Laboratory air quality monitoring management system and method
By using intelligent control analysis module and communication alarm module with wireless communication connection in the laboratory, combined with random forest regression model and Internet of Things gateway, real-time monitoring and regulation of air quality is achieved, solving the inaccurate air quality monitoring in traditional technology. and remove incomplete problems, improving laboratory safety and processing efficiency.
Patent Information
- Application Number
- CN202411071872.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-06
- Publication Date
- 2024-11-12
AI Technical Summary
Traditional laboratory air quality monitoring is difficult to accurately detect the pollution location and type, and manual operation is difficult to completely remove dangerous gases, which poses safety hazards.
Using wireless communication connection intelligent analysis module and communication alarm module, data is collected through sensor networks, a random forest regression model is constructed, air quality risk levels are predicted in real time, and air quality adjustment is performed through the Internet of Things gateway control and execution equipment.
It realizes efficient and accurate monitoring and regulation of laboratory air quality, reduces the number of pollution, improves treatment efficiency and laboratory safety.
Smart Images

Figure CN118937592A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of air quality monitoring, and in particular to a laboratory air quality monitoring management system and method. Background Art
[0002] As the main place for colleges and universities to carry out teaching practice and scientific research, the laboratory is full of activities and has a high mobility of personnel. At the same time, various volatile dangerous goods such as flammable and explosive substances are stored. Safety accidents occur from time to time, seriously threatening the personal safety of teachers and students. For different experimental purposes, a large number of uncontrollable dangerous gases are generated during the experiment, which need to be removed in time. In some specific experiments, there are strict requirements on the experimental environment and laboratory air quality.
[0003] Traditional laboratories adopt a manual safety supervision mode, manually turning on mechanical ventilation equipment such as exhaust fans for ventilation, regularly using air purifiers for air purification, and manually using disinfectants to disinfect the laboratory regularly. These are all managed through manual operations, and it is difficult to accurately detect and monitor the type and location of pollution. It is also difficult to completely remove pollution through manual operations, and there are still hidden dangers. Summary of the invention
[0004] To solve the above problems, the present invention provides a laboratory air quality monitoring management system and method, which can accurately detect the pollution location and pollution type, intelligently identify the risk level and give suggestions, and at the same time start the execution equipment for cleaning.
[0005] To achieve the above object, the technical solution adopted by the present invention is:
[0006] A laboratory air quality monitoring and management system, comprising: an intelligent control analysis module and a communication alarm module connected by wireless communication, wherein:
[0007] The intelligent control analysis module is used to analyze air quality based on the sensor network to predict risks and provide adjustment suggestions, including the following steps:
[0008] A1: Data collection is performed through the sensor network. All sensor data is collected every hour, and information about devices and related activities is recorded during the process. The collected data is cleaned, standardized, and feature engineered.
[0009] A2: Divide the preprocessed data into training set, validation set and test set, build a random forest regression model, use the validation set to evaluate the model, and optimize the model performance by adjusting the parameters of the random forest model based on the evaluation results;
[0010] A3: Input the new laboratory air quality data acquired by the sensor in real time, use the trained random forest regression model to predict the new data, determine the risk level of the laboratory air quality based on the predicted value of the risk indicator and the preset risk threshold, and give corresponding adjustment suggestions for different risk levels;
[0011] A4: In a specific experiment, the threshold of the sensor data in a specific area is set according to the experimental requirements, and the data of the sensor network is compared through the threshold comparison algorithm to determine the risk level;
[0012] The communication alarm module is used to categorize risks that have occurred or are predicted to occur, and promptly notify the control console and the person in charge via communication.
[0013] Furthermore, the preprocessed data is divided into a training set, a validation set and a test set, a random forest regression model is constructed, the model is evaluated using the validation set, and according to the evaluation result, the parameters of the random forest model are adjusted to optimize the model performance, including the following steps:
[0014] A21: Shuffle the preprocessed data and divide the data into training set, validation set, and test set in a ratio of 7:2:1;
[0015] A22: The number of decision trees was initially set to 100, the maximum depth of the decision tree was set to None, the minimum number of samples required for internal node repartitioning was initially set to 2, and the minimum number of samples for leaf nodes was initially set to 1 to determine the model hyperparameters;
[0016] A23: Initialize the random forest regression model, pass the set hyperparameters into the model, input the training set data into the model, and the model begins to learn the relationship between the features in the data and the target variable;
[0017] A24: Input the validation set data into the trained random forest regression model to obtain the corresponding prediction results. According to the prediction results and the true target variable value of the validation set, calculate the mean square error, mean absolute error and determination coefficient;
[0018] A25: Determine the search space for hyperparameters, use the random search method to search within the determined hyperparameter search space, use the validation set data to train and evaluate the model for each set of hyperparameter combinations, compare the evaluation indicators of the models under different hyperparameter combinations, and select the hyperparameter combination that optimizes the model performance;
[0019] A26: Rebuild the random forest regression model using the optimal hyperparameter combination and train it using the training set data.
[0020] Furthermore, the random forest regression model is rebuilt using the optimal hyperparameter combination and trained using the training set data, including the following steps:
[0021] A261: Based on the construction method and interface of the random forest regression model in the framework, start to create a new random forest regression model instance, and accurately input the obtained optimal hyperparameter values into the model construction parameters one by one;
[0022] A262: After the model is built, the prepared training set data is input into the model. The model starts to process the training set data, analyzes the relationship between the input feature values and the corresponding target values, and adjusts the parameters and structure within the model to learn the patterns and rules in the data.
[0023] A263: During the training process, the model is continuously iterated and optimized until the preset training stop condition is reached.
[0024] Furthermore, the risks that have occurred or are predicted to occur are classified into different levels, and the control console and the person in charge are notified in a timely manner by means of communication, including the following steps:
[0025] B1: Based on the numerical range of risk indicators obtained by the risk prediction model, different risk levels are divided and clear threshold ranges are set for each risk level;
[0026] B2: Compare the risk indicator value output by the risk prediction model with the set threshold range, and classify the predicted risk into the corresponding risk level according to the comparison result;
[0027] B3: Determine the method of communicating with the control panel and the person in charge, collect the communication information of the control panel and the person in charge, and when the risk prediction result is determined, write the notification content according to the risk level and send the information in the form of a combination of text messages and monitoring alarms, and record the entire risk notification process.
[0028] Furthermore, it also includes a dynamic monitoring module, which is used to obtain real-time data through a sensor network and display the air quality situation in the corresponding area of the monitoring screen. The dynamic monitoring module includes a sensor network unit, a camera unit and an integrated unit, wherein:
[0029] The sensor network unit is used to acquire real-time data of harmful gases, carbon dioxide, bacteria, temperature, humidity, particulate matter and light in the laboratory; wherein the sensor network unit includes a photoionization sensor, a non-dispersive infrared sensor, a biosensor, a thermistor sensor, a capacitive humidity sensor, a light scattering sensor and a photoresistor sensor;
[0030] The camera unit is used to clearly capture images of various areas of the laboratory through a webcam;
[0031] The integrated unit is used to combine the sensor data obtained by the network sensor unit with the images obtained by the camera unit in a specified area and display them synchronously in a visual form.
[0032] Furthermore, in the designated area, the sensor data acquired by the network sensor unit is combined with the picture acquired by the camera unit to be synchronously displayed in a visual form, including the following steps:
[0033] C1: Remove outliers and noise from the data collected by the sensor network unit, unify the format and standardize it, decode and convert the collected video through the camera unit, and then use image enhancement technology to process it;
[0034] C2: Collect a large amount of laboratory image data, annotate laboratory image data in different regions, and use the annotated data to train the target detection model;
[0035] C3: Apply the trained object detection model to the real-time video stream and match the data collected by the sensor with the corresponding video area detected by the object detection model according to the sensor's installation position in the physical space;
[0036] C4: Design the corresponding visualization form and use image registration and fusion technology to overlay the generated data visualization image into the corresponding video area;
[0037] C5: Design user interaction functions, optimize the real-time, accuracy and stability of the system, regularly collect new data to retrain and update the deep learning model, and regularly maintain and calibrate sensors and hardware equipment.
[0038] Furthermore, the collecting of a large amount of laboratory image data, labeling of laboratory image data of different regions, and training of the target detection model using the labeled data include the following steps:
[0039] C21: Collect a large amount of image and video data containing scenes of the target monitoring area. The data covers different lighting conditions, viewing angles, scene layouts, and personnel activities. Use the image annotation tool Labelmg to annotate different areas in the collected images and video frames. The annotation information includes the bounding box coordinates of the area and the corresponding area category label.
[0040] C22: In the target detection task, select the YOLOv5 target detection model and determine the model's hyperparameters based on the architecture and requirements of the selected model;
[0041] C23: Divide the labeled data set into training set, validation set and test set in a ratio of 7:2:1. Use the training set data to train the built deep learning model. In the process, use the stochastic gradient descent algorithm to update and optimize the model parameters.
[0042] C24: Use validation set data to validate the model regularly during training, adjust model hyperparameters based on validation results, and stop training and save the trained model parameters when the performance of the model on the validation set reaches expectations or no longer improves.
[0043] Furthermore, the design of a corresponding visualization form, using image registration and fusion technology to superimpose the generated data visualization image into the corresponding video area, includes the following steps:
[0044] C41: Visualize various sensor data in the form of heat maps, digital forms, dashboards, pie charts, bar charts and line charts with timelines as reference indicators, and use different colors to show the concentration;
[0045] C42: Reasonably arrange data visualization elements according to the size and position of different areas in the video screen to avoid blocking important information while clearly displaying data;
[0046] C43: Extract SIFT feature points from the video image and the data visualization image, then calculate the transformation matrix between the images through feature point matching, and map the data visualization image to the corresponding position of the video image according to the transformation matrix;
[0047] C44: Use image fusion algorithm to fuse data visualization images with video images. For data visualization elements with high transparency, use weighted averaging method to superimpose visualization images and video images according to certain weights. For opaque data visualization elements, use multi-resolution fusion method to fuse visualization elements with video images on the high-frequency detail layer.
[0048] Furthermore, an Internet of Things gateway connected to the intelligent control analysis module is set up in the laboratory. The air quality risk type is determined by the intelligent control analysis module, and instructions are issued to the execution device through the Internet of Things gateway according to the proportional-integral-differential control algorithm to adjust the air quality.
[0049] A laboratory air quality monitoring and management method comprises the following steps:
[0050] D1: Input the laboratory air quality data obtained by the sensor in real time into the intelligent control analysis module, and use the trained random forest regression model to predict the new data;
[0051] D2: Determine the risk level of laboratory air quality based on the predicted value of the risk indicator and the preset risk threshold, and give corresponding adjustment suggestions for different risk levels. Alternatively, in a specific experiment, set a threshold for the sensor data in a specific area according to the experimental requirements, and perform a threshold comparison algorithm on the sensor network data to determine the risk level.
[0052] D3: When it is determined that the laboratory air quality requirements are not met, the notification content is written according to the risk level and sent in the form of a combination of SMS and monitoring alarms, and the entire risk notification process is recorded;
[0053] D4: Send instructions to the execution device through the IoT gateway to adjust the air quality until the laboratory air quality meets the requirements;
[0054] D5: During the entire process of monitoring the air quality in various areas of the laboratory, the dynamic monitoring module combines the sensor network to obtain real-time data and displays the air quality situation in the corresponding area of the monitoring screen, and displays the dangerous status of various types of data in different colors.
[0055] The beneficial effects of the present invention are:
[0056] 1. The intelligent control analysis module can efficiently and accurately judge the pollution situation through various sensor data, give specific adjustment suggestions based on the pollution situation, and provide guidance for subsequent operations;
[0057] 2. Predict and analyze existing sensor data through the intelligent control analysis module, make preventive judgments in advance before pollution occurs, and handle air quality in time to reduce the number of pollution events;
[0058] 3. The pollution information can be promptly notified to the control console and the responsible person through the communication alarm module, so that the relevant personnel can know the pollution situation at the first time, make arrangements for different pollution levels, and improve the processing efficiency;
[0059] 4. The dynamic monitoring module can intuitively understand the laboratory air quality monitoring situation and the laboratory on-site monitoring picture through the monitoring screen, and the important information can be clearly displayed without obstruction;
[0060] 5. By setting up personnel detection equipment and IoT gateways, the execution equipment can be directly turned on to adjust the air quality when the laboratory air quality is polluted based on IoT technology, the adjustment can be completed efficiently, manual cleaning time can be reduced, and experimental efficiency can be improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0061] Figure 1 It is a schematic diagram of the working process of the intelligent control analysis module of the present invention;
[0062] Figure 2It is a schematic diagram of the working process of the communication alarm module of the present invention;
[0063] Figure 3 It is a flow chart of the laboratory air quality monitoring and management method of the present invention. DETAILED DESCRIPTION
[0064] See also Figure 1 , Figure 2 and Figure 3 As shown, the present invention relates to a laboratory air quality monitoring and management system, including: an intelligent control analysis module and a communication alarm module connected by wireless communication, wherein:
[0065] The intelligent control analysis module is used to analyze air quality based on the sensor network to predict risks and provide adjustment suggestions, including the following steps:
[0066] A1: Data collection is performed through the sensor network. All sensor data is collected every hour, and information about devices and related activities is recorded during the process. The collected data is cleaned, standardized, and feature engineered.
[0067] A2: Divide the preprocessed data into training set, validation set and test set, build a random forest regression model, use the validation set to evaluate the model, and optimize the model performance by adjusting the parameters of the random forest model based on the evaluation results;
[0068] A3: Input the new laboratory air quality data acquired by the sensor in real time, use the trained random forest regression model to predict the new data, determine the risk level of the laboratory air quality based on the predicted value of the risk indicator and the preset risk threshold, and give corresponding adjustment suggestions for different risk levels;
[0069] A4: In a specific experiment, the threshold of the sensor data in a specific area is set according to the experimental requirements, and the data of the sensor network is compared through the threshold comparison algorithm to determine the risk level;
[0070] The communication alarm module is used to categorize risks that have occurred or are predicted to occur, and promptly notify the control console and the person in charge via communication.
[0071] Specifically, in step A1, the collected data is preprocessed. In data cleaning, if a certain data item is missing at a certain moment, it is filled with the average value of the data at adjacent moments. The 3σ principle is used to identify outliers in outlier processing. In data standardization, the Z-Score standardization method is used to convert the data into a distribution with a mean of 0 and a standard deviation of 1. In feature engineering, the ratio between different pollutant concentrations is calculated to construct a comprehensive air quality index AQI. It is assumed that the weights of each pollutant are w i, with a concentration of c i , with a reference concentration of b i , then AQI = max(c1 / b1*w1, c2 / b2*w2,..., c n / b n *w n ).
[0072] In step A3, set the risk threshold, compare the predicted risk index value with the threshold, and divide the risk level. Assume the risk index thresholds are T1 and T2 (T1 < T2), then the risk level is divided as follows:
[0073] 1. When y rf ≤ T1, the risk level is low;
[0074] 2. When T1 < y rf ≤ T2, the risk level is medium;
[0075] 3. When y rf > T2, the risk level is high.
[0076] Give adjustment suggestions. For example, when the risk level is low: maintain the current ventilation rate and operating parameters of the air purification equipment; when the risk level is medium, appropriately increase the ventilation rate by x% (such as 20% - 30%), check the filter of the air purification equipment, and replace it in advance if necessary; when the risk level is high: immediately stop the experimental activities, increase the ventilation rate to the maximum, comprehensively check the air purification equipment, and replace consumables such as filters and purifying agents.
[0077] In step A4, according to specific experimental requirements, set thresholds for harmful gases, carbon dioxide, bacteria, temperature, humidity, particulate matter, and light, compare various data obtained by the sensor network with the set thresholds, and based on the comprehensive threshold comparison results, summarize the factors that exceed the standard and conduct risk level classification. If none of the data exceeds the standard, the risk level is determined to be low; if 1 - 2 items of data exceed the standard, the risk level is determined to be medium; if 3 or more items of data exceed the standard, the risk level is determined to be high. According to the determination result of the risk level, when the risk level is medium or high, the system automatically triggers the communication alarm module to send warning messages to relevant personnel (such as laboratory administrators, experimental operators, etc.) and the console.
[0078] Furthermore, the steps of dividing the pre - processed data into a training set, a validation set, and a test set, constructing a random forest regression model, evaluating the model using the validation set, and optimizing the model performance by adjusting the parameters of the random forest model are as follows:
[0079] A21: Shuffle the pre - processed data, and divide the data into a training set, a validation set, and a test set according to the ratio of 7:2:1;
[0080] A22: The number of decision trees was initially set to 100, the maximum depth of the decision tree was set to None, the minimum number of samples required for internal node repartitioning was initially set to 2, and the minimum number of samples for leaf nodes was initially set to 1 to determine the model hyperparameters;
[0081] A23: Initialize the random forest regression model, pass the set hyperparameters into the model, input the training set data into the model, and the model begins to learn the relationship between the features in the data and the target variable;
[0082] A24: Input the validation set data into the trained random forest regression model to obtain the corresponding prediction results. According to the prediction results and the true target variable value of the validation set, calculate the mean square error, mean absolute error and determination coefficient;
[0083] A25: Determine the search space for hyperparameters, use the random search method to search within the determined hyperparameter search space, use the validation set data to train and evaluate the model for each set of hyperparameter combinations, compare the evaluation indicators of the models under different hyperparameter combinations, and select the hyperparameter combination that optimizes the model performance;
[0084] A26: Rebuild the random forest regression model using the optimal hyperparameter combination and train it using the training set data.
[0085] Specifically, in step A21, the preprocessed data set is randomly shuffled to make the data order irregular, so as to prevent the inherent pattern of data distribution from affecting the subsequent division results; the total amount of data is calculated, and the data amount of the training set, the validation set and the test set is determined according to the ratio of 70%, 20% and 10% respectively; a corresponding amount of data is selected from the front of the shuffled data set as the training set; then, a corresponding amount of data is selected from the front of the remaining data as the validation set; and the remaining data is used as the test set.
[0086] In step A22, in order to determine the model hyperparameters, the number of decision trees is initially determined to be 100 decision trees to form a random forest. The maximum depth of the decision tree is first set to unlimited (None) to allow the decision tree to grow naturally; the minimum number of samples required for internal node re-division is set to 2, that is, when the number of samples in the node is greater than or equal to 2, splitting is considered; the minimum number of leaf node samples is set to 1, that is, a leaf node contains at least 1 sample.
[0087] In step A23, the construction process of the random forest regression model is started. First, according to the construction method and interface of the random forest regression model in the selected machine learning library or framework, a new random forest regression model instance is created; the obtained optimal hyperparameter values are accurately input into the construction parameters of the model in sequence. For example, the number of decision trees is set to 100, the maximum depth is set to 10, the minimum number of samples required for internal node re-division is set to 2, and the minimum number of leaf node samples is set to 1. Through such parameter settings, the model will be built according to the optimal structure and configuration we determine. After the model is built, the prepared training set data is input into the model. The training set data includes multiple samples, each of which has a set of feature values (input features) and corresponding target values (output results); the model starts to process the training set data, and adjusts the parameters and structure inside the model by analyzing the relationship between the input feature values and the corresponding target values to learn the patterns and laws in the data. In the random forest regression model, each decision tree is learned and adjusted based on the training data, and multiple decision trees learn independently from each other through random feature selection and sample sampling. During the training process, the model is continuously iterated and optimized until the preset training stop conditions are reached. These conditions may include reaching a specified number of training rounds, the performance indicators of the model on the training set reaching a certain threshold, or the training error converging to a certain degree.
[0088] In step A24, calculate the mean square error: first, for each sample in the validation set, calculate the square of the difference between its true value and the model predicted value; then, sum these square differences and divide them by the number of samples to obtain the mean square error. Calculate the mean absolute error: for each sample in the validation set, calculate the absolute value of the difference between its true value and the model predicted value; sum these absolute values and divide them by the number of samples to obtain the mean absolute error. Calculate the coefficient of determination: first, calculate the average of the true values of all samples in the validation set; for each sample, calculate the sum of the squares of the difference between its true value and the average value, and the sum of the squares of the difference between the true value and the predicted value; then, calculate according to the formula of the coefficient of determination:
[0089]
[0090] Among them, m is the number of samples in the validation set, y i is the true value, is the predicted value, is the mean of the true values.
[0091] In step A25, the search space of hyperparameters is determined, for example, the possible values of the number of decision trees are [50, 100, 150], the possible values of the maximum depth of the decision tree are [None, 10, 20], the possible values of the minimum number of samples required for internal node re-division are [2, 5, 10], and the possible values of the minimum number of samples for leaf nodes are [1, 2, 4].
[0092] Furthermore, the random forest regression model is rebuilt using the optimal hyperparameter combination and trained using the training set data, including the following steps:
[0093] A261: Based on the construction method and interface of the random forest regression model in the framework, start to create a new random forest regression model instance, and accurately input the obtained optimal hyperparameter values into the model construction parameters one by one;
[0094] A262: After the model is built, the prepared training set data is input into the model. The model starts to process the training set data, analyzes the relationship between the input feature values and the corresponding target values, and adjusts the parameters and structure within the model to learn the patterns and rules in the data.
[0095] A263: During the training process, the model is continuously iterated and optimized until the preset training stop condition is reached.
[0096] Furthermore, the risks that have occurred or are predicted to occur are classified into different levels, and the control console and the person in charge are notified in a timely manner by means of communication, including the following steps:
[0097] B1: Based on the numerical range of risk indicators obtained by the risk prediction model, different risk levels are divided and clear threshold ranges are set for each risk level;
[0098] B2: Compare the risk indicator value output by the risk prediction model with the set threshold range, and classify the predicted risk into the corresponding risk level according to the comparison result;
[0099] B3: Determine the method of communicating with the control panel and the person in charge, collect the communication information of the control panel and the person in charge, and when the risk prediction result is determined, write the notification content according to the risk level and send the information in the form of a combination of text messages and monitoring alarms, and record the entire risk notification process.
[0100] Specifically, in step B1, different risk levels are divided. For example, the risk level can be divided into three levels: low risk, medium risk and high risk; assuming that the value range of the risk index is 0-100, then 0-30 can be set as low risk, 31-60 as medium risk, and 61-100 as high risk.
[0101] In step B3, determine the method of communication with the control desk and the person in charge, such as email, text message, instant messaging software or a dedicated alarm system; collect the communication information of the control desk and the person in charge, such as email address, mobile phone number, instant messaging account, etc., and ensure that this information is accurate and updated. When the risk prediction results are determined, write the notification content according to the risk level. The notification content should include detailed information such as risk level, risk indicator value, possible impact, and adjustment suggestions based on risk level; ensure that the notification content is clear, concise, and accurate, and avoid using vague or ambiguous language.
[0102] According to the established communication channels and collected communication information, use the corresponding tools or systems to send notifications; if asynchronous communication methods such as email or SMS are used, it is necessary to ensure that the notification is sent successfully and record the sending time and status; if real-time communication methods such as instant messaging software or dedicated alarm systems are used, it is necessary to wait for the recipient's confirmation or feedback to ensure that the notification is effectively received. After sending the notification, it is necessary to follow up on the receipt and processing of the notification. If no feedback is received from the recipient within a certain period of time, further measures need to be taken, such as sending the notification again, contacting by phone, etc.; record the entire risk notification process, including the time of notification sending, feedback from the recipient, follow-up measures taken, etc., so as to conduct subsequent analysis and summary, and provide a basis for optimizing the risk notification mechanism.
[0103] Furthermore, it also includes a dynamic monitoring module, which is used to obtain real-time data through a sensor network and display the air quality situation in the corresponding area of the monitoring screen. The dynamic monitoring module includes a sensor network unit, a camera unit and an integrated unit, wherein:
[0104] The sensor network unit is used to acquire real-time data of harmful gases, carbon dioxide, bacteria, temperature, humidity, particulate matter and light in the laboratory; wherein the sensor network unit includes a photoionization sensor, a non-dispersive infrared sensor, a biosensor, a thermistor sensor, a capacitive humidity sensor, a light scattering sensor and a photoresistor sensor;
[0105] The camera unit is used to clearly capture images of various areas of the laboratory through a webcam;
[0106] The integrated unit is used to combine the sensor data obtained by the network sensor unit with the images obtained by the camera unit in a specified area and display them synchronously in a visual form.
[0107] Specifically, the camera unit can also use an industrial camera. The sensor network unit and the camera unit need to be set up in conjunction with each other according to laboratory requirements. A complete sensor network unit needs to be set up within the monitoring range of a single camera. The sensors in the sensor network unit can be adjusted according to the sensing requirements within the monitoring range. For example, if bacterial monitoring is not required, biosensors can be excluded as part of the sensor network layout requirements.
[0108] Furthermore, in the designated area, the sensor data acquired by the network sensor unit is combined with the picture acquired by the camera unit to be synchronously displayed in a visual form, including the following steps:
[0109] C1: Remove outliers and noise from the data collected by the sensor network unit, unify the format and standardize it, decode and convert the collected video through the camera unit, and then use image enhancement technology to process it;
[0110] C2: Collect a large amount of laboratory image data, annotate laboratory image data in different regions, and use the annotated data to train the target detection model;
[0111] C3: Apply the trained object detection model to the real-time video stream and match the data collected by the sensor with the corresponding video area detected by the object detection model according to the sensor's installation position in the physical space;
[0112] C4: Design the corresponding visualization form and use image registration and fusion technology to overlay the generated data visualization image into the corresponding video area;
[0113] C5: Design user interaction functions, optimize the real-time, accuracy and stability of the system, regularly collect new data to retrain and update the deep learning model, and regularly maintain and calibrate sensors and hardware equipment.
[0114] Specifically, in step C1, decoding is to use a corresponding decoder to decode the video into a sequence of original image frames, and convert the decoded image frames from the original format to a common RGB format for subsequent processing; image enhancement is to apply image enhancement techniques such as histogram equalization, contrast stretching or sharpening to improve the quality and clarity of the image; for example, the contrast of the image is increased by histogram equalization to make the details in the image more obvious.
[0115] In step C3, the position information of the sensor is converted into a coordinate system in the video image, and a one-to-one correspondence between the sensor and the video area is established so that spatial mapping can be established. According to the pre-established spatial mapping relationship, the sensor data is accurately associated with the corresponding video area detected by the target detection model. Through the timestamp synchronization mechanism, the temporal consistency of the sensor data and the video frame is ensured to achieve accurate data matching.
[0116] In step C5, a concise and intuitive user interface is designed, including menu bars, toolbars, data display areas, and video playback windows, etc. The user interface is planned to provide users with operation buttons, such as pausing / playing videos, switching the display of different sensor data, zooming in / out on the video screen, etc., allowing users to select different visualization modes, set data thresholds and alarm rules, etc., so that users can filter the displayed data according to conditions such as time range, regional location, or data value. When the user operates, the system can respond immediately and execute the corresponding task. Optimize the data processing algorithm, reduce unnecessary calculation steps and data copy operations, adopt parallel computing technology, such as using GPU to accelerate data processing and model inference processes, and adjust the frequency and buffer size of data acquisition and transmission to balance the real-time nature of the data and the stability of the system.
[0117] Increase the diversity and quantity of training data to improve the generalization ability of the model for different scenarios, introduce more advanced model architectures and training techniques, such as using transfer learning, self-supervised learning and other methods, post-process and filter the output results of the model, reduce the impact of noise and outliers, establish a complete error handling and recovery mechanism, conduct long-term stress testing and stability testing on the system, regularly perform health checks and performance monitoring on the system, and promptly detect and handle performance degradation or abnormalities.
[0118] Develop a regular data collection plan to ensure that the latest laboratory images and sensor data can be obtained, screen, clean and preprocess the newly collected data to make it meet the requirements of model training, and retrain the model. During the retraining process, the structure and hyperparameters of the model can be adjusted according to the changes in the data to improve the performance and adaptability of the model. Develop a regular maintenance plan for sensors and hardware equipment according to the recommendations of the equipment manufacturer and the frequency of use in the laboratory.
[0119] Furthermore, the collecting of a large amount of laboratory image data, labeling of laboratory image data of different regions, and training of the target detection model using the labeled data include the following steps:
[0120] C21: Collect a large amount of image and video data containing scenes of the target monitoring area. The data covers different lighting conditions, viewing angles, scene layouts, and personnel activities. Use the image annotation tool Labelmg to annotate different areas in the collected images and video frames. The annotation information includes the bounding box coordinates of the area and the corresponding area category label.
[0121] C22: In the target detection task, select the YOLOv5 target detection model and determine the model's hyperparameters based on the architecture and requirements of the selected model;
[0122] C23: Divide the labeled data set into training set, validation set and test set in a ratio of 7:2:1. Use the training set data to train the built deep learning model. In the process, use the stochastic gradient descent algorithm to update and optimize the model parameters.
[0123] C24: Use validation set data to validate the model regularly during training, adjust model hyperparameters based on validation results, and stop training and save the trained model parameters when the performance of the model on the validation set reaches expectations or no longer improves.
[0124] Specifically, in step C21, a category label and bounding box coordinates are assigned to each annotated area, and the annotated data is divided into a training set, a validation set, and a test set in a ratio of 8:1:1.
[0125] In step C22, select the stochastic gradient descent optimizer to set the learning rate, set the initial learning rate from a relatively large value, and use the learning rate decay strategy to decay according to the step size. Gradually reduce the learning rate as the training progresses to help the model converge; determine the number of training rounds based on the amount of data and the complexity of the model, conduct a small-scale experiment first, observe the performance of the model under different epochs, and then determine a suitable number of training rounds; select a suitable batch size based on the GPU memory size; select a data augmentation strategy, calculate or manually adjust the size and proportion of the anchor box based on the target size and shape distribution of the data set, use the K-Means clustering method to obtain an anchor box suitable for the data set, and regularly use the validation set to evaluate the performance of the model during the training process. Based on the evaluation results, fine-tune the hyperparameters to optimize the model performance.
[0126] Furthermore, the design of a corresponding visualization form, using image registration and fusion technology to superimpose the generated data visualization image into the corresponding video area, includes the following steps:
[0127] C41: Visualize various sensor data in the form of heat maps, digital forms, dashboards, pie charts, bar charts and line charts with timelines as reference indicators, and use different colors to show the concentration;
[0128] C42: Layout data visualization elements according to the size and position of different areas in the video frame;
[0129] C43: Extract SIFT feature points from the video image and the data visualization image, then calculate the transformation matrix between the images through feature point matching, and map the data visualization image to the corresponding position of the video image according to the transformation matrix;
[0130] C44: Use image fusion algorithm to fuse data visualization images with video images. For data visualization elements with high transparency, use weighted averaging method to superimpose visualization images and video images according to certain weights. For opaque data visualization elements, use multi-resolution fusion method to fuse visualization elements with video images on the high-frequency detail layer.
[0131] Furthermore, an Internet of Things gateway connected to the intelligent control analysis module is set up in the laboratory. The air quality risk type is determined by the intelligent control analysis module, and instructions are issued to the execution device through the Internet of Things gateway according to the proportional-integral-differential control algorithm to adjust the air quality.
[0132] Specifically, it is necessary to install corresponding execution equipment, such as fresh air systems, ultraviolet lamps, and temperature and humidity adjustment equipment (such as air conditioners, humidifiers, dehumidifiers, etc.). Connect sensors and execution equipment to the local IoT gateway or controller via wired (such as Ethernet, RS485, etc.) or wireless (such as Wi-Fi, Bluetooth, Zigbee, etc.). Configure network parameters for the IoT gateway or controller so that it can connect to the local network and ensure smooth network connection with the remote data platform or server. Upload the data collected by the sensor to the remote data platform or server for data analysis and processing; at the same time, enable the data platform or server to send control instructions to the IoT gateway or controller to achieve remote control of the execution equipment.
[0133] The sensor collects the laboratory's air quality data and environmental parameter data in real time at a preset frequency. The collected data is transmitted to the data platform or server through the network for storage and processing. The data analysis module in the data platform or server analyzes and processes the collected data in real time. For example, for particulate matter data, the detected particulate matter concentration is compared with the preset safety standard value; for the number of bacteria, it is determined whether the detected value exceeds the specified upper limit; for temperature and humidity, the actual detected value is compared with the preset appropriate range.
[0134] When the data analysis results show that the particle concentration exceeds the standard, the system determines whether the fresh air system is in an available state (such as no fault, normal power supply, etc.); there are also personnel detection equipment in the laboratory. When the number of bacteria exceeds the standard, the system confirms that the laboratory is unmanned through the personnel detection equipment (such as infrared sensors, access control systems, etc.); when the temperature or humidity exceeds the standard, the system checks the working status and availability of the corresponding adjustment equipment (such as air conditioners, humidifiers, dehumidifiers, etc.). If the particle concentration exceeds the standard and the fresh air system is available, the data platform or server sends a command to the IoT gateway or controller to start the fresh air system, speed up air circulation and filtration, and reduce the particle concentration; if the number of bacteria exceeds the standard and the laboratory is unmanned, the data platform or server sends a command to turn on the ultraviolet lamp for sterilization; after the sterilization operation continues for a period of time (set according to actual needs), the ultraviolet lamp is automatically turned off; when the temperature or humidity exceeds the standard, the data platform or server sends a command according to the actual situation to start the air conditioner for temperature adjustment, or start the humidifier and dehumidifier for humidity adjustment, so that the temperature and humidity of the laboratory reach the preset appropriate range.
[0135] The IoT gateway or controller monitors the operating status and working parameters of each execution device in real time (such as the wind speed of the fresh air system, the working time of the ultraviolet lamp, the operating power of the air conditioner, etc.), and uploads these data to the data platform or server; on the data platform or server side, the operating status and working parameters of the execution device are displayed and monitored in real time to detect equipment failures or abnormal conditions in a timely manner.
[0136] The sensor continuously collects the laboratory's air quality data and environmental parameter data to evaluate the operating effect of the execution equipment; if the air quality or environmental parameters still do not reach the preset standard values after the execution equipment has been running for a period of time, the system will automatically adjust the working parameters of the execution equipment (such as increasing the wind speed of the fresh air system, extending the working time of the ultraviolet lamp, adjusting the temperature setting value of the air conditioner, etc.), or issue an equipment failure alarm to prompt manual inspection and maintenance.
[0137] In the above content, the proportional-integral-differential control algorithm uses the proportion, integration and differentiation to calculate the control quantity according to the system error to control the execution equipment, which is suitable for the regulation and control of parameters such as temperature and humidity.
[0138] A laboratory air quality monitoring and management method comprises the following steps:
[0139] D1: Input the laboratory air quality data obtained by the sensor in real time into the intelligent control analysis module, and use the trained random forest regression model to predict the new data;
[0140] D2: Determine the risk level of laboratory air quality based on the predicted value of the risk indicator and the preset risk threshold, and give corresponding adjustment suggestions for different risk levels. Alternatively, in a specific experiment, set a threshold for the sensor data in a specific area according to the experimental requirements, and perform a threshold comparison algorithm on the sensor network data to determine the risk level.
[0141] D3: When it is determined that the laboratory air quality requirements are not met, the notification content is written according to the risk level and sent in the form of a combination of SMS and monitoring alarms, and the entire risk notification process is recorded;
[0142] D4: Send instructions to the execution device through the IoT gateway to adjust the air quality until the laboratory air quality meets the requirements;
[0143] D5: During the entire process of monitoring the air quality in various areas of the laboratory, the dynamic monitoring module combines the sensor network to obtain real-time data and displays the air quality situation in the corresponding area of the monitoring screen, and displays the dangerous status of various types of data in different colors.
[0144] Specifically, in step D2, the or in a specific experiment, the before and after are in a parallel setting relationship. When it is a conventional experiment, the mode before or can be used all the time; when it belongs to a specific experiment, some sensor types can be selected for threshold setting, and the sensor network data can be compared with the threshold algorithm to determine the risk level. The settings of the remaining sensor types still use the mode before or. In the entire method, the specific risk level division details need to be divided and thresholds set according to the needs of various laboratories and specific experiments. All devices involved in each module in the method are connected using a wireless communication network.
[0145] The above implementation modes are merely descriptions of the preferred implementation modes of the present invention, and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary engineering and technical personnel in the field shall fall within the protection scope determined by the claims of the present invention.
Claims
1. A laboratory air quality monitoring and management system, characterized in that: include: Intelligent control analysis module and communication alarm module connected by wireless communication, including: The intelligent control analysis module is used to analyze air quality based on the sensor network to predict risks and provide adjustment suggestions, including the following steps: A1: Data collection is performed through the sensor network. All sensor data is collected every hour, and information about devices and related activities is recorded during the process. The collected data is cleaned, standardized, and feature engineered. A2: Divide the preprocessed data into training set, validation set and test set, build a random forest regression model, evaluate the model using the validation set, and optimize the model performance by adjusting the parameters of the random forest model based on the evaluation results; A3: Input the new laboratory air quality data acquired by the sensor in real time, use the trained random forest regression model to predict the new data, determine the risk level of the laboratory air quality based on the predicted value of the risk indicator and the preset risk threshold, and give corresponding adjustment suggestions for different risk levels; A4: In a specific experiment, the threshold of the sensor data in a specific area is set according to the experimental requirements, and the data of the sensor network is compared through the threshold comparison algorithm to determine the risk level; The communication alarm module is used to categorize risks that have occurred or are predicted to occur, and promptly notify the control console and the person in charge via communication.
2. A laboratory air quality monitoring and management system according to claim 1, characterized in that: The preprocessed data is divided into a training set, a validation set, and a test set, a random forest regression model is constructed, the model is evaluated using the validation set, and according to the evaluation results, the parameters of the random forest model are adjusted to optimize the model performance, including the following steps: A21: Shuffle the preprocessed data and divide the data into training set, validation set, and test set in a ratio of 7:2:1; A22: The number of decision trees was initially set to 100, the maximum depth of the decision tree was set to None, the minimum number of samples required for internal node repartitioning was initially set to 2, and the minimum number of samples for leaf nodes was initially set to 1 to determine the model hyperparameters; A23: Initialize the random forest regression model, pass the set hyperparameters into the model, input the training set data into the model, and the model begins to learn the relationship between the features in the data and the target variable; A24: Input the validation set data into the trained random forest regression model to obtain the corresponding prediction results. According to the prediction results and the true target variable value of the validation set, calculate the mean square error, mean absolute error and determination coefficient; A25: Determine the search space for hyperparameters, use the random search method to search within the determined hyperparameter search space, use the validation set data to train and evaluate the model for each set of hyperparameter combinations, compare the evaluation indicators of the models under different hyperparameter combinations, and select the hyperparameter combination that optimizes the model performance; A26: Rebuild the random forest regression model using the optimal hyperparameter combination and train it using the training set data.
3. A laboratory air quality monitoring and management system according to claim 2, characterized in that: Rebuild the random forest regression model using the optimal hyperparameter combination and train it using the training set data, including the following steps: A261: Based on the construction method and interface of the random forest regression model in the framework, start to create a new random forest regression model instance, and accurately input the obtained optimal hyperparameter values into the model construction parameters one by one; A262: After the model is built, the prepared training set data is input into the model. The model starts to process the training set data, analyzes the relationship between the input feature values and the corresponding target values, and adjusts the parameters and structure within the model to learn the patterns and rules in the data. A263: During the training process, the model is continuously iterated and optimized until the preset training stop condition is reached.
4. A laboratory air quality monitoring and management system according to claim 1, characterized in that: Risks that have occurred or are predicted to occur are graded and promptly notified to the control center and the responsible person by communication, including the following steps: B1: Based on the numerical range of risk indicators obtained by the risk prediction model, different risk levels are divided and clear threshold ranges are set for each risk level; B2: Compare the risk indicator value output by the risk prediction model with the set threshold range, and classify the predicted risk into the corresponding risk level according to the comparison result; B3: Determine the method of communicating with the control panel and the person in charge, collect the communication information of the control panel and the person in charge, and when the risk prediction result is determined, write the notification content according to the risk level and send the information in the form of a combination of text messages and monitoring alarms, and record the entire risk notification process.
5. A laboratory air quality monitoring and management system according to claim 1, characterized in that: It also includes a dynamic monitoring module, which is used to obtain real-time data through a sensor network and display the air quality situation in the corresponding area of the monitoring screen. The dynamic monitoring module includes a sensor network unit, a camera unit and an integrated unit, wherein: The sensor network unit is used to acquire real-time data of harmful gases, carbon dioxide, bacteria, temperature, humidity, particulate matter and light in the laboratory; wherein the sensor network unit includes a photoionization sensor, a non-dispersive infrared sensor, a biosensor, a thermistor sensor, a capacitive humidity sensor, a light scattering sensor and a photoresistor sensor; The camera unit is used to clearly capture images of various areas of the laboratory through a webcam; The integrated unit is used to combine the sensor data obtained by the network sensor unit with the images obtained by the camera unit in a specified area and display them synchronously in a visual form.
6. A laboratory air quality monitoring and management system according to claim 5, characterized in that: In the designated area, the sensor data acquired by the network sensor unit is combined with the picture acquired by the camera unit to be synchronously displayed in a visual form, including the following steps: C1: Remove outliers and noise from the data collected by the sensor network unit, unify the format and standardize it, decode and convert the collected video through the camera unit, and then use image enhancement technology to process it; C2: Collect a large amount of laboratory image data, annotate laboratory image data in different regions, and use the annotated data to train the target detection model; C3: Apply the trained object detection model to the real-time video stream and match the data collected by the sensor with the corresponding video area detected by the object detection model according to the sensor's installation position in the physical space; C4: Design the corresponding visualization form and use image registration and fusion technology to overlay the generated data visualization image into the corresponding video area; C5: Design user interaction functions, optimize the real-time, accuracy and stability of the system, regularly collect new data to retrain and update the deep learning model, and regularly maintain and calibrate sensors and hardware equipment.
7. A laboratory air quality monitoring and management system according to claim 6, characterized in that: The method of collecting a large amount of laboratory image data, labeling laboratory image data of different regions, and using the labeled data to train the target detection model includes the following steps: C21: Collect a large amount of image and video data containing scenes of the target monitoring area. The data covers different lighting conditions, viewing angles, scene layouts, and personnel activities. Use the image annotation tool Labelmg to annotate different areas in the collected images and video frames. The annotation information includes the bounding box coordinates of the area and the corresponding area category label. C22: In the target detection task, select the YOLOv5 target detection model and determine the model's hyperparameters based on the architecture and requirements of the selected model; C23: Divide the labeled data set into training set, validation set and test set in a ratio of 7:2:
1. Use the training set data to train the built deep learning model. In the process, use the stochastic gradient descent algorithm to update and optimize the model parameters. C24: Use validation set data to validate the model regularly during training, adjust model hyperparameters based on validation results, and stop training and save the trained model parameters when the performance of the model on the validation set reaches expectations or no longer improves.
8. A laboratory air quality monitoring and management system according to claim 6, characterized in that: The design of the corresponding visualization form and the use of image registration and fusion technology to overlay the generated data visualization image into the corresponding video area include the following steps: C41: Visualize various sensor data in the form of heat maps, digital forms, dashboards, pie charts, bar charts and line charts with timelines as reference indicators, and use different colors to show the concentration; C42: Layout data visualization elements according to the size and position of different areas in the video frame; C43: Extract SIFT feature points from the video image and the data visualization image, then calculate the transformation matrix between the images through feature point matching, and map the data visualization image to the corresponding position of the video image according to the transformation matrix; C44: Use image fusion algorithm to fuse data visualization images with video images. For data visualization elements with high transparency, use weighted averaging method to superimpose visualization images and video images according to certain weights. For opaque data visualization elements, use multi-resolution fusion method to fuse visualization elements with video images on the high-frequency detail layer.
9. A laboratory air quality monitoring and management system according to claim 1, characterized in that: The laboratory is equipped with an Internet of Things gateway connected to the intelligent control analysis module. The air quality risk type is determined by the intelligent control analysis module, and instructions are sent to the execution device through the Internet of Things gateway according to the proportional-integral-differential control algorithm to adjust the air quality.
10. A laboratory air quality monitoring and management method, applied to a laboratory air quality monitoring and management system as claimed in any one of claims 1 to 9, characterized in that: The following steps are involved: D1: Input the laboratory air quality data obtained by the sensor in real time into the intelligent control analysis module, and use the trained random forest regression model to predict the new data; D2: Determine the risk level of laboratory air quality based on the predicted value of the risk indicator and the preset risk threshold, and give corresponding adjustment suggestions for different risk levels. Alternatively, in a specific experiment, set a threshold for the sensor data in a specific area according to the experimental requirements, and perform a threshold comparison algorithm on the sensor network data to determine the risk level. D3: When it is determined that the laboratory air quality requirements are not met, the notification content is written according to the risk level and sent in the form of a combination of SMS and monitoring alarms, and the entire risk notification process is recorded; D4: Send instructions to the execution device through the IoT gateway to adjust the air quality until the laboratory air quality meets the requirements; D5: During the entire process of monitoring the air quality in various areas of the laboratory, the dynamic monitoring module combines the sensor network to obtain real-time data and displays the air quality situation in the corresponding area of the monitoring screen, and displays the dangerous status of various types of data in different colors.
Citation Information
Cited By
Operation safety portrait management method and system based on large model
CN119228146A
Laboratory intelligent comprehensive management method and system
CN119359265A
High-rise residential area fire-fighting site automatic generation design method based on deep learning
CN119538389A
