Intelligent oxygen state control method and system based on flow detection

Abstract

The present application relates to the technical field of oxygen therapy monitoring and intelligent control of medical central oxygen supply system, and particularly relates to an intelligent oxygen state control method and system based on flow detection, comprising the following steps: S1, establishing a patient-specific oxygen therapy data storage unit; S2, performing adaptive filtering pretreatment on the original data; S3, extracting the inhale-exhale alternating fluctuation characteristics in the breathing cycle to determine the effective oxygen inhalation state of the patient; S4, starting the effective oxygen inhalation duration accumulation to generate structured oxygen therapy detail data; S5, completing the accurate determination of the stop oxygen inhalation state; S6, establishing a whole-cycle oxygen therapy data traceable link from the patient's admission to discharge; S7, based on the oxygen therapy detail data, connecting the hospital information management system to complete automatic accounting and performing adaptive closed-loop regulation and control of the oxygen flow. Through the double-parameter coupling verification logic of the respiratory physiological fluctuation characteristics and the compliance characteristics of the medical order flow, the present application solves the technical problem of misjudging the open oxygen as effective oxygen inhalation.

Description

Technical Field

[0001] This invention relates to the field of oxygen therapy monitoring and intelligent control technology for medical central oxygen supply systems, and in particular to an intelligent oxygen status control method and system based on flow detection. Background Technology

[0002] Traditional medical oxygen therapy relies on manual inspection, adjustment, and rough recording of oxygen status monitoring and control using float-type and pointer-type flow meters. This method can only determine whether oxygen is turned on by the flow meter scale, but cannot identify the patient's actual oxygen inhalation behavior. This can lead to miscalculation of duration due to oxygen being turned on but not inhaled. Furthermore, oxygen flow and duration data are not electronically recorded, resulting in poor traceability throughout the entire cycle. This directly leads to a lack of accurate basis for medical institutions' billing and accounting. Monitoring oxygen therapy according to doctor's orders also relies solely on manual periodic checks, which is inefficient and prone to missed detections.

[0003] Furthermore, although existing technologies have proposed improvements to external oxygen detection devices to address the aforementioned issues—such as adding flow sensors to collect flow data, using pressure sensors to detect changes in pipeline pressure to differentiate between effective and ineffective oxygen supply, and incorporating storage modules for local data recording, with some solutions even attempting to connect to hospital systems for data uploading—these improvements still suffer from significant technical shortcomings, as detailed below:

[0004] First, the current method of determining oxygen inhalation status only detects pressure fluctuations or flow rate interruptions, which cannot accurately identify actual oxygen inhalation behavior. This results in a high rate of false positives and false negatives, and there are no accurate records of cumulative duration. The accuracy and traceability of data collection cannot meet the core needs of clinical practice and billing.

[0005] Secondly, the existing solutions rely heavily on manual data entry for linking patients with oxygen therapy, lacking automated near-field binding, which easily leads to data mismatch. Furthermore, data transmission is merely a simple upload, failing to automate the calculation of oxygen therapy costs. Additionally, it can only provide simple alerts for abnormal flow rates, without establishing intelligent proactive control of oxygen status based on detection data. Its low level of intelligence makes it unsuitable for the needs of medical institutions for refined oxygen therapy monitoring and control.

[0006] Therefore, it is necessary to design an intelligent oxygen state control method based on flow detection and integrating micro-pressure difference quantitative judgment, precise data correlation and intelligent control. Summary of the Invention

[0007] To solve one of the above-mentioned technical problems, the present invention adopts the following technical solution: an intelligent oxygen status control method based on flow detection, comprising the following steps: S1, adapting and installing an external oxygen detection terminal with an existing medical oxygen flow meter to complete the connection between the oxygen pathway and the respiratory detection pathway, and completing the one-to-one binding between the terminal and the patient's identity information through near-field recognition of a medical handheld mobile nursing PDA to establish a patient-specific oxygen therapy data storage unit.

[0008] S2. Real-time flow data of the oxygen pathway and dynamic air pressure data of the breathing end tubing are collected synchronously through the detection terminal. Adaptive filtering preprocessing is performed on the two raw data to filter out inherent disturbance signals in the airway and fully retain characteristic signals related to breathing behavior.

[0009] S3. Based on the pre-processed dynamic air pressure data, extract the inspiratory-expiratory alternation fluctuation characteristics within the respiratory cycle, and combine the synchronously collected flow data with the compliance characteristics of the preset medical order flow range to perform dual-parameter coupling verification to determine the patient's effective oxygen inhalation status.

[0010] S4. Only when the effective oxygen inhalation status is continuously established will the effective oxygen inhalation duration be accumulated. At the same time, the accumulated duration will be bound to the flow data and timestamp of the corresponding time period in a spatiotemporal dimension to generate structured oxygen therapy details.

[0011] S5. Based on the characteristics of respiratory fluctuations and the temporal variation patterns of flow data, accurately determine the state of cessation of oxygen therapy, terminate the accumulation of effective oxygen therapy duration, and save the detailed oxygen therapy data for the corresponding time period.

[0012] S6. Synchronize and store the time-stamped oxygen therapy details data to the local machine and the hospital's cloud platform to establish a traceable link for the entire cycle of oxygen therapy data from patient admission to discharge.

[0013] S7. Based on the detailed oxygen therapy data, it connects with the hospital information management system to complete automated billing and accounting. At the same time, it compares the flow rate with the doctor's orders to complete the compliance verification. It pushes graded warnings for abnormal status and performs adaptive closed-loop control of oxygen flow.

[0014] Based on any of the above technical solutions, a further optimization is made to perform adaptive filtering preprocessing on the original data in step S2, including the following steps:

[0015] Baseline correction is performed on the synchronously acquired raw flow data and raw gas pressure dynamic data to remove the DC bias caused by pipeline static gas pressure and oxygen steady-state flow rate, resulting in baseline-removed alternating characteristic data.

[0016] Based on the real-time oxygen flow rate and pipeline damping characteristics, the length of the adaptive sliding window is dynamically adjusted. When the oxygen flow rate decreases and the pipeline damping increases, the sliding window length is increased accordingly. When the oxygen flow rate increases and the pipeline damping decreases, the sliding window length is decreased accordingly.

[0017] The alternating characteristic data were processed by sliding window adjustment to filter out high-frequency interference signals caused by inherent disturbances in pipeline airflow and joint vibration, while fully preserving the micro-pressure fluctuation characteristic signals related to the patient's breathing behavior.

[0018] Based on any of the above technical solutions, a further optimization is made: before extracting the inspiratory-expiratory alternation fluctuation characteristics in step S3, an adaptive compensation step for airway attenuation of the respiratory pressure difference signal is also included, specifically including the following steps:

[0019] Step 1: Collect the gas path characteristic parameters of the current oxygen therapy circuit, including the nasal oxygen tube specifications and length, humidification bottle liquid level height, and number of gas path connectors. Combine this with the real-time oxygen flow rate value to construct a gas path damping characteristic matrix.

[0020] Step 2: Using a scenario-specific modified multiple linear regression model, calculate the respiratory pressure differential signal attenuation coefficient under the current airway conditions. The calculation formula is as follows:

[0021] ;

[0022] In the formula: This is the attenuation compensation coefficient for the respiratory pressure difference signal under the current airway conditions, with a value range of (1-3).

[0023] This is the effective length of the nasal cannula. This refers to the relative height of the liquid level in the humidification bottle. This refers to the number of gas line connectors. This is the real-time oxygen flow rate value;

[0024] For model constants, These are the regression coefficients for the characteristic parameters of each gas path. The values ​​of all coefficients are taken from the industry benchmark data of pressure difference attenuation for different gas path parameters in the gas path design specifications for medical central oxygen supply systems.

[0025] Step 3: Use the calculated attenuation compensation coefficient to perform amplitude compensation on the preprocessed dynamic air pressure data to restore the real respiratory pressure difference signal after attenuation through the airway.

[0026] Step 4: Based on the compensated dynamic air pressure data, perform respiratory cycle segmentation and extract the features of inspiratory-expiratory alternation.

[0027] Based on any of the above technical solutions, a further optimization is made to the specific implementation of step S3, which involves extracting the inspiratory-expiratory alternation characteristics within the respiratory cycle, including the following steps:

[0028] The compensated time-series barometric pressure dynamic data is segmented into respiratory cycles to identify the inspiratory start point, inspiratory peak point, inspiratory end point, expiratory peak point, and expiratory end point of each complete respiratory cycle.

[0029] For each complete respiratory cycle, six characteristic parameters were extracted sequentially: peak amplitude of inspiratory negative pressure, duration of inspiratory negative pressure, peak amplitude of expiratory positive pressure, duration of expiratory positive pressure, total duration of respiratory cycle, and inspiratory-expiratory ratio.

[0030] Using the industry statistical benchmark values ​​of normal clinical respiratory physiological characteristics as a reference, the six characteristic parameters were standardized by min-max, and all characteristic values ​​were uniformly mapped to the [0,1] interval;

[0031] Step 4: Arrange the six standardized feature parameters in a fixed order to construct a standardized respiratory fluctuation feature vector for a single respiratory cycle, which serves as the core input feature set for determining effective oxygen intake status.

[0032] Based on any of the above technical solutions, a further optimization is made to the specific implementation of the dual-parameter coupling verification for determining the patient's effective oxygen inhalation status in step S3, which includes the following steps:

[0033] Step 1: Input the standardized respiratory fluctuation feature vector of a single respiratory cycle and the synchronously collected real-time flow data into the preset effective oxygen inhalation state binary classification judgment model.

[0034] Step 2: Calculate the matching degree between real-time flow data and preset medical order flow range to obtain standardized flow compliance characteristics, with a value range of [0,1].

[0035] Step 3: Using a scenario-specific logistic regression model, calculate the initial posterior probability that the current respiratory cycle belongs to an effective oxygen intake state. The calculation formula is as follows:

[0036] ;

[0037] In the formula: This represents the initial posterior probability that the current respiratory cycle is in a state of effective oxygen intake, and its value ranges from [value missing]. ; This indicates an effective oxygen intake status. This indicates an ineffective oxygen administration state;

[0038] This is the standardized 6-dimensional respiratory fluctuation feature vector. The weighted coefficient row vector is the respiratory feature vector. The values ​​of the weighted coefficients for each dimension are derived from the weight allocation specifications for normal respiratory physiological characteristics in the "Guidelines for Clinical Application of Medical Oxygen Therapy", and the sum of the coefficients for each dimension is 1.

[0039] The standardized traffic compliance characteristics have a value range of [value range missing]. ;

[0040] The fusion weighting coefficient for flow compliance characteristics is derived from the priority requirements for oxygen therapy flow control in the "Technical Specification for Central Oxygen Supply System Engineering" GB50751-2012.

[0041] This is the model bias term, and its value is taken from the industry statistical benchmark value of healthy adults in a calm breathing state.

[0042] Step 4: Compare the calculated initial posterior probability with the preset judgment threshold. When the initial posterior probability corresponding to a consecutive preset number of respiratory cycles is greater than or equal to the judgment threshold, the patient is finally determined to be in an effective oxygen inhalation state. The value of the judgment threshold comes from the industry standard for clinical oxygen therapy monitoring.

[0043] Based on any of the above technical solutions, a further optimization is needed: the initial posterior probability output by the logistic regression model needs to be corrected for interference resistance before being used for the final determination. The specific implementation of the correction includes the following steps:

[0044] Obtain the initial posterior probability of the current respiratory cycle from the logistic regression model output, and simultaneously calculate the initial posterior probability that the current respiratory cycle belongs to an ineffective oxygen inhalation state. ;

[0045] A scenario-specific modified Naive Bayes conditional probability model is used to perform anti-interference correction on the initial posterior probability, resulting in the corrected posterior probability of effective oxygen inhalation. The correction formula is as follows:

[0046] ;

[0047] In the formula: The corrected posterior probability of effective oxygen intake state, with a value range of [value missing]. ;

[0048] The class conditional probability of the respiratory fluctuation feature vector under effective oxygen inhalation is derived from industry statistical distribution data of respiratory features in clinical effective oxygen therapy scenarios;

[0049] The class conditional probability of the respiratory fluctuation feature vector under ineffective oxygen inhalation is derived from industry statistical distribution data of air pressure characteristics under ineffective scenarios such as pipeline disturbance and oxygen supply without inhalation.

[0050] The corrected posterior probability replaces the initial posterior probability output by the original model and serves as the core basis for the final determination of the effective oxygen inhalation status.

[0051] The determination of effective oxygen intake status is based on the corrected posterior probability.

[0052] Based on any of the above technical solutions, a further optimization is made to the specific implementation of step S5, which involves determining the state of oxygen inhalation cessation and terminating the duration accumulation. This includes the following steps:

[0053] The final determination results of effective oxygen inhalation status for consecutive respiratory cycles are collected in chronological order to construct a time sequence of oxygen inhalation status.

[0054] Using a scenario-specific modified Hidden Markov Model, the state transition probabilities of the oxygen inhalation state time series are calculated to identify the type of the current oxygen inhalation state. The formula for calculating the one-step state transition probability is as follows:

[0055] ;

[0056] In the formula: From the moment of oxygen inhalation At the time The one-step transition probability;

[0057] , These are mutually exclusive states within a preset set of oxygen inhalation states, including three states: effective oxygen inhalation, short pause inhalation, and stopped oxygen inhalation.

[0058] For a moment The corresponding oxygen inhalation state, For a moment The corresponding oxygen inhalation state;

[0059] The values ​​of the state transition probability matrix are derived from industry statistical patterns of patient respiratory state switching during clinical oxygen therapy;

[0060] When the Hidden Markov Model determines that the oxygen inhalation state time sequence continuously enters the oxygen inhalation stop state, the flow threshold verification program is started.

[0061] Verify whether the real-time flow data exceeds the preset threshold range, and whether the duration of the excess exceeds the maximum tolerable duration of respiratory interruption as specified in the "Clinical Nursing Operation Specifications";

[0062] When the flow rate threshold verification passes, the patient is determined to be in a state of oxygen cessation, the effective oxygen inhalation time accumulation is immediately terminated, and all structured oxygen therapy details within this oxygen therapy cycle are sealed in an unalterable manner.

[0063] Based on any of the above technical solutions, a further optimization is made to the specific implementation of step S4, which involves accumulating effective oxygen inhalation time and completing spatiotemporal binding, including the following steps:

[0064] Receive the final determination result of the effective oxygen inhalation status, and start the effective oxygen inhalation duration accumulation program only when the effective oxygen inhalation status determination remains valid;

[0065] In accordance with the minimum time unit stipulated in the medical institution oxygen therapy data management specifications, the oxygen therapy process is divided into continuous equal-length segments, which serve as the minimum unit for time accumulation and data binding;

[0066] Within each time slice, calculate the average oxygen flow rate for that period and simultaneously mark the effective oxygen inhalation status, start timestamp, and end timestamp for that time slice.

[0067] The timestamp, average flow rate, and effective oxygen inhalation status identifier of each time slice are bound one-to-one to generate structured oxygen therapy detailed data with a three-dimensional correlation between time, flow rate, and status.

[0068] Using time slices as the step size, the total duration of all time slices marked as effective oxygen inhalation is accumulated to obtain the real-time effective oxygen inhalation cumulative duration, ensuring that each time segment has corresponding flow data and status judgment basis that can be traced.

[0069] Based on any of the above technical solutions, a further optimization is made to the specific implementation of establishing a traceable link for the patient's full-cycle oxygen therapy data in step S6, including the following steps:

[0070] A globally unique timestamp and a unique patient identifier are added to the sealed structured oxygen therapy details data to generate an unalterable standardized oxygen therapy data block;

[0071] Standardized oxygen therapy data blocks are synchronously written to the terminal's local storage unit. The local storage adopts a hierarchical cyclic overwrite mechanism that complies with the "Medical Data Storage Security Specification", prioritizing the retention of oxygen therapy data during periods of abnormal status and periods of medical order adjustment.

[0072] According to the preset synchronization cycle, the standardized oxygen therapy data blocks stored locally are encrypted using the national cryptographic algorithm and then uploaded to the hospital's cloud platform;

[0073] The cloud platform assigns the received oxygen therapy data blocks to the corresponding patient's exclusive oxygen therapy data file based on the patient's unique identity, and establishes a full-cycle oxygen therapy timeline database from admission to discharge.

[0074] Based on a full-cycle oxygen therapy time-series database, a traceable link is established for multi-dimensional retrieval by patient identity, time interval, and oxygen therapy status, supporting data traceability and compliance verification at all times and in all dimensions.

[0075] The present invention also provides an intelligent oxygen status control system based on flow detection, characterized in that the system stores a computer program, which, when executed by a processor, implements the control method described above.

[0076] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0077] 1. This invention achieves accurate differentiation between effective oxygen inhalation and ineffective oxygen supply by using a dual-parameter coupled verification logic of respiratory physiological fluctuation characteristics and medical order flow compliance characteristics, thus solving the technical problem of misjudging effective oxygen inhalation when oxygen is turned on but not inhaled in the prior art.

[0078] Meanwhile, through the patient-specific dynamic adaptation mechanism of model parameters, it can accurately adapt to the atypical respiratory characteristics of special populations such as children, elderly patients, and patients with severe respiratory diseases, greatly improving the accuracy, sensitivity and adaptability of effective oxygenation status determination to all population scenarios, and technically reducing billing disputes and clinical monitoring failures caused by misjudgment and missed judgment.

[0079] 2. This invention employs an adaptive sliding window filter linked to real-time flow rate and gas path characteristics. This filter removes inherent disturbances in the gas flow while fully preserving the ±Pa level micro-pressure respiratory characteristics. Furthermore, it compensates for signal amplitude attenuation caused by changes in gas path parameters using a multiple linear regression model. Finally, it suppresses residual interference signals using a Bayesian conditional probability model. This solves the problem of feature extraction failure caused by the submersion and attenuation of respiratory signals in low-flow oxygen therapy scenarios in existing technologies. It achieves distortion-free extraction of respiratory signals across the entire flow range, providing a reliable input basis for subsequent status determination and significantly improving the operational stability and reliability of the protocol in complex clinical scenarios.

[0080] 3. This invention accumulates the duration only when the effective oxygen inhalation state is continuously established, and binds each duration with the corresponding timestamp, flow data, and state determination result. Through hierarchical cyclic overlay storage, end-to-end encrypted transmission using national cryptographic algorithms, and tamper-proof data block sealing, a traceable link for the entire cycle of oxygen therapy data from patient admission to discharge is established.

[0081] At the same time, it achieves seamless integration of oxygen therapy data with the hospital's HIS and billing systems, enabling automated and compliant billing without the need for manual secondary processing, thus reducing the non-nursing workload of nursing staff.

[0082] 4. This invention accurately distinguishes the three main causes of abnormal flow rates through multi-dimensional data. It only performs stepwise adaptive closed-loop compensation for non-human-caused flow drift caused by fluctuations in central oxygen supply pressure. For abnormalities such as human-caused adjustments and pipeline leaks that cannot be automatically corrected, it pushes the information to the appropriate medical staff terminals according to the severity level. This solves the technical defects of existing technologies that cannot cope with flow deviations caused by pipeline pressure fluctuations, untimely response to abnormalities, and doctor-patient conflicts caused by indiscriminate forced intervention. It realizes steady-state automatic control of oxygen therapy flow and precise closed-loop handling of abnormal events, which not only ensures compliance with medical orders and patient treatment safety during oxygen therapy, but also significantly reduces the burden of ward rounds and emergency response for medical staff. Attached Figure Description

[0083] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or components are generally identified by similar reference numerals. In the drawings, the elements or components are not necessarily drawn to scale.

[0084] Figure 1 This is a schematic diagram of the workflow of the intelligent oxygen state control method of the present invention.

[0085] Figure 2 This is a flowchart illustrating the adaptive compensation step for airway attenuation of the respiratory pressure differential signal according to the present invention.

[0086] Figure 3 This is a schematic diagram of the process for extracting the alternating inspiratory-expiratory fluctuation characteristics within the respiratory cycle according to the present invention. Detailed Implementation

[0087] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and are therefore merely examples and should not be used to limit the scope of protection of the present invention. The specific process of the present invention is as follows: Figures 1-3 As shown in the image.

[0088] Example 1: An intelligent oxygen state control method based on flow detection, comprising the following steps:

[0089] S1. Adapt and install the external oxygen detection terminal with the existing medical oxygen flow meter to connect the oxygen pathway and the respiratory detection pathway. Use a medical handheld mobile nursing PDA for near-field recognition to complete the one-to-one binding of the terminal and the patient's identity information, and establish a patient-specific oxygen therapy data storage unit.

[0090] S2. Real-time flow data of the oxygen pathway and dynamic air pressure data of the breathing end tubing are collected synchronously through the detection terminal. Adaptive filtering preprocessing is performed on the two raw data to filter out inherent disturbance signals in the airway and fully retain characteristic signals related to breathing behavior.

[0091] S3. Based on the pre-processed dynamic air pressure data, extract the inspiratory-expiratory alternation fluctuation characteristics within the respiratory cycle, and combine the synchronously collected flow data with the compliance characteristics of the preset medical order flow range to perform dual-parameter coupling verification to determine the patient's effective oxygen inhalation status.

[0092] S4. Only when the effective oxygen inhalation status is continuously established will the effective oxygen inhalation duration be accumulated. At the same time, the accumulated duration will be bound to the flow data and timestamp of the corresponding time period in a spatiotemporal dimension to generate structured oxygen therapy details.

[0093] S5. Based on the characteristics of respiratory fluctuations and the temporal variation patterns of flow data, accurately determine the state of cessation of oxygen therapy, terminate the accumulation of effective oxygen therapy duration, and save the detailed oxygen therapy data for the corresponding time period.

[0094] S6. Synchronize and store the time-stamped oxygen therapy details data to the local machine and the hospital's cloud platform to establish a traceable link for the entire cycle of oxygen therapy data from patient admission to discharge.

[0095] S7. Based on the detailed oxygen therapy data, it connects with the hospital information management system to complete automated billing and accounting. At the same time, it compares the flow rate with the doctor's orders to complete the compliance verification. It pushes graded warnings for abnormal status and performs adaptive closed-loop control of oxygen flow.

[0096] The external oxygen detection terminal in step S1 is a medical oxygen therapy monitoring terminal known in the art. Near-field identification uses known NFC / RFID technology and is bound to the patient's wristband hospital number (a hospital's universally unique identifier). The dedicated storage unit is established based on a known database partitioning architecture and can be created directly on the terminal and in the hospital cloud.

[0097] The adaptive filtering preprocessing in step S2 uses a sliding window filtering algorithm. The input is the synchronously acquired raw flow and pressure time series data, and the output is the characteristic signal after removing interference. The sampling frequency conforms to the industry standard for medical physiological signal acquisition and can be directly implemented through the terminal MCU.

[0098] The dual-parameter coupling verification in step S3 is based on multi-feature fusion judgment logic. The input is respiratory fluctuation feature and flow compliance feature, and the output is the judgment result of effective oxygen inhalation status.

[0099] The duration accumulation in step S4 is bound to the space-time dimension. Based on the well-known time slice data statistical method, the timing is started only when the valid state is triggered. The binding logic is the one-to-one correspondence of the time stamp, flow value, and status flag, which conforms to the medical data statistical specification.

[0100] The determination of the oxygen inhalation stop state in step S5 is based on the time sequence state recognition logic. The input is the state and flow data of consecutive periods, and the output is the determination result of oxygen inhalation stop.

[0101] The dual storage and traceable link in step S6 is based on the medical data encrypted storage and database retrieval architecture, and uses the national cryptographic algorithm to complete the encrypted transmission.

[0102] The system docking, hierarchical warning and closed-loop regulation in step S7 are based on the hospital information system docking protocol, hierarchical event handling logic and closed-loop PID control algorithm. It can be directly docked with the existing HIS / LIS systems in the hospital, and the flow regulation is realized through the micro electric control valve supporting the terminal.

[0103] The adaptive filtering preprocessing in step S2 of this solution provides a high-quality signal without interference for the respiratory feature extraction in step S3, improving the accuracy of state determination from the front end; the dual-parameter coupling state determination in step S3 provides the only trigger basis for the effective duration accumulation in step S4, fundamentally solving the problem of incorrect billing for ineffective oxygen supply; the space-time dimension data binding in step S4 provides a structured core data basis for the full-cycle traceability in step S6 and the automatic billing in step S7; the accurate determination of the oxygen inhalation stop state in step S5 provides an accurate time node for the termination of duration accumulation and data sealing, avoiding incorrect timing interruption caused by short-term oxygen inhalation stop; the full-cycle traceable link in step S6 provides complete data support for the billing calculation and doctor's order verification in step S7; the closed-loop flow regulation in step S7 in turn ensures the stability of the flow compliance characteristics in step S3, forming a closed-loop coordination.

[0104] This technical solution achieves non-invasive compatibility with existing equipment, solving the industry problems of high cost and difficult deployment in intelligent transformation of hospital oxygen therapy monitoring. Its design principle is that installation can be completed simply by connecting the terminal in series with the outlet of the existing flow meter, without changing the original flow meter structure or requiring professional installation. It is compatible with existing float-type and pointer-type oxygen flow meters on the market. Unlike existing technologies that require replacing the original flow meter, this solution achieves intelligent upgrades to oxygen therapy monitoring without increasing hospital transformation costs, possessing extremely high clinical application value. This technical solution achieves precise dual-parameter determination of effective oxygen inhalation status, solving the existing technical problem of misjudging effective oxygen inhalation when oxygen is on but not inhaled. Its design principle involves coupling and verifying the alternating inspiratory and expiratory fluctuations of the patient's actual breathing behavior with the compliance characteristics of the prescribed flow rate, a core requirement of clinical oxygen therapy. Only when both conditions are met simultaneously—the presence of genuine respiratory physiological characteristics and the flow rate meeting the prescribed requirements—is it considered effective oxygen inhalation. This fundamentally distinguishes between ineffective oxygen supply and genuine oxygen inhalation, significantly reducing the misjudgment rate and protecting the legitimate rights and interests of both doctors and patients.

[0105] This technical solution also enables full-cycle traceability management of oxygen therapy data, solving the industry problems of easy loss, easy tampering, and lack of evidence for oxygen therapy data. It constructs a loss prevention mechanism with local + cloud dual storage, a tamper prevention mechanism with national cryptographic encryption, and a traceability mechanism with three-dimensional binding of time, flow rate, and status. Each oxygen inhalation duration has corresponding flow rate data, status judgment criteria, and timestamps. The full-cycle data from admission to discharge can be quickly retrieved through multiple dimensions such as patient identification, time interval, and oxygen therapy status. It can not only be used for automated billing, but also for medical order compliance verification and medical dispute evidence collection, and is better adapted to the industry standards for medical data management.

[0106] This technical solution achieves adaptive closed-loop steady-state control of oxygen therapy flow, solving the industry problem of flow deviation from medical orders caused by pressure fluctuations in central oxygen supply pipelines. By constructing a full-process control mechanism of anomaly detection, cause identification, automatic compensation, and closed-loop verification, it can accurately distinguish between three types of anomalies: non-human-caused flow drift, patient self-adjustment, and pipeline leakage. It only performs stepwise dynamic compensation for non-human-caused drift caused by central oxygen supply pressure fluctuations, avoiding the stimulation of the patient's respiratory tract by sudden increases or decreases in flow. At the same time, real-time flow verification ensures that the flow is always maintained within the range allowed by medical orders, eliminating the need for frequent manual adjustments by medical staff, significantly reducing nursing workload, and ensuring the safety and compliance of oxygen therapy.

[0107] Based on any of the above technical solutions, a further optimization is made to perform adaptive filtering preprocessing on the original data in step S2, including the following steps:

[0108] Baseline correction is performed on the synchronously acquired raw flow data and raw gas pressure dynamic data to remove the DC bias caused by pipeline static gas pressure and oxygen steady-state flow rate, resulting in baseline-removed alternating characteristic data.

[0109] Based on the real-time oxygen flow rate and pipeline damping characteristics, the length of the adaptive sliding window is dynamically adjusted. When the oxygen flow rate decreases and the pipeline damping increases, the sliding window length is increased accordingly. When the oxygen flow rate increases and the pipeline damping decreases, the sliding window length is decreased accordingly.

[0110] The alternating characteristic data were processed by sliding window adjustment to filter out high-frequency interference signals caused by inherent disturbances in pipeline airflow and joint vibration, while fully preserving the micro-pressure fluctuation characteristic signals related to the patient's breathing behavior.

[0111] The algorithm of this scheme is divided into a three-layer linear processing architecture. The first layer is the baseline correction layer, the second layer is the sliding window dynamic adjustment layer, and the third layer is the sliding filter processing layer. The three layers are connected sequentially, with the output of the previous layer serving as the input of the next layer, and the hierarchical relationship is clear. Furthermore, the input-output relationship of each layer is clear: the input of the baseline correction layer is the synchronously acquired raw flow time-series data and raw gas pressure dynamic time-series data. The sampling frequency of the two data streams is synchronized, conforming to the industry standard for medical physiological signal acquisition, and the output is alternating characteristic data after removing DC bias; the input of the sliding window dynamic adjustment layer is the real-time oxygen flow value and pipeline gas path damping characteristics, and the output is the sliding window length adapted to the current operating conditions; the input of the sliding filter processing layer is the alternating characteristic data and the dynamically adjusted sliding window length, and the output is preprocessed data that filters out interference and retains respiratory characteristics, used in the subsequent respiratory feature extraction stage.

[0112] Furthermore, the baseline correction in this scheme employs a well-known DC component removal algorithm. It calculates the moving average of the time-series data as the baseline, subtracts the baseline value from the original data to obtain the alternating component, a common method for physiological signal preprocessing. The sliding window filtering uses a well-known moving average filtering algorithm, a common method for denoising time-series signals. The method for determining the dynamic adjustment coefficient of the sliding window length in this scheme is clear. The adjustment of the window length is negatively correlated with the oxygen flow rate and airway damping characteristics. The adjustment coefficient is derived from statistical data on airflow disturbance frequencies under different flow rates and airway damping conditions in the engineering technical specifications for medical central oxygen supply systems. When the oxygen flow rate decreases, the amplitude of the micro-pressure signal generated by respiration is smaller, and the corresponding disturbance frequency is lower; therefore, the window length needs to be increased to filter out low-frequency disturbances. When the oxygen flow rate increases, the amplitude of the respiratory signal is larger, and the disturbance frequency is higher; therefore, the window length needs to be decreased to retain high-frequency respiratory feature details. All adjustment coefficients can be directly determined using benchmark data from industry standards.

[0113] This technical solution designs an adaptive sliding window filtering scheme that is linked to real-time flow rate and airway damping characteristics. This scheme works logically and synergistically with other technical features in the overall invention, providing a high-quality signal foundation for subsequent respiratory feature extraction and effective oxygen inhalation status determination, thus substantially contributing to improving the overall accuracy of the solution. In existing technologies, the preprocessing of oxygen therapy tubing pressure signals uses filtering algorithms with fixed window lengths. This approach has an inherent, unavoidable flaw: when the window length is too large, although it can effectively filter out interference signals, it will overwhelm the micro-pressure respiratory features in low-flow-rate scenarios, leading to subsequent feature extraction failures and missed detections of effective oxygen inhalation; when the window length is too small, although it can retain respiratory feature details, it cannot effectively filter out airflow disturbances in high-flow-rate scenarios, causing interference signals to be introduced into subsequent feature extraction and leading to misjudgments of effective oxygen inhalation.

[0114] In clinical oxygen therapy scenarios, the patient's oxygen flow rate is dynamically adjusted according to their condition, and the airway damping of the tubing also changes dynamically due to differences in the length of the nasal oxygen tube, the number of connectors, and the liquid level in the humidification bottle. Fixed window filtering cannot adapt to complex and ever-changing clinical scenarios, which is one of the core reasons for the insufficient accuracy of existing technologies.

[0115] The technical features of this solution are deeply synergistic with other aspects of the overall inventive concept: the preprocessing output of this solution is the core input for respiratory feature extraction, and the quality of the preprocessed signal directly determines the accuracy of feature extraction; the adaptive window adjustment of this solution, together with the airway attenuation compensation, forms a dual synergy in front-end signal processing, jointly solving the problem of respiratory signal recognition in low-flow scenarios; the interference filtering effect of this solution, together with the Naive Bayes probability correction, forms a dual anti-interference mechanism of front-end signal filtering and back-end probability correction, jointly reducing the misjudgment rate in complex scenarios.

[0116] The algorithm of this scheme takes the flow rate and airway parameters collected in real time in clinical settings as input, and the output directly determines the retention effect and interference filtering effect of the respiratory signal. It makes a substantial technical contribution to improving the accuracy of effective oxygen inhalation status determination and solving existing technical problems. It is not an isolated abstract algorithm, and the overall concept is not a simple adjustment of existing filtering algorithms.

[0117] This technical solution achieves dynamic adaptive adjustment of the filtering window, balancing interference filtering and feature preservation under different flow rate scenarios, and solves the industry problem that fixed window filtering cannot adapt to variable flow rate oxygen therapy scenarios. Based on the airflow disturbance patterns and respiratory signal characteristics under different clinical flow rates, a negative correlation model between window length, flow rate, and airway damping is established. In low flow rate scenarios, a large window is used to filter out low-frequency disturbances, while in high flow rate scenarios, a small window is used to preserve high-frequency respiratory details. Regardless of how the oxygen therapy flow rate is adjusted, optimal filtering effect can be achieved, significantly improving the scenario adaptability of the preprocessing solution. This technical solution achieves accurate extraction of respiratory alternating features, solving the problem of DC bias masking micro-pressure respiratory signals. By removing the DC bias caused by pipeline static pressure and steady-state oxygen flow rate through baseline correction, only the alternating pressure signal related to respiratory behavior is retained. The ±Pa level micro-respiratory signals that were originally submerged in the DC component are extracted, providing a core foundation for subsequent respiratory feature recognition. Even in low flow rate oxygen therapy scenarios of 1-2L / min, the patient's respiratory signal can be accurately captured, greatly expanding the applicability of the solution.

[0118] This technical solution achieves distortion-free preservation of respiratory features, solves the problem that existing filtering algorithms are prone to causing respiratory signal distortion. The dynamic adjustment of the sliding window only targets the frequency range of the interference signal and will not cause distortion to the core parameters of the respiratory signal such as characteristic amplitude, duration, and rhythm. The preprocessed signal can completely restore the patient's true respiratory physiological characteristics, ensuring the authenticity and accuracy of subsequent feature extraction, and ensuring the accuracy of effective oxygen inhalation status determination from the front-end signal processing stage.

[0119] This technical solution achieves a lightweight implementation with low computing power requirements. It can run directly on the low-power MCU of the terminal without increasing hardware costs. The baseline correction and sliding window filtering used in this solution are both lightweight linear calculation algorithms without complex nonlinear operations. The computing power requirements are low, and it can run directly on the low-power MCU of the external detection terminal in real time without increasing the hardware cost and power consumption of the terminal. It is suitable for scenarios where the terminal is used continuously for a long time in wards.

[0120] Based on any of the above technical solutions, a further optimization is made: before extracting the inspiratory-expiratory alternation fluctuation characteristics in step S3, an adaptive compensation step for airway attenuation of the respiratory pressure difference signal is also included, specifically including the following steps:

[0121] Step 1: Collect the gas path characteristic parameters of the current oxygen therapy circuit, including the nasal oxygen tube specifications and length, humidification bottle liquid level height, and number of gas path connectors. Combine this with the real-time oxygen flow rate value to construct a gas path damping characteristic matrix.

[0122] Step 2: Using a scenario-specific modified multiple linear regression model, calculate the respiratory pressure differential signal attenuation coefficient under the current airway conditions. The calculation formula is as follows:

[0123] ;

[0124] In the formula: This is the attenuation compensation coefficient for the respiratory pressure difference signal under the current airway conditions, with a value range of (1-3).

[0125] This is the effective length of the nasal cannula. This refers to the relative height of the liquid level in the humidification bottle. This refers to the number of gas line connectors. This is the real-time oxygen flow rate value;

[0126] For model constants, These are the regression coefficients for the characteristic parameters of each gas path. The values ​​of all coefficients are taken from the industry benchmark data of pressure difference attenuation for different gas path parameters in the gas path design specifications for medical central oxygen supply systems.

[0127] Step 3: Use the calculated attenuation compensation coefficient to perform amplitude compensation on the preprocessed dynamic air pressure data to restore the real respiratory pressure difference signal after attenuation through the airway.

[0128] Step 4: Based on the compensated dynamic air pressure data, perform respiratory cycle segmentation and extract the features of inspiratory-expiratory alternation.

[0129] The model of this scheme is divided into three layers: an input layer, a linear regression calculation layer, and an output layer. The input and output relationships between the layers are clear: the input layer consists of four directly collectable gas path characteristic parameters, namely the effective length L of the nasal oxygen tube, the relative height H of the humidification bottle liquid level, the number of gas path connectors N, and the real-time oxygen flow rate Q. These four parameters together constitute the gas path damping characteristic matrix, which serves as the input to the model. The linear regression calculation layer completes linear weighted calculations through preset regression coefficients and constant terms, and is the core processing layer of the model. The output layer is the respiratory pressure difference signal attenuation compensation coefficient λ under the current gas path conditions, which serves as the core parameter for subsequent gas pressure signal amplitude compensation. Secondly, the determination method of all coefficients in the formula is clear: θ0 is a model constant term, whose value comes from the pressure difference attenuation benchmark value under the benchmark gas path conditions (standard length nasal oxygen tube, standard liquid level humidification bottle, single connector, standard flow rate) in the "Design Specification for Gas Path of Medical Central Oxygen Supply System". It is the basic compensation coefficient for the inherent attenuation of the gas path. The regression coefficient for the length of the nasal cannula is derived from the baseline data of friction loss and pressure differential attenuation rate of medical nasal cannulas of different lengths in the standard, representing the incremental attenuation of respiratory signal per unit length of nasal cannula. The regression coefficient for the humidification bottle liquid level is derived from the baseline data of gas resistance and pressure difference attenuation rate of the humidification bottle at different liquid levels in the standard, representing the increase in respiratory signal attenuation caused by each unit change in liquid level height. The regression coefficient for the number of airway connectors is derived from the benchmark data of local resistance loss and differential pressure attenuation rate for different numbers of airway connectors in the standard, representing the increase in respiratory signal attenuation caused by each additional connector. The regression coefficient for oxygen flow rate is derived from benchmark data on the friction loss and pressure differential attenuation rate of oxygen flow at different flow rates in the standard, representing the incremental attenuation of the respiratory signal per unit change in flow rate; all coefficients can be directly determined through publicly available industry standards. Furthermore, the implementation logic of this solution is clear, and the gas path characteristic parameters are all hardware parameters that can be directly obtained during clinical installation, requiring no additional detection devices; the calculation of the attenuation compensation coefficient can be directly completed on the terminal MCU, using a lightweight linear regression algorithm with low computational requirements; the amplitude compensation is achieved by multiplying the preprocessed dynamic air pressure data by the attenuation compensation coefficient λ, thus restoring the true respiratory pressure differential signal after gas path attenuation, making the implementation logic simple and clear.

[0130] The content not described in this solution is common knowledge in the field. For example, the calculation of air path damping and the principles of friction resistance and local resistance are common knowledge in fluid mechanics. The construction and implementation of the multiple linear regression model is a common method in the field of statistical learning. Based on the content recorded in this document, those skilled in the art can directly construct the corresponding model architecture and completely reproduce this solution.

[0131] This technical solution is the first to design an adaptive compensation scheme for airway attenuation based on a multiple linear regression model in the oxygen therapy status determination link. It works logically and synergistically with other technical features in the overall invention concept, solving the problem of missed detection of effective oxygen inhalation caused by respiratory signal attenuation in low-flow oxygen therapy scenarios. It makes a substantial core contribution to improving the determination accuracy of the overall solution.

[0132] In existing technologies, all oxygen therapy status determination schemes assume that the collected air pressure signal is the actual signal generated by the patient's breathing. They ignore the fact that airway parameters such as the length of the nasal oxygen tube, the liquid level of the humidification bottle, the number of connectors, and the oxygen flow rate in the clinical oxygen therapy circuit can attenuate the micro-pressure signal generated by breathing. Especially in low-flow oxygen therapy scenarios of 1-2L / min, the amplitude of the breathing signal itself is only ±2-5Pa. After airway attenuation, the amplitude can drop by more than 60%, and it is submerged by the airflow disturbance in the tubing, resulting in the failure of subsequent breathing feature extraction and the missed detection of effective oxygen inhalation status. This is the core reason for the low accuracy of existing technologies in low-flow scenarios, and it is also a common problem that the industry has long ignored and failed to solve.

[0133] This technical solution achieves adaptive compensation for the attenuation of respiratory pressure differential signals in the airway, fundamentally solving the problem of missed detection of effective oxygen inhalation caused by respiratory signal attenuation in low-flow oxygen therapy scenarios. Based on the fluid dynamics of medical gas tubing, a multiple linear regression model is used to quantify the impact of four core parameters—nasal cannula length, humidification bottle level, number of connectors, and oxygen flow rate—on respiratory signal attenuation. The corresponding compensation coefficients are accurately calculated, restoring the attenuated micro-pressure signal to the true respiratory signal. Even in low-flow oxygen therapy scenarios of 1-2 L / min, the patient's respiratory characteristics can be completely preserved, significantly improving the accuracy of effective oxygen inhalation detection in low-flow scenarios.

[0134] The regression coefficient determination method based on industry standards can be directly applied. All regression coefficient values ​​are derived from publicly available benchmark data in the "Design Specification for Gas Path of Medical Central Oxygen Supply System". This specification is a mandatory national standard for the design, construction, and acceptance of medical central oxygen supply systems in China. The gas path attenuation and resistance loss data in it are all industry-recognized benchmark values. There is no need to conduct on-site testing and calibration for each device. During installation, only the corresponding gas path parameters need to be input to complete adaptive compensation, which greatly reduces the difficulty and workload of clinical deployment and has extremely high promotion value.

[0135] This technical solution achieves full-scenario gas path attenuation adaptation, solving the problem of existing solutions' inability to adapt to different clinical installation scenarios. The model covers all core gas path parameters in clinical oxygen therapy circuits that cause respiratory signal attenuation. Regardless of the length of the nasal oxygen tubing, the level of the humidification bottle, the number of tubing connectors, or the oxygen therapy flow rate used in clinical practice, it can accurately calculate the corresponding compensation coefficient, achieving adaptive compensation across all scenarios. Unlike existing solutions that can only adapt to fixed gas path conditions, this significantly improves the solution's scenario versatility and clinical adaptability. This technical solution achieves lightweight, distortion-free signal compensation, without distorting the characteristics of the respiratory signal, ensuring the accuracy of subsequent state determination. This solution uses linear amplitude compensation, only proportionally restoring the amplitude of the air pressure signal, without changing the core characteristics of the respiratory signal such as peak time, duration, and rhythm. The compensated signal is consistent with the patient's actual respiratory signal, without introducing additional distortion and interference, ensuring the authenticity of subsequent respiratory feature extraction, and guaranteeing the accuracy of effective oxygen inhalation state determination from the signal source.

[0136] Based on any of the above technical solutions, a further optimization is made to the specific implementation of step S3, which involves extracting the inspiratory-expiratory alternation characteristics within the respiratory cycle, including the following steps:

[0137] The compensated time-series barometric pressure dynamic data is segmented into respiratory cycles to identify the inspiratory start point, inspiratory peak point, inspiratory end point, expiratory peak point, and expiratory end point of each complete respiratory cycle.

[0138] For each complete respiratory cycle, six characteristic parameters were extracted sequentially: peak amplitude of inspiratory negative pressure, duration of inspiratory negative pressure, peak amplitude of expiratory positive pressure, duration of expiratory positive pressure, total duration of respiratory cycle, and inspiratory-expiratory ratio.

[0139] Using the industry statistical benchmark values ​​of normal clinical respiratory physiological characteristics as a reference, the six characteristic parameters were standardized by min-max, and all characteristic values ​​were uniformly mapped to the [0,1] interval;

[0140] Step 4: Arrange the six standardized feature parameters in a fixed order to construct a standardized respiratory fluctuation feature vector for a single respiratory cycle, which serves as the core input feature set for determining effective oxygen intake status.

[0141] The processing hierarchy of this technical solution is clear, and it is divided into a four-layer linear processing architecture. The first layer is the respiratory cycle segmentation layer, the second layer is the feature parameter extraction layer, the third layer is the feature standardization processing layer, and the fourth layer is the feature vector construction layer. The four layers are connected in sequence, and the output of the previous layer is the input of the next layer, with a clear hierarchical relationship.

[0142] Secondly, the input-output relationships of each level are clearly defined: the input of the respiratory cycle segmentation layer is the time-series dynamic air pressure data after airway attenuation compensation, and the output is the key node identifiers for each complete respiratory cycle, including the inspiratory start point, inspiratory peak point, inspiratory end point, expiratory peak point, and expiratory end point; the input of the feature parameter extraction layer is the segmented single respiratory cycle air pressure data and key node identifiers, and the output is 6 respiratory feature parameters; the input of the feature standardization processing layer is 6 original feature parameters, and the output is standardized feature values ​​mapped to the [0,1] interval; the input of the feature vector construction layer is the standardized 6 feature values, and the output is a standardized respiratory fluctuation feature vector of fixed dimension, which serves as the core input of the subsequent effective oxygen inhalation status determination model.

[0143] All algorithms in this scheme are well-known physiological signal processing algorithms in the field. The respiratory cycle segmentation adopts a well-known algorithm that combines zero-crossing detection and peak detection, which is a general method for respiratory signal cycle segmentation. The min-max normalization processing is a well-known data normalization algorithm in the field, which is a general method for processing input features of machine learning models. Then, the extraction methods for all feature parameters in this scheme are clear: peak amplitude of inspiratory negative pressure, which is the difference between the minimum air pressure during the inspiratory phase and the baseline value, represents the intensity of the inspiratory action; duration of inspiratory negative pressure, which is the time interval from the start point of inspiratory to the end point of inspiratory, represents the duration of a single inspiratory breath; peak amplitude of expiratory positive pressure, which is the difference between the maximum air pressure during the expiratory phase and the baseline value, represents the intensity of the expiratory action; duration of expiratory positive pressure, which is the time interval from the end point of inspiratory to the end point of expiratory, represents the duration of a single expiratory breath; total duration of respiratory cycle, which is the time interval from the start point of inspiratory to the start point of inspiratory in the next respiratory cycle, represents the total duration of a single complete breath; and the inspiratory-expiratory ratio, which is the ratio of the duration of inspiratory negative pressure to the duration of expiratory positive pressure, represents the rhythmic characteristics of breathing. These six feature parameters fully cover the three core dimensions of respiratory action: intensity, duration, and rhythm, and are core indicators for assessing respiratory behavior in clinical practice. The extraction methods for all parameters are well-known methods for respiratory physiological signal analysis.

[0144] The benchmark values ​​for standardization in this scheme are clear. The maximum and minimum values ​​of min-max standardization are derived from the industry statistical benchmark values ​​of physiological characteristics of healthy adults in a calm breathing state in the "Guidelines for Clinical Application of Medical Oxygen Therapy", which can be directly determined through publicly available industry standards.

[0145] This technical solution designs a 6-dimensional full feature extraction scheme covering three major dimensions: breathing intensity, duration, and rhythm. It forms a logical coordination and synergistic effect with other technical features in the overall invention concept, providing comprehensive and standardized feature input for subsequent effective oxygen inhalation status determination. It solves the problem of indistinguishable real breathing from tubing disturbance from the feature dimension, and makes a substantial contribution to improving the determination accuracy of the overall solution.

[0146] In existing technologies, the determination of oxygen therapy status relies solely on the presence of pressure fluctuations. This approach has an inherent and unavoidable flaw: interference scenarios such as pipeline vibration, airflow impact, and leaks at joints can all generate pressure fluctuations. A single feature cannot distinguish these interference signals from the signals generated by the patient's actual breathing, which can easily lead to misjudgments of effective oxygen therapy status. At the same time, a single feature cannot be adapted to the atypical respiratory characteristics of different populations. For example, the respiratory amplitude and rhythm of children and patients with severe respiratory diseases differ from those of healthy adults, making it highly susceptible to missed detections.

[0147] The 6-dimensional features extracted in this scheme fully cover all the core physiological characteristics of respiratory behavior. Real respiratory behavior necessarily satisfies the physiological laws of amplitude, duration, and rhythm simultaneously. Signals generated by tubing disturbances cannot simultaneously match the physiological distribution of the 6-dimensional features, thus fundamentally distinguishing real breathing from interference signals. The algorithmic and technical features of this scheme are functionally mutually supportive and have a strong interactive relationship. The algorithm's input comes from the restored real respiratory physiological signal, and the output feature vector directly determines the effectiveness of the effective oxygenation state determination model. It makes a substantial technical contribution to distinguishing real breathing from interference signals and reducing the false positive rate. It is not an isolated feature extraction operation, and the overall concept is not a simple enumeration of existing physiological features.

[0148] This technical solution achieves full-dimensional feature extraction of respiratory behavior, fundamentally solving the problem of misjudgment caused by the inability of existing single features to distinguish between real breathing and tubular disturbances. The extracted 6-dimensional features fully cover the three core physiological dimensions of respiratory action: intensity (peak amplitude of inhalation / exhalation), duration (duration of inhalation / exhalation, total duration of respiratory cycle), and rhythm (inspiratory-to-expiratory ratio). Real human respiratory behavior must simultaneously conform to the physiological laws of these three dimensions. However, interference signals such as tubular disturbances and airflow impacts can only produce random air pressure fluctuations and cannot simultaneously match the physiological distribution of the 6-dimensional features. By combining multi-dimensional features, real breathing can be accurately distinguished from interference signals, significantly reducing the misjudgment rate.

[0149] This technical solution standardizes feature parameters, resolving the issue of decreased model accuracy caused by inconsistent feature dimensions across different patients and scenarios. Using industry statistical benchmarks of normal clinical respiratory physiological characteristics as a reference, the min-max standardization method maps six feature parameters with different dimensions and value ranges to the [0,1] interval. This eliminates the influence of feature dimensions caused by differences in respiratory characteristics among different patients and different oxygen therapy scenarios, providing standardized input for subsequent judgment models and ensuring the stability and universality of model judgments across different scenarios and populations. Furthermore, this technical solution achieves precise segmentation of single respiratory cycles, resolving feature distortion caused by cross-cycle feature extraction. By accurately identifying the inspiratory and expiratory start and end points, continuous temporal pressure data is segmented into independent complete respiratory cycles. This ensures that each set of feature parameters corresponds to a single complete respiratory behavior, avoiding distortion of duration and amplitude parameters caused by cross-cycle feature extraction, and guaranteeing that the extracted feature parameters accurately reflect the patient's respiratory physiological state.

[0150] Based on any of the above technical solutions, a further optimization is made to the specific implementation of the dual-parameter coupling verification for determining the patient's effective oxygen inhalation status in step S3, which includes the following steps:

[0151] Step 1: Input the standardized respiratory fluctuation feature vector of a single respiratory cycle and the synchronously collected real-time flow data into the preset effective oxygen inhalation state binary classification judgment model.

[0152] Step 2: Calculate the matching degree between real-time flow data and preset medical order flow range to obtain standardized flow compliance characteristics, with a value range of [0,1].

[0153] Step 3: Using a scenario-specific logistic regression model, calculate the initial posterior probability that the current respiratory cycle belongs to an effective oxygen intake state. The calculation formula is as follows:

[0154] ;

[0155] In the formula: This represents the initial posterior probability that the current respiratory cycle is in a state of effective oxygen intake, and its value ranges from [value missing]. ; This indicates an effective oxygen intake status. This indicates an ineffective oxygen administration state;

[0156] This is the standardized 6-dimensional respiratory fluctuation feature vector. The weighted coefficient row vector is the respiratory feature vector. The values ​​of the weighted coefficients for each dimension are derived from the weight allocation specifications for normal respiratory physiological characteristics in the "Guidelines for Clinical Application of Medical Oxygen Therapy", and the sum of the coefficients for each dimension is 1.

[0157] The standardized traffic compliance characteristics have a value range of [value range missing]. ;

[0158] The fusion weighting coefficient for flow compliance characteristics is derived from the priority requirements for oxygen therapy flow control in the "Technical Specification for Central Oxygen Supply System Engineering" GB50751-2012.

[0159] This is the model bias term, and its value is taken from the industry statistical benchmark value of healthy adults in a calm breathing state.

[0160] Step 4: Compare the calculated initial posterior probability with the preset judgment threshold. When the initial posterior probability corresponding to a consecutive preset number of respiratory cycles is greater than or equal to the judgment threshold, the patient is finally determined to be in an effective oxygen inhalation state. The value of the judgment threshold comes from the industry standard for clinical oxygen therapy monitoring.

[0161] The logistic regression model used in this scheme has a clear hierarchy; it is a single-hidden-layer binary classification model, belonging to the classic statistical learning classification algorithm known in the field. The model is divided into four layers: input layer, linear transformation layer, sigmoid activation layer, and output layer. These four layers are sequentially connected, with the output of the previous layer serving as the input of the next, clearly defining the hierarchical relationship and data flow. The method for determining all coefficients and parameters in the formula is clear: weighted coefficient row vectors... The weighted coefficients comprise six dimensions, each corresponding to a weight for one of the six respiratory characteristics. The values ​​are derived from the weighting guidelines for normal respiratory physiological characteristics in the "Clinical Application Guidelines for Medical Oxygen Therapy." The peak amplitude of inspiratory negative pressure is the core characteristic for determining respiratory behavior, with the highest weight. The inspiratory-to-expiratory ratio (IPR) is a rhythmic characteristic with the second highest weight. The sum of the six coefficients is 1, which can be directly determined through industry standards. The fusion weighted coefficients... The weights for flow rate compliance characteristics are derived from the priority requirements for oxygen therapy flow rate control in the "Technical Specifications for Central Oxygen Supply System Engineering." In clinical practice, adhering to prescribed flow rates is a core requirement for safe oxygen therapy. The value of the bias term is balanced with the total weight of the respiratory features and can be directly determined through national standards; The values ​​are derived from industry statistical benchmarks for healthy adults in a calm breathing state, representing the model benchmark offset under normal breathing conditions, and can be directly determined through industry statistical data. The judgment thresholds are derived from industry standards for clinical oxygen therapy monitoring, representing the clinically recognized lower limit of confidence for effective judgment, and can be directly determined through industry standards.

[0162] Furthermore, the implementation logic of this solution is clear, and it exhibits traffic compliance characteristics. The calculation method is a well-known interval matching degree calculation method in this field. When the real-time traffic is within the range of the medical order, The value is 1, and when it exceeds the range, it decreases linearly according to the excess ratio, making the logic simple and clear; the continuous multi-cycle judgment mechanism can avoid false triggering caused by a single respiratory disturbance, which is consistent with the continuous characteristics of clinical respiratory behavior; all calculations can be completed directly on the terminal low-power MCU, and the logistic regression model is a lightweight algorithm with low computing power requirements and no additional hardware support is needed.

[0163] This technical solution designs a logistic regression probabilistic determination scheme that couples respiratory physiological characteristics and medical order flow compliance characteristics. This scheme works in conjunction with other technical features in the overall invention concept to solve the existing technical problem of misjudging oxygen administration when oxygen is turned on but not inhaled from the core of the determination logic. This plays a decisive role in achieving the overall invention objective.

[0164] This solution breaks through the conventional thinking by adopting a probabilistic binary classification judgment model. It couples the physiological characteristics of the patient's actual breathing with the core requirements of clinical oxygen therapy (compliance of the doctor's order flow rate). Only when both conditions are met simultaneously, such as the presence of breathing characteristics that conform to physiological laws and the flow rate that meets the doctor's order requirements, will the patient be judged as receiving effective oxygen therapy. This fundamentally solves the problem of misjudging ineffective oxygen supply.

[0165] This technical solution achieves probabilistic determination through dual-parameter coupling, fundamentally solving the existing technical problem that fixed threshold determination cannot distinguish between ineffective oxygen supply and genuine oxygen inhalation. By using a logistic regression model, it integrates multi-dimensional respiratory physiological characteristics with flow compliance characteristics, quantifying the posterior probability that the current state belongs to effective oxygen inhalation. Only when the probability reaches a clinically acceptable threshold will it be determined as effective oxygen inhalation. At the same time, combined with a continuous multi-cycle determination mechanism, it ensures that only continuous genuine breathing behavior will be recognized as effective oxygen inhalation, fundamentally eliminating the problem of misjudgment of oxygen supply without inhalation and protecting the legitimate rights and interests of both medical staff and patients.

[0166] This technical solution also achieves deep integration of respiratory physiological characteristics and medical order compliance requirements, simultaneously addressing two major clinical needs: effective oxygen inhalation identification and monitoring of prescribed oxygen therapy. By using the medical order flow rate compliance characteristic as one of the core inputs of the model, it enjoys the same weight as respiratory physiological characteristics. This not only identifies whether a patient is actually inhaling oxygen but also simultaneously determines whether the oxygen flow rate meets the medical order requirements. Only when both conditions are met is the oxygen inhalation considered effective. This resolves billing disputes while enabling real-time monitoring of prescribed oxygen therapy, thus addressing two core clinical needs in one fell swoop, unlike existing technologies that only achieve a single function.

[0167] Based on any of the above technical solutions, a further optimization is needed: the initial posterior probability output by the logistic regression model needs to be corrected for interference resistance before being used for the final determination. The specific implementation of the correction includes the following steps:

[0168] Obtain the initial posterior probability of the current respiratory cycle from the logistic regression model output, and simultaneously calculate the initial posterior probability that the current respiratory cycle belongs to an ineffective oxygen inhalation state. ;

[0169] A scenario-specific modified Naive Bayes conditional probability model is used to perform anti-interference correction on the initial posterior probability, resulting in the corrected posterior probability of effective oxygen inhalation. The correction formula is as follows:

[0170] ;

[0171] In the formula: The corrected posterior probability of effective oxygen intake state, with a value range of [value missing]. ;

[0172] The class conditional probability of the respiratory fluctuation feature vector under effective oxygen inhalation is derived from industry statistical distribution data of respiratory features in clinical effective oxygen therapy scenarios;

[0173] The class conditional probability of the respiratory fluctuation feature vector under ineffective oxygen inhalation is derived from industry statistical distribution data of air pressure characteristics under ineffective scenarios such as pipeline disturbance and oxygen supply without inhalation.

[0174] The corrected posterior probability replaces the initial posterior probability output by the original model and serves as the core basis for the final determination of the effective oxygen inhalation status.

[0175] The determination of effective oxygen intake status is based on the corrected posterior probability.

[0176] The Naive Bayes conditional probability model used in this scheme has a clear hierarchy. It is a posterior probability correction model based on Bayes' theorem and belongs to the well-known classic statistical learning algorithm in this field. The model is divided into three layers: the input layer, the Bayes posterior correction layer, and the output layer. The three layers are connected sequentially, and the output of the previous layer is the input of the next layer. The hierarchical relationship and data flow are clear.

[0177] Secondly, the input-output relationship of each level is clear: the input layer contains three sets of core data, the first set being the initial posterior probability of effective oxygen inhalation output by the logistic regression model. The second group is the calculated initial posterior probability of ineffective oxygen inhalation. The third group consists of the class-conditional probabilities of the respiratory fluctuation feature vector under effective / ineffective states. and The three sets of data together constitute the input of the model; the Bayesian posterior correction layer, based on Bayes' theorem, uses class-conditional probabilities to weight and correct the initial posterior probability, outputting the corrected posterior probability; the output layer is the corrected posterior probability of the effective oxygen inhalation state. It replaces the original initial posterior probability and serves as the core basis for the final determination of effective oxygen inhalation status.

[0178] Furthermore, the method for determining all parameters in the formula is clear: The class conditional probability of the respiratory fluctuation feature vector under effective oxygen inhalation is represented by the probability of the current respiratory feature vector occurring when the patient is in a true effective oxygen inhalation state. The value is taken from the recognized normal respiratory feature probability distribution and can be directly determined through publicly available industry statistical standards. The class conditional probability is the respiratory fluctuation feature vector under ineffective oxygen inhalation. It represents the probability of the current respiratory feature vector occurring when the patient is in an ineffective oxygen inhalation state. The value is derived from industry statistical distribution data of air pressure characteristics under ineffective scenarios such as tubing disturbance, oxygen supply without inhalation, and airflow impact. It can be directly determined through publicly available industry statistical standards. Both class conditional probabilities are benchmark values ​​obtained based on industry big data statistics.

[0179] The implementation logic of this solution is clear. Based on Bayes' theorem for posterior correction, it essentially performs a secondary verification of the initial judgment result through the distribution of effective / ineffective scenario features in clinical statistics. When the current feature vector matches the feature distribution of effective oxygen inhalation, the posterior probability of effective oxygen inhalation is increased; when it matches the feature distribution of ineffective scenarios, the posterior probability of effective oxygen inhalation is decreased, thus achieving anti-interference correction. All calculations are simple linear multiplication and division operations with low computing power requirements. They can be completed directly on the low-power MCU of the terminal in real time without additional hardware support.

[0180] This technical solution constructs a dual anti-interference mechanism of front-end signal filtering and back-end probability correction, further reducing the misjudgment rate in complex clinical scenarios and making a substantial contribution to improving the overall stability and accuracy of the judgment. In existing technologies, interference processing for oxygen therapy signals is entirely concentrated in the front-end signal filtering stage. Filtering algorithms remove interference signals caused by tubing disturbances and airflow impacts. However, this approach has an inherent and unavoidable drawback: front-end filtering can only filter out interference with frequencies different from the respiratory signal. For random disturbances with frequencies similar to the respiratory signal, the filtering algorithm cannot distinguish them and will retain them, leading to misjudgments in subsequent feature extraction and state determination. At the same time, front-end filtering cannot solve the initial judgment bias caused by changes in airway parameters and abnormal patient respiratory characteristics, which easily leads to false positives.

[0181] This solution breaks through this conventional thinking. In the back-end probability determination stage, the initial determination result is corrected a second time by using the distribution of effective / ineffective scene features in clinical statistics. This distinguishes between real respiratory features and residual interference signals from a probabilistic perspective, complementing the front-end filtering and achieving dual anti-interference.

[0182] The front-end adaptive filtering of this technical solution is responsible for filtering out high-frequency / low-frequency interference that is different from the respiratory signal frequency. The back-end Naive Bayes correction is responsible for distinguishing residual interference with a frequency similar to the respiratory signal from a probabilistic perspective. By matching the feature distribution, the interference signal retained by the front-end filtering is identified, and the initial posterior probability is corrected. Even if there is a disturbance similar to the respiratory signal, its confidence in effective oxygen inhalation can be reduced through probability correction, avoiding misjudgment and greatly improving the stability of judgment in complex clinical scenarios.

[0183] Based on the determination of class-conditional probabilities using industry statistical distribution data, this solution can be applied directly without additional data collection or model training. The class-conditional probabilities are derived from publicly available industry statistical distribution data in the field of clinical oxygen therapy. These data are industry-recognized benchmark values ​​obtained from big data statistics of hundreds of thousands of clinical scenarios, covering the characteristic distribution of most clinically effective / ineffective scenarios. No additional data collection or model training is required for specific use cases. Anti-interference correction can be directly achieved after installation, significantly reducing the threshold for implementation.

[0184] The technical solution achieves distortion-free probability correction, which does not negatively affect the judgment of real breathing signals, while effectively suppressing misjudgments of interference signals. The correction model is based on Bayes' theorem, weighting the initial probability solely by the matching degree of the feature distribution. When the current feature matches the distribution of real breathing, the corrected probability will slightly increase or remain stable, without inhibiting the judgment of effective oxygen intake. When the current feature matches the distribution of interference signals, the corrected probability will significantly decrease, effectively suppressing misjudgments. This achieves a precise correction effect that preserves authenticity and suppresses interference, unlike existing one-size-fits-all threshold adjustment schemes.

[0185] Based on any of the above technical solutions, a further optimization is made to the specific implementation of step S5, which involves determining the state of oxygen inhalation cessation and terminating the duration accumulation. This includes the following steps:

[0186] The final determination results of effective oxygen inhalation status for consecutive respiratory cycles are collected in chronological order to construct a time sequence of oxygen inhalation status.

[0187] Using a scenario-specific modified Hidden Markov Model, the state transition probabilities of the oxygen inhalation state time series are calculated to identify the type of the current oxygen inhalation state. The formula for calculating the one-step state transition probability is as follows:

[0188] ;

[0189] In the formula: From the moment of oxygen inhalation At the time The one-step transition probability;

[0190] , These are mutually exclusive states within a preset set of oxygen inhalation states, including three states: effective oxygen inhalation, short pause inhalation, and stopped oxygen inhalation.

[0191] For a moment The corresponding oxygen inhalation state, For a moment The corresponding oxygen inhalation state;

[0192] The values ​​of the state transition probability matrix are derived from industry statistical patterns of patient respiratory state switching during clinical oxygen therapy;

[0193] When the Hidden Markov Model determines that the oxygen inhalation state time sequence continuously enters the oxygen inhalation stop state, the flow threshold verification program is started.

[0194] Verify whether the real-time flow data exceeds the preset threshold range, and whether the duration of the excess exceeds the maximum tolerable duration of respiratory interruption as specified in the "Clinical Nursing Operation Specifications";

[0195] When the flow rate threshold verification passes, the patient is determined to be in a state of oxygen cessation, the effective oxygen inhalation time accumulation is immediately terminated, and all structured oxygen therapy details within this oxygen therapy cycle are sealed in an unalterable manner.

[0196] The Hidden Markov Model used in this scheme has a clear hierarchy. It is a 3-state first-order Hidden Markov Model, which is a well-known classic time series analysis algorithm in this field. The model is divided into four layers: observation layer, hidden state layer, state transition probability matrix layer, and output layer. The four layers are connected in sequence, and the hierarchical relationship and data flow are clear.

[0197] Secondly, the input-output relationships at each level are clear: the input to the observation layer is the final determination of the effective oxygenation state in a continuous respiratory cycle, and the constructed oxygenation state time sequence serves as the model's observation value, which only includes two mutually exclusive results: effective oxygenation and ineffective oxygenation; the hidden state layer sets three clinically real mutually exclusive oxygenation states: effective oxygenation, short pause inhalation, and cessation of oxygenation, representing the patient's true oxygen therapy state and are the hidden variables that the model needs to identify; the state transition probability matrix layer defines the one-step transition probabilities between the three hidden states. The 3×3 state transition probability matrix is ​​the core parameter of the model; the output layer is the true hidden state corresponding to the current time sequence, that is, whether the patient is currently in an effective oxygen inhalation, a short pause in oxygen inhalation, or a complete stop in oxygen inhalation, which is used for subsequent duration accumulation termination and data sealing.

[0198] Furthermore, the method for determining the state transition probability matrix in the formula is clear: each transition probability in the matrix... All probabilities are derived from industry statistical patterns of respiratory state switching during clinical oxygen therapy, representing industry-recognized baseline values ​​for state switching probabilities. Specifically, the probability of transitioning from an effective oxygen inhalation state to a short-term pause in oxygen inhalation state is derived from statistical analysis of the frequency of state switching in clinical scenarios such as patients drinking water, turning over, coughing, and briefly getting out of bed; the probability of transitioning from a short-term pause in oxygen inhalation state back to an effective oxygen inhalation state is derived from statistical analysis of patients resuming oxygen inhalation after a brief procedure; the probability of transitioning from a short-term pause in oxygen inhalation state to a state of no oxygen inhalation is derived from statistical analysis of patients ending oxygen therapy and extubation; the probability of directly transitioning from an effective oxygen inhalation state to a state of no oxygen inhalation is low, consistent with the clinical pattern that patients will inevitably have a short pause in oxygen inhalation before ending oxygen therapy; the probability of transitioning from a state of no oxygen inhalation back to an effective oxygen inhalation state is derived from statistical analysis of patients restarting oxygen inhalation. All transition probabilities can be directly determined through publicly available clinical industry statistical standards, requiring no additional experiments or data collection.

[0199] This technical solution solves the problem of misjudging the inability of existing technologies to distinguish between a patient's brief pause in oxygen intake and a true cessation of oxygen intake, ensuring the accuracy of the cumulative effective oxygen intake duration and making a substantial contribution to the achievement of the overall invention objective.

[0200] In existing technologies, the determination of oxygen cessation states all use a fixed threshold logic of 0 flow rate + no pressure fluctuation at a single moment. This method has an inherent and unavoidable flaw: In clinical practice, patients may experience brief pauses in oxygen administration due to actions such as drinking water, turning over, going to the toilet, or suctioning, with pause durations ranging from tens of seconds to tens of minutes. Existing technologies may misjudge these brief pauses as oxygen cessation, accumulating the termination time. When the patient resumes oxygen administration, the timing restarts, resulting in incomplete accumulation of effective oxygen administration time and causing billing disputes. At the same time, the tolerance time of the fixed threshold cannot be adapted to all clinical scenarios. Setting the time too short will lead to frequent misjudgments, while setting the time too long will cause a lag in the determination of oxygen cessation and failure to promptly save data.

[0201] This approach analyzes the state switching patterns of continuous respiratory cycles from a temporal perspective. It uses a hidden Markov model to identify the patient's true hidden state, distinguishing between short pauses in breathing and true cessation of oxygen inhalation. The accumulation of duration will only be terminated when the model determines that the patient has continuously entered a state of cessation of oxygen inhalation and the flow rate exceeds the threshold for the duration required by the standard. This fundamentally avoids misjudgments caused by short pauses in breathing.

[0202] The technical solution implements a dual stop determination mechanism of time-series status recognition and flow rate threshold verification, which takes into account both the accuracy and timeliness of the determination: first, the oxygen cessation status is identified from the time-series dimension through a hidden Markov model, and then a secondary confirmation is performed through flow rate threshold and duration verification. The dual determination mechanism avoids misjudgment caused by short-term cessation of oxygen inhalation and avoids the lag in oxygen cessation determination. It ensures the integrity of the accumulated duration and can complete the sealing of oxygen therapy data in a timely manner, which is in line with the actual workflow of clinical nursing and is different from the existing single threshold determination scheme.

[0203] Based on any of the above technical solutions, a further optimization is made to the specific implementation of step S4, which involves accumulating effective oxygen inhalation time and completing spatiotemporal binding, including the following steps:

[0204] Receive the final determination result of the effective oxygen inhalation status, and start the effective oxygen inhalation duration accumulation program only when the effective oxygen inhalation status determination remains valid;

[0205] In accordance with the minimum time unit stipulated in the medical institution oxygen therapy data management specifications, the oxygen therapy process is divided into continuous equal-length segments, which serve as the minimum unit for time accumulation and data binding;

[0206] Within each time slice, calculate the average oxygen flow rate for that period and simultaneously mark the effective oxygen inhalation status, start timestamp, and end timestamp for that time slice.

[0207] The timestamp, average flow rate, and effective oxygen inhalation status identifier of each time slice are bound one-to-one to generate structured oxygen therapy detailed data with a three-dimensional correlation between time, flow rate, and status.

[0208] Using time slices as the step size, the total duration of all time slices marked as effective oxygen inhalation is accumulated to obtain the real-time effective oxygen inhalation cumulative duration, ensuring that each time segment has corresponding flow data and status judgment basis that can be traced.

[0209] The processing hierarchy of this solution is clear, and it is divided into a five-layer linear processing architecture, namely the effective state verification layer, the time slice partitioning layer, the time slice data marking layer, the three-dimensional data binding layer, and the duration accumulation layer. The five layers are connected in sequence, and the output of the previous layer is the input of the next layer. The hierarchical relationship and data flow are clear.

[0210] The methods for determining all parameters in this solution are clear: the smallest time unit of the time slice comes from the smallest unit of oxygen therapy data statistics specified in the "Specifications for Oxygen Therapy Data Management in Medical Institutions," which is an industry-recognized standard for medical data statistics and can be directly determined through industry standards; the calculation method for the average oxygen flow rate is the arithmetic mean calculation method known in the field, that is, the arithmetic mean of all flow rate collection values ​​within a single time slice, and the calculation logic is simple and clear; the effective oxygen inhalation status is identified by a binary identifier, where 1 represents that the time slice is in an effective oxygen inhalation state and 0 represents an ineffective state, and the identification rules are clear and unambiguous; the timestamp adopts the well-known Unix timestamp, which is a globally unified time encoding method, ensuring the global uniqueness and accuracy of the time identifier.

[0211] This technical solution, together with other technical features in the overall inventive concept, forms a logical coordination and synergistic effect, fundamentally solving the industry problems of including invalid oxygen supply time in billing, lack of basis for time accumulation, and untraceable data, and making a substantial contribution to the realization of the overall inventive purpose.

[0212] This solution uses effective status-triggered timing, which only accumulates time when the effective oxygen inhalation status is continuously established, thus eliminating the inclusion of invalid time from the source. At the same time, through the three-dimensional binding of time, flow rate and status, each time period is bound to the corresponding flow rate data and status determination result, ensuring that the billing for each minute has a clear basis, is transparent and traceable.

[0213] The oxygen therapy process is divided into equal-length minimum time slices. The timestamp, average flow rate, and effective status identifier of each time slice are bound one-to-one to form structured detailed data. For each minute of effective duration, the oxygen flow rate and the basis for determining the effective oxygen status can be found. Patients and hospitals can access and view the data at any time. The billing details are transparent and verifiable, which is different from the existing schemes that only record the total duration and total flow rate.

[0214] Based on any of the above technical solutions, a further optimization is made to the specific implementation of establishing a traceable link for the patient's full-cycle oxygen therapy data in step S6, including the following steps:

[0215] A globally unique timestamp and a unique patient identifier are added to the sealed structured oxygen therapy details data to generate an unalterable standardized oxygen therapy data block;

[0216] Standardized oxygen therapy data blocks are synchronously written to the terminal's local storage unit. The local storage adopts a hierarchical cyclic overwrite mechanism that complies with the "Medical Data Storage Security Specification", prioritizing the retention of oxygen therapy data during periods of abnormal status and periods of medical order adjustment.

[0217] According to the preset synchronization cycle, the standardized oxygen therapy data blocks stored locally are encrypted using the national cryptographic algorithm and then uploaded to the hospital's cloud platform;

[0218] The cloud platform assigns the received oxygen therapy data blocks to the corresponding patient's exclusive oxygen therapy data file based on the patient's unique identity, and establishes a full-cycle oxygen therapy timeline database from admission to discharge.

[0219] Based on a full-cycle oxygen therapy time-series database, a traceable link is established for multi-dimensional retrieval by patient identity, time interval, and oxygen therapy status, supporting data traceability and compliance verification at all times and in all dimensions.

[0220] The processing hierarchy of this technical solution is clear, and it is divided into a five-layer linear processing architecture, namely, the standardized data block generation layer, the local hierarchical storage layer, the encrypted transmission layer, the cloud archiving layer, and the multi-dimensional traceability link construction layer. The five layers are connected in sequence, and the output of the previous layer is the input of the next layer. The hierarchical relationship and data flow are clear.

[0221] The methods for determining all parameters and rules in this solution are clear: the patient's unique identifier uses the patient's hospital number, a universally accepted unique identifier in domestic hospitals, ensuring the uniqueness of the patient's identity; the globally unique timestamp uses the well-known Unix timestamp, ensuring that the timestamp of each data block is globally unique; the priority division rules of the hierarchical cyclic coverage mechanism come from the hierarchical management requirements of medical data in the "Medical Data Storage Security Specification," classifying oxygen therapy data during abnormal state periods and medical order adjustment periods as high-priority data, and ordinary effective oxygen inhalation period data as ordinary priority data, which can be directly determined through industry standards; the encryption algorithm uses the SM4 national cryptographic symmetric encryption algorithm issued by the State Cryptography Administration, which is the legally recognized algorithm for encrypted transmission of medical data in China and complies with the encryption requirements of the "Medical Data Security Management Specification"; the preset synchronization period comes from the data upload period requirements stipulated in the "Medical Institution Oxygen Therapy Data Management Specification," which can be directly determined through industry standards; the full-cycle oxygen therapy time-series database adopts a well-known time-series database architecture, a general architecture for medical IoT data storage, supporting fast retrieval by time dimension.

[0222] This technical solution, together with other technical features in the overall inventive concept, forms a logical coordination and synergistic effect, solving the industry problems of oxygen therapy data being easily lost, easily tampered with, difficult to trace, and non-compliant, and making a substantial contribution to the realization of the overall inventive purpose.

[0223] This solution's multi-dimensional traceability provides complete data support for clinical medical order verification, medical quality control, and evidence collection in medical disputes, realizing the full-scenario value of oxygen therapy data.

[0224] This technical solution enables the immutable, full-cycle archiving of oxygen therapy data, solving the industry problems of data loss and tampering in existing solutions, and complies with the regulatory requirements for medical data management.

[0225] Each oxygen therapy data block is attached with a globally unique timestamp and a unique patient identification identifier. At the same time, hash value verification ensures that the data block cannot be tampered with after its generation, thus guaranteeing the authenticity and integrity of the data from the source of data generation. The cloud platform establishes a dedicated oxygen therapy data file for each patient, and archives all oxygen therapy data from the patient's admission to discharge in chronological order, forming an immutable full-cycle time-series database, which complies with the requirements of the Basic Medical and Health Care and Health Promotion Law and the Medical Data Security Management Standard regarding the immutability and traceability of original medical data records.

[0226] Based on any of the above technical solutions, the following optimization is made: the specific implementation of automated billing and accounting and tiered early warning in step S7, including the following steps:

[0227] When patients are discharged and settle their accounts, the cloud platform extracts all structured oxygen therapy details marked as effective oxygen inhalation from the patient's full-cycle oxygen therapy time-series database;

[0228] According to the billing standards for different oxygen flow levels in the hospital's fee management regulations, the total cost of effective oxygen inhalation is automatically calculated by time period, and a time-stamped, traceable billing details list is generated.

[0229] During oxygen therapy, the cloud platform continuously compares real-time oxygen flow data and effective oxygen inhalation status with the doctor's prescription requirements, and identifies abnormal flow and status events in real time.

[0230] Based on the degree of deviation from the doctor's order in terms of flow rate, duration of abnormality, and type of abnormal condition, the warning level is classified in accordance with the clinical oxygen therapy abnormal event classification and handling guidelines;

[0231] Different levels of early warning information are pushed to the terminals of medical staff with corresponding permissions, so as to realize the graded and precise response and closed-loop management of abnormal states.

[0232] The processing hierarchy of this solution is clear, and it is divided into two major parallel processing branches: the automated discharge billing and accounting branch and the real-time oxygen therapy abnormality classification and early warning branch. Both branches are based on the patient's full-cycle oxygen therapy time series database. Each branch is further divided into a three-layer linear processing architecture, with clear hierarchical relationships and data flow.

[0233] Secondly, the hierarchical structure and input / output relationship of the automated billing and accounting branch are clear: The first layer is the billing data extraction layer, which takes the patient's discharge settlement instruction as input and outputs structured oxygen therapy details for all patients marked as having effective oxygen inhalation status, derived from the established full-cycle oxygen therapy time-series database; the second layer is the time-segmented cost accounting layer, which takes the effective oxygen inhalation details as input and the flow-level billing standards in the hospital's fee management regulations as input and outputs the time-segmented cost details and total cost; the third layer is the cost detail generation layer, which takes the time-segmented cost details as input and outputs a time-stamped, traceable cost detail list that can be directly connected to the hospital's billing system.

[0234] Furthermore, the hierarchical structure and input-output relationship of the real-time graded early warning branch are clear: The first layer is the abnormal event identification layer, with inputs including real-time oxygen therapy flow data, effective oxygen inhalation status, and doctor's prescribed medical orders, and outputs including identified flow abnormalities and status abnormalities; the second layer is the early warning level classification layer, with inputs including the characteristic parameters of the abnormal event (degree of deviation, duration, and type of abnormality), and outputs the corresponding early warning level, with classification rules derived from the "Clinical Oxygen Therapy Abnormal Event Grading and Handling Guidelines"; the third layer is the graded push and handling layer, with inputs including the early warning level and corresponding abnormal event details, and outputs the early warning information pushed to the terminals of medical staff with corresponding permissions, realizing closed-loop handling of abnormal events.

[0235] This technical solution, together with other technical features in the overall inventive concept, forms a logical coordination and synergistic effect, solving the industry problem that oxygen therapy data cannot be directly applied and requires secondary manual processing, and making a substantial contribution to the realization of the overall inventive purpose.

[0236] This solution deeply integrates oxygen therapy monitoring data with the hospital's core business scenarios. On the one hand, based on structured detailed data of effective oxygen inhalation, it realizes automated billing and accounting, generating traceable billing details without manual intervention. On the other hand, based on real-time oxygen therapy data and medical orders, it realizes real-time identification, graded early warning, and precise push notifications for abnormal events, ensuring that medical staff can respond to abnormal events in a timely manner and guarantee the safety of oxygen therapy.

[0237] The tiered early warning system of this solution works in synergy with adaptive closed-loop flow control. For non-human-caused flow drift anomalies, the closed-loop control automatically corrects them. For anomalies that cannot be automatically corrected, such as those caused by human adjustment or patient cessation of inhalation, early warning information is pushed to medical staff, forming an anomaly closed-loop handling system that combines automatic correction with human intervention. The automated billing and tiered early warning results of this solution are also synchronized back to the patient's full-cycle oxygen therapy data file, improving the comprehensive information of the data and forming a closed loop between business and data.

[0238] This technical solution achieves fully automated billing and accounting based on detailed data of effective oxygen inhalation, solving the industry problems of low efficiency, large errors, and easy disputes caused by existing manual billing. Its design principle lies in directly extracting structured detailed data of all effective oxygen inhalation statuses from the patient's full-cycle oxygen therapy time-series database. According to the hospital's flow rate tier billing standards, it automatically calculates the cost for each effective oxygen inhalation duration by time period, generating a timestamped billing details list. Each charge has corresponding oxygen inhalation time, flow rate, and status information, making it transparent and traceable. It eliminates the need for manual data entry and calculation by nursing staff, significantly reducing non-nursing workload and avoiding human billing errors, thus fundamentally preventing billing disputes and improving the hospital's billing management efficiency.

[0239] This technical solution enables real-time identification and tiered early warning of abnormal oxygen therapy events, addressing the safety hazards of existing technologies that cannot monitor adherence to medical orders in real time and fail to detect abnormalities promptly. The cloud platform continuously compares real-time oxygen therapy flow data and effective oxygen intake status with the doctor's prescribed requirements, providing 24 / 7 monitoring. Once abnormal events such as flow deviations from the prescribed levels or loss of effective oxygen intake are detected, the system immediately identifies and assigns an early warning level, ensuring real-time detection of abnormal events. Unlike existing technologies that issue indiscriminate warnings across the entire spectrum, this solution tiers warnings based on the severity of the abnormality, avoiding interference with medical staff with ineffective warnings, improving the effectiveness of early warnings and the response efficiency of medical staff, and ensuring patient safety and compliance with medical orders regarding oxygen therapy.

[0240] Example 2: Compared with Example 1, this example also includes the following technical features:

[0241] Based on any of the above technical solutions, a further optimization is made: the parameters of the logistic regression model can be dynamically adjusted according to individual patient characteristics, and the specific implementation includes the following steps:

[0242] During the initial oxygen therapy phase for patients, basic characteristic information such as the patient's age, disease type, and oxygen therapy method is collected.

[0243] The patient's basic characteristics are matched with the corresponding population's baseline data to determine the appropriate feature weight adjustment coefficients.

[0244] Based on the adjustment coefficient, the weighted coefficient row vector of the respiratory feature vector Weighted coefficient of traffic compliance characteristics Perform dynamic adaptation and adjustment;

[0245] Simultaneously, the threshold for determining effective oxygenation status is adjusted according to the severity of the patient's condition, and the configuration parameters of the binary classification model are updated.

[0246] The updated model parameters are used to determine the effective oxygenation status, which is adapted to the atypical respiratory characteristics of special populations such as children, patients with severe respiratory diseases, and elderly patients.

[0247] The input-output relationships at each level are clear: the patient individual feature acquisition layer takes into account the patient's basic clinical information upon admission and outputs three core individual features: age, disease type, and oxygen therapy method. These are all publicly available information directly obtainable from clinical medical records, requiring no additional examination. The baseline data matching layer takes into account the patient individual features and outputs baseline respiratory physiological features for the corresponding population and the appropriate feature weight adjustment coefficients. The baseline data comes from the "Standardized Respiratory Physiological Features of Oxygen Therapy for Different Populations". The feature weight adjustment layer takes into account the original model weighting coefficients and adjustment coefficients and outputs a row vector of the adapted respiratory feature weighting coefficients. Weighted coefficient integrated with traffic compliance The input to the threshold adaptation layer is the severity of the patient's condition, and the output is the adapted effective oxygenation status determination threshold. The input to the updated model application layer is the adapted model parameters, and the output is the updated binary classification determination model, which is used for subsequent effective oxygenation status determination. Furthermore, the methods for determining all parameters and baseline data in this scheme are clear: The "Standardized Respiratory Physiological Characteristics of Different Populations under Oxygen Therapy" is a recognized industry standard in the fields of respiratory medicine and clinical oxygen therapy in China. It clearly stipulates the baseline data of respiratory physiological characteristics for different populations, including children, elderly patients, patients with chronic obstructive pulmonary disease, patients with severe respiratory failure, and patients receiving home oxygen therapy. This includes the normal range of characteristics such as inspiratory negative pressure amplitude, respiratory rate, and inspiratory-to-expiratory ratio for different populations, as well as the weight of different characteristics in judging respiratory behavior. This data can be directly obtained from publicly available industry standards. The feature weight adjustment coefficient is determined based on the proportion of difference between the baseline data of the patient's population and the baseline data of healthy adults. For example, children have a faster respiratory rate and a lower inspiratory negative pressure amplitude, so the weight of inspiratory duration and inspiratory amplitude characteristics is adjusted accordingly. The adjustment logic is based on the baseline data in the industry standards. The adaptation rules for the judgment threshold are determined based on the severity of the patient's condition. For patients with severe respiratory diseases, the judgment threshold is lowered to avoid missed detection of effective oxygen therapy status. For patients receiving ordinary oxygen therapy, the standard threshold is maintained to avoid misjudgment. The threshold adjustment range comes from the industry standards for clinical oxygen therapy monitoring.

[0248] Then, the implementation logic of this solution is clear. The collection of individual patient characteristics is completed in the oxygen therapy initialization stage. Only scanning the patient's wristband is needed to obtain the corresponding medical record information from the hospital's HIS system, without the need for manual entry. The adaptation and adjustment of model parameters is a linear weighted adjustment. The calculation logic is simple and clear and can be completed instantly on the terminal MCU without the need for cloud computing power support. The adapted model parameters directly replace the fixed parameters of the original model without changing the core architecture and calculation logic of the model. It can be seamlessly compatible with the preceding feature extraction and subsequent probability correction stages.

[0249] This approach is based on individual characteristics such as patient age, disease type, and oxygen therapy method. It matches the respiratory physiological baseline data of the corresponding population and dynamically adjusts the feature weights, fusion coefficients, and judgment thresholds of the model. This enables the model to accurately adapt to the atypical respiratory characteristics of different populations and significantly improves the accuracy of judgment for different patient groups.

[0250] The adaptation and adjustment of this solution is based on the logistic regression binary classification model, without changing the core architecture of the model, and can be seamlessly integrated into the original decision chain. The adapted model parameters of this solution work synergistically with the Naive Bayes anti-interference correction, and the class conditional probability of the corrected model can be adapted to the feature distribution of the corresponding population in a synchronous manner, further improving the judgment accuracy of special populations. The individual patient characteristics of this solution come from the patient medical record information in the hospital HIS system, which is bound to the patient identity to form a synergy. One binding and automatic adaptation are required without manual intervention. The adapted model of this solution provides more accurate single-cycle judgment results for time-series state recognition and duration accumulation, improving the overall accuracy of the solution.

[0251] This technical solution enables dynamic adaptation of model parameters to individual patients, solving the problem of missed detection and misjudgment caused by the inability of existing general fixed parameter models to adapt to atypical respiratory characteristics of special populations, and greatly expanding the applicable scenarios of the solution.

[0252] This technical solution achieves refined adaptation of the model's core parameters, balancing the sensitivity and specificity of judgments across different populations and avoiding the accuracy degradation caused by a one-size-fits-all parameter adjustment. Instead of uniformly adjusting all parameters, it differentiates the feature weights across different dimensions based on the varying respiratory physiological characteristics of different populations. For example, for patients with severe respiratory diseases, the weight of respiratory rhythm features is increased while the weight of respiratory amplitude features is decreased, because the respiratory amplitude of severely ill patients is unstable, but their respiratory rhythm still conforms to physiological laws. For pediatric patients, the weight of respiratory rate-related features is increased to accommodate children's faster breathing rhythm. Through refined parameter adjustments, the sensitivity and specificity of judgments across different populations are ensured, avoiding both missed detections and false positives.

[0253] Based on any of the above technical solutions, a further optimization is made to the specific implementation of the adaptive closed-loop control of oxygen flow in step S7, which includes the following steps:

[0254] When the oxygen flow rate is detected to be continuously deviating from the range allowed by the doctor's orders, extract the dynamic data of air pressure and the pressure data of the central oxygen supply pipeline during the period of abnormal flow rate to identify the cause of the abnormality.

[0255] The causes of abnormal flow can be categorized into three main types: non-human-caused flow drift caused by pressure fluctuations in the central oxygen supply pipeline, patient self-regulation, and pipeline leakage.

[0256] When the abnormal flow is determined to be a non-human-caused flow drift caused by pressure fluctuations in the central oxygen supply pipeline, a corresponding step-by-step flow compensation control command is generated based on the magnitude of the flow deviation.

[0257] The control command is sent to the flow regulation execution unit of the external oxygen detection terminal to dynamically compensate and regulate the oxygen flow.

[0258] Real-time collection of compensated oxygen flow data is compared and verified with the doctor's prescribed flow range. When the flow returns to the range allowed by the doctor's prescription, compensation adjustment is stopped to maintain steady-state oxygen therapy flow.

[0259] The anomaly cause identification layer takes as input real-time flow data, dynamic air pressure data, and central oxygen supply pipeline pressure data during the period of flow anomaly, and outputs characteristic parameters of the flow anomaly cause. The anomaly type differentiation layer takes as input the characteristic parameters of the anomaly cause and outputs the cause type of flow anomaly, which is divided into three categories: non-human-caused flow drift caused by central oxygen supply pressure fluctuations, patient self-regulation, and pipeline leakage. The control command generation layer takes as input the flow deviation magnitude of non-human-caused flow drift and outputs a stepped flow compensation control command. The regulation execution layer takes as input the compensation control command and outputs the valve position adjustment action of the flow regulation execution unit to achieve dynamic compensation of oxygen flow. The closed-loop verification layer takes as input the compensated real-time flow data and outputs a regulation stop command or compensation amount correction command to form a closed-loop control.

[0260] This technical solution solves the industry problem of flow deviation from medical orders caused by pressure fluctuations in central oxygen supply pipelines, realizes steady-state automatic control of oxygen therapy flow, ensures the safety and compliance of oxygen therapy with medical orders, and makes a substantial contribution to the realization of the overall invention objective.

[0261] In existing technologies, oxygen flow regulation in hospital wards relies on manual adjustment of float-type flow meters by medical staff. This adjustment cannot achieve automatic steady-state control. Furthermore, the pressure in the hospital's central oxygen supply pipeline fluctuates due to factors such as simultaneous oxygen use in multiple wards on the same floor and peak oxygen usage periods. This directly causes the oxygen flow at the patient's end to deviate from the doctor's prescribed range. Insufficient flow leads to inadequate oxygen therapy, while excessive flow can cause safety risks such as oxygen toxicity. Medical staff can only manually adjust the flow during regular ward rounds, unable to address pressure fluctuations in real time, posing serious safety hazards for oxygen therapy. Simultaneously, some existing automatic flow regulation solutions use indiscriminate full-volume regulation, failing to distinguish between non-human-caused drift due to central oxygen supply pressure fluctuations and manual adjustments by patients or their families. This forcibly pulls the patient's self-adjusted flow back to the prescribed range, causing patient dissatisfaction and failing to meet the actual needs of clinical diagnosis and treatment.

[0262] This solution first accurately identifies the causes of abnormal flow, distinguishing between three main types: non-human-caused flow drift, patient self-adjustment, and pipeline leakage. It automatically compensates and adjusts only for non-human-caused drift caused by fluctuations in central oxygen supply pressure. For cases of patient self-adjustment, no mandatory intervention is performed; only early warning information is sent to medical staff. This achieves steady-state control of flow while respecting the patient's autonomous operation, meeting actual clinical needs.

[0263] The abnormal flow rate triggering mechanism of this solution originates from the real-time comparison results of medical order compliance. Prior real-time flow monitoring and medical order comparison provide the abnormality trigger signal. The abnormality cause identification mechanism, based on the effective oxygenation status assessment results and dynamic air pressure data, distinguishes the cause of the abnormal flow rate by checking whether respiratory fluctuation characteristics are normal. The adaptive closed-loop control mechanism, in conjunction with tiered early warning, automatically restores non-human-caused flow rate drift that can be automatically corrected to the medical order range through closed-loop control, without manual intervention. For abnormalities such as patient self-adjustment and tubing leaks that cannot be automatically corrected, tiered early warning information is pushed to medical staff, forming an abnormality closed-loop handling system of automatic correction + manual intervention. The steady-state flow control mechanism, in turn, ensures the stability of flow compliance characteristics and improves the accuracy of effective oxygenation status assessment, forming a positive technological synergy.

[0264] This technical solution achieves differentiated flow regulation based on anomaly cause identification, solving the problem of indiscriminate forced intervention in existing automatic adjustment schemes and meeting the actual needs of clinical diagnosis and treatment. Its design principle lies in first accurately identifying the causes of flow anomalies through multi-dimensional data, and then automatically compensating for non-human-induced flow drift caused by fluctuations in central oxygen supply pressure. For manual adjustments made by patients or their families, no forced intervention is performed; only early warning information is sent to medical staff for communication and handling. This achieves steady-state flow control while avoiding conflict with patients, closely aligning with actual clinical scenarios and differing from existing indiscriminate forced adjustment schemes, thus possessing strong clinical applicability.

[0265] This technical solution implements a stepped dynamic flow compensation, avoiding the irritation to the patient's airway caused by sudden increases or decreases in flow, thus ensuring the safety of oxygen therapy. Its design principle lies in employing a step-by-step, fine-tuning adjustment method. The adjustment step size is determined based on the magnitude of flow deviation; a larger deviation requires a slightly larger step size, and a smaller deviation requires a very small step size. After each adjustment, the process is paused to allow the flow to stabilize before verification and the next adjustment. This gradually brings the flow back to the range allowed by the doctor's orders, avoiding sudden increases or decreases in flow caused by a single large adjustment. This reduces irritation to the patient's airway, making it particularly suitable for patients with severe respiratory diseases, children, and other special populations with sensitive airways. It ensures the safety and comfort of the oxygen therapy process and complies with clinical oxygen therapy safety guidelines.

[0266] Example 3: Compared with Example 1, this example also includes the following technical features: The present invention also provides an intelligent oxygen state control system based on flow detection, wherein the system stores a computer program, and the computer program is executed by a processor to implement the control method as described above.

[0267] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention. For those skilled in the art, any alternative improvements or transformations made to the implementation of the present invention fall within the protection scope of the present invention.

[0268] Any aspects of this invention not described in detail are well-known to those skilled in the art.

Claims

1. An intelligent oxygen state control method based on flow detection, characterized in that, Includes the following steps: S1. Complete the connection between the oxygen pathway and the respiratory monitoring pathway, bind the oxygen monitoring terminal to the patient's identity information one-to-one, and establish a patient-specific oxygen therapy data storage unit. S2. Real-time flow data of the oxygen pathway and dynamic air pressure data of the breathing end tubing are collected synchronously through the detection terminal. Adaptive filtering preprocessing is performed on the two raw data to filter out inherent disturbance signals in the air pathway and retain characteristic signals related to breathing behavior. S3. Extract the inspiratory-expiratory alternation fluctuation characteristics within the respiratory cycle, combine the synchronously collected flow data with the compliance characteristics of the preset medical order flow range, perform dual-parameter coupling verification, and determine the patient's effective oxygen inhalation status. S4. Only when the effective oxygen inhalation status is continuously established, the effective oxygen inhalation duration is accumulated. At the same time, the accumulated duration is bound to the flow data and timestamp of the corresponding time period in a spatiotemporal dimension to generate structured oxygen therapy details. S5. Based on the characteristics of respiratory fluctuations and the temporal variation patterns of flow data, accurately determine the state of cessation of oxygen inhalation, terminate the accumulation of effective oxygen inhalation time, and seal the detailed oxygen therapy data for the corresponding time period. S6. Synchronize and store the time-stamped oxygen therapy details data to the local machine and the hospital's cloud platform to establish a traceable link for the entire cycle of oxygen therapy data from patient admission to discharge. S7. Based on the detailed oxygen therapy data, it connects with the hospital information management system to complete automated billing and accounting. At the same time, it compares the flow rate with the doctor's orders to complete the compliance verification. It pushes graded warnings for abnormal status and performs adaptive closed-loop control of oxygen flow.

2. The intelligent oxygen state control method based on flow detection according to claim 1, characterized in that, Step S2 involves adaptive filtering preprocessing of the raw data, including the following steps: Baseline correction is performed on the synchronously acquired raw flow data and raw gas pressure dynamic data to remove the DC bias caused by pipeline static gas pressure and oxygen steady-state flow rate, resulting in baseline-removed alternating characteristic data. Based on the real-time oxygen flow rate and pipeline damping characteristics, the length of the adaptive sliding window is dynamically adjusted. When the oxygen flow rate decreases and the pipeline damping increases, the sliding window length is increased accordingly. When the oxygen flow rate increases and the pipeline damping decreases, the sliding window length is decreased accordingly. The alternating characteristic data were processed by sliding window adjustment to filter out high-frequency interference signals caused by inherent disturbances in pipeline airflow and joint vibration, while fully preserving the micro-pressure fluctuation characteristic signals related to the patient's breathing behavior.

3. The intelligent oxygen state control method based on flow detection according to claim 2, characterized in that, Before extracting the inspiratory-expiratory alternation fluctuation features in step S3, an adaptive compensation step for airway attenuation of the respiratory pressure differential signal is also included, specifically including the following steps: Collect the gas path characteristic parameters of the current oxygen therapy circuit, including the nasal oxygen cannula specification and length, humidification bottle liquid level height, and number of gas path connectors, and construct a gas path damping characteristic matrix by combining it with the real-time oxygen flow value. A scenario-specific modified multiple linear regression model is used to calculate the attenuation coefficient of the respiratory pressure difference signal under the current airway conditions. The calculation formula is as follows: ； In the formula: This is the attenuation compensation coefficient for the respiratory pressure difference signal under the current airway conditions, with a value range of (1-3). This is the effective length of the nasal cannula. This refers to the relative height of the liquid level in the humidification bottle. This refers to the number of gas line connectors. This is the real-time oxygen flow rate value; For model constants, These are the regression coefficients of the characteristic parameters of each gas path; The calculated attenuation compensation coefficient is used to perform amplitude compensation on the preprocessed dynamic air pressure data to restore the true respiratory pressure difference signal after attenuation through the airway. Based on the compensated dynamic air pressure data, respiratory cycle segmentation and inspiratory-expiratory alternation fluctuation features are extracted.

4. The intelligent oxygen state control method based on flow detection according to claim 3, characterized in that, The specific implementation of extracting the inspiratory-expiratory alternation fluctuation characteristics within the respiratory cycle in step S3 includes the following steps: The compensated time-series barometric pressure dynamic data is segmented into respiratory cycles to identify the inspiratory start point, inspiratory peak point, inspiratory end point, expiratory peak point, and expiratory end point of each complete respiratory cycle. For each complete respiratory cycle, six characteristic parameters were extracted sequentially: peak amplitude of inspiratory negative pressure, duration of inspiratory negative pressure, peak amplitude of expiratory positive pressure, duration of expiratory positive pressure, total duration of respiratory cycle, and inspiratory-expiratory ratio. Using the industry statistical benchmark values ​​of normal clinical respiratory physiological characteristics as a reference, the six characteristic parameters were standardized by min-max, and all characteristic values ​​were uniformly mapped to the [0,1] interval; The six standardized feature parameters are arranged in a fixed order to construct a standardized respiratory fluctuation feature vector for a single respiratory cycle, which serves as the core input feature set for determining effective oxygen intake status.

5. The intelligent oxygen state control method based on flow detection according to claim 4, characterized in that, The specific implementation of the two-parameter coupling verification to determine the patient's effective oxygen inhalation status in step S3 includes the following steps: The standardized respiratory fluctuation feature vector of a single respiratory cycle and the synchronously collected real-time flow data are input into the preset binary classification judgment model for effective oxygen inhalation status. Calculate the matching degree between real-time flow data and preset medical order flow range to obtain standardized flow compliance characteristics, with a value range of [0,1]. A scenario-specific logistic regression model is used to calculate the initial posterior probability that the current respiratory cycle belongs to an effective oxygen intake state. The calculation formula is as follows: ； In the formula: This represents the initial posterior probability that the current respiratory cycle is in a state of effective oxygen intake, and its value ranges from [value missing]. ; This indicates an effective oxygen intake status. This indicates an ineffective oxygen administration state; This is the standardized 6-dimensional respiratory fluctuation feature vector. This is a row vector of weighted coefficients for the respiratory feature vector, where the values ​​of the weighted coefficients for each dimension satisfy the condition that the sum of the coefficients for each dimension is 1. The standardized traffic compliance characteristics have a value range of [value range missing]. ; The fusion weighting coefficient for flow compliance characteristics is derived from the priority requirements for oxygen therapy flow control; This is the model bias term, and its value is taken from the industry statistical benchmark value of healthy adults in a calm breathing state. The calculated initial posterior probability is compared with a preset judgment threshold. When the initial posterior probability corresponding to a preset number of consecutive respiratory cycles is greater than or equal to the judgment threshold, the patient is finally determined to be in an effective oxygen inhalation state.

6. The intelligent oxygen state control method based on flow detection according to claim 5, characterized in that, The initial posterior probability output by the logistic regression model needs to be corrected for interference resistance before being used for the final decision. The specific implementation of the correction includes the following steps: Obtain the initial posterior probability of the current respiratory cycle from the logistic regression model output, and simultaneously calculate the initial posterior probability that the current respiratory cycle belongs to an ineffective oxygen inhalation state. ; A scenario-specific modified Naive Bayes conditional probability model is used to perform anti-interference correction on the initial posterior probability, resulting in the corrected posterior probability of effective oxygen inhalation. The correction formula is as follows: ； In the formula: The corrected posterior probability of effective oxygen intake state, with a value range of [value missing]. ; The class conditional probability of the respiratory fluctuation feature vector under effective oxygen inhalation is derived from industry statistical distribution data of respiratory features in clinical effective oxygen therapy scenarios; The class conditional probability of the respiratory fluctuation feature vector under ineffective oxygen inhalation is derived from industry statistical distribution data of air pressure characteristics under ineffective scenarios such as pipeline disturbance and oxygen supply without inhalation. The corrected posterior probability replaces the initial posterior probability output by the original model and serves as the core basis for the final determination of the effective oxygen inhalation status. The determination of effective oxygen intake status is based on the corrected posterior probability.

7. The intelligent oxygen state control method based on flow detection according to claim 1, characterized in that, Step S5, which involves determining the state of oxygen inhalation cessation and terminating the duration accumulation, includes the following steps: The final determination results of effective oxygen inhalation status for consecutive respiratory cycles are collected in chronological order to construct a time sequence of oxygen inhalation status. Using a scenario-specific modified Hidden Markov Model, the state transition probabilities of the oxygen inhalation state time series are calculated to identify the type of the current oxygen inhalation state. The formula for calculating the one-step state transition probability is as follows: ； In the formula: From the moment of oxygen inhalation At the time The one-step transition probability; , These are mutually exclusive states within a preset set of oxygen inhalation states, including three states: effective oxygen inhalation, short pause inhalation, and stopped oxygen inhalation. For a moment The corresponding oxygen inhalation state, For a moment The corresponding oxygen inhalation state; When the Hidden Markov Model determines that the oxygen inhalation state time sequence continuously enters the oxygen inhalation stop state, the flow threshold verification program is started. Verify whether the real-time flow data exceeds the preset threshold range, and whether the duration of the excess exceeds the maximum tolerable duration of respiratory interruption; When the flow rate threshold verification passes, the patient is determined to be in a state of oxygen cessation, the effective oxygen inhalation time accumulation is immediately terminated, and all structured oxygen therapy details within this oxygen therapy cycle are sealed in an unalterable manner.

8. The intelligent oxygen state control method based on flow detection according to claim 1, characterized in that, Step S4, which involves accumulating effective oxygen inhalation time and completing spatiotemporal binding, includes the following steps: Receive the final determination result of the effective oxygen inhalation status, and start the effective oxygen inhalation duration accumulation program only when the effective oxygen inhalation status determination remains valid; The oxygen therapy process is divided into continuous equal-length segments, which serve as the smallest unit for time accumulation and data binding; Within each time slice, calculate the average oxygen flow rate for that period and simultaneously mark the effective oxygen inhalation status, start timestamp, and end timestamp for that time slice. The timestamp, average flow rate, and effective oxygen inhalation status identifier of each time slice are bound one-to-one to generate structured oxygen therapy detailed data with a three-dimensional correlation between time, flow rate, and status. Using time slices as the step size, the total duration of all time slices marked as effective oxygen inhalation is accumulated to obtain the real-time cumulative duration of effective oxygen inhalation.

9. The intelligent oxygen state control method based on flow detection according to claim 1, characterized in that, The specific implementation of establishing a traceable data link for the patient's entire oxygen therapy cycle in step S6 includes the following steps: A globally unique timestamp and a unique patient identifier are added to the sealed structured oxygen therapy details data to generate an unalterable standardized oxygen therapy data block; Standardized oxygen therapy data blocks are synchronously written to the terminal's local storage unit. The local storage adopts a hierarchical cyclic overwrite mechanism, prioritizing the retention of oxygen therapy data during periods of abnormal status and periods of medical order adjustment. According to the preset synchronization cycle, the standardized oxygen therapy data blocks stored locally are encrypted using the national cryptographic algorithm and then uploaded to the hospital's cloud platform; The cloud platform assigns the received oxygen therapy data blocks to the corresponding patient's exclusive oxygen therapy data file based on the patient's unique identity, and establishes a full-cycle oxygen therapy timeline database from admission to discharge. Based on a full-cycle oxygen therapy time-series database, a traceable link is established for multi-dimensional retrieval by patient identity, time interval, and oxygen therapy status, supporting data traceability and compliance verification at all times and in all dimensions.

10. An intelligent oxygen status control system based on flow detection, characterized in that, The system stores a computer program, which, when executed by a processor, implements the control method as described in any one of claims 1-9.

Images