A real-time monitoring system for the tightness of anesthesia tubing connections

Abstract

This invention provides a real-time monitoring system for the tightness of anesthesia tubing connections, relating to the field of intelligent detection. It includes an information acquisition method, a model prediction method, and a model self-learning method. The information acquisition method measures the pressure and flow rate of the fluid within the connecting tube. The model prediction method uses a pre-trained detachment detection model to output the probability of detachment. The model self-learning method calculates the severity of model errors in the prediction method through a model error hazard calculation method, thus achieving self-learning. This method addresses the difficulties in observing and detecting tubing detachment or loose connections during general anesthesia, and the problems of lag and low timeliness in existing technologies regarding tubing detachment and loose connections.

Description

[0001] This invention is a divisional application of patent application number 202310891769.6, entitled "A method for intelligently detecting the dislodgement of intravenous tubing connections during general anesthesia". Technical Field

[0002] This invention relates to the field of intelligent detection technology, and more specifically to a real-time monitoring system for the tightness of anesthesia tubing connections. Background Technology

[0003] Intravenous injection involves connecting an anesthesia catheter to a pre-placed intravenous catheter on the patient's body surface. The anesthetic is then delivered through the catheter to the catheter, achieving general anesthesia. During general anesthesia, problems can arise, such as the catheter becoming loose or dislodged from the catheter due to patient movement (e.g., during the initial or pre-anesthesia stages due to anxiety or restlessness) or improper operation. This is especially problematic when the patient's arms are wrapped around their sides, making it difficult to observe the connection between the catheter and the catheter. Improperly introducing the anesthetic can lead to leakage, insufficient dosage, and shallow anesthesia, causing hemodynamic fluctuations, intraoperative awareness issues, and movement disturbances, thus affecting the normal progress of the surgery. Furthermore, shallow anesthesia can increase pain sensation during surgery, increasing surgical risks. Additionally, leaked anesthetic can diffuse into the operating room, posing a risk of inhalation by medical staff and disrupting the surgical process. Currently, the connection status between the anesthesia catheter and the intravenous catheter is mainly determined by the anesthesiologist's observation. One method involves visual inspection, which is complex and cumbersome (for example, when the patient's arms are wrapped around their sides, it's difficult to observe the connection), time-consuming, laborious, and prone to slowing down the surgical process. Another method involves monitoring the patient's vital signs during anesthesia, but this method has a certain lag (i.e., it can only be determined whether the connection between the anesthesia catheter and the intravenous catheter has loosened or dislodged after the anesthetic has been absorbed and vital signs have appeared), making it impossible to promptly address any loosening or dislodging, thus slowing down the surgical process and potentially affecting the patient's health. Summary of the Invention

[0004] To address the problems existing in the prior art, the present invention aims to provide a method for intelligently detecting the detachment of intravenous tubing connections during general anesthesia. This method solves the problems that are difficult to observe and detect during general anesthesia, such as tubing detachment or loose connections, and that existing technologies have a lag and low timeliness in detecting tubing detachment and loose connections.

[0005] Another object of the present invention is to provide a real-time monitoring system for the tightness of anesthesia tubing connections.

[0006] The objective of this invention is achieved through the following technical solution:

[0007] A method for intelligently detecting the dislodgement of intravenous tubing connections during general anesthesia, characterized by comprising an information acquisition method, a model prediction method, and a model self-learning method;

[0008] The information acquisition method specifically involves periodically measuring the pressure and flow rate of the liquid within the connecting pipe using a pressure sensor and a liquid flow meter pre-installed on the connecting pipe, with a measurement period of [number missing]. The pressure and flow rates measured periodically are then structured.

[0009] The specific model prediction method is as follows: A pre-trained dropout detection model is used to classify and detect structured data, and the output is... In the formula, O y This indicates the probability of the connector detaching. O n This indicates the probability that the connector has not detached.

[0010] The model self-learning method specifically involves: obtaining the severity of model errors in the model prediction method through a model error severity calculation method, and determining the probability of connector detachment in the model prediction method. O y Greater than the preset danger threshold O w At that time, the training set is reconstructed, the dropout detection model is fine-tuned, and self-learning is completed.

[0011] As a preferred embodiment of this application, the specific steps for the pressure sensor to measure the pressure value inside the connecting pipe are as follows: [The following text appears to be unrelated and possibly from a different source:] ...arranged sequentially at intervals on the connecting pipe... n There are 1 pressure sensor, and the distance between any two adjacent pressure sensors is equal. n The pressure values ​​measured by the pressure sensors at the same time were as follows: The pressure value of the connecting pipe within the corresponding cycle is:

[0012] .

[0013] As a preferred embodiment of this application, the specific steps for the liquid flow meter to measure the flow velocity in the connecting pipe are as follows: First, the measurement cycle... Divided into m a short period of time At the same time, obtain each The flow rate of the liquid flow meter within the time period is recorded as follows: Then, the cross-sectional area of ​​the connecting pipe corresponding to the pre-determined liquid flow meter is used. S , get each The flow rate over a given time period, i.e.:

[0014] ;

[0015] Finally, the flow rate within the corresponding measurement period is obtained:

[0016] .

[0017] As a preferred embodiment of this application, the data structuring specifically involves: forming two-dimensional data pairs of pressure values ​​and flow velocities measured in the same period, which serve as node data for the periodic sampling time; and sorting multiple node data according to the sampling time order to form time series data; specifically:

[0018]

[0019] In the formula: t The time to be evaluated is [time period], and the total sampling period is [period]. ; Indicates the number of moments before the time to be evaluated. i The pressure value was measured once; Indicates the number of moments before the time to be evaluated. i The flow rate was measured once;

[0020] Select the time to be evaluated and the time preceding it. N Secondary node data serves as time-series data for the time point to be evaluated, or structured data.

[0021] As a preferred embodiment of this application, the shedding detection model adopts a deep neural network model, which mainly includes a time series fusion stage and a classification detection stage;

[0022] The time series fusion stage consists of two networks with the same structure. LSTM Network components, defined as pressure LSTM Network and flow rate LSTM The network, whose inputs are measured pressure values ​​of node data input in time sequence. and measuring flow rate Their outputs are pressure value features and flow velocity features of the same dimension; then, a concat network layer is used to concatenate the pressure value features and flow velocity features to obtain the output time series features of the first stage.

[0023] In the classification and detection stage, a ResNet network is used. Its input is the time-series features output from the first stage, and its output is the classification and detection result; that is, the input is time-series data, and the output is... The probability of judgment is used to complete the judgment;

[0024] The pre-training process uses experimentally generated labeled data to train the detachment detection model. The experimentally generated labeled data consists of a large amount of time-series data collected according to the information collection method. The labeled data is labeled by the labelers to determine whether there are cases of connector detachment within the corresponding time period of the time-series data. All time-series data are classified into positive examples and negative examples.

[0025] Positive examples are all time series data where connector tube detachment occurs, while negative examples are all time series data where connector tube detachment does not occur.

[0026] The positive and negative examples are split into training and testing data according to the proportions to complete the pre-training of the dropout detection model.

[0027] As a preferred embodiment of this application, during the pre-training process, training data accounts for 70% of the total data.

[0028] As a preferred embodiment of this application, in the model prediction method, based on the output of the detachment detection model, the probability of the connecting tube detaching is... O y Greater than the preset alarm threshold O b In case of connection failure, a warning will be issued and an alarm message will be sent.

[0029] As a preferred embodiment of this application, the... O b It is 0.5.

[0030] As a preferred embodiment of this application, in the method for intelligent detection of detachment of intravenous tubing connections during general anesthesia, the anesthesiologist obtains the alarm information pushed by the detachment detection model, promptly checks whether the connecting tube has detached, and records the input time series data and feedback results of the alarm; wherein, the feedback result is specifically: the connecting tube has detached or the connecting tube has not detached.

[0031] As a preferred embodiment of this application, the method for calculating the harm of model error is specifically as follows:

[0032]

[0033] In the formula: w p This indicates that the feedback result is recorded as the "connector tube detachment" event weight. w n This indicates the weight of the "connecting tube not detached" event recorded as the feedback result; N Indicates the current record number, Nr Indicates the minimum number of valid records;

[0034] If the first iIf the record result is "connecting tube detached", then Conversely, ;

[0035] ;

[0036] e This represents the hazard factor of model error; the larger the value, the higher the degree of harm caused by model error.

[0037] As a preferred embodiment of this application, the specific steps for reconstructing the training set are as follows:

[0038] First, the newly recorded time-series data are categorized into new positive examples and new negative examples based on their corresponding feedback results. New positive examples are data where the feedback result for all newly recorded time-series data indicates that the connector has detached, while new negative examples are data where the feedback result for all newly recorded time-series data indicates that the connector has not detached. Next, the new positive examples and new negative examples are proportionally split into new training data and new test data. The original training data and test data are then redistributed, and the new training data and the redistributed training data are merged into training data, and the new test data and the redistributed test data are merged into test data. Finally, the learning rate is adjusted to fine-tune the detachment detection model.

[0039] Considering that in real-world applications of connecting tubes, there may be situations where the detachment detection model fails to issue a warning, yet anesthesiologists detect the detachment during routine inspections—a situation that has significant implications in practical applications—the preferred embodiment of this application includes an update for missed detections in the model error hazard calculation method. Specifically:

[0040]

[0041] In the formula: w l This indicates the weight of the missed detection record. Nl Indicates the number of missed detection records. Nlr This indicates the threshold for the number of missed detection records.

[0042] The present invention has the following technical effects:

[0043] This application achieves accurate pressure and flow rate measurements of the connecting tube through segmented pressure and phased flow rate acquisition, resulting in high precision and minimal error. This avoids false alarms and missed alarms caused by pressure or flow rate errors. By using the pressure and flow rates of the connecting tube to form a data pair, intelligent detection of tube detachment is achieved. This provides timely and effective warnings for tube detachment or loose connections, prompting anesthesiologists to reconnect the tube promptly. This ensures the safety of general anesthesia and prevents surgical delays and secondary injuries to patients due to anesthetic leakage. This intelligent method for detecting detached intravenous tubing connections in general anesthesia can be applied to various clinical environments. Furthermore, the model possesses self-learning and self-updating capabilities, making it widely applicable and suitable for multiple scenarios. It is more intelligent and precise, capable of learning and avoiding missed detections, thus preventing the harm caused by such omissions. Attached Figure Description

[0044] Figure 1 This is a flowchart of the intelligent detection method for detachment of intravenous tubing connections during general anesthesia, as described in this application.

[0045] Figure 2 This is a structural block diagram of the system in the embodiments of this application.

[0046] The module includes: 100, data acquisition module; 101, pressure sensor; 102, pressure sensing module; 103, liquid flow meter; 104, flow velocity detection module; 200, data processing module; 201, communication module; 300, shedding detection module; 400, early warning module; 500, data storage module; 600, self-learning module; 700, display module; and 800, feedback module. Detailed Implementation

[0047] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0048] Example 1:

[0049] like Figure 1 As shown, a method for intelligently detecting the dislodgement of intravenous tubing connections during general anesthesia is characterized by including an information acquisition method, a model prediction method, and a model self-learning method.

[0050] The specific information acquisition method involves periodically measuring the pressure and flow rate of the liquid within the connecting pipe using a pressure sensor and a liquid flow meter pre-installed on the connecting pipe. The measurement period is [number missing]. In this embodiment ;

[0051] The specific steps for measuring the pressure value inside the connecting pipe using a pressure sensor (a common conduit pressure sensor in this field is acceptable, and its installation and measurement method also adopts existing technology, which can be understood by those skilled in the art) are as follows: Arrange the pressure sensor sequentially at intervals on the connecting pipe. n There are 1 pressure sensor, and the distance between any two adjacent pressure sensors is equal. n The pressure values ​​measured by the pressure sensors at the same time were as follows: Then the pressure value of the connecting pipe in the corresponding period is:

[0052] .

[0053] The specific steps for measuring the flow velocity in the connecting pipe using a liquid flow meter (a common pipeline liquid flow meter in this field is acceptable, and its installation and measurement methods also adopt existing technology, which can be understood by those skilled in the art) are as follows: First, the measurement cycle... Divided into m a short period of time ( The settings should be configured according to the actual situation, for example, in this embodiment. In practical applications, to ensure the accuracy of flow rate measurement, m (Not less than 10), and obtain each The flow rate of the liquid flow meter within the time period is recorded as follows: Then, the cross-sectional area of ​​the connecting pipe corresponding to the pre-determined liquid flow meter is used. S , get each The flow rate over a given time period, i.e.:

[0054] ;

[0055] Finally, the flow rate within the corresponding measurement period is obtained:

[0056] .

[0057] The periodically measured pressure and flow velocity values ​​are then structured. Specifically, pressure and flow velocity values ​​measured in the same period are paired into two-dimensional data pairs, serving as node data for the periodic sampling time. Multiple node data points are then sorted according to their sampling time order to form a time series data.

[0058]

[0059] In the formula: t The time to be evaluated is [time period], and the total sampling period is [period]. ; Indicates the number of moments before the time to be evaluated. i The pressure value was measured once; Indicates the number of moments before the time to be evaluated.i The flow rate was measured once;

[0060] Select the time to be evaluated and the time preceding it. N Secondary node data serves as time-series data for the time point to be evaluated, or structured data.

[0061] The model prediction method specifically involves using a pre-trained dropout detection model to classify and detect structured data, and then outputting the results. In the formula, O y This indicates the probability of the connector detaching. O n This indicates the probability that the connector has not detached.

[0062] The shedding detection model adopts a deep neural network model, which mainly includes a time series fusion stage and a classification detection stage.

[0063] The time series fusion stage consists of two networks with the same structure. LSTM Network components, defined as pressure LSTM Network and flow rate LSTM The network, whose inputs are measured pressure values ​​of node data input in time sequence. and measuring flow rate Their outputs are pressure value features and flow velocity features of the same dimension; then, a concat network layer is used to concatenate the pressure value features and flow velocity features to obtain the output time series features of the first stage.

[0064] In the classification and detection phase, a ResNet network is used. Its input is the time-series features output from the first phase, and its output is the classification and detection result; that is, the input is time-series data, and the output is... The probability of judgment is used to complete the judgment;

[0065] Pre-training uses experimentally generated labeled data to train the disconnection detection model. The experimentally generated labeled data consists of a large amount of time-series data collected according to the information collection method. All time-series data are labeled by labelers (using common labeling tools in the field) to determine whether there are disconnection cases of connector tubes within the corresponding time periods of the time-series data. All time-series data are classified into positive examples and negative examples.

[0066] Positive examples are all time series data where connector tube detachment occurs, while negative examples are all time series data where connector tube detachment does not occur.

[0067] According to the ratio (in this embodiment, the ratio of training data to test data, i.e., positive example data to negative example data is 7:3, that is, training data accounts for 70% of the total data), the positive example data and negative example data are split into training data and test data to complete the pre-training of the dropout detection model.

[0068] Based on the output of the detachment detection model, the probability of the connecting tube detaching is... O y Greater than the preset alarm threshold O b (In this embodiment, O b When the value is 0.5, a pre-warning for connector detachment will be issued and an alarm message will be pushed.

[0069] The anesthesiologist receives the alarm information pushed by the detachment detection model, promptly checks whether the connecting tube has detached, and records the input time series data and feedback results of the alarm; the feedback results are specifically: connecting tube detached or connecting tube not detached.

[0070] The model self-learning method specifically involves obtaining the severity of model errors in the prediction method through a model error severity calculation method. The model error severity calculation method is as follows:

[0071]

[0072] In the formula: w p The feedback result is recorded as the "connector tube detachment" event weight in this embodiment. w p It is -0.6. w n The feedback result is recorded as the "connecting pipe not detached" event weight in this embodiment. w n It is 0.4; N Indicates the current record number, Nr This represents the minimum number of valid records, in this embodiment. Nr It is 20;

[0073] If the first i If the record result is "connecting tube detached", then Conversely, ;

[0074] ;

[0075] e This represents the hazard factor of model error; the larger the value, the higher the degree of harm caused by model error.

[0076] When the probability of connector detachment in the model prediction methodO y Greater than the preset danger threshold O w Time (in this embodiment, O w =0, meaning the model error harm result is greater than 0, indicating that the model evaluation result benefit is less than the loss, and the model needs to be optimized. Reconstruct the training set, fine-tune the dropout detection model, and complete self-learning.

[0077] The update of the shedding detection model (i.e., fine-tuning the shedding detection model) is as follows:

[0078] First, the newly recorded time-series data are categorized into new positive examples and new negative examples based on their corresponding feedback results. New positive examples are data where the feedback result for all newly recorded time-series data indicates that the connector has detached, while new negative examples are data where the feedback result for all newly recorded time-series data indicates that the connector has not detached. Next, the new positive examples and new negative examples are proportionally split into new training data and new test data. The original training data and test data are then redistributed, and the new training data and the redistributed training data are merged into training data, and the new test data and the redistributed test data are merged into test data. Finally, the learning rate is adjusted to fine-tune the detachment detection model, completing self-learning.

[0079] Regarding the loss function in the training of the dropout detection model, based on the inconsistency in the severity of errors in the model's judgment of positive and negative samples, the weight of positive samples is 0.6 and the weight of negative samples is 0.4.

[0080] Example 2:

[0081] Considering that in real-world applications of connecting tubes, there may be situations where the detachment detection model fails to issue a warning, yet anesthesiologists detect the detachment during routine inspections—a situation with significant implications in practical applications—the model error hazard calculation method, based on the scheme in Example 1, includes an update to account for missed detections. Specifically:

[0082]

[0083] In the formula: w l The weight of the missed detection record is represented in this embodiment. w l It is 0.7. Nl This indicates the number of missed detection records, determined based on the actual situation. Nlr Indicates the threshold for the number of missed detection records, in this embodiment Nlr It is 10.

[0084] Example 3:

[0085] like Figure 2 As shown, a real-time monitoring system for the tightness of anesthesia tubing connections is used to detect the detachment of general anesthesia intravenous tubing connections as described in Embodiment 1 or Embodiment 2 above. The system includes a data acquisition module 100, a data processing module 200, a detachment detection module 300, an early warning module 400, a data storage module 500, and a self-learning module 600.

[0086] The data acquisition module 100 includes a pressure sensor 101, a pressure sensing module 102, a liquid flow meter 103, and a flow velocity detection module 104. The pressure sensor 101 is electrically connected to the pressure sensing module 102, thereby obtaining the measured values ​​of multiple pressure sensors 101 and converting them into the pressure value of the connecting pipe within the corresponding period. The specific method is the same as that described in Example 1; the liquid flow meter 103 is installed on the anesthesia connecting tube and is used to test the liquid flow rate of the anesthesia connecting tube. The liquid flow meter 103 is electrically connected to the flow rate detection module 104 and is used to convert the liquid flow rate into flow rate. The specific method is the same as that described in Example 1.

[0087] The pressure sensing module 102 and the flow velocity detection module 104 are electrically connected to the data processing module 200, respectively, and are used to structure the pressure value and flow velocity data. The specific method is the same as that described in Example 1.

[0088] The shedding detection module 300, i.e. the shedding detection model, is connected to the data processing module 200 via the communication module 201 (the communication module 201 enables remote communication between the shedding detection module 300 and the data processing module 200, thereby reducing the space occupied by the shedding detection module 300, which is located at the anesthesia connection tube and requires a large amount of computation and training, and is located externally, thus reducing the space occupied by the patient). It is used to perform the model prediction method as described in Example 1 using the structured data in the data processing module 200.

[0089] The output of the detachment detection module 300 is connected to the input of the early warning module 400 and the data storage module 500, respectively, for early warning output and data storage; the early warning output method is as described in Example 1; at the same time, the monitoring system also includes a display module 700 (which can be a common medical display screen model, which can be understood by those skilled in the art), the output of the data processing module 200 and the early warning module 400 are connected to the input of the display module 700, for real-time display of the pipeline pressure value and flow rate obtained by the data processing module 200 at the current moment, and also for displaying early warning information.

[0090] The output of the data storage module 500 is connected to the input of the self-learning module 600 for model self-learning; the output of the self-learning module 600 is connected to the input of the dropout detection module 300 for updating the dropout detection module 300 (i.e., the dropout detection module 300 performs fine-tuning of the detection model). The self-learning of the self-learning module 600 and the updating of the dropout detection module 300 are performed using the method described in Example 1.

[0091] Example 4:

[0092] Based on the scheme of Embodiment 3, a real-time monitoring system for the tightness of anesthesia tubing connections further includes a feedback module 800 with human-computer interaction functionality. The output of the feedback module 800 is connected to the input of the data storage module 500. The anesthesiologist detects the loosening status of the anesthesia tubing based on the output of the detachment detection module 300, and records the input time series data of the alarm and the feedback result in the data storage module 500 through the feedback module 800. Specifically, the output of the feedback module 800 indicates whether the anesthesia tubing is loose or not, which is achieved by setting a button with human-computer interaction functionality.

[0093] Therefore, the model error hazard calculation method in the self-learning module 600 also includes the update of missed detections, and the update method is consistent with that in Example 2.

Claims

1. A real-time monitoring system for the tightness of anesthesia tubing connections, characterized in that: It includes a data acquisition module, a data processing module, a shedding detection module, an early warning module, a data storage module, and a self-learning module; The data acquisition module includes a pressure sensor, a pressure sensing module, a liquid flow meter, and a flow velocity detection module. The pressure sensor and the pressure sensing module are electrically connected to obtain the measured values ​​from multiple pressure sensors and convert them into the pressure value of the connecting pipe within the corresponding cycle. It employs an intelligent detection method to identify detached intravenous tubing connections during general anesthesia; a liquid flow meter is installed on the anesthesia connection tube and is electrically connected to a flow velocity detection module to convert liquid flow rate into flow velocity. ; The intelligent detection method for intravenous catheter disconnection during general anesthesia includes information acquisition, model prediction, and model self-learning. The model prediction method involves using a pre-trained disconnection detection model to classify and detect structured data, and then outputting the results. ; O y This indicates the probability of the connector detaching. O n This indicates the probability that the connector has not detached. The shedding detection model adopts a deep neural network model, which mainly includes a time series fusion stage and a classification detection stage; The time series fusion stage consists of two networks with the same structure. LSTM Network components, defined as pressure LSTM Network and flow rate LSTM The network, whose inputs are measured pressure values ​​of node data input in time sequence. and measuring flow rate The output consists of pressure and flow velocity features of the same dimension. Then, a concat network layer is used to concatenate the pressure and flow velocity features to obtain the first stage output time series features. The pressure sensing module and the flow velocity detection module are electrically connected to the data processing module, which is used to structure the pressure value and flow velocity data. The shedding detection module, or shedding detection model, is connected to the data processing module via the communication module and is used to process the structured data in the data processing module. The output of the shedding detection module is connected to the input of the early warning module and the data storage module, respectively, for early warning output and data storage; the monitoring system also includes a display module, and the data processing module is connected to the output of the early warning module and the input of the display module, for real-time display of the pipeline pressure and flow rate obtained by the data processing module at the current moment, and also for displaying early warning information; The output of the data storage module is connected to the input of the self-learning module for model self-learning; the output of the self-learning module is connected to the input of the dropout detection module for updating the dropout detection module. The model self-learning method is as follows: The severity of model error in the prediction method is obtained through a model error severity calculation method. The model error severity calculation method is as follows: In the formula: w p , w n These represent the event weights for the feedback results recorded as "connecting tube detached" and "connecting tube not detached," respectively. N Indicates the current record number, Nr Indicates the minimum number of valid records; If the first i If the record result is "connecting tube detached", then Conversely, ; ； e This represents the hazard factor of model error; the larger the value, the higher the degree of harm caused by model error. The real-time monitoring system acquires various data by intelligently detecting the dislodgement of intravenous tubing connections during general anesthesia.

2. The real-time monitoring system for the tightness of anesthesia tubing connections according to claim 1, characterized in that: The information acquisition method specifically involves periodically measuring the pressure and flow rate of the liquid within the connecting pipe using a pressure sensor and a liquid flow meter pre-installed on the connecting pipe. The measurement period is [number missing]. ; The specific steps for the pressure sensor to measure the pressure value inside the connecting pipe are as follows: Arrange the pressure sensor sequentially at intervals on the connecting pipe. n There are 1 pressure sensor, and the distance between any two adjacent pressure sensors is equal. n The pressure values ​​measured by the pressure sensors at the same time were as follows: The pressure value of the connecting pipe within the corresponding cycle is: ； The specific steps for measuring the flow velocity in the connecting pipe with a liquid flow meter are as follows: First, set the measurement cycle... Divided into m a short period of time At the same time, obtain each The flow rate of the liquid flow meter within the time period is recorded as follows: Then, the cross-sectional area of ​​the connecting pipe corresponding to the pre-determined liquid flow meter is used. S , obtain each The flow rate over a given time period, i.e.: ； Finally, the flow rate within the corresponding measurement period is obtained: ； The pressure and flow rates measured periodically are then structured into data, specifically as follows: Pressure and flow velocity values ​​measured in the same period are paired to form two-dimensional data pairs, which serve as node data for the periodic sampling time. Multiple node data are sorted according to the sampling time order to form time series data; specifically: ； In the formula: t The time to be evaluated, the total sampling period, i.e. the measurement period is ; Indicates the number of moments before the time to be evaluated. i The pressure value was measured once; Indicates the number of moments before the time to be evaluated. i The flow rate was measured once; Select the time to be evaluated and the time preceding it. N’ Secondary node data serves as time-series data for the time point to be evaluated, or structured data. The classification detection stage of the aforementioned dropout detection model employs a ResNet network. Its input is the time-series features output from the first stage, and its output is the classification detection result; that is, the input is time-series data, and the output is... The probability of judgment is used to complete the judgment; The pre-training process uses experimentally generated labeled data to train the detachment detection model. The experimentally generated labeled data consists of a large amount of time-series data collected according to the information collection method. The labeled data is labeled by the labelers to determine whether there are cases of connector detachment within the corresponding time period of the time-series data. All time-series data are classified into positive examples and negative examples. Positive examples are all time series data where connector tube detachment occurs, while negative examples are all time series data where connector tube detachment does not occur. The positive and negative examples are split into training and test data according to the proportion to complete the pre-training of the dropout detection model. In the model prediction method, the probability of the connecting pipe detaching is determined based on the output of the detachment detection model. O y Greater than the preset alarm threshold O b In case of connector detachment, a warning will be issued and an alarm message will be pushed. When the probability of connector detachment in the model prediction method O y Greater than the preset danger threshold O w At that time, the training set is reconstructed, the dropout detection model is fine-tuned, and self-learning is completed; The output of the self-learning module is connected to the input of the dropout detection module for updating the dropout detection module, specifically: First, the newly recorded time-series data is categorized into new positive examples and new negative examples based on their corresponding feedback results. New positive examples are data where the feedback result for all newly recorded time-series data indicates that the connector has detached, while new negative examples are data where the feedback result for all newly recorded time-series data indicates that the connector has not detached. Next, the new positive and negative examples are proportionally split into new training data and new test data. The original training and test data are then redistributed, with the new training data and the redistributed training data merged into a single training dataset, and the new test data and the redistributed test data merged into a single test dataset. Finally, the learning rate is adjusted to fine-tune the detachment detection model, completing the self-learning process. The method for calculating the hazard of model errors includes updates for missed detections, specifically: In the formula: w l This indicates the weight of the missed detection record. Nl Indicates the number of missed detection records. Nlr This indicates the threshold for the number of missed detection records.

3. The real-time monitoring system for the tightness of anesthesia tubing connections according to claim 1, characterized in that: The real-time monitoring system also includes a feedback module with human-computer interaction functionality. The output of the feedback module is connected to the input of the data storage module. Based on the output of the dislodgement detection module, the anesthesiologist detects the dislodgement status of the anesthesia connecting tube and records the input time-series data of the alarm signal triggered by the dislodgement detection module and the feedback result in the data storage module. Specifically, the output of the feedback module indicates whether the anesthesia connecting tube is dislodged or not, which is achieved by setting up a button with human-computer interaction functionality.

Images