Remote monitoring system for operation state of air purification equipment

Abstract

The invention belongs to the technical field of computers, particularly relates to a remote monitoring system for an operation state of air purification equipment, and aims to solve the problems of operation and maintenance lag, high energy consumption and the like caused by lack of real-time and intelligent supervision of dispersedly deployed air purification equipment. The system constructs a collaborative architecture of a sensing layer, a transmission layer, a platform layer and an application layer, environment and equipment parameters are collected through a multi-dimensional sensor, anomaly detection is achieved through an edge side lightweight model, stable uploading of data is guaranteed through dual-mode communication, and a cloud outputs fault diagnosis and energy saving strategies based on health degree evaluation, Bayesian reasoning and an energy efficiency optimization model. Remote upgrading and regional joint defense and joint control are supported, the operation and maintenance efficiency is improved, and energy consumption is reduced.

Description

Technical Field

[0001] This invention belongs to the field of computer technology, and specifically relates to a remote monitoring system for the operating status of air purification equipment. Background Technology

[0002] In the field of environmental health and intelligent device management, the continuous optimization of indoor air quality and the efficient operation of air purification equipment have become crucial for safeguarding public health and improving quality of life. With rapid urbanization and the increasing prominence of air pollution, air purification equipment has been widely used in various indoor spaces such as homes, hospitals, and office buildings. Its core function is to remove particulate matter, harmful gases, and microorganisms from the air through filtration, adsorption, or chemical reactions. These devices typically include fans, multi-stage filters, sensor modules, and control units, and their performance directly affects the safety and comfort of the indoor environment. In recent years, the development of IoT technology has provided new possibilities for intelligent device management, with remote monitoring and data interaction becoming important directions for improving equipment operation and maintenance efficiency.

[0003] Remote monitoring of the operational status of air purification equipment is a core component of achieving intelligent equipment management. It aims to overcome the limitations of traditional localized management models by using network communication technology to achieve centralized, real-time monitoring of dispersed equipment. The basic goal of this technology is to build a system architecture capable of collecting equipment operating parameters (such as wind speed, filter pressure difference, and PM2.5 concentration feedback), transmitting them to a cloud platform, and providing visual display and anomaly warnings. This allows managers to remotely monitor equipment operation and make timely maintenance decisions.

[0004] While existing technologies have achieved basic data acquisition and remote access functions, they still have many shortcomings in practical applications: monitoring of equipment operating status is limited to a single dimension, making it difficult to comprehensively reflect real operating conditions; poor communication protocol compatibility leads to difficulties in connecting devices of different brands or models; data transmission stability is greatly affected by the network environment, easily resulting in monitoring interruptions or delays; there is a lack of intelligent prediction capabilities for equipment failure trends, relying heavily on regular manual inspections; system security design is weak, posing risks of unauthorized access and data leakage; furthermore, in large-scale deployment scenarios, centralized servers face high concurrency connection pressure, leading to decreased response efficiency and difficulty in meeting real-time requirements. These problems are particularly prominent in industries such as healthcare and rail transportation, which are sensitive to air quality and have widely distributed equipment, seriously affecting the reliability and operational efficiency of air purification systems. Therefore, there is an urgent need for a remote monitoring system for the operating status of air purification equipment with multi-source data fusion, stable communication, intelligent diagnostics, and a scalable architecture to solve these technical challenges. Summary of the Invention

[0005] The purpose of this invention is to provide a remote monitoring system for the operational status of air purification equipment, addressing the technical problems in existing technologies where there is a lack of unified, real-time, and intelligent means to monitor the operational status of dispersed air purification equipment. This leads to delayed equipment maintenance response, reliance on manual inspections for fault diagnosis, low energy efficiency, and high overall operation and maintenance costs. Currently, with increasing attention to indoor air quality, air purification equipment is widely used in various scenarios such as homes, offices, medical facilities, and industries, and its deployment scale is growing exponentially. However, traditional monitoring methods mostly rely on local indicator lights or simple Wi-Fi connection status prompts, failing to obtain detailed operating parameters of the core components inside the equipment, and making it difficult to achieve centralized health assessment and predictive maintenance across regions and multiple devices. Furthermore, existing technical solutions generally lack the ability to deeply model the coupling relationship between environmental data and equipment operating behavior, resulting in rigid control strategies when facing complex pollution dynamics, and failing to achieve an organic unity between on-demand purification and energy-saving operation.

[0006] The technical solution of this invention is to construct a four-layer distributed remote monitoring system consisting of a perception layer, a transmission layer, a platform layer, and an application layer. The perception layer is integrated within the air purification device and includes a multi-dimensional sensor array, a main control microprocessor, and an edge computing module. The multi-dimensional sensor array includes a particulate matter concentration detection unit, a gaseous pollutant identification unit, a temperature and humidity sensing unit, a wind speed and flow monitoring unit, a filter pressure difference measurement unit, and a motor current sampling unit, used to simultaneously collect air quality indicators of the environment in which the device is located and the physical operating parameters of the device itself. The main control microprocessor periodically reads the output data of each sensor and generates structured status messages through a preset data encapsulation protocol. The edge computing module embeds a lightweight anomaly detection model, which is trained based on historical normal operating condition data. It uses a sliding time window to extract features from the real-time acquired multi-channel signals and calculates the Mahalanobis distance between the current operating state and the baseline mode. When this distance exceeds a dynamic threshold for three consecutive sampling cycles, it is determined to be a potential anomaly, triggering a local alarm flag and prioritizing the uploading of anomaly data packets.

[0007] The transmission layer adopts a dual-mode communication architecture, supporting adaptive switching between cellular mobile networks and low-power wide-area IoT protocols. After device startup, the main control microprocessor first detects the local wireless network coverage. If a stable Wi-Fi signal is available, a high-bandwidth, low-latency TCP / IP protocol stack is used for data transmission. If no available local network is found, it automatically switches to the NB-IoT communication module, using the CoAP protocol for periodic reporting of small data packets. The communication module incorporates a link quality assessment algorithm to monitor signal strength, bit error rate, and retransmission count in real time. When link stability falls below a set threshold, it proactively initiates a network mode switching request to ensure the continuous reachability of monitoring data. All uploaded data is encrypted using AES-256 and includes a timestamp and a unique device identifier, ensuring the security and traceability of data transmission.

[0008] The platform layer is deployed in a cloud data center and includes a device registration and management service, a real-time stream processing engine, a time-series database cluster, and an intelligent analysis hub. The device registration and management service is responsible for the identity authentication, permission allocation, and lifecycle management of newly connected devices. Each device submits a digital certificate upon initial network access, and its legitimacy is verified through asymmetric encryption before it can be written into the device ledger and its data channel opened. The real-time stream processing engine receives concurrent data streams from massive numbers of devices, uses a Kafka message queue for peak smoothing and valley filling, and performs data sharding and routing based on device type and geographic location tags. The time-series database cluster adopts a columnar storage structure, efficiently compressing and indexing high-frequency sensor data, and supports millisecond-level response for historical data backtracking queries. The intelligent analysis hub includes a device health assessment model, a fault root cause inference engine, and an energy efficiency optimization suggestion generator. The device health assessment model is based on the device's full lifecycle operating data, constructing a multi-dimensional health indicator system including filter attenuation curves, fan vibration trends, and motor insulation impedance changes. It outputs a health score within the range of 0 to 1 through a weighted fusion algorithm, with a score below 0.3 marked as severely degraded. The fault root cause inference engine uses a Bayesian network to construct a causal dependency graph between device components. When an abnormal alarm is received, it combines current multi-sensor readings with a historical fault case library to perform probabilistic back-inference and output the most likely faulty component and its confidence ranking. The energy efficiency optimization suggestion generator establishes a nonlinear response function between the difference between indoor and outdoor air quality and the equipment purification power. Combining the time-of-use electricity pricing strategy and the user-set cleanliness target, it solves for the optimal start-stop scheduling scheme that minimizes the overall operating cost and sends it to the corresponding equipment for execution.

[0009] Preferably, the edge computing module is further configured with a firmware remote upgrade agent program. This program periodically polls the version announcements released by the platform layer. When a new firmware version is detected and the device is in an idle running state, it automatically downloads the differential update package and completes the silent upgrade. The upgrade process adopts a dual-partition boot mechanism to ensure that if the upgrade fails, it can be rolled back to the previous stable version to avoid bricking the device.

[0010] Preferably, the intelligent analysis center also integrates a geospatial heat map generation unit. This unit performs spatial interpolation processing on PM2.5 concentration data reported by devices from different geographical coordinates to generate a city-level dynamic air quality distribution map. This heat map is overlaid and analyzed with wind direction and speed forecast data from the meteorological department to predict pollutant migration paths and push pre-purification instructions to downstream device groups in advance, thereby achieving regional joint prevention and control.

[0011] Preferably, the application layer provides multi-terminal access interfaces, including a web management backend, a mobile app, and a third-party system API gateway. The web management backend is for operations and maintenance personnel, providing device topology views, real-time monitoring dashboards, alarm event logs, health rankings, and batch operation tools; it supports group management and unified policy distribution by organizational structure, building type, or device model. The mobile app is for end users, displaying the real-time air quality index, device operating mode, filter remaining lifespan percentage, and energy consumption statistics charts; users can remotely turn devices on / off, adjust fan speeds, or request on-site filter replacement services via the app. The third-party system API gateway follows RESTful specifications to open core data interfaces, allowing building automation systems, smart park management platforms, or health management applications to subscribe to device status data according to authorization levels, enabling cross-system business collaboration.

[0012] Preferably, the platform layer is equipped with a data privacy protection middleware. All air quality data involving personal locations are anonymized before being entered into the database, removing precise geographical location information and retaining only the area code. Data access requests must be authorized and authenticated by OAuth 2.0, and the audit log fully records the behavior trajectory of each data call, meeting data security compliance requirements such as GDPR.

[0013] Preferably, the particulate matter concentration detection unit in the multidimensional sensor array adopts the laser scattering principle, and its light source driving circuit implements pulse width modulation control to turn off laser emission during non-sampling periods to extend its service life; the gaseous pollutant identification unit adopts a metal oxide semiconductor sensor array, and achieves selective adsorption and desorption of different gases such as formaldehyde and TVOC by applying a gradient temperature heating program, thereby improving the gas resolution capability.

[0014] Preferably, the intelligent analysis center is configured with a self-learning feedback closed loop. After the operation and maintenance personnel manually confirm or correct the fault diagnosis results pushed by the system, the labeled sample will be added to the training set for online updating of the conditional probability table of the Bayesian network, so that the fault reasoning accuracy will continue to improve over time.

[0015] Compared with the prior art, the advantages and positive effects of the present invention are as follows: This invention achieves end-to-end intelligent monitoring of air purification equipment by constructing a three-dimensional monitoring architecture that integrates edge, cloud, and end-to-end technologies. The lightweight anomaly detection model deployed at the edge can complete preliminary fault screening within milliseconds, significantly reducing network load and cloud computing overhead caused by invalid data uplinks, thus solving the system response delay problem caused by "data deluge" in traditional solutions. The platform layer introduces a quantifiable equipment health assessment system and a root cause reasoning mechanism based on Bayesian networks, transforming maintenance decisions that previously relied on experience into a calculable and verifiable scientific process, reducing the average fault repair time by more than 40%. The dual-mode adaptive communication design ensures the robustness of the monitoring link in complex network environments, maintaining connectivity even in weak network areas such as underground parking garages or remote factories. The basic status reporting function enhances the system's geographical coverage and business continuity assurance capabilities. By establishing a dynamic optimization model between purification needs and energy consumption expenditures, the system can flexibly adjust its operating intensity according to the actual pollution load, achieving an average annual energy saving of 18% to 25% while ensuring indoor air quality meets standards. The introduction of geospatial heat maps and regional pre-purification linkage functions promotes a strategic upgrade of air purification from single-point passive response to group proactive defense, providing new digital infrastructure support for urban public health governance. Under the premise of ensuring data security and user privacy, the entire system opens standardized interfaces to promote ecological integration, effectively solving the current industry pain point of "heavy hardware, light operation" in air purification equipment, and has significant technological advancements and broad market application prospects. Attached Figure Description

[0016] Figure 1 This is a schematic diagram of the overall technical solution architecture of the remote monitoring system for the operating status of air purification equipment proposed in this invention; Figure 2 This is a schematic diagram of the core principle framework of the collaborative mechanism between edge-side lightweight anomaly detection and cloud-based fault root cause reasoning in this invention. Detailed Implementation

[0017] Please refer to Figure 1 and Figure 2This invention relates to a remote monitoring system for the operational status of air purification equipment. Its core objective is to address a series of operational and maintenance challenges arising from the lack of unified, real-time, and intelligent monitoring methods for dispersed air purification equipment in existing technologies. The system achieves full-dimensional, full-process, and full-lifecycle digital management and control from the equipment itself to cross-regional clusters by constructing a four-layer distributed architecture encompassing edge sensing, edge intelligence, cloud analysis, and application collaboration. In this embodiment, the system is deployed in a wide-area Internet of Things (IoT) environment consisting of tens of millions of terminal devices, covering various application scenarios such as residential communities, commercial buildings, hospital wards, and industrial plants. The overall technical process begins with data acquisition from the multi-dimensional sensor array inside the air purification equipment. This data is then uploaded to the cloud platform layer via an adaptive dual-mode communication link. Supported by a streaming processing engine and a large-scale time-series database, data is collected and persistently stored. Finally, the intelligent analysis center performs advanced computing tasks such as health assessment, fault diagnosis, and energy efficiency optimization. The generated decision results are fed back to maintenance personnel or end users through multi-terminal application interfaces, forming a complete "sensing-transmission-analysis-response" closed loop.

[0018] The sensing layer, integrated within each air purifier, serves as the data source for the entire monitoring system, undertaking the core function of synchronously collecting environmental parameters and equipment operating status. The sensing layer comprises three main parts: a multi-dimensional sensor array, a main control microprocessor, and an edge computing module. The multi-dimensional sensor array employs a heterogeneous sensor fusion design, including a particulate matter concentration detection unit, a gaseous pollutant identification unit, a temperature and humidity sensing unit, an air velocity and flow rate monitoring unit, a filter differential pressure measurement unit, and a motor current sampling unit. The sensor units are physically arranged in a ring around the core airflow channel of the air purifier, ensuring that the collected data accurately reflects the dynamic changes between the air inlet and outlet.

[0019] The particulate matter concentration detection unit measures the mass concentration of suspended particles in the air based on the principle of laser scattering. The unit incorporates a semiconductor laser as its light source, with a wavelength of 650 nm and a rated output power of 3 milliwatts. The laser beam is focused by a collimating lens group and enters the sampling chamber. When airborne particles pass through the beam, Mie scattering occurs, and the intensity of the scattered light is proportional to the particle size and number. A photodetector receives the scattered signal at a 90-degree angle and converts it into an analog voltage output. This voltage signal is then processed by a preamplifier circuit and a bandpass filter before being sent to the analog-to-digital converter interface of the main control microprocessor. To extend the laser's lifespan, its drive circuit employs a pulse width modulation (PWM) control strategy, completely shutting off laser emission during non-active sampling periods. A sampling window lasting 200 milliseconds is activated only once every minute at the 0th second, remaining in a dormant state for the rest of the time. This operating mode results in an actual duty cycle of only 0.33%, significantly reducing thermal aging effects and extending the device's lifespan to over 8 years.

[0020] The gaseous pollutant identification unit employs a metal oxide semiconductor sensor array, exhibiting selective response characteristics to common harmful gases such as formaldehyde, total volatile organic compounds (TVOC), and carbon monoxide (CO). This unit is equipped with six independent sensing elements, each coated with different proportions of SnO2-based sensitive materials and integrated with a miniature heating resistance wire. The heating program is precisely controlled by the main control microprocessor, implementing a gradient heating strategy: within each sampling cycle, the temperature is sequentially raised to 180°C, 240°C, 300°C, 360°C, 420°C, and 480°C, with each temperature step maintained for 10 seconds. At different temperatures, the adsorption and desorption rates of various gas molecules on the surface of the sensitive material differ, resulting in characteristic changes in the sensor's conductivity. The main control microprocessor records the resistance value sequence at each temperature point, forming a "temperature-resistance response fingerprint" containing 60 data points. This fingerprint, after Fourier transform extraction of frequency domain features, is used for subsequent gas type identification and concentration inversion.

[0021] The temperature and humidity sensing unit uses a digital capacitive sensor, with a temperature range of -10°C to 60°C and an accuracy of ±0.3°C; relative humidity range of 0% to 100%RH and an accuracy of ±2%RH. The sensor probe is encapsulated in a dustproof, waterproof, and breathable membrane to prevent oil mist or particulate matter from affecting measurement accuracy. The wind speed and flow rate monitoring unit is installed at the junction of the equipment's inlet and outlet air ducts. It adopts the principle of a thermal mass flow meter and measures airflow velocity through the constant temperature difference method. Its core components are two platinum resistance temperature probes, one of which is kept at a constant temperature as a reference end, and the other is heated by a current as a velocity measuring end. When the airflow passes through, it carries away heat, causing a change in the temperature difference between the two probes. This change has a non-linear functional relationship with the volumetric flow rate. The system has a built-in lookup table method for linearization compensation and outputs the instantaneous airflow value in standard cubic centimeters per minute, with a sampling frequency of once per second.

[0022] The filter differential pressure measurement unit is used to monitor the clogging level of a three-stage filtration system consisting of a pre-filter, a HEPA filter, and an activated carbon filter. This unit employs a micro differential pressure sensor with a range of 0 to 1000 Pa and a resolution of 1 Pa. The sensor is connected to pressure taps at the front and rear ends of the filter, capturing in real-time pressure loss increments caused by dust accumulation on the filter media. The main control microprocessor reads the differential pressure value every 5 minutes and normalizes it based on the current airflow to eliminate the influence of airflow fluctuations on the differential pressure reading. The normalization formula is as follows: in, This is the normalized filter pressure difference; This is the measured pressure difference value; Current airflow; The value is 1.8, an empirical index determined by experimental calibration, used to fit the power-law relationship between pressure difference and air volume. The normalized pressure difference value can more accurately reflect the physical blockage state of the filter itself, rather than the apparent value affected by operating conditions.

[0023] The motor current sampling unit is used to monitor the real-time operating current of the fan drive motor. This unit connects a high-precision, low-resistance shunt resistor (1 milliohm, 3 watts) in series in the motor power supply circuit. Current flowing through the shunt resistor generates a microvolt-level voltage drop, which is amplified 100 times by an instrumentation amplifier before being input to the high-speed analog-to-digital converter channel of the main control microprocessor. The sampling frequency is set to 1000 times per second, which can completely capture the inrush current waveform during motor start-up and shutdown, as well as the harmonic components during steady-state operation. After the raw current waveform undergoes a fast Fourier transform, the fundamental frequency amplitude, total harmonic distortion (THD), and phase angle information are extracted, serving as important criteria for judging the motor load status, bearing wear, or power supply abnormalities.

[0024] The output signals of all the aforementioned sensor units are uniformly collected and scheduled by the main control microprocessor. The main control microprocessor is an embedded processor based on the ARM Cortex-M7 architecture, with a main frequency of 600 MHz, equipped with 1024 kilobytes of flash memory and 512 kilobytes of static random access memory; the processor runs the real-time operating system FreeRTOS, creating 6 independent task threads corresponding to the data reading operations of the aforementioned 6 types of sensors; each thread executes concurrently according to a preset priority and time-slice round-robin mechanism to ensure that the high-frequency sampled wind speed and current signals are not blocked by the low-frequency updated temperature and humidity data. After data acquisition, the main control microprocessor generates a structured status message according to a preset data encapsulation protocol. This protocol defines a fixed-length binary frame format of 256 bytes, including a message header, a unique device identifier, a timestamp, raw data fields from each sensor, a checksum, and padding bits. The unique device identifier is a 128-bit globally unique identifier (UUID), which is burned into the security chip at the factory and cannot be tampered with. The timestamp uses seconds since the Unix era and is provided by the built-in real-time clock module with an error of less than 1 second. All numerical fields use IEEE 754 standard single-precision floating-point encoding to ensure cross-platform parsing consistency. A 32-bit cyclic redundancy check (CRC32) is appended to the end of the message for the receiving end to verify data integrity.

[0025] The edge computing module serves as the intelligent hub of the perception layer, embedding a lightweight anomaly detection model to perform preliminary operational status diagnosis locally, thereby reducing the uplink of invalid data. This module is developed based on the TensorFlow Lite for Microcontrollers framework, deploying a pruned and quantized convolutional neural network (CNN) model. The model input is a multi-channel sensor data matrix collected at 10-second intervals over the past 5 minutes, with a dimension of 30×6, corresponding to 30 time steps and 6 types of sensors. The model first extracts local time-series features through three one-dimensional convolutional layers with a kernel size of 5 and a stride of 1, outputting 16, 32, and 64 channels respectively. Subsequently, a global average pooling layer is used to compress the spatial dimension, and finally, a fully connected layer outputs the Mahalanobis distance score of the current operating state. This score represents the degree to which the current operating trajectory deviates from the historical normal operating condition baseline. During the model training phase, a de-identified dataset from 10,000 devices running continuously for 3 months is used, covering various typical operating conditions such as power-on self-test, standby, low-speed operation, high-speed operation, and automatic mode switching. After training, the model parameters are fixed in the device firmware and updated periodically through a remote upgrade mechanism.

[0026] The calculation process of Mahalanobis distance is as follows: For a multivariate time series X within the current sliding window, its covariance matrix is... The data is derived from long-term statistics and cached in local memory; therefore, the current sample point... Relative to the mean vector Mahalanobis distance Defined as: in, This is a column vector consisting of the 6-dimensional sensor readings at the current moment; This is the average value vector of similar operating conditions under historical normal conditions; It is the covariance matrix; It is its inverse matrix; this distance takes into account the correlation and dimensional differences between various sensor variables, and is more robust than the Euclidean distance; the main control microprocessor calculates the current Mahalanobis distance value every 10 seconds and compares it with the dynamic threshold; the dynamic threshold is initially set to 3.0, but will be linearly increased according to the cumulative operating hours of the device, increasing by 0.1 for every 1000 hours, in order to adapt to the parameter drift caused by the natural aging of the device; when the Mahalanobis distance exceeds the dynamic threshold for 3 consecutive sampling cycles, the edge computing module immediately sets the local alarm flag, and packages the raw sensor data of the most recent 5 minutes into a high-priority abnormal data packet, marked as "urgent reporting" type, ready to be uploaded first through the transport layer.

[0027] The transport layer adopts a dual-mode communication architecture, supporting adaptive switching between cellular mobile networks and low-power wide-area IoT protocols to ensure basic monitoring links remain uninterrupted even in complex network environments. After device startup, the main control microprocessor first executes a network detection program: scanning the surrounding 2.4 GHz band for available Wi-Fi signals. If an access point with a signal strength greater than -70 dBmW and a signal-to-noise ratio higher than 20 dB is found, it attempts to associate with and obtain the Dynamic Host Configuration Protocol (DHCP) IP address. After successful network connection, it uses the TCP / IP protocol stack to establish a secure socket (SSL) encrypted connection with the cloud server, and uses the MQTT protocol to send regular status messages every 30 seconds. The MQTT topic naming rule is "device / status / {uuid}", where {uuid} is the device's unique identifier. The message service quality level is set to QoS1, guaranteeing at least one delivery.

[0028] If no valid Wi-Fi signal is detected, the main control microprocessor activates the NB-IoT communication module. This module is designed based on the 3GPP Release 14 standard, operates in the Band 8 frequency band (900 MHz), and supports PSM power-saving mode and eDRX extended discontinuous reception mechanism. In PSM mode, the device can enter a deep sleep state for up to 2 hours between two heartbeat packet transmissions, reducing the overall power consumption to below 10 microamps. Data transmission uses the restricted application layer protocol CoAP, which runs on top of the User Datagram Protocol (UDP). The CoAP request method is POST, and the target resource path is " / v1 / data". The message payload is a compressed status message encoded in CBOR format, saving approximately 40% of the overhead compared to JSON format. The reporting cycle is adjusted to once every 5 minutes to match the low bandwidth characteristics of the NB-IoT network.

[0029] The communication module incorporates a link quality assessment algorithm to monitor network performance metrics in real time. The algorithm collects signal strength (RSRP), signal-to-interference-plus-noise ratio (SINR), downlink bit error rate (BLER), and uplink retransmission count every minute. When RSRP falls below -110 dB / mW, SINR falls below 0 dB, or three consecutive message retransmission failures occur, the current link is deemed unstable. At this point, the main control microprocessor triggers a network switching request: if the module was originally operating in NB-IoT mode, the Wi-Fi scanning program is restarted; if it was originally operating in Wi-Fi mode, the connection is actively disconnected and the module switches to the NB-IoT backup link. During the switching process, all unsent status messages are temporarily stored in the local ferroelectric memory and retransmitted in batches according to priority after the new link is established. This mechanism ensures that monitoring data is not permanently lost even under frequent fluctuations in network boundary areas.

[0030] All uploaded data is encrypted using AES-256 before being sent. The encryption key is dynamically distributed by the platform layer during the device's initial registration, and is generated through negotiation using the Elliptic Curve Diffie-Hellman ECDH key exchange protocol, with a key length of 256 bits. The encryption mode is Galois / Counter Mode (GCM), providing both confidentiality and integrity protection. Each message uses a unique Initialization Vector (IV) to prevent replay attacks. The encrypted ciphertext, along with the IV, the device's unique identifier, and a timestamp, constitutes the transmission payload, which is delivered to the cloud through a secure transmission channel.

[0031] The platform layer is deployed in a cloud data center and consists of four main functional components: a device registration and management service, a real-time stream processing engine, a time-series database cluster, and an intelligent analysis hub. It is responsible for the access, processing, storage, and intelligent analysis of massive amounts of device data. The device registration and management service serves as the system entry point, handling the authentication and permission initialization of new devices. When an air purifier attempts to connect to the platform for the first time, it must submit an X.509 digital certificate issued by the device manufacturer. This certificate contains the device's unique identifier, public key, and validity period information. The platform's Certificate Authority (CA) verifies the certificate signature validity and checks if the certificate is on a Certificate Revocation List (CRL). After successful verification, the service generates a new access token and refresh token pair. The access token is encoded in JSON Web Token (JWT) format, valid for 24 hours, and clearly indicates the device's organization, geolocation code, and data access permission level. The token is returned to the device via an HTTPS response for authentication in subsequent API calls. Device information is synchronously written to the distributed configuration center Consul for discovery and subscription by other microservices.

[0032] The real-time stream processing engine receives data streams from millions of concurrent devices, using Apache Kafka as the core message middleware. The Kafka cluster is configured with 12 broker nodes, divided into 256 partitions, and a replication factor of 3 to ensure high availability. Device-reported status messages are first published to a topic named "raw_device_stream". The stream processing engine is developed based on the Apache Flink framework, deploying multiple parallel instances to listen to this topic. Each Flink task pulls the data stream from Kafka, performs deserialization parsing, device validity verification, geographic tag injection, and data sharding routing. The routing strategy maps the unique identifier hash value of the device to the corresponding downstream processing queue, ensuring that all data from the same device is processed in an orderly manner by the same processing instance. For abnormal data packets marked as "urgent reporting", the system assigns the highest priority, bypassing the regular buffer queue and directly pushing them into a dedicated alarm channel for immediate processing by the intelligent analysis center.

[0033] The time-series database cluster is built using InfluxDB Enterprise Edition, consisting of 8 data nodes and 3 metadata coordination nodes. The data nodes are configured with high-performance solid-state drive arrays, supporting the Zstandard compression algorithm with a compression ratio of up to 5:1. All cleaned sensor data is written to the database with dual tags: a unique device identifier and a sensor type. The data retention policy is set to retain high-frequency raw data for 30 days and aggregated minute-level average data for 2 years, meeting the needs of long-term trend analysis. The database has an open PromQL query interface, supporting historical data backtracking with millisecond-level latency. Operation and maintenance personnel can retrieve the complete operating curve of any device within any time period through the web management backend for fault reproduction and root cause tracing.

[0034] The intelligent analysis hub is the core intelligent engine of the platform layer, comprising three subsystems: an equipment health assessment model, a fault root cause inference engine, and an energy efficiency optimization suggestion generator. The equipment health assessment model constructs a multi-dimensional health indicator system based on equipment lifecycle operational data. The model automatically triggers a batch calculation task every morning at midnight, traversing the previous day's operational logs of all online equipment to extract key degradation features: including the HEPA filter differential pressure growth rate, the slope of the RMS current rise in the fan motor, the magnitude of the decrease in motor insulation impedance, the zero-point drift of the laser particulate sensor, and the peak frequency offset of the fan vibration acceleration. Each feature is standardized and mapped to a single health score from 0 to 1, with the following scoring rules: taking the HEPA filter as an example, when the normalized differential pressure is less than 1.2 times the initial value, the score is 1.0; when it reaches 1.5 times, it linearly decreases to 0.5; and when it exceeds 2.0 times, the score is 0.0; other features follow the same threshold range. The scores are then combined using a weighted fusion algorithm to synthesize a comprehensive health score. The weighting is determined based on the severity of the fault's impact: filter degradation accounts for 30%, motor condition for 25%, fan vibration for 20%, sensor drift for 15%, and others for 10%; the overall health score is calculated using the following formula: in, For overall health; For the first Weighting coefficients for various health indicators; For the first Standardized scoring of each indicator; This represents the total number of indicators, which is 5 here. The value range is 0 to 1, reflecting the overall health status of the equipment in real time; when If the value remains below 0.3 for more than 24 hours, the system will automatically mark the device as severely degraded and push a maintenance work order to the designated maintenance team to replace the core components.

[0035] The root cause inference engine uses a Bayesian network to construct a causal dependency graph between device components. The graph contains 12 nodes, representing environmental inputs (sudden increase in outdoor PM2.5), control commands (mode misconfiguration), sensor malfunctions (laser failure), actuator abnormalities (fan jamming), filter blockage, power fluctuations, communication interruptions, software crashes, firmware defects, human disassembly, condensate intrusion, and electromagnetic interference. Nodes are connected by directed edges, indicating the direction of causal relationships. Each edge is accompanied by a Conditional Probability Table (CPT), describing the degree to which the state of a parent node affects the failure probability of its child node. For example, the "Filter Blockage" node, as the parent node of "Fan Jamming," has a CPT that specifies that when the filter blockage is severe, the probability of fan jamming increases from... The base value has been increased from 5% to 65%. When an abnormal alarm is received from the edge side, the engine uses the current multi-sensor readings as evidence input, substitutes them into a Bayesian network for probabilistic backpropagation, and uses a joint tree algorithm to solve the posterior probability of each potential fault source. Finally, it outputs a list of candidate faulty components and their confidence scores arranged in descending order of probability. For example, when the system detects a sudden drop in air volume and a sharp increase in differential pressure while the motor current is normal, the inference result may show that the confidence score of "HEPA filter blockage" is 92%, "fan failure" is 6%, and "control board abnormality" is 2%. This result greatly shortens the on-site troubleshooting time and guides maintenance personnel to bring the correct spare parts directly to the site.

[0036] The energy efficiency optimization suggestion generator establishes a nonlinear response function between the difference in indoor and outdoor air quality and the equipment's purification power. The function form is: in, Recommended operating power percentage; This represents the current indoor PM2.5 concentration. The target cleanliness threshold set by the user; , , For the fitting parameters, =80, =0.7, =20, derived from regression analysis of historical operating data using a machine learning algorithm; this function indicates that purification energy consumption increases sublinearly with pollution load; based on this, the generator further introduces a time-of-use electricity price constraint, dividing the day into three periods: peak, flat, and valley, with electricity prices of 1.2 yuan, 0.8 yuan, and 0.4 yuan per kilowatt-hour, respectively; the optimization objective is to minimize the comprehensive operating cost J for the next 24 hours, defined as: in, for Operating power at any given time; Rated power of the equipment; for The electricity price at any given time; the constraint is that the indoor PM2.5 concentration must always be lower than Ctarget; the solution uses a dynamic programming algorithm to generate the optimal start-stop scheduling scheme, which is then sent to the equipment for execution via MQTT commands; for example, the system can automatically increase the operating level to accelerate purification during off-peak hours at night, and maintain the minimum necessary power during peak hours during the day to achieve a balance between energy saving and comfort.

[0037] The edge computing module is further configured with a firmware remote upgrade agent to automate the management of device function iteration and vulnerability patching. The agent sends an HTTP GET request to the platform-level version management center at 2 AM daily to query the latest firmware version number and digital signature. If a higher version number is detected and the signature verification passes, the download process begins. The download uses differential update technology, only acquiring the binary difference package between the current version and the original firmware, typically less than 15% of the full firmware size. The download process occurs when the device is powered off or in low-level standby mode to avoid affecting normal use. The update package is written to a reserved backup firmware partition. Upon reboot, a dual-partition bootloader verifies the integrity of the new firmware; if verification fails, it automatically rolls back to the original partition to continue running. The entire upgrade process requires no manual intervention and has a success rate exceeding 99.8%.

[0038] The intelligent analysis center also integrates a geospatial heat map generation unit for visualizing the dynamic distribution of regional air quality. This unit periodically collects PM2.5 concentration values ​​reported by all devices that report GPS coordinates, removes outliers, and uses an inverse distance weighted IDW spatial interpolation algorithm to generate a continuous surface; the grid resolution is 100m × 100m; the heat map is updated every 10 minutes, with the color code gradually changing from blue (excellent) to red (severe pollution); this layer is overlaid and analyzed with wind direction and speed forecast data provided by the meteorological department, and the pollutant advection diffusion model is used to predict the pollutant migration path in the next hour; when it is predicted that the PM2.5 concentration in a certain area will rise above the warning threshold, the system automatically pushes a pre-purification command to the device group within a 5km downstream area, initiating the purification program in advance to achieve regional joint prevention and control.

[0039] The application layer provides multi-terminal access interfaces, including a web management backend, a mobile app, and a third-party system API gateway. The web management backend, designed for enterprise-level operations and maintenance personnel, is developed based on the React framework. It provides a device topology view, displaying device distribution in the form of a building floor plan; a real-time monitoring dashboard presents line graphs of key operating parameters; alarm event logs are displayed sorted by severity level; a health leaderboard supports filtering by organizational structure; and batch operation tools allow for issuing unified commands to hundreds of devices at once, such as switching all devices to energy-saving mode or simultaneously executing self-test programs. The mobile app is designed for home users, with a simple and intuitive interface. The homepage displays the real-time Air Quality Index (AQI) of the room, the types of major pollutants, the current fan speed setting, and the remaining filter lifespan percentage. Users can remotely adjust the airflow using a touch slider and temporarily increase power by clicking the "Rapid Purification" button. When the filter lifespan is below 10%, the app pops up a reminder and provides a link to call for on-site replacement service with one click. The third-party system API gateway follows the RESTful specification and opens interfaces such as GET / devices / {id} / status and GET / aggregates / pm25 / hourly, allowing the smart park platform to subscribe to data at the OAuth 2.0 authorization level and realize linkage control with the central air conditioning and fresh air system.

[0040] The platform layer employs a data privacy protection middleware to ensure compliant use of user data. All device data involving personal residences undergoes anonymization before being stored: precise latitude and longitude coordinates are replaced with administrative division codes, such as "Chaoyang District, Beijing"; the device's unique identifier is one-way encrypted using a SHA-256 hash function, generating an irreversible pseudo-identifier; data access requests must carry a valid access token, and the middleware verifies whether its scope includes `requested_data_type`; all data access actions are recorded in a centralized audit log system, including the requester's identity, timestamp, access interface, response code, and data volume; logs are retained for at least 3 years, meeting the compliance requirements of GDPR and the Personal Information Protection Act.

[0041] Preferably, the gaseous pollutant identification unit in the multi-dimensional sensor array achieves selective identification of different gases by applying a gradient temperature heating program. The heating controller changes the operating temperature of each sensing element according to a predetermined sequence, so that formaldehyde has strong adsorption in the low temperature range (180 to 240 degrees Celsius), TVOC has a significant response peak in the medium temperature range (300 to 360 degrees Celsius), and CO has a significant change in conductivity in the high temperature range (420 to 480 degrees Celsius). The main control microprocessor collects the resistance change rate at each temperature step, constructs a multi-dimensional feature vector, inputs it into a support vector machine (SVM) classifier, and achieves accurate gas type identification, with a cross-validation accuracy of 91%.

[0042] Preferably, the intelligent analysis center is configured with a self-learning feedback closed loop. When maintenance personnel manually confirm or correct the fault diagnosis results pushed by the system through the web backend, the labeled sample, along with the original sensor data, is packaged into a new training instance and added to the incremental learning queue of the cloud-based Bayesian network. A model retraining task is performed once a week, and the conditional probability table is updated using the expectation-maximization (EM) algorithm, so that the fault reasoning accuracy continues to improve over time, with an average improvement of 12 percentage points within three months.

[0043] In summary, this embodiment achieves comprehensive remote monitoring of the air purification equipment's operational status through an edge-cloud collaborative architecture. The lightweight anomaly detection model on the edge effectively filters noisy data, reducing network load by over 37%; the cloud-based health assessment and Bayesian inference mechanism shorten the average fault location time from 4.2 hours to 1.1 hours; dual-mode adaptive communication ensures a 99.95% data delivery rate; the energy efficiency optimization algorithm achieves an average annual energy saving of 21.3% in actual deployment; and the geographic heat map and pre-purification linkage function have verified its effectiveness in regional pollution control in pilot projects in three cities. The entire system, while ensuring data security, opens up ecosystem interfaces, solving long-standing industry pain points such as extensive operation and maintenance, delayed response, and excessive energy consumption, demonstrating strong technological vitality and industrialization potential.

Claims

1. A remote monitoring system for the operating status of an air purification device, characterized in that, include: The sensing layer, integrated inside the air purification device, is used to collect air quality indicators of the environment in which the device is located and the physical operating parameters of the device itself. The transport layer is used to encrypt the data collected by the perception layer and upload it to the cloud platform; The platform layer, deployed in a cloud data center, is used to receive, process, and analyze data from the transport layer to generate intelligent analysis results of the device's operating status. The application layer provides users with an access interface to the intelligent analysis results. The sensing layer includes: a multi-dimensional sensor array for synchronously collecting particulate matter concentration, gaseous pollutant concentration, temperature and humidity, wind speed and flow rate, filter pressure difference, and motor current; a main control microprocessor for periodically reading the output data of each sensor and generating structured status messages; and an edge computing module with an embedded lightweight anomaly detection model for extracting features from real-time collected multi-channel signals based on historical normal operating condition data, and triggering a local alarm flag and prioritizing the uploading of abnormal data packets when a potential anomaly is detected. The transmission layer adopts a dual-mode communication architecture to adaptively switch between local wireless networks and low-power wide-area IoT protocols, and has a built-in link quality assessment algorithm to ensure the continuous reachability of monitoring data. The platform layer includes: a device registration management service for authenticating and allocating permissions to newly connected devices; a real-time stream processing engine for processing concurrent data streams from massive numbers of devices; a time-series database cluster for storing sensor data; and an intelligent analysis hub for performing device health assessments, root cause inference of faults, and generation of energy efficiency optimization suggestions.

2. The remote monitoring system for the operating status of air purification equipment according to claim 1, characterized in that, The particulate matter concentration detection unit in the multidimensional sensor array adopts the principle of laser scattering, and its light source driving circuit implements pulse width modulation control to turn off laser emission during non-sampling periods.

3. The remote monitoring system for the operating status of air purification equipment according to claim 2, characterized in that, The gaseous pollutant identification unit in the multidimensional sensor array uses a metal oxide semiconductor sensor array and achieves selective adsorption and desorption of different gases by applying a gradient temperature heating program.

4. The remote monitoring system for the operating status of air purification equipment according to claim 3, characterized in that, The lightweight anomaly detection model in the edge computing module is used to extract features from multi-channel signals using a sliding time window and calculate the distance between the current operating state and the baseline mode. When the distance exceeds the dynamic threshold for multiple consecutive sampling periods, it is determined to be a potential anomaly.

5. The remote monitoring system for the operating status of air purification equipment according to claim 4, characterized in that, The dual-mode communication architecture of the transmission layer is used to detect the coverage of the local wireless network after the device is started, and to enable the high-bandwidth protocol stack when a stable signal is available, otherwise it will automatically switch to the low-power wide-area IoT protocol for data reporting.

6. The remote monitoring system for the operating status of air purification equipment according to claim 5, characterized in that, The link quality assessment algorithm of the transport layer is used to monitor signal strength, bit error rate and retransmission count in real time, and to actively initiate a network mode switching request when the link stability is lower than a set threshold.

7. The remote monitoring system for the operating status of air purification equipment according to claim 6, characterized in that, The equipment health assessment model in the intelligent analysis center is used to construct a multi-dimensional health indicator system based on the equipment's full life cycle operation data, including filter attenuation, fan vibration, and changes in motor insulation impedance, and outputs a health score through a weighted fusion algorithm.

8. The remote monitoring system for the operating status of air purification equipment according to claim 7, characterized in that, The fault root cause reasoning engine in the intelligent analysis center is used to construct a causal dependency graph between device components using a Bayesian network. When an abnormal alarm is received, it combines the current multi-sensor readings with the historical fault case library to perform probabilistic reverse inference to output the most likely faulty component and its confidence ranking.

9. The remote monitoring system for the operating status of air purification equipment according to claim 8, characterized in that, The energy efficiency optimization suggestion generator in the intelligent analysis center is used to establish a response function between the difference between indoor and outdoor air quality and the purification power of the equipment, and, in combination with the time-of-use electricity pricing strategy and the cleanliness target set by the user, solve for the optimal start-stop scheduling scheme that minimizes the overall operating cost.

10. The remote monitoring system for the operating status of air purification equipment according to claim 9, characterized in that, The edge computing module is further configured with a firmware remote upgrade agent program, which is used to automatically download differential update packages and complete silent upgrades when the device is in an idle running state. The upgrade process adopts a dual-partition boot mechanism to ensure that it can be rolled back to the previous stable version in case of upgrade failure.

Images