Online calibration method for high-flow respiratory humidification therapeutic apparatus

Abstract

The invention discloses an on-line calibration method for a high-flow respiratory humidification therapeutic apparatus, and belongs to the technical field of electronic measurement, and the method comprises the steps: carrying out multi-round data collection by collection equipment, carrying out cross-parameter cooperative calibration, predicting the future health state of the equipment through LSTM, and calculating the fault probability of the equipment based on a prediction error. The technical problems that in the prior art, under environmental parameter cross interference, dynamic calibration precision is insufficient, and hidden faults of equipment cannot be predicted early are solved, a cross-parameter compensation algorithm is adopted, a temperature-humidity-oxygen concentration nonlinear compensation model is established, the environmental interference error is reduced to be within + / -0.5%, and the accuracy of dynamic calibration is improved. And an LSTM-attention mechanism prediction model is adopted to realize fault probability prediction in the next 90 days.

Description

Technical Field

[0001] The invention belongs to the technical field of electronic measurement, and in particular relates to an online calibration method for a high-flow respiratory humidification therapeutic apparatus. Background Art

[0002] As a respiratory therapy device, the high-flow respiratory humidification therapy device has a better oxygenation effect than ordinary oxygen therapy, which can make patients more comfortable during the treatment process.

[0003] The current calibration technology of high-flow respiratory humidification therapy devices generally has the following defects:

[0004] Limitations of single-parameter calibration: Traditional methods only calibrate flow or oxygen concentration independently, ignoring the coupling effects of multiple parameters such as temperature and humidity, resulting in error accumulation when the environment changes (for example, a temperature fluctuation of 1°C can cause an oxygen concentration measurement deviation of 0.3% to 0.8%).

[0005] Static compensation model: The existing calibration coefficients are fixed locally on the device and cannot be dynamically updated based on historical data. After long-term use, the compensation becomes invalid due to sensor aging.

[0006] Passive fault warning: Relying on threshold alarm mechanism, it can only trigger response after the parameter exceeds the limit, and lacks early degradation prediction capability (the existing system cannot provide warning 30 days in advance). Summary of the invention

[0007] The purpose of the present invention is to provide an online calibration method for a high-flow respiratory humidification therapy device, which solves the technical problems of insufficient dynamic calibration accuracy under cross-interference of environmental parameters and inability to predict hidden equipment failures in early stages in traditional technologies.

[0008] To achieve the above object, the present invention adopts the following technical solution:

[0009] A method for online calibration of a high-flow respiratory humidification therapeutic device comprises the following steps:

[0010] Step 1: Deploy a central server at the data processing layer and deploy collection devices at the data collection layer. The central server and the collection devices communicate with each other through the communication network of the data transmission layer. After the collection device is initialized, it sends a link request to the central server, and the link request contains the device ID.

[0011] The central server monitors the connection request of the collection device and verifies the device ID. After the verification is correct, it establishes a connection with the collection device;

[0012] Step 2: The acquisition device performs a primary calibration data acquisition on the flow rate and oxygen concentration of the high-flow respiratory humidification and treatment device under test. Specifically, multiple rounds of multi-point acquisitions are performed within a preset time, and environmental data is acquired simultaneously to obtain a detection data set. The detection data set includes a flow rate data sequence set, an oxygen concentration data sequence set, a temperature data sequence set, a humidity data sequence set, and a barometric pressure data sequence set;

[0013] Step 3: After the acquisition device performs outlier screening and preprocessing on the detection data set, it performs local storage. After completing a primary calibration data acquisition, it performs cross-parameter collaborative calibration calculations, including calculating temperature-oxygen concentration cross-compensation and calculating humidity-oxygen concentration cross-compensation. It determines whether to trigger re-calibration based on the results of the cross-parameter collaborative calibration calculations: If triggered, re-calibration data acquisition is performed according to the method in Step 2; if not triggered, the detection data set is sent to the central server;

[0014] Step 4: The central server establishes an LSTM prediction model and trains the LSTM prediction model using historical data; it inputs the obtained detection data set into the LSTM prediction model, calculates the device failure probability, generates a device health report, and predicts the future device life;

[0015] Step 5: The central server determines whether the result of the device health report exceeds the calibration range. If it exceeds, it generates and saves a maintenance report, and at the same time sends a maintenance message to the acquisition device, which is displayed to the maintenance personnel through the human-machine interface.

[0016] Preferably, the acquisition device includes a main controller, a human-machine interface, an AD module, a flow sensor, a temperature and humidity sensor, a data memory, a communication module, a barometric pressure sensor, a power supply module, a power supply interface, a reference voltage module, and an oxygen concentration sensor. The human-machine interface, AD module, flow sensor, temperature and humidity sensor, data memory, communication module, and barometric pressure sensor are all connected to the main controller, and the oxygen concentration sensor and the reference voltage module are both connected to the AD module;

[0017] The power supply module supplies power to the main controller, human-machine interface, AD module, flow sensor, data memory, communication module, barometric pressure sensor, reference voltage module, and oxygen concentration sensor;

[0018] The power supply module is connected to the power supply interface, and the power supply interface is connected to an external regulated power supply.

[0019] Preferably, when performing Step 2, the specific steps are as follows:

[0020] The acquisition device locally pre-stores multiple rounds of acquisition strategies, specifically including:

[0021] Preset a primary calibration data acquisition strategy: Perform Z times of intensive sampling within a preset time window;

[0022] Preset the number of rounds Y for one-time calibration data acquisition;

[0023] In each round, data acquisition is performed using the spatial distribution sampling method, that is, 7 characteristic points are selected within the output range of the high-flow respiratory humidifier being measured;

[0024] The collected detection data set specifically includes:

[0025] Flow data sequence set Q ref ; Oxygen concentration data sequence set C O2,set , Temperature data sequence set T set , Humidity data sequence set H set , Air pressure data sequence set P set ; where set represents the complete data sequence set formed by multi-point acquisition in one round, Q represents the flow data, C represents the oxygen concentration data, T represents the temperature data, H represents the humidity data, and P represents the air pressure data.

[0026] Preferably, when performing step 3, the specific steps are as follows:

[0027] Step 3-1: Remove outliers through sliding window filtering and the three-standard-deviation method;

[0028] Step 3-2: Store the processed detection data set in local storage;

[0029] Step 3-3: After completing one-time calibration data acquisition, perform cross-parameter collaborative calibration calculation on the detection data set, specifically including calculating temperature-oxygen concentration cross-compensation and calculating humidity-oxygen concentration cross-compensation:

[0030] Calculate temperature-oxygen concentration cross-compensation:

[0031] Calculate the temperature error:

[0032] ΔT = T meas - T re f ;

[0033] where meas represents the original measurement data directly read by the sensor, and ref represents the preset standard reference value;

[0034] Calculate the influence of temperature on oxygen concentration:

[0035] C O2,corrected = C O2,meas + k T × ΔT × Q ref ;

[0036] where k Tis the temperature-oxygen concentration cross-influence coefficient, which is a preset value; corrected represents the corrected value after environmental compensation;

[0037] If |ΔT| > ε T , then trigger re-calibration, where ε T is the preset temperature error threshold;

[0038] Calculate the humidity-oxygen concentration cross-compensation:

[0039] Calculate the humidity error:

[0040] ΔH = H meas - H re f ;

[0041] Calculate the influence of humidity on oxygen concentration:

[0042] C O2,corrected = C O2,meas + k H ×ΔH;

[0043] where k H is the humidity-oxygen concentration cross-influence coefficient, which is a preset value;

[0044] If |ΔH| > ε H , then trigger re-calibration, where ε H is the preset humidity error threshold;

[0045] Step 3-4: According to the result of Step 3-3, if re-calibration is triggered, then return to Step 2 to re-collect data; otherwise, send the processed detection data set to the central server.

[0046] Preferably, when performing Step 4, the specific steps are as follows:

[0047] Step 4-1: Perform preprocessing of normalization on the data in the detection data set, set the time window m, and use m historical data points to predict the data of the next k time points;

[0048] Step 4-2: Establish an LSTM prediction model, and use the flow data, oxygen concentration data, temperature data, humidity data, and air pressure data in the historical data as inputs to train the LSTM prediction model to establish an input set:

[0049] X t = [Q t , C O2,t , T t , H t , P t ;

[0050] where t represents the time window;

[0051] Predict the equipment error in the next k periods:

[0052] X pred,t+k = f LSTM (X t-m ,..., X t );

[0053] Among them, m represents the length of the time window, k represents the prediction step, pred represents the predicted value, which is the prediction result output by the LSTM model; f LSTM (·) represents the LSTM model;

[0054] Loss function L:

[0055]

[0056] Among them, true represents the actual measurement and verification value, and N represents the number of training batches;

[0057] Step 4-4: Predict the equipment failure probability through the LSTM prediction model. The specific calculation formula is as follows:

[0058] Calculate the prediction error:

[0059] ΔX pred = X pred,t+k - X true,t+k ;

[0060] Among them, k represents the prediction step;

[0061] Calculate the equipment failure probability:

[0062]

[0063] Among them, Φ(·) is the cumulative distribution function of the standard normal distribution; σ is the standard deviation of the measurement noise, and ε max is the preset maximum allowable error. According to P f ′ ail 's calculation result, predict the life of the measured high-flow respiratory humidifier, generate an equipment health report. If the predicted equipment life is lower than the preset threshold, the central server sends maintenance information to the acquisition equipment.

[0064] An online calibration method for a high-flow respiratory humidifier according to the present invention solves the technical problems of insufficient dynamic calibration accuracy under cross-interference of environmental parameters and inability to early predict hidden equipment failures in the traditional technology. The present invention adopts a cross-parameter compensation algorithm, reduces the environmental interference error to within ±0.5% by establishing a temperature-humidity-oxygen concentration nonlinear compensation model, and adopts an LSTM-attention mechanism prediction model to achieve the prediction of the failure probability in the next 90 days. Brief Description of the Drawings

[0065] Figure 1 is the system architecture diagram of the present invention;

[0066] Figure 2 is the schematic block diagram of the acquisition device of the present invention;

[0067] Figure 3 is the main flowchart of the present invention. Detailed implementation manners

[0068] Comprising Figures 1 - 3 An online calibration method for a high-flow respiratory humidification treatment apparatus as shown, comprising the following steps:

[0069] Step 1: Deploy a central server in the data processing layer and an acquisition device in the data acquisition layer. The central server and the acquisition device communicate with each other through the communication network in the data transmission layer; after the acquisition device is initialized, it sends a link request to the central server, and the link request includes the device ID;

[0070] The central server listens to the link request of the acquisition device and performs device ID verification. After verification is correct, a link with the acquisition device is established;

[0071] The acquisition device includes a main controller, a human-machine interface, an AD module, a flow sensor, a temperature and humidity sensor, a data memory, a communication module, a pressure sensor, a power supply module, a power supply interface, a reference voltage module, and an oxygen concentration sensor. The human-machine interface, the AD module, the flow sensor, the temperature and humidity sensor, the data memory, the communication module, and the pressure sensor are all connected to the main controller, and the oxygen concentration sensor and the reference voltage module are both connected to the AD module;

[0072] The power supply module supplies power to the main controller, the human-machine interface, the AD module, the flow sensor, the data memory, the communication module, the pressure sensor, the reference voltage module, and the oxygen concentration sensor;

[0073] The power supply module is connected to the power supply interface, and the power supply interface is connected to an external regulated power supply.

[0074] The model of the main controller is STM32H743ZI;

[0075] The human-machine interface is an LCD display screen and buttons, which are respectively connected to the STM32H743ZI through the UART interface and the IO port;

[0076] The model of the AD module is AD5542C;

[0077] The model of the flow sensor is SFM3300, the model of the barometric pressure sensor is BMP388, and the model of the temperature and humidity sensor is AHT30. The flow sensor, barometric pressure sensor, and temperature and humidity sensor communicate with the main controller through different I2C buses respectively;

[0078] The model of the data memory is W25Q128JV, which communicates with the main controller through the SPI interface; the model of the communication module is ESP32-WROOM-32, which communicates with the main controller through the UART interface; the model of the power module is MP2451, the model of the reference voltage module is ADR433B, and the model of the oxygen concentration sensor is O2 / M-08.

[0079] In this embodiment, the hardware parameters of the acquisition device are as follows:

[0080] Main control chip STM32H743ZI: Adopts the Cortex-M7 core (480MHz), supports double-precision FPU, and meets the real-time data processing requirements (computing latency < 1μs).

[0081] SFM3300 flow sensor: Based on the thermal mass flow principle, the measuring range is 0 - 200SLM, and the accuracy is ±1.5% RD (superior to the medical device standard of ±3%).

[0082] O2 / M-08 oxygen concentration sensor: Electrochemical principle, response time T90 < 10s, cross-sensitivity < 1% (CO2 interference suppression).

[0083] ESP32 dual-mode communication: Supports Wi-Fi (802.11b / g / n) and Bluetooth BLE 5.0, ensuring local control can still be maintained when the network is disconnected.

[0084] Reference voltage ADR433B: Provides a reference voltage of 3.3V ± 0.02%, ensuring that the effective resolution of the 16-bit ADC of AD5542C reaches 14.5 bits (ENOB).

[0085] The main control chip STM32H743ZI presets the unique device identification ID.

[0086] Step 2: The acquisition device performs a calibration data acquisition on the flow rate and oxygen concentration of the high-flow humidified therapeutic ventilator to be measured. Specifically, it performs multiple rounds of multi-point acquisitions within a preset time, and at the same time acquires environmental data to obtain a detection data set. The detection data set includes a flow rate data sequence set, an oxygen concentration data sequence set, a temperature data sequence set, a humidity data sequence set, and a barometric pressure data sequence set;

[0087] The specific steps are as follows:

[0088] The acquisition device locally pre-stores multiple rounds of acquisition strategies, specifically including:

[0089] Preset the first calibration data acquisition strategy: conduct Z dense samplings within a preset time window;

[0090] Preset the number of rounds Y of the first calibration data acquisition;

[0091] In each round, use the spatial distribution sampling method for data acquisition, that is, select 7 characteristic points within the output range of the measured high-flow breathing humidification treatment device;

[0092] The collected detection data set specifically includes:

[0093] Flow data sequence set Q ref ; Oxygen concentration data sequence set C O2,set , Temperature data sequence set T set , Humidity data sequence set H set , Air pressure data sequence set P set ; Among them, set represents the complete data sequence set formed by multi-point acquisition in one round, Q represents the flow data, C represents the oxygen concentration data, T represents the temperature data, H represents the humidity data, and P represents the air pressure data.

[0094] In this embodiment, the specific strategy for multi-round multi-point data acquisition is as follows:

[0095] Characteristic point selection: Select 7 points (5, 10, 20, 30, 40, 50, 60 L / min) according to logarithmic distribution within the output range of the treatment device (such as 5 - 60 L / min) to cover the full-range non-linear characteristics;

[0096] Sampling mechanism: The acquisition time window for each round is 15 minutes, the sampling rate can be 50 Hz (anti-aliasing filter cut-off frequency 25 Hz), and the single-round data volume = 15 × 60 × 50 = 45,000 points / parameter.

[0097] Synchronous temperature and humidity acquisition: The AHT30 sensor has a built-in temperature and humidity compensation algorithm, and the humidity accuracy is ±2% RH within the range of -40 to 85 °C.

[0098] Air pressure correction: Obtain the atmospheric pressure through the BMP388, which can be used for altitude compensation of the oxygen concentration value (ΔC_O2 = 0.1% / 100 m).

[0099] Step 3: After the acquisition device performs outlier screening and preprocessing on the detection data set, perform local storage, and after finishing one calibration data acquisition, perform cross-parameter collaborative calibration calculation, including calculating the temperature-oxygen concentration cross-compensation and calculating the humidity-oxygen concentration cross-compensation, and judge whether to trigger re-calibration according to the cross-parameter collaborative calibration calculation result: If triggered, perform re-calibration data acquisition according to the method in Step 2; if not triggered, send the detection data set to the central server;

[0100] The specific steps are as follows:

[0101] Step 3-1: Remove outliers through sliding window filtering and the three-sigma method;

[0102] In this embodiment, the sliding window filtering has a window width of 100 points (2 seconds of data), and median filtering is used to remove pulse interference.

[0103] The three-sigma method calculates the MAD (Median Absolute Deviation) for each window of data to replace the standard deviation, improving the anti-noise ability.

[0104] Step 3-2: Store the processed detection data set in local storage;

[0105] Step 3-3: After finishing one calibration data acquisition, perform cross-parameter collaborative calibration calculation on the detection data set, specifically including calculating temperature-oxygen concentration cross-compensation and calculating humidity-oxygen concentration cross-compensation:

[0106] Calculate temperature-oxygen concentration cross-compensation:

[0107] Calculate the temperature error:

[0108] ΔT = T meas - T ref ;

[0109] Where meas represents the original measurement data directly read by the sensor, and ref represents the preset standard reference value;

[0110] Calculate the influence of temperature on oxygen concentration:

[0111] C O2,corrected = C O2,meas + k T ×ΔT×Q ref ;

[0112] Where k T is the temperature-oxygen concentration cross-influence coefficient, which is a preset value; corrected represents the corrected value after environmental compensation;

[0113] If |ΔT| > ε T , then trigger re-calibration, where ε T is the preset temperature error threshold;

[0114] Calculate humidity-oxygen concentration cross-compensation:

[0115] Calculate the humidity error:

[0116] ΔH = H meas - H ref ;

[0117] Calculate the influence of humidity on oxygen concentration:

[0118] C O2,corrected = C O2,meas + k H ×ΔH;

[0119] where k H is the humidity-oxygen concentration cross-influence coefficient, which is a preset value;

[0120] If |ΔH| > ε H , then trigger re-calibration, where ε H is the preset humidity error threshold;

[0121] Step 3-4: According to the result of Step 3-3, if re-calibration is triggered, return to Step 2 to re-collect data; otherwise, send the processed detection data set to the central server.

[0122] Step 4: The central server establishes an LSTM prediction model and trains the LSTM prediction model using historical data; input the obtained detection data set into the LSTM prediction model, calculate the equipment failure probability, generate an equipment health report, and predict the future equipment life;

[0123] The specific steps are as follows:

[0124] Step 4-1: Preprocess the data in the detection data set by normalization, set the time window m, and use m historical data points to predict the data of the next k time points;

[0125] Step 4-2: Establish an LSTM prediction model, and use the flow data, oxygen concentration data, temperature data, humidity data, and air pressure data in the historical data as inputs to train the LSTM prediction model, and establish an input set:

[0126] X t = [Q t , C O2,t , T t , H t , P t ;

[0127] where t represents the time window;

[0128] Predict the equipment error in the next k periods:

[0129] X pred,t+k = f LSTM (X t-m ,..., X t );

[0130] Among them, m represents the time window length, k represents the prediction step, pred represents the predicted value, which is the prediction result output by the LSTM model; f LSTM (·) represents the LSTM model;

[0131] Loss function L:

[0132]

[0133] Among them, true represents the actual measurement verification value, and N represents the number of training batches;

[0134] Step 4-4: Predict the device failure probability through the LSTM prediction model. The specific calculation formula is as follows:

[0135] Calculate the prediction error:

[0136] ΔX pred = X pred,t+k - X true,t+k ;

[0137] Among them, k represents the prediction step;

[0138] Calculate the device failure probability:

[0139]

[0140] Among them, Φ(·) is the cumulative distribution function of the standard normal distribution; σ is the standard deviation of the measurement noise, and ε max is the preset maximum allowable error. According to P f ′ ail 's calculation result, predict the life of the measured high-flow respiratory humidifier and generate a device health report. If the predicted device life is lower than the preset threshold, the central server sends a maintenance message to the acquisition device.

[0141] In this embodiment, if P f ′ ail is greater than 80%, it is considered that the measured device needs to be replaced.

[0142] Step 5: The central server determines whether the result of the device health report exceeds the calibration range. If it exceeds, generate a maintenance report and save it. At the same time, send a maintenance message to the acquisition device, and the acquisition device displays it to the maintenance personnel through the human-machine interface.

[0143] An online calibration method for a high-flow breathing humidification treatment device according to the present invention solves the technical problems of insufficient dynamic calibration accuracy under cross-interference of environmental parameters and inability to early predict latent device failures in the traditional technology. The present invention adopts a cross-parameter compensation algorithm. By establishing a non-linear compensation model for temperature-humidity-oxygen concentration, the environmental interference error is reduced to within ±0.5%. An LSTM-attention mechanism prediction model is adopted to realize the prediction of the failure probability in the next 90 days.

Claims

1. A method for online calibration of a high-flow respiratory humidification therapeutic apparatus, characterized in that: The steps include: Step 1: Deploy a central server at the data processing layer and deploy collection devices at the data collection layer. The central server and the collection devices communicate with each other through the communication network of the data transmission layer. After the collection device is initialized, it sends a link request to the central server, and the link request contains the device ID. The central server monitors the connection request of the collection device and verifies the device ID. After the verification is correct, it establishes a connection with the collection device; Step 2: The acquisition device performs a calibration data acquisition on the flow rate and oxygen concentration of the high-flow respiratory humidification therapy device under test, specifically, multiple rounds of multi-point acquisition are performed within a preset time, and environmental data is collected at the same time to obtain a detection data set, which includes a flow data sequence set, an oxygen concentration data sequence set, a temperature data sequence set, a humidity data sequence set, and an air pressure data sequence set; Step 3: After the acquisition device screens and preprocesses the detection data set for abnormal values, it stores it locally and performs cross-parameter collaborative calibration calculation after completing one calibration data acquisition, including calculating temperature-oxygen concentration cross compensation and humidity-oxygen concentration cross compensation. It determines whether to trigger re-calibration based on the cross-parameter collaborative calibration calculation result: if triggered, re-calibration data acquisition is performed according to the method in step 2; if not triggered, the detection data set is sent to the central server; Step 4: The central server establishes an LSTM prediction model and uses historical data to train the LSTM prediction model; the acquired detection data set is input into the LSTM prediction model to calculate the equipment failure probability, generate an equipment health report, and predict the future equipment life; Step 5: The central server determines whether the result of the equipment health report exceeds the calibration range. If so, a maintenance report is generated and saved, and the maintenance information is sent to the collection device, which displays it to the maintenance personnel through the human-machine interface.

2. The method for online calibration of a high-flow respiratory humidification therapeutic apparatus as claimed in claim 1, characterized in that: The acquisition device includes a main controller, a human-machine interface, an AD module, a flow sensor, a temperature and humidity sensor, a data storage device, a communication module, an air pressure sensor, a power module, a power interface, a reference voltage module and an oxygen concentration sensor. The human-machine interface, the AD module, the flow sensor, the temperature and humidity sensor, the data storage device, the communication module and the air pressure sensor are all connected to the main controller, and the oxygen concentration sensor and the reference voltage module are all connected to the AD module; The power module supplies power to the main controller, human-machine interface, AD module, flow sensor, data storage, communication module, air pressure sensor, reference voltage module and oxygen concentration sensor; The power module is connected to the power interface, and the power interface is connected to an external regulated power supply.

3. The online calibration method of a high-flow respiratory humidification therapeutic apparatus as claimed in claim 1, characterized in that: When executing step 2, the specific steps are as follows: The collection device pre-stores multiple rounds of collection strategies locally, including: Preset a calibration data acquisition strategy: perform Z times of intensive sampling within the preset time window; Preset the number of rounds Y for collecting calibration data once; In each round, the spatial distribution sampling method is used for data collection, that is, 7 characteristic points are selected within the output range of the high-flow respiratory humidification therapy device under test; The collected detection data sets specifically include: Traffic data series set Q ref ; Oxygen concentration data series set C O2,set , temperature data series set T set , humidity data series set H set , air pressure data series set P set ; Among them, set represents a complete data sequence set formed by a round of multi-point collection, Q represents flow data, C represents oxygen concentration data, T represents temperature data, H represents humidity data, and P represents air pressure data.

4. The method for online calibration of a high-flow respiratory humidification therapeutic apparatus as claimed in claim 3, characterized in that: When executing step 3, the specific steps are as follows: Step 3-1: Remove outliers through sliding window filtering and triple standard deviation method; Step 3-2: The processed detection data set is stored in local storage; Step 3-3: After completing a calibration data collection, perform cross-parameter collaborative calibration calculation on the detection data set, specifically including calculating temperature-oxygen concentration cross compensation and calculating humidity-oxygen concentration cross compensation: Calculate the temperature-oxygen concentration cross compensation: Calculate the temperature error: ΔT=T meas -T ref ; Among them, meas represents the original measurement data directly read by the sensor, and ref represents the preset standard reference value; Calculate the effect of temperature on oxygen concentration: C O2,corrected =C O2,meas +k T ×ΔT×Q ref ; Among them, k T is the temperature-oxygen concentration cross-influence coefficient, which is the preset value; corrected indicates the corrected value after environmental compensation; If ΔT|>ε T , then recalibration is triggered, where ε T is the preset temperature error threshold; Calculate humidity-oxygen concentration cross compensation: Calculate humidity error: ΔH=H meas -H ref ; Calculate the effect of humidity on oxygen concentration: C O2,corrected =C O2,meas +k H ×ΔH; Among them, k H is the humidity-oxygen concentration cross-influence coefficient, which is the preset value; If ΔH>ε H , then recalibration is triggered, where ε H is the preset humidity error threshold; Step 3-4: According to the result of step 3-3, if recalibration is triggered, return to step 2 and re-collect data; otherwise, send the processed detection data set to the central server.

5. The online calibration method of a high-flow respiratory humidification therapeutic apparatus as claimed in claim 4, characterized in that: When executing step 4, the specific steps are as follows: Step 4-1: Perform normalization preprocessing on the data in the test data set, set a time window m, and use m historical data points to predict the data at k future time points; Step 4-2: Establish an LSTM prediction model, use the flow data, oxygen concentration data, temperature data, humidity data and air pressure data in the historical data as input to train the LSTM prediction model and establish the input set: X t =[Q t ,C O2,t ,T t ,H t ,P t ]; Where t represents the time window; Predict the equipment error for the next k periods: X pred,t+k =f LSTM (X t-m ,...,X t ); Where m represents the time window length, k represents the prediction step length, and pred represents the predicted value, which is the prediction result output by the LSTM model; f LSTM (·) indicates LSTM model; Loss function L: Among them, true represents the actual measured verification value, and N represents the number of training batches; Step 4-4: Use the LSTM prediction model to predict the probability of equipment failure. The specific calculation formula is as follows: Calculate the prediction error: ΔX pred =X pred,t+k -X true,t+k ; Among them, k represents the prediction step length; Calculate the probability of equipment failure: where Φ(·) is the cumulative distribution function of the standard normal distribution; σ is the standard deviation of the measurement noise, and ε max To preset the maximum allowable error, according to P′ fail The calculation results predict the life of the high-flow respiratory humidification therapy device under test and generate a device health report. If the predicted device life is lower than the preset threshold, the central server sends maintenance information to the collection device.

Images