Data processing method and system of wireless passive temperature sensor

Abstract

The invention discloses a data processing method and system for a wireless passive temperature sensor. The method comprises four stages of data acquisition layer preprocessing, edge node dynamic filtering, cloud adaptive calibration and intelligent early warning: extracting multi-dimensional temperature characteristic parameters through orthogonal demodulation, and eliminating singular values by adopting double windows; a filtering algorithm is dynamically switched based on noise statistical characteristics, and anti-interference and low-delay balance is achieved; constructing a multi-dimensional drift compensation model, and dynamically updating calibration parameters through a particle swarm optimization algorithm; and constructing a four-level early warning mechanism based on the absolute temperature, the change rate and the fluctuation degree. The system comprises a sensing layer, an edge layer, a cloud layer and an application layer, and realizes data acquisition-processing-calibration-early warning full-link cooperation. The method solves the problems of signal distortion, poor filtering adaptability, insufficient drift compensation and early warning rigidity in the prior art, and is suitable for long-term precision monitoring of special scenes such as high-voltage equipment and rotating machinery.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of sensor data processing, and in particular to a data processing method and system of a wireless passive temperature sensor. BACKGROUND

[0002] At present, the common data processing methods of wireless passive temperature sensors mainly include RFID (Radio Frequency Identification) technology, SAW (Surface Acoustic Wave) technology, and energy collection-based processing mode. The sensor based on RFID technology obtains energy through radio frequency signals and transmits temperature data. After the reader reads the data, it generally performs simple filtering processing to remove noise interference, and then transmits the data to the backend system for further analysis. The temperature sensor using the SAW technology converts temperature information into frequency signals based on the principle that the propagation characteristics of the SAW change with temperature. After the receiving device receives the signal, it interprets the frequency signal into temperature data through a specific algorithm. The sensor based on energy collection collects energy from the environment (such as solar energy, radio frequency energy, etc.). In the data processing process, the collected energy is first managed to ensure the stable operation of the sensor, and then the collected temperature data is processed for preliminary amplification, digitization, etc. and then transmitted.

[0003] The existing technology mainly relies on the advantages of no power supply requirement and strong anti-electromagnetic interference capability, and is widely used in industrial monitoring field. However, there are the following key problems: Signal distortion problem is prominent: in wireless transmission, multipath effect and metal shielding cause signal attenuation (attenuation can reach 20-30 dB), original data contains a large amount of pulse noise and random drift, temperature measurement error is often more than ±3℃, which cannot meet the demand of precise monitoring (such as transformer winding temperature requirement ±1℃).

[0004] Poor adaptability of filtering algorithm: traditional fixed threshold filtering (such as mean filtering) is invalid when the noise characteristics change suddenly (such as sudden electromagnetic interference), and the data efficiency is less than 85%; and complex filtering algorithm (such as Kalman filtering) causes processing delay > 200ms due to the limitation of algorithm power when deployed at the edge.

[0005] Long-term drift compensation is missing: the core components of the sensor (such as SAW resonator, temperature sensitive capacitor) produce aging drift over time (more than 1 year), and the drift can reach 2-5℃. The existing system relies on manual periodic calibration (period 3-6 months), which has high maintenance cost and unreliable data during calibration interval.

[0006] The early warning mechanism is rigid: a single fixed threshold is used to trigger early warning, without considering the temperature change rate (such as abnormal heating of equipment often accompanied by a rate > 5℃ / min), resulting in a false alarm rate > 5% and a high risk of missed reporting. SUMMARY

[0007] In order to overcome the defects of the prior art, the application provides a data processing method and system of a wireless passive temperature sensor, which realizes high-precision, low-delay and full-life-cycle reliable processing of wireless passive temperature data through a technical architecture of "layered processing + dynamic adaptation + intelligent calibration".

[0008] In one aspect, the application provides a data processing method of a wireless passive temperature sensor, which comprises the following steps: (1) Data acquisition layer preprocessing: extracting temperature characteristic values from the radio frequency signals output by the sensor, and removing singular values deviating from the mean value ±3σ through a sliding window, wherein the window size is 5 sampling points; (2) Edge node dynamic filtering: calculating a data fluctuation coefficient k, and switching the filtering algorithm according to the k value: using Kalman filtering when k≤0.05, using wavelet threshold filtering when 0.05 (3) Cloud adaptive calibration: calibrating the preprocessed temperature based on a drift function f(t)=a·t²+b·t+c, and iteratively optimizing the coefficients a, b and c every 1000 groups of data; (4) Intelligent early warning: performing quadrature demodulation on the radio frequency signals output by the sensor, extracting the frequency offset Δf, signal strength RSSI and phase difference φ, and calculating the initial temperature; removing singular values through a double window: outputting an effective data sequence D; triggering four levels of early warning based on three-dimensional indicators, starting local prompts at the first level of early warning, pushing APP messages at the second level of early warning, triggering dispatch at the third level of early warning, and stopping the equipment at the fourth level of early warning, and the second and above levels of early warning need to be verified for 3 consecutive sampling periods.

[0009] In some embodiments, in step (4), the three-dimensional indicators include: absolute temperature T, change rate v and fluctuation degree σ.

[0010] In some embodiments, in step (1), the radio frequency signal is in the UHF frequency band 902-928MHz, and in the mapping formula, k1=0.01℃ / Hz, k2=-0.1℃ / dBm, k3=5℃ / rad, and b=25℃.

[0011] In some embodiments, in step (2), the improved wavelet threshold filtering adopts a hard threshold λ=1.5σ√(2lnn), and the high-frequency coefficients are reconstructed by weighting factors (0.8-1.2).

[0012] In some embodiments, in step (3), the iteration number of the PSO algorithm is =100, the learning factor c1=c2=2, and the inertia weight ω linearly decreases from 0.9 to 0.4.

[0013] In some embodiments, the fluctuation σ in step (4) is the standard deviation of temperature within 10 minutes, the normal threshold σ0=0.5°C, and the rate threshold v0 is preset according to the type of equipment (v0=2°C / min for motor bearings and v0=2°C / min for transformer windings).

[0014] In another aspect, the application provides a data processing system for a wireless passive temperature sensor, which comprises: The perception layer includes UHF RFID temperature tags and readers, the size of the temperature tags is ≤15×15×3mm, the temperature range is -55°C to 150°C, the frequency of the readers is 902-928MHz, and the reading distance is 0.5-3m; The edge layer includes industrial-grade gateways that deploy dynamic filtering algorithm libraries and local caches; The cloud layer includes cloud-native architecture server clusters, including data platforms, algorithm engines, and rule engines; The application layer includes local touchscreens, web platforms, mobile APPs, and linked devices.

[0015] In some embodiments, the edge layer and the cloud layer communicate through the MQTT protocol, the data transmission rate is ≥100kbps, and the packet loss rate is ≤0.1%.

[0016] In some embodiments, the data platform of the cloud layer supports writing ≥100,000 data per second, the storage period is ≥5 years, and supports temperature curve backtracking and report generation.

[0017] According to the above embodiments, the beneficial effects include: Improved measurement accuracy: after dynamic filtering and adaptive calibration, the temperature measurement error is reduced from ±3°C to ±0.8°C, the long-term (3 years) drift compensation accuracy is ≤±0.5°C, and the precision industrial monitoring requirements are met.

[0018] Optimized processing efficiency: the single processing delay of the edge node is ≤30ms, which is reduced by 85% compared to the traditional method (200ms); the data efficiency is improved from 85% to 99.7%, and the anti-interference ability is significantly enhanced.

[0019] Reduced maintenance cost: cloud adaptive calibration replaces manual calibration, the maintenance period is extended to more than 3 years, and the annual maintenance cost of a single sensor is reduced by 70%.

[0020] Enhanced early warning reliability: four-level early warning mechanism combined with multi-parameter judgment, false alarm rate reduced from 5% to 0.3%, and missed alarm rate close to 0, ensuring timely response to equipment abnormalities. BRIEF DESCRIPTION OF DRAWINGS

[0021] Figure 1 A flowchart of a data processing method for a wireless passive temperature sensor. Detailed Implementation

[0022] The present application will now be described in further detail with reference to the accompanying drawings and specific embodiments. Similar elements in different embodiments are referred to by related similar element reference numerals. In the following embodiments, many details are described to facilitate a better understanding of the present application. However, those skilled in the art will readily recognize that some features may be omitted in different situations, or may be replaced by other elements, materials, or methods. In some cases, certain operations related to the present application are not shown or described in the specification. This is to avoid obscuring the core parts of the present application with excessive description. For those skilled in the art, detailed description of these related operations is not necessary; they can fully understand the related operations based on the description in the specification and general technical knowledge in the art.

[0023] The serial numbers assigned to components in this document, such as "first" and "second," are used only to distinguish the objects being described and have no sequential or technical meaning. The terms "connection" and "linkage" used in the application, unless otherwise specified, include both direct and indirect connections (linkages).

[0024] This application provides a data processing method and system for a wireless passive temperature sensor, which solves the problems of signal distortion, poor filter adaptability, insufficient drift compensation and rigid early warning, so that the temperature measurement error is ≤±0.8℃, the processing delay is ≤50ms, the long-term drift compensation accuracy is ≤±0.5℃, and the false alarm rate is <0.5%.

[0025] The data processing method for wireless passive temperature sensors includes four stages: data acquisition layer preprocessing, edge node dynamic filtering, cloud-based adaptive calibration, and intelligent early warning. The key technologies for each stage are as follows: (1) Data acquisition layer preprocessing (anti-interference enhancement) Signal feature extraction: The radio frequency signal output by the sensor (taking UHF RFID as an example, carrier frequency 920MHz±4MHz) is quadrature demodulated to extract temperature-sensitive feature parameters: frequency offset Δf (unit Hz), signal strength RSSI (unit dBm), and phase difference φ (unit rad); establish the mapping relationship between feature parameters and temperature: T0=k1·Δf +k2·RSSI + k3·φ + b (k1, k2, k3 are calibration coefficients, b is the intercept).

[0026] Dual-window singularity removal: Set a sliding main window (window size N=10 sampling points, sampling frequency f=10Hz) and an auxiliary window (N=3 points): The main window calculates the mean μ1 and standard deviation σ1, and removes outliers where |x - μ1| > 3σ1; The auxiliary window calculates the mean μ2 and the range R of the main window output data, and removes the mutation point with |R|>2℃ (adapted to the scene of rapid temperature change); The output pre-processing sequence D=[d1,d2,...,d n ] ensures that the singular value removal rate is greater than or equal to 99%.

[0027] (2) Dynamic filtering of edge nodes (adaptive noise suppression) Real-time analysis of noise characteristics: calculate the statistical characteristics of the sequence D: The fluctuation coefficient k=σ / μ (σ is the standard deviation and μ is the mean); Noise type discrimination index: kurtosis K (K>3 is pulse noise and K≈3 is Gaussian noise); Change rate v.

[0028] Multi-algorithm dynamic switching: construct a filtering algorithm decision tree: When k≤0.05 and K≈3 (stationary Gaussian noise): perform Kalman filtering, process noise variance Q=0.005, observation noise variance R=0.02; When 0.05 When v≤1℃ / s (slow change + strong interference): perform EMD-ICA joint filtering, EMD decomposition gets 8-layer IMF components, ICA separates noise dominant components and removes them to reconstruct the signal; When v>1℃ / s (rapid change): perform adaptive median filtering, the window size is dynamically adjusted according to v (when v=5℃ / s, the window size=5×5).

[0029] Real-time optimization: the edge uses fixed-point arithmetic to optimize the filtering algorithm, the single processing time is less than or equal to 30ms, and the local cache stores the latest 500 groups of data (without loss when power failure).

[0030] (3) Cloud adaptive calibration (full life cycle drift compensation) Multi-dimensional drift model construction: based on sensor factory data and running history, establish a drift compensation function: T_drift (t, T a , H) = a1t 2 + a2t + b1T a + b2H + c Where: t is the running time (h), T aH represents ambient temperature (°C), H represents relative humidity (%), and a1, a2, b1, b2, and c represent calibration parameters.

[0031] Dynamic calibration process: The edge node uploads 100 sets of filtered data T' and environmental parameters (T) to the cloud every hour. a H); The cloud-based drift model is used to calculate the compensation amount: ΔT_drift = T_drift(t,T) a , H); Actual temperature calibration: T = T' - ΔT_drift; Model self-updating: Every 5000 sets of data, the parameters are iteratively optimized using the particle swarm optimization algorithm (PSO) (iteration number = 100, learning factor c1 = c2 = 2, inertia weight ω linearly decreases from 0.9 to 0.4) to ensure that the model error is ≤0.3℃.

[0032] Abnormal data repair: For data loss caused by transmission interruption (≤5 consecutive missing sets), an LSTM-based time series prediction model is used to fill the gaps, with a prediction error ≤0.5℃.

[0033] Intelligent early warning (multi-parameter linkage judgment) The radio frequency signal output by the sensor is quadrature demodulated to extract the frequency offset Δf, signal strength RSSI, and phase difference φ. The initial temperature is calculated using the mapping formula T0=k1·Δf + k2·RSSI + k3·φ + b. A dual-window approach is used to remove outliers: a 10-point main window removes gross errors that deviate from the mean by ±3σ, and a 3-point auxiliary window removes abrupt changes with a range >2℃, outputting the effective data sequence D; Calculate the fluctuation coefficient k = σ / μ, kurtosis K, and rate of change v of D; The filtering algorithm is dynamically switched based on k, K, and v: Kalman filtering is used when k ≤ 0.05 and K ≈ 3; improved wavelet threshold filtering is used when 0.05 < k ≤ 0.2 and K > 3; EMD-ICA joint filtering is used when v ≤ 1℃ / s; and adaptive median filtering is used when v > 1℃ / s. The preprocessed temperature T' is output, and the processing delay is ≤ 30ms. Establish the drift compensation function T_drift(t, T) a , H) = a1t 2 + a2t + b1T a + b2H + c, where t is the running time, T a H represents ambient temperature, and H represents humidity. Calculate the actual temperature T = T' - T_drift(t, T) a For every 5000 sets of data, the PSO algorithm is used to optimize the parameters a1, a2, b1, b2, and c. Four levels of early warning are triggered based on three-dimensional indicators (absolute temperature T, rate of change v, and fluctuation σ). Level 1 early warning initiates local notification, Level 2 early warning pushes APP message, Level 3 early warning triggers work order dispatch, and Level 4 early warning links equipment shutdown. Moreover, Level 2 and above early warnings require verification for 3 consecutive sampling cycles.

[0034] Early warning parameter system: Constructing a three-dimensional early warning indicator system: Absolute temperature T: Normal operating range of the equipment [T_min, T_max]; Temperature change rate v: Normal rate threshold v0 (e.g., motor bearing v0=2℃ / min, transformer winding v0=2℃ / min). Temperature fluctuation σ t Temperature standard deviation within 10 minutes, normal threshold σ0 = 0.5℃.

[0035] Level 4 early warning mechanism: Early warning level Triggering condition Response measure Level 1 (prompt) [T∈[T_min-1, T_min)∪(T_max, T_max+1] and v≤v0] Local LED flashes, log records Level 2 (attention) [T∈[T_min-2, T_min-1)∪(T_max+1, T_max+2] or v∈(v0, 2v0]]] Push APP message, sound and light alarm Level 3 (warning) [T∈[T_min-3, T_min-2)∪(T_max+2, T_max+3] or v∈(2v0, 5v0]]] Trigger monitoring system pop-up, dispatch processing Level 4 (emergency) [T∈[T_min-3, T_max+3] and v≤5v0 and σ t > 2σ0]]> Linkage equipment shutdown, start emergency plan Early warning verification mechanism: Before a Level II or higher early warning is triggered, the conditions must be met for three consecutive sampling cycles (0.3s) to reduce false alarms caused by instantaneous interference.

[0036] Correspondingly, the data processing system of the wireless passive temperature sensor includes a sensing layer, an edge layer, a cloud layer, and an application layer. The hardware and software architecture is as follows: Perception layer Wireless passive temperature sensor: adopts UHF RFID tag (size 15×15×3mm), built-in temperature-sensitive LC resonant circuit, working frequency band 902-928MHz, reading distance 0.5-3m, temperature measurement range -55℃~150℃, sampling interval adjustable from 100ms to 10s; Reader: Integrates RF front-end (output power 30dBm) and signal processing unit, supports simultaneous identification of multiple tags (≤50 tags / second), and connects to edge nodes via Ethernet / RS485.

[0037] edge layer Hardware: Industrial-grade edge gateway (CPU: ARM Cortex-A53 quad-core 1.2GHz, memory: 1GB DDR4, storage: 8GB eMMC), supporting LoRaWAN / NB-IoT wireless communication; Software: Embedded Linux system, deploying a filtering algorithm library (optimized implementation in C language), a local data buffer (SQLite database), and a communication protocol stack (MQTT), with an average power consumption of ≤5W.

[0038] Cloud layer Server cluster: Adopting a cloud-native architecture (Kubernetes orchestration), including: Data platform: Distributed database (MongoDB + InfluxDB), supporting 100,000 data entries per second, with a storage period of ≥5 years; Algorithm engine: Deploy drift compensation model and prediction model (based on TensorFlow framework), GPU accelerated computing (NVIDIA T4); Rules engine: Configures alert logic and linkage strategies, and supports visual rule configuration; Interface: Provides RESTful API and WebSocket interfaces, supporting integration with third-party systems.

[0039] Application layer Local terminal: Industrial touch screen (10.1 inches), displaying real-time temperature curves and early warning information; Remote terminals: Web platform (B / S architecture), mobile APP (iOS / Android), supporting data query, report generation, and remote control; Linked equipment: audible and visual alarm (110dB), relay module (AC220V / DC24V output), emergency handling device.

[0040] In a specific embodiment, taking the monitoring of 110kV transformer winding temperature as an example, the implementation steps are explained in detail: Step 1: System Deployment Sensing layer: Three UHF RFID temperature tags (model: TP-RFID-01) are embedded in each of the high-voltage windings (A, B, and C phases) of the transformer, with the tags 5mm away from the winding surface; the reader (model: RD-9200) is installed on the outer wall of the transformer tank, 1.5m away from the tags, and the sampling frequency is set to 5Hz (200ms / time).

[0041] Edge layer: Deploy an edge gateway (model: EG-200) in the substation control cabinet, connect the reader via RS485, configure the LoRa communication module (frequency band 470MHz), and set the local cache capacity to 1000 groups.

[0042] Cloud layer: Deployed on a private power cloud, the data platform is configured with 10TB of storage, and the algorithm engine loads a transformer-specific drift model (T_drift(t, T)). a , H) = 2e-8t² + 1e-5t + 0.02T a + 0.01H + 0.05).

[0043] Application layer: The substation monitoring center deploys a web platform, maintenance personnel are equipped with a mobile APP, and audible and visual alarms and emergency trip relays (connected to the fourth-level early warning interface) are installed next to the transformer.

[0044] Step 2: Data Processing Flow Preprocessing: The reader receives the tag's RF signal and extracts Δf=1250Hz, RSSI=-65dBm, and φ=1.2rad. Using the mapping formula (k1=0.01℃ / Hz, k2=-0.1℃ / dBm, k3=5℃ / rad, b=25℃), T0=62.3℃ is obtained. Through a 10-point main window (μ1=62.1℃, σ1=0.8℃) and a 3-point auxiliary window (μ2=62.2℃, R=0.5℃), it is confirmed that there are no singular values, and the output is D=[62.3,62.1,62.2,...].

[0045] Dynamic filtering: Calculate k=0.012 (σ=0.73, μ=61.8), K=2.9 (approximate Gaussian noise), trigger Kalman filtering; after filtering, T'=62.0℃, processing time 28ms.

[0046] Cloud calibration: The sensor has been running for 2000 hours, ambient temperature T a =35℃, humidity H=60%, calculate ΔT_drift=2e-8×2000² + 1e-5×2000 + 0.02×35 + 0.01×60 + 0.05=0.23℃; actual temperature T=62.0-0.23=61.77℃.

[0047] Warning judgment: Transformer winding normal range [T_min=40℃, T_max=70℃], v=0.3℃ / min (≤v0=2℃ / min), σ t =0.4℃ (≤σ0=0.5℃), which is considered a normal state, and no warning is triggered.

[0048] Step 3: Verification of Abnormal Scenarios When a partial short circuit in a transformer causes a sudden rise in winding temperature: t=10:00:00, T'=71.2℃, after filtering, a first-level warning is triggered (LED flashing); At t=10:00:10, the temperature rises to 73.5℃, v=13℃ / min (>2v0=4℃ / min), triggering a level 3 warning (monitoring platform pop-up, dispatching maintenance order). At t=10:00:20, the temperature rises to 78℃, v=27℃ / min (>5v0=10℃ / min), triggering a level 4 warning, the emergency trip relay operates, cutting off the transformer power supply to prevent the accident from escalating.

[0049] The above examples illustrate this application only to aid understanding and are not intended to limit its scope. Those skilled in the art to which this application pertains can make various simple deductions, modifications, or substitutions based on the ideas presented.

Claims

1. A data processing method for a wireless passive temperature sensor, characterized in that, Comprising the following steps: (1) Data acquisition layer preprocessing: extracting temperature characteristic values from the sensor output radio frequency signal, removing outliers deviating from the mean value ±3σ through a sliding window, wherein the window size is 5 sampling points; (2) Edge node dynamic filtering: calculating the data fluctuation coefficient k, and switching the filtering algorithm according to the k value: k≤0.05 uses Kalman filtering, 0.05 (3) Cloud adaptive calibration: based on the drift function f (t)=a·t²+b·t+c, calibrate the preprocessed temperature, and iterate the coefficients a, b, and c every 1000 groups of data; (4) Intelligent early warning: orthogonal demodulation of the sensor output radio frequency signal, extracting frequency offset Δf, signal strength RSSI, and phase difference φ, calculating the initial temperature; using a double window to remove outliers: outputting the effective data sequence D; based on three-dimensional indicators, triggering four levels of early warning, the first level of early warning starts local prompt, the second level of early warning pushes APP message, the third level of early warning triggers dispatch, and the fourth level of early warning links equipment shutdown, and the second and above early warning needs to be verified for 3 consecutive sampling periods.

2. The method of claim 1, wherein, In step (4), the three-dimensional indicators include: absolute temperature T, change rate v, and fluctuation degree σ.

3. The method of claim 1, wherein, In step (1), the radio frequency signal is UHF band 902-928MHz, the mapping formula is k1=0.01℃ / Hz, k2=-0.1℃ / dBm, k3=5℃ / rad, and b=25℃.

4. The method of claim 1, wherein, In step (2), the improved wavelet threshold filtering uses hard threshold λ=1.5σ√(2lnn), and the high frequency coefficients are reconstructed by weighting factor (0.8-1.2).

5. The method of claim 1, wherein, In step (3), the iteration number of PSO algorithm is =100, the learning factor c1=c2=2, and the inertia weight ω decreases linearly from 0.9 to 0.

4.

6. The method of claim 2, wherein, In step (4), the fluctuation degree σ is the temperature standard deviation within 10 minutes, the normal threshold σ0=0.5℃, and the rate threshold v0 is preset according to the device type (motor bearing v0=2℃ / min, transformer winding v0=2℃ / min).

7. A system for implementing the method of any one of claims 1-6, characterized by Comprising: Sensing layer: containing UHF RFID temperature tag and reader, the temperature tag size ≤15×15×3mm, -55℃~150℃, the reader frequency is 902-928MHz, the reading distance is 0.5-3m; Edge layer: industrial gateway, deploying dynamic filtering algorithm library and local cache; Cloud layer: cloud native architecture server cluster, containing data platform, algorithm engine, and rule engine; Application layer: local touch screen, Web platform, mobile APP, and linked equipment.

8. The system of claim 7, wherein, The edge layer and cloud layer communicate through MQTT protocol, the data transmission rate is ≥100kbps, and the packet loss rate is ≤0.1%.

9. The system of claim 7, wherein, The cloud layer data platform supports writing ≥100,000 data per second, the storage period is ≥5 years, supports temperature curve backtracking and report generation.

Images