Artificial intelligence system for synchronous regulation of laboratory environment and field

Abstract

The invention discloses an artificial intelligence system for synchronous regulation of a laboratory environment and a field. The artificial intelligence system comprises a data acquisition module, a data cleaning and preprocessing module, an environment database module, an artificial intelligence prediction module, an environment regulation control module, a central control and execution module and a communication and feedback module, the data acquisition module is used for acquiring multi-dimensional time sequence environment data from a laboratory and a target field environment; the data cleaning and preprocessing module processes the data and completes time alignment and resampling; the environment database module stores the standardized data; the artificial intelligence prediction module obtains a prediction value; the environment regulation control module generates a regulation control strategy and drives corresponding execution equipment to carry out environment parameter regulation; the central control and execution module drives the environment adjustment control module to adjust the environment; and the communication and feedback module realizes data communication and state feedback of each module, and constructs a complete closed-loop control system. A more reliable experiment environment is provided for scientific research work.

Description

Technical Field

[0001] This invention relates to the field of interdisciplinary technology of artificial intelligence and intelligent environmental control, specifically to an artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions. Background Technology

[0002] While environmental simulation technology has made some progress in current scientific research, it still faces numerous limitations. For example, studies on material aging, biological culture simulation, and climate adaptability experiments require a high degree of fidelity to real-world conditions in a laboratory environment. The complexity of real-world environments makes laboratory simulations difficult to achieve. These limitations significantly hinder in-depth exploration in many research fields. Taking rubber aging research as an example, in studying the response of rubber to ultraviolet radiation and temperature and humidity, the inability to accurately simulate the complex changes in ultraviolet radiation and temperature and humidity in the natural environment makes it difficult to accurately reveal the aging results of rubber under real-world conditions caused by variations in temperature, humidity, and ultraviolet radiation. This makes many studies requiring a high degree of fidelity to real-world ultraviolet radiation and temperature and humidity conditions in a laboratory environment difficult to replicate, hindering the effective output and practical application of related research results.

[0003] Factors such as ultraviolet (UV) intensity, temperature, and humidity in the natural environment are constantly changing, and their patterns of change vary significantly across different regions and time periods. Research on rubber aging and the effects of UV radiation also suffers from the limitations of environmental simulation. Rubber aging in natural environments is a complex process, with UV radiation playing a crucial role. However, laboratory simulations struggle to accurately reproduce the actual aging conditions caused by the combined effects of UV radiation, temperature, humidity, and oxygen on rubber. For example, it's impossible to precisely simulate the differences in UV intensity and duration across seasons and regions, and the interplay between these differences and fluctuations in environmental temperature and humidity on the aging rate and degree of rubber. This makes it difficult for researchers to comprehensively and accurately grasp the intrinsic relationship between rubber aging and UV radiation, temperature, and humidity, thus hindering the development of new anti-aging rubber materials and impacting the transformation of related research findings into practical production and applications. Furthermore, in laboratory simulations, traditional environmental control equipment often only allows for fixed UV intensity and temperature / humidity settings, failing to dynamically adapt to real-time changes in the environment. This results in significant discrepancies between simulation results and actual conditions, severely affecting the accuracy and reliability of research findings.

[0004] Traditional ultraviolet light and temperature and humidity control technologies have the following shortcomings:

[0005] (1) Laboratory ultraviolet (UV) control systems are typically based on simple threshold control logic. This involves setting upper and lower limits for UV intensity or spectral parameters, and the control equipment activates to adjust when environmental parameters exceed these limits. This control method lacks the ability to predict UV change trends, resulting in significant lag in the adjustment process. For example, during periods of rapid UV intensity increase on sunny days, the dimming equipment only starts working after the laboratory UV intensity exceeds the set upper limit. During the dimming process, the continuously rising external UV intensity may cause the laboratory UV intensity to exceed the upper limit again within a short period, resulting in a constantly fluctuating and unstable UV intensity. Furthermore, traditional systems cannot effectively integrate multi-source UV data, making it difficult to accurately simulate complex UV environments.

[0006] (2) Traditional temperature and humidity control technologies have shortcomings. Laboratory temperature and humidity control systems are usually based on simple threshold control logic. That is, an upper and lower limit value for temperature or humidity is set, and when the environmental parameters exceed this range, the control equipment starts to adjust. This control method lacks the ability to predict environmental change trends, and the adjustment process has obvious lag. For example, in a hot and humid environment in summer, the dehumidification equipment only starts working after the laboratory humidity exceeds the set upper limit. During the dehumidification process, due to the continuous rise in the external humidity, the laboratory humidity may exceed the upper limit again in a short period of time, resulting in the temperature and humidity being in an unstable and fluctuating state. In addition, traditional systems cannot effectively integrate multi-source environmental data, making it difficult to achieve accurate simulation of complex environments. Summary of the Invention

[0007] To overcome the shortcomings of the existing technology, this invention provides an artificial intelligence system for synchronous adjustment of laboratory environment and field environment. It uses artificial intelligence technology to realize real-time synchronous adjustment of laboratory ultraviolet environment and field environment, providing a more reliable experimental environment for scientific research.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] An artificial intelligence system for synchronous regulation of laboratory environment and field, comprising a data acquisition module, a data cleaning and preprocessing module, an environmental database module, an artificial intelligence prediction module, an environmental regulation and control module, a central control and execution module, and a communication and feedback module;

[0010] The data acquisition module is used to acquire raw data from the laboratory and target site environments, including key parameters such as ultraviolet intensity, temperature, and humidity; and to acquire multidimensional time-series environmental data.

[0011] The data cleaning and preprocessing module performs format unification, outlier removal, missing value imputation, noise reduction, and smoothing on the collected raw data, and completes time alignment and resampling to ensure the consistency and usability of various types of data on the timeline; it outputs high-quality standardized data as the input basis for subsequent AI modeling.

[0012] The environmental database module stores the standardized data output by the data cleaning and preprocessing module;

[0013] The artificial intelligence prediction module makes predictions based on the environmental database module and the multidimensional time series environmental data, and obtains predicted values.

[0014] The environmental regulation and control module generates a regulation and control strategy based on the difference between the predicted value and the current value, and drives the corresponding execution device to adjust the environmental parameters.

[0015] The central control and execution module adopts a PLC or DCS architecture as the system control core. It is responsible for receiving input from the data acquisition module, calling the prediction results of the artificial intelligence prediction module, executing the control strategy of the environmental adjustment and control module, distributing control instructions, and driving the environmental adjustment and control module to perform environmental adjustment in real time.

[0016] The communication and feedback module enables data communication and status feedback between modules, supports remote data transmission, device control command issuance, and operation status feedback, and constructs a complete closed-loop control system.

[0017] The data acquisition module deploys high-precision ultraviolet intensity sensors, temperature sensors, and humidity sensors, and connects to remote data sources via meteorological department APIs to form a multi-dimensional, multi-time-period environmental data acquisition system. The sensors are deployed in key reaction areas of the laboratory and light-sensitive areas on-site to ensure that the sampled data is representative and timely.

[0018] High-precision ultraviolet intensity sensors are deployed above the experimental operating table, near the sample processing area, in the center of the light-receiving area, and in the equipment exposure area to monitor the ambient ultraviolet intensity in real time. The ultraviolet intensity sensors convert the collected ultraviolet intensity signals into standard electrical signals and transmit them to the central control and execution module in real time through wired or wireless data communication modules.

[0019] The environmental database module constructs a structured environmental database that stores preprocessed historical and real-time environmental parameter data, including ultraviolet intensity, spectral distribution, temperature, and humidity. This database serves as the core data source for training and predicting artificial intelligence models and supports system retrospective analysis and continuous optimization.

[0020] The artificial intelligence prediction module uses recurrent neural networks (RNNs) and long short-term memory networks (LSTMs) to model multidimensional time-series environmental data; the operation steps are as follows:

[0021] Step 1. Read data and set input variables; The system first reads the environmental dataset that has been cleaned and preprocessed by the data cleaning and preprocessing module, which includes the laboratory temperature, humidity and ultraviolet index as prediction targets;

[0022] Step 2. Construct time series sample windows to train the time series model. The system divides the preprocessed environmental data into fixed-length sliding windows in chronological order. Data from multiple consecutive time moments within each window are used as model input, and the target value at the next time moment outside the window is used as model output. Supervised learning samples are generated in this way.

[0023] Step 3. Divide all samples into training and testing sets in chronological order (e.g., 80% for training and 20% for testing), and do not shuffle the order of the samples, preserving their temporal dependency structure;

[0024] Step 4. Construct the LSTM network structure:

[0025] The model uses a two-layer stacked LSTM structure:

[0026] The first layer contains 64 LSTM units, and the output sequence is used by the next layer; the second layer contains 32 LSTM units and outputs the final state; then there is a 16-node fully connected layer (ReLU activation) for feature transformation; the last layer is a linear output layer that predicts three target variables (temperature, humidity, and UV index).

[0027] Step 5. Model compilation and training: Mean squared error is used as the loss function. During the process, 50 training cycles are performed using training set samples, with 16 data points processed in each batch. The model performance is monitored in real time on the validation set to prevent overfitting.

[0028] Step 6. Model Evaluation and Testing; After the model is trained, evaluate its performance on the test set and output the mean absolute error key indicator to ensure that the model has good prediction accuracy.

[0029] Step 7. Prediction and Visualization: The system uses the trained model to predict the test set data and compares the prediction results with the actual values ​​graphically to intuitively judge the accuracy of the model in temperature prediction and other aspects.

[0030] Predict the trends of ultraviolet intensity, spectral distribution, temperature, and humidity changes over several time steps in the future;

[0031] Through feature extraction, sample construction, model training and optimization, high-precision and strong generalization prediction capabilities are achieved, providing data support for intelligent regulation. The specific implementation plan is as follows:

[0032] Step 1 Feature Extraction: Extract key features related to environmental parameters from historical and real-time environmental data to enhance the model's ability to identify complex time-series patterns;

[0033] Step 2 Sample Construction: Using the sliding time window method, environmental data from multiple consecutive time moments are used as input, and the target value at the next time moment or several future time moments is used as output to construct supervised learning samples, preserving time dependencies and improving the model's ability to capture dynamic changes;

[0034] Step 3: Model Training and Optimization: Construct a deep learning model using a recurrent neural network (RNN) and a long short-term memory network (LSTM), and use the mean squared error as the loss function for training to avoid overfitting and improve the model's generalization ability;

[0035] Step 4: Predictive Applications: The trained model is evaluated on the validation and test sets using accuracy metrics to ensure its stability and accuracy on unseen data. The optimized model can be integrated into the control system to predict future trends in UV intensity, spectral distribution, temperature, and humidity in real time, serving as a basis for automatic adjustment decisions.

[0036] The environmental regulation and control module includes:

[0037] UV intensity adjustment: The intensity is adjusted by regulating the current or voltage of the UV lamp assembly (UV = k * V * I);

[0038] Spectral adjustment: Data is collected using a spectral sensor and the spectrally tunable ultraviolet light source is adjusted to achieve spectral consistency;

[0039] Temperature regulation: The temperature is dynamically adjusted through air conditioning, electric heaters, compression or absorption refrigeration systems;

[0040] Humidity control: Precise humidification or dehumidification control can be achieved through steam humidifiers, ultrasonic humidifiers, refrigeration dehumidifiers, and rotary dehumidifiers.

[0041] The beneficial effects of this invention are:

[0042] The system technical solution of this invention achieves the effect that, through the coordinated work of the above modules, the system can realize high-precision, dynamic, and real-time synchronous adjustment of key parameters such as ultraviolet intensity, spectral distribution, temperature, and humidity in the laboratory environment and the field environment, significantly improving the realism, stability, and controllability of experimental simulation.

[0043] Through the coordinated operation of the above modules, the system can achieve high-precision, dynamic, and real-time synchronous adjustment of key parameters such as ultraviolet intensity, spectral distribution, temperature, and humidity in both the laboratory and field environments. Specifically, this is manifested in:

[0044] 1. High realism of experimental simulation: Real field data is obtained through multi-source data acquisition and access to meteorological platforms; future environmental trends are predicted based on LSTM models, enabling the laboratory to make adjustments in advance; UV intensity and spectral adjustment modules are combined to reconstruct UV irradiation conditions consistent with the field; through temperature and humidity prediction and compensation control, the laboratory environment matches the field in both macroscopic and dynamic trends.

[0045] 2. High controllability of experimental conditions: The sensor layout, control strategy and equipment type can be flexibly adjusted according to experimental needs; through the central control system, multi-channel data and prediction results are integrated to achieve joint control of ultraviolet light and temperature, thereby improving the overall precision of regulation.

[0046] In summary, this invention can not only reproduce the real-world environment, but also maintain the long-term stability of environmental parameters and programmable adjustment capabilities, greatly improving the reliability and repeatability of experimental results. It provides technical support for scientific research and industrial applications that require complex environmental simulation, such as material aging research, biological cultivation, and climate adaptability testing. Attached Figure Description

[0047] Figure 1 This is a schematic diagram illustrating the automatic data acquisition process of the present invention.

[0048] Figure 2 This is a diagram illustrating the learning and training process for artificial intelligence.

[0049] Figure 3 This is a schematic diagram of data cleaning and preprocessing.

[0050] Figure 4 A diagram illustrating the prediction process for building a model.

[0051] Figure 5 This is a schematic diagram of the codes for adjusting ultraviolet intensity and temperature and humidity. Detailed Implementation

[0052] The present invention will now be described in further detail with reference to the accompanying drawings.

[0053] An artificial intelligence system for synchronous regulation of laboratory environment and field, comprising a data acquisition module, a data cleaning and preprocessing module, an environmental database module, an artificial intelligence prediction module, an environmental regulation and control module, a central control and execution module, and a communication and feedback module;

[0054] The data acquisition module is used to acquire raw data from the laboratory and target site environments, including key parameters such as ultraviolet intensity, temperature, and humidity; and to acquire multidimensional time-series environmental data.

[0055] The data cleaning and preprocessing module performs format unification, outlier removal, missing value imputation, noise reduction, and smoothing on the collected raw data, and completes time alignment and resampling to ensure the consistency and usability of various types of data on the timeline; it outputs high-quality standardized data as the input basis for subsequent AI modeling.

[0056] The environmental database module stores the standardized data output by the data cleaning and preprocessing module;

[0057] The artificial intelligence prediction module makes predictions based on the environmental database module and the multidimensional time series environmental data, and obtains predicted values.

[0058] The environmental regulation and control module generates a regulation and control strategy based on the difference between the predicted value and the current value, and drives the corresponding execution device to adjust the environmental parameters.

[0059] The central control and execution module adopts a PLC or DCS architecture as the system control core. It is responsible for receiving input from the data acquisition module, calling the prediction results of the artificial intelligence prediction module, executing the control strategy of the environmental adjustment and control module, distributing control instructions, and driving the environmental adjustment and control module to perform environmental adjustment in real time.

[0060] The communication and feedback module enables data communication and status feedback between modules, supports remote data transmission, device control command issuance, and operation status feedback, and constructs a complete closed-loop control system.

[0061] The data acquisition module deploys high-precision ultraviolet intensity sensors, temperature sensors, and humidity sensors, and connects to remote data sources via meteorological department APIs to form a multi-dimensional, multi-time-period environmental data acquisition system. The sensors are deployed in key reaction areas of the laboratory and light-sensitive areas on-site to ensure that the sampled data is representative and timely.

[0062] High-precision ultraviolet intensity sensors are deployed above the experimental operating table, near the sample processing area, in the center of the light-receiving area, and in the equipment exposure area to monitor the ambient ultraviolet intensity in real time. The ultraviolet intensity sensors convert the collected ultraviolet intensity signals into standard electrical signals and transmit them to the central control and execution module in real time through wired or wireless data communication modules.

[0063] The environmental database module constructs a structured environmental database that stores preprocessed historical and real-time environmental parameter data, including ultraviolet intensity, spectral distribution, temperature, and humidity. This database serves as the core data source for training and predicting artificial intelligence models and supports system retrospective analysis and continuous optimization.

[0064] The artificial intelligence prediction module uses recurrent neural networks (RNNs) and long short-term memory networks (LSTMs) to model multidimensional time-series environmental data; the operation steps are as follows:

[0065] Step 1. Read data and set input variables; The system first reads the environmental dataset that has been cleaned and preprocessed by the data cleaning and preprocessing module, which includes the laboratory temperature, humidity and ultraviolet index as prediction targets;

[0066] Step 2. Construct time series sample windows to train the time series model. The system divides the preprocessed environmental data into fixed-length sliding windows in chronological order. Data from multiple consecutive time moments within each window are used as model input, and the target value at the next time moment outside the window is used as model output. Supervised learning samples are generated in this way.

[0067] Step 3. Divide all samples into training and testing sets in chronological order (e.g., 80% for training and 20% for testing), and do not shuffle the order of the samples, preserving their temporal dependency structure;

[0068] Step 4. Construct the LSTM network structure:

[0069] The model uses a two-layer stacked LSTM structure:

[0070] The first layer contains 64 LSTM units, and the output sequence is used by the next layer; the second layer contains 32 LSTM units and outputs the final state; then there is a 16-node fully connected layer (ReLU activation) for feature transformation; the last layer is a linear output layer that predicts three target variables (temperature, humidity, and UV index).

[0071] Step 5. Model compilation and training: Mean squared error is used as the loss function. During the process, 50 training cycles are performed using training set samples, with 16 data points processed in each batch. The model performance is monitored in real time on the validation set to prevent overfitting.

[0072] Step 6. Model Evaluation and Testing; After the model is trained, evaluate its performance on the test set and output the mean absolute error key indicator to ensure that the model has good prediction accuracy.

[0073] Step 7. Prediction and Visualization: The system uses the trained model to predict the test set data and compares the prediction results with the actual values ​​graphically to intuitively judge the accuracy of the model in temperature prediction and other aspects.

[0074] Predict the trends of ultraviolet intensity, spectral distribution, temperature, and humidity changes over several time steps in the future;

[0075] Through feature extraction, sample construction, model training and optimization, high-precision and strong generalization prediction capabilities are achieved, providing data support for intelligent regulation.

[0076] The specific implementation plan is as follows:

[0077] Step 1 Feature Extraction: Extract key features related to environmental parameters from historical and real-time environmental data to enhance the model's ability to identify complex time-series patterns;

[0078] Step 2 Sample Construction: Using the sliding time window method, environmental data from multiple consecutive time moments are used as input, and the target value at the next time moment or several future time moments is used as output to construct supervised learning samples, preserving time dependencies and improving the model's ability to capture dynamic changes;

[0079] Step 3: Model Training and Optimization: Construct a deep learning model using a recurrent neural network (RNN) and a long short-term memory network (LSTM), and use the mean squared error as the loss function for training to avoid overfitting and improve the model's generalization ability;

[0080] Step 4: Predictive Applications: The trained model is evaluated on the validation and test sets using accuracy metrics to ensure its stability and accuracy on unseen data. The optimized model can be integrated into the control system to predict future trends in UV intensity, spectral distribution, temperature, and humidity in real time, serving as a basis for automatic adjustment decisions.

[0081] The environmental regulation and control module includes:

[0082] UV intensity adjustment: The intensity is adjusted by regulating the current or voltage of the UV lamp assembly (UV = k * V * I);

[0083] Spectral adjustment: Data is collected using a spectral sensor and the spectrally tunable ultraviolet light source is adjusted to achieve spectral consistency;

[0084] Temperature regulation: The temperature is dynamically adjusted through air conditioning, electric heaters, compression or absorption refrigeration systems;

[0085] Humidity control: Precise humidification or dehumidification control can be achieved through steam humidifiers, ultrasonic humidifiers, refrigeration dehumidifiers, and rotary dehumidifiers.

[0086] A method for operating an artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions includes the following steps;

[0087] Step 1: The data acquisition and learning stage of artificial intelligence:

[0088] To achieve real-time synchronous adjustment of laboratory temperature, humidity, and ultraviolet environment with the field, a large data sample library was first built. Through the data acquisition module, the required local historical ultraviolet intensity, temperature, humidity, and other data were obtained from multiple channels, such as the official database of the local meteorological department and the sensor network deployed on site.

[0089] During the data acquisition phase, to ensure the accuracy and integrity of the raw data, the system performs the following cleaning and preprocessing operations on the collected historical UV intensity, temperature, humidity, and other raw environmental data:

[0090] First, standardize the time format and units of measurement for data to ensure comparability and compatibility across different data sources. Specifically, this includes standardizing all timestamps to a unified format and unifying the units for temperature, humidity, and UV intensity to the system-defined standard units (°C, %RH, mW / m²). 2 ).

[0091] Secondly, statistical analysis methods are used to identify and process potential outliers in the data. Box plots are applied to detect outliers exceeding the interquartile range threshold, or the standard deviation range is used to determine extreme deviations. For identified outliers, replacement with the mean or median of adjacent valid values ​​can be chosen, or interpolation algorithms can be used for correction.

[0092] In terms of handling missing values, the system first identifies null values, illegal values, and invalid data (such as "-999" or "0") in the original data. Then, based on the data characteristics and time series patterns, it uses linear interpolation, forward / backward imputation, K-nearest neighbor interpolation, or time series model prediction to fill in the missing values ​​appropriately in order to maintain data continuity.

[0093] In addition, to reduce the interference of high-frequency noise in the raw data on model training, the system performs data smoothing. After completing data cleaning and denoising, the system also needs to perform time alignment processing on data from different sensors, unify the sampling frequency and perform resampling operations to ensure the consistency of various parameters on the time axis.

[0094] Step 2: Input the processed data into the artificial intelligence model for training:

[0095] After data cleaning and preprocessing, this invention establishes an environmental parameter database to store historical and real-time collected multi-source heterogeneous data such as ultraviolet intensity, spectral distribution, temperature, and humidity, serving as the foundational data for training and predicting artificial intelligence models. Based on this, recurrent neural networks (RNNs) and long short-term memory networks (LSTMs) are used to model and predict time-series data.

[0096] First, an input sequence is constructed based on historical data in the database, and features such as intraday rate of change, extreme values, and mean values ​​are extracted. Supervised learning samples are generated through a sliding time window, and the data is divided into training set, validation set, and test set.

[0097] Subsequently, a network structure composed of single or multiple RNN / LSTM units is designed, and the Dropout mechanism and regularization method are combined to improve the generalization ability of the model. During the model training process, loss functions such as mean squared error are selected, and the parameters are iterated using Adam or RMSprop optimizers. The network structure and hyperparameter combination are continuously optimized through K-fold cross-validation.

[0098] During the validation and testing phase, the model's predictive accuracy and stability on unseen data are evaluated.

[0099] Finally, the trained high-precision model is integrated into the environmental control system to achieve real-time prediction of future ultraviolet radiation intensity and temperature and humidity, guiding the intelligent synchronous control of the laboratory environment. The environmental control system enables real-time prediction of ultraviolet radiation intensity and temperature and humidity, supporting subsequent intelligent control and regulation.

[0100] Step 3: After the database is established and preliminary simulations have verified that it has a certain level of prediction accuracy and stability, the system enters the stage of real-time monitoring and intelligent control of environmental parameters.

[0101] In terms of ultraviolet (UV) intensity regulation, the system deploys high-precision UV intensity sensors in key reaction areas within the laboratory (such as above the experimental workbench and near the sample processing area) and in areas of the target environment most sensitive to UV changes (such as the center of the light-receiving area and the equipment exposure area) to monitor the ambient UV intensity in real time. These UV intensity sensors possess high temporal resolution and high sensitivity, capable of converting the collected UV intensity signals into standard electrical signals and transmitting them to the central control system in real time via wired or wireless data communication modules.

[0102] The environmental control module adopts a programmable logic controller (PLC) or distributed control system (DCS) architecture, receives ultraviolet (UV) intensity signals from various monitoring points, and compares the UV intensity difference between the laboratory and the field in real time. When the system detects a significant deviation between the laboratory UV intensity and the target field, it will combine the prediction results of the artificial intelligence prediction module for future UV intensity with the system's preset control strategy to activate the UV regulation module.

[0103] After the control system receives the real-time data from the ultraviolet intensity sensor, it first compares the current ultraviolet intensity in the laboratory with the real-time ultraviolet intensity at the target site and calculates the difference ΔU between the two.

[0104] At the same time, the artificial intelligence prediction model deployed in the system is invoked to predict the changing trend of ultraviolet intensity at the target site within a certain number of time steps in the future, generating a prediction sequence as a preliminary basis for regulation.

[0105] Based on the above prediction results and the current difference ΔU, the control system generates control commands in conjunction with the system's preset control strategy. The preset control strategy includes, but is not limited to, proportional-integral-derivative (PID) control rules, fuzzy control logic, or expert rule tables, used to determine the magnitude and direction of the current control response.

[0106] The system determines the dynamic adjustment trend of ultraviolet output power based on the rising or falling trend of the predicted value, and combines this with the current error ΔU. V The magnitude of the amplitude determines the strength of the regulatory response.

[0107] Specifically, the control system adjusts the input voltage (V) or current (I) of the ultraviolet lamp assembly to achieve the output intensity U. V The system dynamically adjusts the UV lamp intensity using the formula U = k * V * I, where k is the equipment characteristic coefficient of the UV lamp assembly. The system can be programmed with a target following curve for UV intensity, fitting the predicted target intensity to the current laboratory intensity, and achieving precise adjustment through closed-loop control. The adjustment process can employ PWM pulse width modulation, analog signal output, or digital signal control to regulate power, ensuring the stability and response speed of the UV output intensity.

[0108] When the AI ​​prediction module determines that a significant change in ultraviolet (UV) intensity is imminent, the control system can intervene in advance to adjust the process, starting or pre-setting the output power of the UV equipment, thereby improving the system's response speed and robustness to sudden changes. This prediction-driven control mechanism enables the system to have feedforward regulation capabilities, effectively improving the accuracy of synchronous tracking of the target site by the laboratory environment.

[0109] By adjusting the current or voltage of the ultraviolet lamp using the formula UV=k*V*I, the luminous intensity is automatically adjusted based on the difference in ultraviolet intensity, ensuring that the ultraviolet intensity in the laboratory can quickly and accurately follow changes in the ultraviolet intensity on site.

[0110] Regarding spectral modulation:

[0111] Ultraviolet (UV) spectral data are collected from both the laboratory and the target site using spectral sensors. During the UV spectral synchronization adjustment process, high-precision spectral sensors are deployed in both the laboratory and the target site to acquire UV band data (such as U+) in real time. VA U VB U VC The spectral distribution data of (etc.) are transmitted via a data transmission module to a central control and execution module, where they are analyzed and processed by an artificial intelligence model.

[0112] The artificial intelligence prediction module, based on historical and real-time spectral data, predicts the trend of ultraviolet spectral changes at the target site over several future time steps and outputs a reference spectral curve S. _ref(λ) And compared with the actual spectral distribution S in the current laboratory _lab(λ) Compare and calculate the spectral difference ΔS within the wavelength range. (λ) =S _ref(λ) –S _lab(λ) ;

[0113] The difference function ΔS (λ) The differences in energy distribution at different wavelengths λ between the laboratory ultraviolet spectrum and the target site were characterized.

[0114] This ensures that the ultraviolet (UV) spectral distribution in the laboratory is consistent with that in the field. This guarantees that the UV environment in the laboratory and the field are identical.

[0115] Regarding temperature and humidity, once the database is established and the preliminary simulation verification is mature, real-time control can be put into the monitoring and regulation stage. In terms of temperature regulation, high-precision temperature sensors are deployed in the laboratory and the target site for real-time monitoring of temperature data.

[0116] The sensor converts the collected temperature signal into an electrical signal and transmits it to the control system in real time via a data transmission network.

[0117] When a deviation between the laboratory temperature and the ambient temperature is detected, the control system controls temperature control equipment such as air conditioning units and electric heaters based on the prediction results of the artificial intelligence model and the preset control strategy. For the refrigeration process, a compression refrigeration system or an absorption refrigeration system is used, and the refrigeration capacity is automatically adjusted according to the temperature difference by regulating the operating frequency of the refrigeration compressor or the flow rate of the refrigerant. During the heating process, precise temperature compensation is achieved by adjusting the power of the electric heater, ensuring that the laboratory temperature can quickly and accurately follow changes in the ambient temperature.

[0118] Regarding humidity control:

[0119] High-precision humidity sensors are used to collect humidity data in the laboratory and on-site. When humidification is needed, a steam humidifier or ultrasonic humidifier operates under the control of a humidity controller. The humidity controller precisely adjusts the humidification level based on the humidity difference calculated by an artificial intelligence model, causing water molecules to be released into the air in the form of steam or tiny droplets, increasing the ambient humidity. The specific calculation steps are as follows:

[0120] During humidity control, the system deploys high-precision humidity sensors in key areas of the laboratory and at the target site to collect relative humidity data in real time, which is denoted as RH. _lab (Laboratory) and RH _site (On-site). Sensor data is converted into standard electrical signals by the signal processing module and transmitted to the central control system via the data communication network.

[0121] The central control and execution module, along with the communication and feedback module, invokes the deployed artificial intelligence model to predict the humidity change trend of the target site over a future period, outputting a short-term predicted humidity value RH_. pred_site(t+1,t+2,.) At the same time, considering the current humidity (RH) in the laboratory... lab (t), calculate the expected humidity difference between the laboratory and the field within the prediction time window;

[0122] The specific calculation steps are as follows:

[0123] 1. Real-time data input:

[0124] • Obtain laboratory humidity (RH) lab (t)

[0125] • Obtain on-site humidity prediction values ​​(RH) pred_site (t+n) (Predicted by AI model)

[0126] • Set the predictive control time window to n (e.g., 10 minutes).

[0127] 2. Difference Calculation:

[0128] • Calculate the target humidity (RH) target (t), can be selected as:

[0129]

[0130] • Calculate the humidity difference:

[0131] ΔRH(t)=RH target (t)-RH lab (t)

[0132] 3. Control Decision Generation:

[0133] • If ΔRH(t) > 0, the system determines that humidification is needed;

[0134] The controller calculates the required humidification amount Q based on the difference in humidification volume, combined with the equipment adjustment curve and the space volume. humid (t), as follows:

[0135] Qhumid(t) = f(ΔRH(t), V room T lab )

[0136] Where V room T is the volume of the experimental space. lab Let f be the current room temperature, and f be the humidification function derived from meteorological formulas.

[0137] 4. Device driving and execution:

[0138] • The humidity controller outputs a control signal to a steam humidifier or an ultrasonic humidifier;

[0139] • Control parameters include atomization power, spray frequency, and on-time duration;

[0140] • Control methods can include PWM modulation, analog voltage output, or digital control interface.

[0141] Using the above methods, the system can calculate reasonable humidification targets and control strategies based on artificial intelligence prediction results and the current environmental conditions, and drive the humidification equipment to release water molecules into the air in the form of steam or fine mist droplets, accurately improving the humidity level of the laboratory and ensuring that it remains synchronized with the target site environment.

[0142] like Figure 1 The diagram shows automatic data acquisition via sensors and networked machines.

[0143] The system described in this invention achieves real-time automatic acquisition of environmental parameters in the laboratory and target sites by deploying high-precision environmental sensors and embedded network terminals. At the same time, the system connects to a meteorological platform interface to periodically acquire historical and real-time data of the target area. All data is converted into a standard format through a unified interface and uploaded to the central control system or cloud platform, providing high-quality data support for artificial intelligence modeling and synchronous environmental adjustment.

[0144] like Figure 2 The diagram illustrates the training approach for artificial intelligence.

[0145] The artificial intelligence system described in this invention models and trains multi-dimensional time-series environmental data such as ultraviolet radiation intensity, temperature, and humidity based on recurrent neural networks (RNNs) and long short-term memory networks (LSTMs). During the training phase, a multi-layer LSTM network structure is constructed, and Dropout and regularization are introduced to prevent overfitting. Mean squared error is used as the loss function, and the Adam optimizer is employed for weight updates. After training, the system evaluates the test set. Once the model demonstrates good predictive accuracy and stability, it is integrated into an environmental control system for real-time prediction of future environmental trends. This prediction serves as the basis for generating control commands, enabling feedforward adjustment of environmental parameters.

[0146] like Figure 3 The data is cleaned and preprocessed as shown.

[0147] First, the time format and units of measurement across all data sources are standardized, and timestamps, temperature, humidity, and UV intensity are converted to system-wide units. Second, outliers are identified using box plots or standard deviation methods and corrected through interpolation of adjacent data, mean replacement, or median replacement. To reduce high-frequency noise interference, the data is further smoothed. Finally, the system performs time alignment and resampling operations on data collected from different sensors to ensure consistency and comparability of all parameters across the time axis, providing high-quality, continuous data input for subsequent model training.

[0148] like Figure 4 The diagram shows the construction of an RNN / LSTM for time series prediction.

[0149] This invention, after cleaning and preprocessing environmental data, employs Recurrent Neural Networks (RNNs) and Long Short-Term Memory Networks (LSTMs) to model and predict multivariate time-series data, enabling prediction of future trends in environmental parameters such as UV intensity, temperature, and humidity. First, the system constructs a feature input sequence and target prediction values ​​based on the processed data, and converts continuous historical data into supervised learning samples using a sliding time window. Then, a deep network structure composed of multiple LSTM units is constructed, with the input being the environmental parameter sequence from several past time steps and the output being the predicted value for the next time step. The model structure includes a two-layer LSTM layer, a fully connected transform layer, and a linear output layer, and introduces Dropout and L2 regularization to suppress overfitting. During training, mean squared error is used as the loss function, and the Adam optimizer is used for parameter iteration. After fitting the model to the training set, its prediction accuracy and stability are evaluated using validation and test sets, and K-fold cross-validation is used to further optimize the hyperparameter combination. After training, the model is integrated into the system, receiving the latest data in real time and predicting the environmental state for several future time steps, serving as the basis for generating adjustment instructions and achieving prediction-driven feedforward control.

[0150] like Figure 5 The diagram shows the automatic on-site environmental adjustment of the predicted data.

[0151] In this invention system, the future trends of ultraviolet intensity, temperature, and humidity predicted by an artificial intelligence model (such as LSTM) will serve as an important basis for laboratory environmental regulation. After receiving the predicted values ​​output by the model, the control system automatically compares them with the current measured data in the laboratory, calculates the difference between the predicted target value and the current situation, determines whether adjustment is needed, and automatically drives the corresponding equipment to execute control commands.

[0152] Code 1: Automatic Data Collection

[0153]

[0154]

[0155] ""Collect local and API data"

[0156] local_data=get_sensor_data()

[0157] api_data=get_weather_api_data()

[0158] timestamp=time.strftime("%Y-%m-%d%H:%M:%S")

[0159] data = {

[0160] "timestamp":timestamp,

[0161] "local_temperature":local_data["temperature"],

[0162] "local_humidity":local_data["humidity"],

[0163] "local_uv_index":local_data["uv_index"],

[0164] "api_temperature":api_data["temperature"]if api_data else None,

[0165] "api_humidity":api_data["humidity"]if api_data else None,

[0166] "api_uv_index":api_data["uv_index"]if api_data else None,

[0167] }

[0168] return data

[0169] #Simulated Data Collection

[0170] data_list=[collect_data()for_in range(100)]

[0171] df = pd.DataFrame(data_list)

[0172] df.to_csv("environment_data.csv",index=False)

[0173] This section of code is used to collect environmental data in real time from both the laboratory and the field. It acquires temperature, humidity, and simulated ultraviolet (UV) data from local sensors, while simultaneously calling a weather API to obtain real-time temperature, humidity, and UV index data for the corresponding location. All data is compiled into a single record, containing a timestamp, local data, and API data. After iterative execution, a batch of data is generated and saved as a CSV file named `environment_data.csv`, providing the raw data foundation for subsequent modeling and prediction.

[0174] Code 2: Data Cleaning and Preprocessing

[0175] #Read data

[0176] df=pd.read_csv("environment_data.csv")

[0177] #Handling Missing Values

[0178] imputer=SimpleImputer(strategy="mean")

[0179] df.iloc[:,1:]=imputer.fit_transform(df.iloc[:,1:])

[0180] #Normalized data

[0181] from sklearn.preprocessing import MinMaxScaler

[0182] scaler = MinMaxScaler()

[0183] df.iloc[:,1:]=scaler.fit_transform(df.iloc[:,1:])

[0184] #Convert Time Column

[0185] df["timestamp"]=pd.to_datetime(df["timestamp"])

[0186] df.set_index("timestamp",inplace=True)

[0187] #Save the processed data

[0188] df.to_csv("cleaned_environment_data.csv")

[0189] print("Data cleaning complete")

[0190] This module reads the previously collected and saved data file and preprocesses it. First, it uses SimpleImputer to impute missing data by applying the mean. Then, it uses MinMaxScaler to normalize all data except for timestamps, ensuring they fall within a uniform numerical range, which is beneficial for improving model training performance. Finally, it converts the timestamps to datetime type and sets them as indexes, saving the result as cleaned_environment_data.csv to ensure the data format is suitable for time series modeling.

[0191] Code 3: Constructing RNN / LSTM for time series prediction

[0192] import tensorflow as tf

[0193]

[0194] X_train.shape[2])),

[0195] LSTM(32,return_sequences=False),

[0196] Dense(16,activation="relu"),

[0197] Dense(3)# predicts temperature, humidity, and UV index. ])

[0199] model.compile(optimizer="adam",loss="mse",metrics="["mae"])

[0200] model.summary()

[0201] #Training the model

[0202] history = model.fit(

[0203] X_train,y_train,epochs=50,batch_size=16,validation_data=(X_test,y_test),verbose=1 )

[0205] import matplotlib.pyplot as plt

[0206] #Evaluation Model

[0207] loss,mae=model.evaluate(X_test,y_test)

[0208] print(f"Test set MAE:{mae}")

[0209] #Making predictions

[0210] y_pred=model.predict(X_test)

[0211] #Visualization Results

[0212] plt.figure(figsize=(10,5))

[0213] plt.plot(y_test[:,0],label="True Temperature")

[0214]

[0215] This section constructs and trains an LSTM neural network model to predict future on-site temperature, humidity, and UV intensity. First, local environmental features are extracted from cleaned data as input, and on-site data is used as the target. A sliding time window is constructed to generate time-series training samples. After splitting the training and testing data using `train_test_split`, a model containing two LSTM layers and a fully connected layer is constructed. After training, the model can predict future on-site environmental conditions based on historical local environmental parameters, and the comparison between actual and predicted values ​​can be visualized.

[0216] Code 4. Automatic code adjustment

[0217]

[0218] "Control equipment"

[0219] GPIO.output(pin,GPIO.HIGH if action=="on"else GPIO.LOW)

[0220] # Load the LSTM prediction model model = load_model("lstm_environment_model.h5")

[0221] # Get the latest sensor data latest_data = df.iloc[-10:,:3].values.reshape(1,10,3)

[0222] predicted_temp, predicted_humidity, predicted_uv = model.predict(latest_data)[0] # Calculate environmental difference

[0223] temp_diff=predicted_temp-df["site_temp"].iloc[-1]

[0224] humidity_diff=predicted_humidity-df["site_humidity"].iloc[-1]

[0225] uv_diff=predicted_uv-df["site_uv"].iloc[-1]

[0226] #Temperature Control

[0227] if temp_diff>1:

[0228] control_device(AC_PIN,"on") # Cool down elif temp_diff<-1:

[0229] control_device(HEATER_PIN,"on") #Heating

[0230] #Humidity control

[0231] if humidity_diff <-5:

[0232] control_device(HUMIDIFIER_PIN,"on")#humidificationelif humidity_diff>5:

[0233] control_device(DEHUMIDIFIER_PIN,"on")#Dehumidification

[0234]

[0235] This code integrates model prediction and automatic control to achieve dynamic adjustment of environmental parameters. It uses a trained LSTM model to predict the most recent 10 sets of sensor data and compares the results with actual field data to obtain the difference. Based on the difference in temperature, humidity, and UV index, it determines whether to activate the air conditioner, electric heater, humidifier, dehumidifier, or UV lamp, thereby achieving intelligent adjustment of the laboratory environment. The adjustment logic triggers control devices based on set thresholds, and the switching operations of hardware devices are implemented through GPIO interfaces, ultimately constructing a closed-loop control system based on AI prediction.

Claims

1. An artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions, characterized in that, It includes a data acquisition module, a data cleaning and preprocessing module, an environmental database module, an artificial intelligence prediction module, an environmental regulation and control module, a central control and execution module, and a communication and feedback module; The data acquisition module is used to acquire raw data from the laboratory and target site environments. The raw data includes key parameters such as ultraviolet intensity, temperature, and humidity. Obtain multidimensional time-series environmental data; The data cleaning and preprocessing module performs format unification, outlier removal, missing value imputation, noise reduction, and smoothing on the collected raw data, and completes time alignment and resampling to ensure the consistency and usability of various types of data on the timeline; it outputs high-quality standardized data as the input basis for subsequent AI modeling. The environmental database module stores the standardized data output by the data cleaning and preprocessing module; The artificial intelligence prediction module makes predictions based on the environmental database module and the multidimensional time series environmental data, and obtains predicted values. The environmental regulation and control module generates a regulation and control strategy based on the difference between the predicted value and the current value, and drives the corresponding execution device to adjust the environmental parameters. The central control and execution module adopts a PLC or DCS architecture as the system control core. It is responsible for receiving input from the data acquisition module, calling the prediction results of the artificial intelligence prediction module, executing the control strategy of the environmental adjustment and control module, distributing control instructions, and driving the environmental adjustment and control module to perform environmental adjustment in real time. The communication and feedback module enables data communication and status feedback between modules, supports remote data transmission, device control command issuance, and operation status feedback, and constructs a complete closed-loop control system.

2. The artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions according to claim 1, characterized in that, The data acquisition module deploys high-precision ultraviolet intensity sensors, temperature sensors, and humidity sensors, and connects to remote data sources via meteorological department APIs to form a multi-dimensional, multi-time-period environmental data acquisition system. The sensors are deployed in key reaction areas of the laboratory and light-sensitive areas on-site to ensure that the sampled data is representative and timely.

3. The artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions according to claim 2, characterized in that, High-precision ultraviolet intensity sensors are deployed above the experimental operating table, near the sample processing area, in the center of the light-receiving area, and in the equipment exposure area to monitor the ambient ultraviolet intensity in real time. The ultraviolet intensity sensors convert the collected ultraviolet intensity signals into standard electrical signals and transmit them to the central control and execution module in real time through wired or wireless data communication modules.

4. The artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions according to claim 1, characterized in that, The environmental database module constructs a structured environmental database that stores preprocessed historical and real-time environmental parameter data, including ultraviolet intensity, spectral distribution, temperature, and humidity. This database serves as the core data source for training and predicting artificial intelligence models and supports system retrospective analysis and continuous optimization.

5. The artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions according to claim 1, characterized in that, The artificial intelligence prediction module uses recurrent neural networks (RNNs) and long short-term memory networks (LSTMs) to model multidimensional time-series environmental data; the operation steps are as follows: Step 1. Read data and set input variables; The system first reads the environmental dataset that has been cleaned and preprocessed by the data cleaning and preprocessing module, which includes the laboratory temperature, humidity and ultraviolet index as prediction targets; Step 2. Construct time series sample windows to train the time series model. The system divides the preprocessed environmental data into fixed-length sliding windows in chronological order. Data from multiple consecutive time moments within each window are used as model input, and the target value at the next time moment outside the window is used as model output. Supervised learning samples are generated in this way. Step 3. Divide all samples into training and testing sets in chronological order (e.g., 80% for training and 20% for testing), and do not shuffle the order of the samples, preserving their temporal dependency structure; Step 4. Construct the LSTM network structure: The model uses a two-layer stacked LSTM structure: The first layer contains 64 LSTM units, and the output sequence is used by the next layer; the second layer contains 32 LSTM units and outputs the final state; then there is a 16-node fully connected layer for feature transformation; the last layer is a linear output layer that predicts three target variables. Step 5. Model compilation and training: Mean squared error is used as the loss function. During the process, 50 training cycles are performed using training set samples, with 16 data points processed in each batch. The model performance is monitored in real time on the validation set to prevent overfitting. Step 6. Model Evaluation and Testing; After the model is trained, evaluate its performance on the test set and output the mean absolute error key indicator to ensure that the model has good prediction accuracy. Step 7. Prediction and Visualization: The system uses the trained model to predict the test set data and compares the prediction results with the actual values ​​graphically to intuitively judge the accuracy of the model in temperature prediction and other aspects.

6. The artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions according to claim 5, characterized in that, Predict the trends of ultraviolet intensity, spectral distribution, temperature, and humidity changes over several time steps in the future; Through feature extraction, sample construction, model training and optimization, high-precision and strong generalization prediction capabilities are achieved, providing data support for intelligent regulation. The specific implementation plan is as follows: Step 1: Extract key features related to environmental parameters from historical and real-time environmental data to enhance the model's ability to identify complex time-series patterns; Step 2: Using the sliding time window method, environmental data from multiple consecutive time moments are used as input, and the target value at the next time moment or several future time moments is used as output to construct supervised learning samples; Step 3: Construct a deep learning model using a recurrent neural network (RNN) and a long short-term memory network (LSTM), and train it using mean squared error as the loss function; Step 4: The trained model is evaluated on the validation and test sets using accuracy metrics to ensure its stability and accuracy on unseen data. The optimized model can be integrated into the control system to predict future trends in UV intensity, spectral distribution, temperature, and humidity in real time, serving as the basis for automatic adjustment decisions.

7. The artificial intelligence system for synchronous adjustment of laboratory environment and on-site conditions according to claim 1, characterized in that, The environmental regulation and control module includes: UV intensity adjustment: The intensity is adjusted by regulating the current or voltage of the UV lamp assembly (UV = k * V * I); Spectral adjustment: Data is collected using a spectral sensor and the spectrally tunable ultraviolet light source is adjusted to achieve spectral consistency; Temperature control: The temperature is dynamically adjusted through air conditioning, electric heaters, compression or absorption refrigeration systems; Humidity control: Precise humidification or dehumidification control is achieved through steam humidifiers, ultrasonic humidifiers, refrigeration dehumidifiers, and rotary dehumidifiers.

Images