045] In one embodiment, to generate targeted optimization strategies, the cloud-based decision platform includes a model management unit. The model management unit is configured as follows: For any selected edge-aware gateway, monitor the time series of confidence ratings reported by that gateway for the target consumable. The model management unit is a background service in the cloud that can periodically scan the database or listen to event streams to maintain a confidence rating sequence for each "gateway-consumable" pair. .

[0046] The root cause analysis process is triggered when the time series indicates that the confidence rating remains below a predetermined number of consecutive consumable change events. For example, setting a confidence threshold... ,continuous Rating of this event All less than This triggers root cause analysis for the "gateway-consumables" pair.

[0047] The root cause analysis process involves performing root cause analysis based on multiple characteristic dimensions of recent consumable change events with low confidence ratings. These characteristic dimensions include at least the sequence of weight matching errors and the environmental data sequence at the time the event occurred. Based on the results of the root cause analysis, event discrimination parameters for optimizing dynamic adaptive thresholds are generated and distributed via the communication unit. Characteristic dimensions also include operation time and operation interval. The environmental data sequence refers to the temperature and humidity sensor data sequences synchronously received from the gateway by the cloud. Root cause analysis aims to identify which feature(s) are strongly correlated with low confidence (or high weight matching error).

[0048] Based on the above, proactively tracing the possible causes of low scores in the cloud (such as whether the error sequence exhibits a certain pattern or whether it is synchronized with environmental changes) is a prerequisite for generating accurate optimization parameters (such as adjusting thresholds and increasing environmental compensation), which transforms the system's optimization behavior from empirical or blind to targeted and intelligent.

[0049] In one embodiment, to determine whether a low-confidence problem is caused by environmental factors, the root cause analysis process includes: Sequence for calculating weight matching error With at least one environmental parameter data in the environmental data sequence (e.g., relative humidity sequence) The correlation over time. Any existing correlation calculation method can be used, such as the Pearson correlation coefficient. Perform the calculation: ,in and These are the mean values ​​of the sequences.

[0050] If the correlation between the weight matching error and the environmental parameter data exceeds a preset correlation threshold, the low-confidence problem is determined to be related to environmental interference, and event discrimination parameters with an environmental compensation strategy are generated; otherwise, event discrimination parameters without an environmental compensation strategy are generated. (Preset correlation threshold) It can be set to 0.6 or 0.7. If... If the error is strongly correlated with the environment, an environment compensation strategy module needs to be added to the event discrimination parameter package. Otherwise, the problem is considered to stem from random errors or operational habits, and the generated parameter package only includes the basic threshold. The adjustment amount.

[0051] Based on the above, by calculating statistical correlations, a quantitative basis for determining environmental disturbances is provided. This avoids subjective assumptions, ensuring that the triggering of environmental compensation strategies is a data-driven decision based on sound evidence. It clarifies the branches of the optimization strategy: for environmentally relevant issues, a more complex compensation model is employed; for non-environmental issues, simple threshold tuning is performed. This differentiated approach improves optimization efficiency and accuracy.

[0052] Once the problem is determined to be environmentally relevant, this relationship needs to be modeled and transformed into a compensation strategy that can be deployed to the edge for execution. Therefore, in one embodiment, event discrimination parameters with an environmental compensation strategy are generated, including: Based on the correspondence between weight matching error and environmental parameter data in historical data, a mapping relationship between environment and error is fitted through regression analysis, and this mapping relationship is incorporated into the event discrimination parameters.

[0053] For example, select the weight matching error corresponding to a recent batch of low-confidence events. and ambient humidity Any existing regression analysis method, such as a univariate linear regression model, can be used for fitting: The parameters are estimated using the least squares method. (Intercept) and (Slope). So, what is the mapping relationship between the environment and the error? This can be represented as the regression line, i.e. In practical applications, multinomial regression or piecewise linear regression can also be used to fit more complex relationships. The event discrimination parameters will include the type of this mapping relationship (e.g., "linear") and the coefficients. , ).

[0054] Based on the above, the systemic weight error that may occur under the current conditions is predicted by using real-time environmental parameters (such as humidity). This mapping relationship itself is the core environmental compensation strategy, which enables the edge side to anticipate and offset environmental interference in advance, rather than passively bearing its impact.

[0055] In one embodiment, to achieve the effect of combating environmental interference, the fusion processing unit is further configured as follows: It receives event discrimination parameters with mapping relationships and reconstructs the mapping relationships. Specifically, the edge software parses the parameter packet and, based on its type and coefficients, instantiates the same mapping function (mapping relationship) in memory. .

[0056] Before each weight matching determination, the current environmental parameter vector is obtained in real time. For example, read the current humidity value from the connected temperature and humidity sensor. .

[0057] Calculate the environmental compensation amount based on the reconstructed mapping relationship: Specifically, the current humidity is substituted into a humidity-related mapping function. The prediction error value is calculated, and this value is used as the compensation amount for the dynamic threshold.

[0058] The received benchmark dynamically adaptive threshold Combined with the environmental compensation amount, a final judgment threshold applicable to the current environment is obtained: And use this final judgment threshold to determine weight matching. Additive synthesis is used here, meaning that when the environment causes the prediction error to be positive ( Relax the judgment threshold ( (to increase), in order to tolerate the systematic biases caused by the environment.

[0059] Based on the above, the strategy of utilizing cloud-based data at the edge and combining it with real-time sensor data is clarified to dynamically adjust core judgment parameters (thresholds). Through this feedforward compensation mechanism, the system can proactively adapt to environmental changes, reduce the impact of environmental factors on the accuracy of weight determination, and thus maintain stable high performance in changing physical environments, thereby achieving long-term reliable monitoring.

[0060] In real-world scenarios with unstable networks, it is necessary to ensure the continuous and stable execution of edge-side optimization logic to prevent the system from reverting to an unoptimized state or causing chaos due to brief interruptions in cloud connectivity. Therefore, in one embodiment, the edge-aware gateway includes a parameter execution management unit, which is configured as follows: Maintain a local parameter version queue to cache multiple consecutively received versions of event discrimination parameters in the order they are received. Specifically, allocate a queue data structure in the edge gateway's persistent storage (such as eMMC). Whenever a new event discrimination parameter packet is successfully received and verified from the cloud, assign it an incrementing version number (e.g., V1.0, V1.1), and store its metadata and content at the tail of the queue. The queue capacity is configurable, for example, storing the 5 most recent versions.

[0061] When performing time-series fusion analysis, the event discrimination parameters are selected based on the latest version in the local parameter version queue. Specifically, when the fusion processing unit needs to obtain parameters such as dynamic thresholds, it queries the parameter execution management unit. The parameter execution management unit always returns the content of the parameter package with the largest version number in the queue.

[0062] When the connection between the edge-aware gateway and the cloud decision-making platform is interrupted, the update of the parameter version queue is stopped, and the latest version of parameters cached before the interruption is locked to continue performing time-series fusion analysis until communication is restored. Specifically, the communication unit continuously monitors the network link status. Once it detects that the connection with the cloud heartbeat cannot be synchronized for an extended period (e.g., more than 30 seconds), it sends a connection interruption flag to the parameter execution management unit. The parameter execution management unit will then refuse to receive any new parameters (even if the network is briefly restored) and will always provide the fusion processing unit with the latest parameters in the queue at the time of the interruption. Only when the communication unit confirms that a stable connection with the cloud has been fully restored will the interruption flag be cleared and the queue update capability be restored.

[0063] Based on the above, the optimization strategy is remembered even after power outages by locally caching multiple versions of parameters. An interrupt locking mechanism ensures the consistency of the judgment logic during network fluctuations, preventing jumps in judgment criteria due to parameter acquisition failures or out-of-order delivery. This ensures that cloud-based optimization results can be stably implemented and remain effective, improving the system's availability in complex network environments.

[0064] To address variations in the reliability of different sensors, in one embodiment, the confidence rating acquisition process includes: The weight matching determination result is mapped to the first confidence component. The analysis results based on the spectral energy distribution of the vibration spectrum obtained after time alignment are mapped to a second confidence component. Specifically, the mapping method could be: if the weight matching is successful, then... ,otherwise Analyzing the energy distribution of the vibration spectrum allows for the calculation of a score characterizing operational stability using any existing method. For example, it can be used to calculate the actual energy in a preset impact frequency band (e.g., above 50 Hz). The reference energy threshold of this frequency band in a static state The ratio, and mapped to Through this mapping, The value ranges from 0 to 1, and the higher the value, the closer the vibration characteristics are to the normal operating mode.

[0065] Receive the synthesized weight vector sent by the cloud-based decision-making platform as part of the event discrimination parameters. The final confidence rating is then synthesized using the following formula. ; in, and The composition weights are dynamically adjustable, and .

[0066] Specifically, the synthesized weight vector is generated by the model management unit of the cloud-based decision platform and encapsulated in the event discrimination parameter data packet sent to the edge-aware gateway. When calculating the confidence rating, the edge-side fusion processing unit parses the latest event discrimination parameters from its locally cached data packet. and The current value is used as the basis for calculation.

[0067] Based on this, during root cause analysis, when it is determined that a low-confidence problem is related to a systematic bias in a specific sensor data dimension (weight or vibration), the synthetic weight vector is adjusted. The weights of the corresponding dimensions are used to generate new event discrimination parameters to reduce the influence of unreliable sensor dimensions on confidence rating. Specifically, when the model management unit performs the aforementioned correlation analysis ("calculating the temporal correlation between the weight matching error sequence and at least one environmental parameter data in the environmental data sequence"), it not only analyzes the correlation between the weight matching error and the environment, but also extends the analysis to the two confidence components. and The correlation between their respective historical sequences and the overall low-confidence problem. For example, define a weight adjustment strategy: if within multiple consecutive periods, The values ​​are consistently low (e.g., all below 0.6) and consistent with the overall confidence level. The lower time periods highly overlap, and If the data remains normal, the model management unit determines that there is a systematic bias in the "weight" data dimension. In this case, the model management unit will calculate new composite weights and reduce them. The value, and correspondingly increased. The value ensures This still holds true. For example, changing the weight from... Adjust to This new set of weight vectors will be sent to the corresponding edge-aware gateways as part of the new version of event discrimination parameters. After the edge gateways apply the new weights, unreliable weight dimension data will be used when synthesizing confidence ratings. The proportion of weight sensor failure will decrease, and the overall system evaluation will be more tolerant of weight sensor failure.

[0068] Based on the above, the cloud-based system diagnoses which sensor source has a systematic unreliability problem by monitoring the long-term performance of each confidence component. Then, by dynamically adjusting the weight parameters in the final synthesis formula, it proactively weakens the impact of unreliable sources at the decision-making level. This provides an alternative system fault-tolerance mechanism: when front-end threshold calibration cannot fully compensate for sensor degradation, back-end weight adjustment can serve as an effective supplement, jointly ensuring the robustness of the output results. This elevates the system's intelligence from optimizing a single criterion to optimizing a comprehensive evaluation strategy. By dynamically reconstructing the trust system of sensor data, it achieves greater flexibility and intelligence, closer to human decision-making, and enhances the long-term reliability and autonomy of the entire monitoring system under complex and non-ideal conditions.

[0069] Furthermore, the data collection, processing, and use involved in this application embodiment are all conducted in compliance with relevant laws, regulations, and privacy policies. Users of the smart locker should have legal management rights over the items stored within the locker, and the system operation does not involve the collection or processing of unauthorized personal information.

[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0071] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0072] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A smart cabinet status monitoring system based on the Internet of Things, characterized in that, include: An edge-aware gateway, deployed in a smart cabinet, is configured to collect multi-source sensor data within the smart cabinet, perform time-series fusion analysis on the multi-source sensor data, and generate consumable change events with confidence ratings. The cloud-based decision-making platform is communicatively connected to the edge sensing gateway and configured to receive and aggregate consumable change events from at least one of the edge sensing gateways, and run an adaptive prediction model based on historical event sequences and the confidence rating to generate event discrimination parameters and send the event discrimination parameters to the edge sensing gateway. The edge-aware gateway is further configured to calibrate the subsequent time-series fusion analysis process based on the received event discrimination parameters.

2. The IoT-based intelligent cabinet status monitoring system according to claim 1, characterized in that, The edge-aware gateway includes: The sensing unit is configured to respond to cabinet door status change events, synchronously trigger and collect RFID scanning data, weight sensing data and vibration sensing data; The fusion processing unit, connected to the sensing unit, is configured to perform time alignment and logical correlation analysis on the RFID scanning data, the weight sensing data, and the vibration sensing data to determine the validity of the consumable access operation and generate a consumable change event with the confidence rating. The communication unit is connected to both the fusion processing unit and the cloud decision platform, and is configured to report the consumable change event to the cloud decision platform and receive the event discrimination parameters from the cloud decision platform.

3. The IoT-based intelligent cabinet status monitoring system according to claim 2, characterized in that, The fusion processing unit is further configured to: Receive the time-aligned set of RFID tags, weight sampling sequence, and vibration spectrum; The RFID tag set of the smart cabinet is compared at two moments: when the door is opened and when it is closed, and the RFID tags that have changed are identified. When a changed RFID tag is identified, and the weight sampling sequence has a step change value within the corresponding time period, and the step change value and the nominal weight of the single consumable corresponding to the changed RFID tag satisfy a preset integer multiple relationship, and the vibration spectrum does not have an impact energy component exceeding the threshold within the time period, a valid consumable change operation is determined to have occurred. The confidence rating of the consumable change event is calculated based on the weight matching error and the energy distribution of the vibration spectrum, wherein the weight matching error is used to characterize the deviation between the step change value and the theoretical weight change value.

4. The IoT-based intelligent cabinet status monitoring system according to claim 3, characterized in that, The weight matching error is calculated and determined through the following process: The nominal weight of a single target consumable is determined based on the changes in the RFID tag set. and expected number of changes Calculate the theoretical weight change: ; Extract the actual weight change value from the weight sampling sequence. ; Calculate the weight matching error: ; Based on the aforementioned weight matching error, a weight matching determination is made; The weight matching determination includes: obtaining a dynamic adaptive threshold. The dynamic adaptive threshold is generated and distributed through the cloud decision platform based on the event discrimination parameters; the weight matching error is compared with the state adaptive threshold, if... If the weight match is successful, a positive weight factor is generated to improve the confidence rating; otherwise, a negative weight factor is generated to lower the confidence rating.

5. The IoT-based intelligent cabinet status monitoring system according to claim 4, characterized in that, The cloud-based decision-making platform includes a model management unit, which is configured as follows: For any selected edge sensing gateway, monitor the time series of confidence ratings for the target consumable reported by that edge sensing gateway; When the time series indicates that the confidence rating is consistently below the confidence threshold for a predetermined number of consecutive consumable change events, the root cause analysis process is triggered. The root cause analysis process includes performing root cause analysis based on multiple feature dimensions of recent consumable change events with low confidence ratings. The feature dimensions include at least the sequence of weight matching errors and the environmental data sequence at the time of the event. Based on the results of the root cause analysis, event discrimination parameters for optimizing the dynamic adaptive threshold are generated and sent through the communication unit.

6. The IoT-based intelligent cabinet status monitoring system according to claim 5, characterized in that, The root cause analysis process includes: Calculate the temporal correlation between the sequence of weight matching errors and at least one environmental parameter data in the environmental data sequence; If the correlation between the weight matching error and the environmental parameter data exceeds a preset correlation threshold, the low confidence problem is determined to be related to environmental interference, and the event discrimination parameters with an environmental compensation strategy are generated; otherwise, the event discrimination parameters without an environmental compensation strategy are generated.

7. The IoT-based intelligent cabinet status monitoring system according to claim 6, characterized in that, The generation of the event discrimination parameters with an environmental compensation strategy includes: Based on the correspondence between the weight matching error and environmental parameter data in historical data, a mapping relationship between the environment and the error is fitted through regression analysis, and this mapping relationship is incorporated into the event discrimination parameters.

8. The IoT-based intelligent cabinet status monitoring system according to claim 7, characterized in that, The fusion processing unit is further configured to: Receive the event discrimination parameters having the mapping relationship, and reconstruct the mapping relationship; Before each weight matching determination, the current environmental parameter vector is obtained in real time. ; Calculate the environmental compensation amount based on the reconstructed mapping relationship: ; The received benchmark dynamically adaptive threshold Combined with the environmental compensation amount, a final judgment threshold applicable to the current environment is obtained: The final determination threshold is then used to determine the weight matching.

9. The IoT-based intelligent cabinet status monitoring system according to any one of claims 1 to 8, characterized in that, The edge-aware gateway includes a parameter execution management unit, which is configured as follows: Maintain a local parameter version queue, which is used to cache multiple consecutively received versions of the event discrimination parameters in the order of receipt; When performing the time-series fusion analysis, the event discrimination parameters of the latest version in the local parameter version queue are followed; When the connection between the edge-aware gateway and the cloud-based decision-making platform is interrupted, the update of the parameter version queue is stopped, and the latest version of parameters cached before the interruption is locked to continue the timing fusion analysis until communication is restored.

10. The IoT-based intelligent cabinet status monitoring system according to any one of claims 1 to 8, characterized in that, The process of obtaining the confidence rating includes: The weight matching determination result is mapped to the first confidence component. The analysis results based on the spectral energy distribution of the vibration spectrum obtained after time alignment are mapped to a second confidence component. ; Receive the composite weight vector issued by the cloud decision-making platform as part of the event discrimination parameters. The final confidence rating is then synthesized using the following formula. ; in, and The composition weights are dynamically adjustable, and .

Images