Coenzyme Q10 soft capsule dynamic storage environment control system based on AI

Abstract

The invention relates to the technical field of medicine storage, and discloses an AI-based coenzyme Q10 soft capsule dynamic storage environment control system, which comprises a data acquisition module used for acquiring temperature data, humidity data, illumination intensity data and medicine state data; the data preprocessing module is used for cleaning the collected data, removing abnormal values and noise data, and performing normalization processing to obtain preprocessed data; the model prediction module is used for inputting the preprocessed data into an LSTM (Long Short Term Memory) model, predicting the change trend of environmental parameters in a future time window, simulating a degradation path of the activity of the coenzyme Q10 and outputting a prediction result; the environment regulation and control module is used for generating a control strategy according to the prediction result and realizing dynamic adjustment of environment parameters through an execution mechanism; according to the invention, the dynamic and accurate regulation and control of the storage environment are realized, the manual intervention is reduced, the labor cost is reduced, and the overall efficiency of storage management is improved.

Description

Technical Field

[0001] This invention relates to the field of pharmaceutical storage technology, specifically to an AI-based dynamic storage environment control system for coenzyme Q10 soft capsules. Background Technology

[0002] Coenzyme Q10 soft capsules are a commonly used medicine for adjuvant treatment of cardiovascular diseases, hepatitis, and cancer. However, coenzyme Q10 is chemically unstable and is easily degraded during storage due to environmental factors such as temperature, humidity, and light, leading to a decrease in the content of its active ingredients and a decline in drug quality. Traditional storage environment control is mostly statically set and cannot be dynamically and precisely adjusted according to real-time changes in the storage environment and the state of the drug itself, making it difficult to meet the strict requirements of coenzyme Q10 soft capsules for storage environment. Summary of the Invention

[0003] The purpose of this invention is to solve the above problems by designing an AI-based dynamic storage environment control system for coenzyme Q10 soft capsules.

[0004] The first aspect of this invention provides an AI-based dynamic storage environment control system for coenzyme Q10 soft capsules, the system comprising:

[0005] The data acquisition module is used to collect temperature data, humidity data, light intensity data, and drug status data;

[0006] The data preprocessing module is used to clean the collected data, remove outliers and noisy data, and normalize the data to obtain preprocessed data.

[0007] The model prediction module is used to input preprocessed data into the LSTM model, predict the changing trends of environmental parameters within the future time window, simulate the degradation pathway of coenzyme Q10 activity, and output the prediction results.

[0008] The environmental control module is used to generate control strategies based on the prediction results and to dynamically adjust environmental parameters through actuators.

[0009] Optionally, in a first implementation of the first aspect of the present invention, the data acquisition module includes:

[0010] The initialization submodule is used to initialize the temperature sensor, humidity sensor, light sensor and drug status monitoring sensor. The temperature sensor, humidity sensor and light sensor are evenly distributed in different positions in the storage area, and the drug status monitoring sensor is installed at the detection point of the drug storage rack.

[0011] The data acquisition submodule is used to generate acquisition signals according to a preset sampling frequency, trigger each sensor to start data acquisition, and obtain temperature data, humidity data, light intensity data, and drug status data.

[0012] The transmitting submodule is used to convert the collected data into digital signals and transmit them wirelessly to the data preprocessing module.

[0013] Optionally, in a second implementation of the first aspect of the present invention, the acquisition submodule includes:

[0014] Temperature sensors detect changes in ambient temperature via thermocouples and convert them into electrical signals to obtain temperature data.

[0015] Humidity sensors detect humidity by measuring changes in capacitance, thus obtaining humidity data.

[0016] A light sensor obtains light intensity data by measuring the change in the resistance of a photoresistor.

[0017] The drug status sensor acquires drug status data through near-infrared spectral scanning.

[0018] Optionally, in a third implementation of the first aspect of the present invention, the data preprocessing module includes:

[0019] The judgment submodule is used to calculate the mean and standard deviation of the collected data using the 3σ criterion algorithm, mark data that are out of range as outliers, determine the sensor failure for consecutive outliers and issue a prompt, and remove isolated outliers.

[0020] The noise reduction submodule is used to filter noise from the data after removing outliers using adaptive wavelet denoising, so as to obtain the noise-reduced data.

[0021] The normalization submodule is used to normalize the denoised data using the min-max normalization algorithm to obtain the processed data.

[0022] Optionally, in a fourth implementation of the first aspect of the present invention, the noise reduction submodule includes:

[0023] The Daubechies wavelet basis function is used to perform multi-scale decomposition on the outlier-removed data to separate high-frequency noise from low-frequency effective signals. The wavelet coefficients with concentrated noise energy are located by modulus maxima detection. The high-frequency noise is compressed using a soft threshold function. The inverse transform of the denoised wavelet coefficients is then performed to obtain the denoised data.

[0024] Optionally, in a fifth implementation of the first aspect of the present invention, the model prediction module includes:

[0025] The input submodule is used to divide the preprocessed data into a fixed-length sliding window sequence to obtain time series data, and input the time series data into the LSTM model, which contains two LSTM branch networks.

[0026] The weighted summation submodule is used by the first LSTM branch network to progressively extract time-dependent features from environmental parameters in time series data. It uses attention weights to perform weighted summation on the hidden states of the LSTM layer, and then concatenates the hidden state of the current time step with the fully connected layer to generate a context vector.

[0027] The prediction submodule is used to predict the changing trends of environmental parameters within a future time window based on context vectors through a fully connected output layer.

[0028] The dynamic adjustment submodule combines the output of the first LSTM branch network with the drug status data in the time series data and inputs it into another LSTM branch network. It models the nonlinear relationship between environmental parameters and active degradation through the attention mechanism, dynamically adjusts the attention weights, simulates the degradation path of coenzyme Q10 activity, and outputs the probability distribution of active degradation within the future time window to obtain the prediction results.

[0029] Optionally, in the sixth implementation of the first aspect of the present invention, the first LSTM branch network contains two LSTM layers, each LSTM layer contains 128 hidden units, and the LSTM layers gradually extract the time-dependent features of environmental parameters by dynamically updating the hidden state.

[0030] An attention mechanism is added to the output of the LSTM layer. The hidden state of the last layer of the LSTM layer is input into a fully connected layer to generate an attention energy vector of the same length as the time step. The energy vector is normalized into attention weights by the softmax function. The attention mechanism adopts Bahdanau attention.

[0031] Optionally, in the seventh implementation of the first aspect of the present invention, the first LSTM branch network contains two LSTM layers, each LSTM layer contains 128 hidden units, and the LSTM layers gradually extract the time-dependent features of environmental parameters by dynamically updating the hidden state.

[0032] An attention mechanism is added to the output of the LSTM layer. The hidden state of the last layer of the LSTM layer is input into a fully connected layer to generate an attention energy vector of the same length as the time step. The energy vector is normalized into attention weights by the softmax function. The attention mechanism adopts Bahdanau attention.

[0033] Optionally, in an eighth implementation of the first aspect of the present invention, the generation submodule includes:

[0034] The target difference of environmental parameters is determined, and the deviation and rate of change of the deviation between the prediction results output by the LSTM model and the target difference are used as inputs. The control quantity is output through the fuzzy rule base to generate a control strategy.

[0035] A second aspect of this invention provides an AI-based method for controlling the dynamic storage environment of coenzyme Q10 soft capsules, the method comprising the following steps:

[0036] Collect temperature data, humidity data, light intensity data, and drug status data;

[0037] The collected data is cleaned to remove outliers and noisy data, and then normalized to obtain preprocessed data.

[0038] The preprocessed data is input into the LSTM model to predict the changing trends of environmental parameters within the future time window, and to simulate the degradation pathway of coenzyme Q10 activity, and output the prediction results.

[0039] Control strategies are generated based on the prediction results, and environmental parameters are dynamically adjusted through actuators.

[0040] The technical solution provided by this invention involves collecting temperature data, humidity data, light intensity data, and drug status data; cleaning the collected data to remove outliers and noise, and normalizing the data to obtain preprocessed data; inputting the preprocessed data into an LSTM model to predict the changing trends of environmental parameters within a future time window and simulate the degradation path of coenzyme Q10 activity, outputting the prediction results; generating a control strategy based on the prediction results, and dynamically adjusting the environmental parameters through an actuator; this invention can monitor storage environment parameters and the status of coenzyme Q10 soft capsules in real time, and achieve dynamic and precise control of the storage environment through intelligent analysis and prediction using an LSTM model, so as to maximize the quality and stability of coenzyme Q10 soft capsules, reduce manual intervention, lower labor costs, avoid human error, facilitate data analysis and management decisions for managers, and improve the overall efficiency of storage management. Attached Figure Description

[0041] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0042] Figure 1 A schematic diagram of the AI-based dynamic storage environment control system for coenzyme Q10 soft capsules provided in an embodiment of the present invention;

[0043] Figure 2 This is a schematic diagram of the structure of the data acquisition module provided in an embodiment of the present invention;

[0044] Figure 3 This is a schematic diagram of the data preprocessing module provided in an embodiment of the present invention. Detailed Implementation

[0045] The terms “first,” “second,” “third,” “fourth,” etc. (if present) in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms “comprising” or “having,” and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, apparatus, product, or device that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0046] For ease of understanding, the specific process of the embodiments of the present invention is described below. Please refer to [link / reference]. Figure 1 A schematic diagram of the structure of the AI-based dynamic storage environment control system for coenzyme Q10 soft capsules provided in this embodiment of the invention. The system includes:

[0047] The data acquisition module is used to collect temperature data, humidity data, light intensity data, and drug status data;

[0048] The data preprocessing module is used to clean the collected data, remove outliers and noisy data, and normalize the data to obtain preprocessed data.

[0049] The model prediction module is used to input preprocessed data into the LSTM model, predict the changing trends of environmental parameters within the future time window, simulate the degradation pathway of coenzyme Q10 activity, and output the prediction results.

[0050] The environmental control module is used to generate control strategies based on the prediction results and to dynamically adjust environmental parameters through actuators.

[0051] Please see Figure 2 In this embodiment, the data acquisition module includes:

[0052] The initialization submodule is used to initialize the temperature sensor, humidity sensor, light sensor and drug status monitoring sensor. The temperature sensor, humidity sensor and light sensor are evenly distributed in different positions in the storage area, and the drug status monitoring sensor is installed at the detection point of the drug storage rack.

[0053] The data acquisition submodule is used to generate acquisition signals according to a preset sampling frequency, trigger each sensor to start data acquisition, and obtain temperature data, humidity data, light intensity data, and drug status data.

[0054] The transmitting submodule is used to convert the collected data into digital signals and transmit them wirelessly to the data preprocessing module.

[0055] In this embodiment, the data acquisition submodule includes: a temperature sensor that senses changes in ambient temperature through a thermocouple and converts them into electrical signals to obtain temperature data; a humidity sensor that senses humidity through changes in capacitance to obtain humidity data; a light sensor that obtains light intensity through changes in the resistance of a photoresistor to obtain light intensity data; and a drug status sensor that obtains drug status data through near-infrared spectral scanning.

[0056] Please see Figure 3 In this embodiment, the data preprocessing module includes:

[0057] The judgment submodule is used to calculate the mean and standard deviation of the collected data using the 3σ criterion algorithm, mark data that are out of range as outliers, determine the sensor failure for consecutive outliers and issue a prompt, and remove isolated outliers.

[0058] The noise reduction submodule is used to filter noise from the data after removing outliers using adaptive wavelet denoising, so as to obtain the noise-reduced data.

[0059] The normalization submodule is used to normalize the denoised data using the min-max normalization algorithm to obtain the processed data.

[0060] In this embodiment, the denoising submodule includes: using the Daubechies wavelet basis function to perform multi-scale decomposition on the data after removing outliers, separating high-frequency noise from low-frequency effective signals, locating wavelet coefficients with concentrated noise energy through modulus maxima detection, compressing high-frequency noise using a soft threshold function, and performing inverse transform on the denoised wavelet coefficients to obtain the denoised data.

[0061] In this embodiment, the model prediction module includes:

[0062] The input submodule is used to divide the preprocessed data into a fixed-length sliding window sequence to obtain time series data, and input the time series data into the LSTM model, which contains two LSTM branch networks.

[0063] The weighted summation submodule is used by the first LSTM branch network to progressively extract time-dependent features from environmental parameters in time series data. It uses attention weights to perform weighted summation on the hidden states of the LSTM layer, and then concatenates the hidden state of the current time step with the fully connected layer to generate a context vector.

[0064] The prediction submodule is used to predict the changing trends of environmental parameters within a future time window based on context vectors through a fully connected output layer.

[0065] The dynamic adjustment submodule combines the output of the first LSTM branch network with the drug status data in the time series data and inputs it into another LSTM branch network. It models the nonlinear relationship between environmental parameters and active degradation through the attention mechanism, dynamically adjusts the attention weights, simulates the degradation path of coenzyme Q10 activity, and outputs the probability distribution of active degradation within the future time window to obtain the prediction results.

[0066] In this embodiment, the first LSTM branch network contains two LSTM layers, each containing 128 hidden units. The LSTM layers gradually extract the time-dependent features of environmental parameters by dynamically updating the hidden state.

[0067] An attention mechanism is added to the output of the LSTM layer. The hidden state of the last layer of the LSTM layer is input into a fully connected layer to generate an attention energy vector of the same length as the time step. The energy vector is normalized into attention weights by the softmax function. The attention mechanism adopts Bahdanau attention.

[0068] In this embodiment, the environmental control module includes:

[0069] The generation submodule is used to obtain the prediction results output by the LSTM model, generate control strategies using fuzzy control algorithms, convert the control strategies into control instructions for the actuators, and send them to the corresponding devices through the PLC controller.

[0070] The trigger submodule is used to collect environmental parameters in real time after execution. If the environmental parameters do not reach the preset threshold, the control strategy generation will be retried.

[0071] In this embodiment, the generation submodule includes: determining the target difference of environmental parameters, taking the deviation and rate of change of the deviation between the prediction result output by the LSTM model and the target difference as input, outputting the control quantity through the fuzzy rule base, and generating a control strategy.

[0072] In this embodiment, a sensor deployment plan is formulated based on the spatial structure of the storage area, such as the warehouse length, width, height, and shelf layout: temperature and humidity sensors are evenly distributed according to the "regional gridding" principle, with monitoring points set at the four corners, center, and upper, middle, and lower shelves of the storage area, ensuring that each 10-square-meter area is covered by at least one set of temperature and humidity sensors; light sensors are mainly deployed near windows, under lighting equipment, and on the backlit surfaces of shelves to form a light intensity monitoring gradient; drug status monitoring sensors are installed at each detection point on each layer of the drug storage shelf, maintaining a fixed distance of 30-50cm from the soft capsule storage location to avoid the influence of shelf obstruction during detection; after completing the physical department After deployment, the submodule automatically initializes the parameters of each sensor: it uniformly sets the device ID to distinguish different monitoring points, configures the sampling trigger mode as timed-wake-up, and performs baseline calibration on the sensors using standard calibration tools such as constant temperature and humidity chambers and standard light source instruments. The temperature sensor uses 0℃, 25℃, and 50℃ as calibration points, the humidity sensor uses 30%RH, 60%RH, and 90%RH as calibration points, the light sensor uses 100lux, 500lux, and 1000lux as calibration points, and the drug status monitoring sensor performs spectral baseline calibration using a known activity coenzyme Q10 soft capsule sample. Finally, an initialization report is generated, recording the deployment location, calibration error, and communication connection status of each sensor.

[0073] The data acquisition submodule has a built-in sampling frequency adjustment unit, which can flexibly set the acquisition interval according to the stability of the storage environment. The default initial frequency is once every 5 minutes, and it can automatically switch to once every 2 minutes if the environment fluctuates greatly. When the preset acquisition time point is reached, the submodule generates a standardized acquisition signal, which includes the device ID, acquisition timestamp and trigger command, and sends it synchronously to each sensor through a wired connection: After receiving the signal, the temperature sensor activates the internal thermocouple sensing element to convert the thermoelectric potential corresponding to the ambient temperature into an analog electrical signal; the humidity sensor senses the water vapor content in the air through a capacitive sensing film and outputs a voltage signal corresponding to the humidity; the photoresistor of the light sensor adjusts its resistance according to the change in light intensity to form a change in current signal; and the drug status monitoring sensor activates the near-infrared spectral scanning module to perform spectral acquisition on the coenzyme Q10 soft capsules on the storage rack every 3 seconds to capture the absorption peak signal at a specific wavelength.

[0074] After receiving the raw data from each sensor, the module first performs format conversion: converting the analog electrical signals of temperature, humidity, and light intensity into decimal digital signals, and converting the spectral signals of the drug's state into a feature value array, and adding a unique identifier to each data set. After conversion, the submodule establishes a communication link through the LoRa wireless transmission module and sends data using a packet-verification mechanism. When the amount of a single data set exceeds the transmission packet limit, it is automatically split into multiple data packets and marked with sequence numbers, with a CRC checksum appended to the end of each data packet. During transmission, the transmission status is monitored in real time: if a successful reception response is received from the data preprocessing module, the next data transmission process begins; if a transmission timeout or verification failure occurs, a retransmission mechanism is automatically triggered, and a transmission exception log is recorded.

[0075] In this embodiment, the core component of the temperature sensor is a thermocouple, which consists of a closed loop composed of conductors made of two different materials. When the storage environment temperature changes, the two junctions of the thermocouple generate a thermoelectric potential due to the temperature difference, and the magnitude of this thermoelectric potential is proportional to the temperature difference. The signal processing circuit inside the temperature sensor captures this thermoelectric potential signal in real time and converts it into a corresponding temperature electrical signal through a pre-calibrated conversion coefficient. For example, when the storage environment temperature rises from 23°C to 25°C, the thermoelectric potential generated by the thermocouple changes accordingly. The signal processing circuit responds quickly to this change and converts it into an electrical signal corresponding to 25°C, ultimately generating temperature data in °C. The entire conversion process is completed within 0.1 seconds, ensuring the real-time nature of the temperature data.

[0076] The humidity sensor employs a capacitive sensing structure. Its sensing element consists of two metal electrodes and a polymer film sandwiched between them. This film is hygroscopic, and its dielectric constant changes with variations in the ambient humidity. When the moisture content in the storage environment changes, the polymer film absorbs or releases moisture, causing a corresponding change in the dielectric constant and consequently, the capacitance. When humidity increases, the film absorbs more water, increasing the dielectric constant and thus the capacitance. Conversely, when humidity decreases, the film releases moisture, decreasing the dielectric constant and the capacitance. The humidity sensor's detection circuit converts the capacitance change into a voltage signal, which is then converted into a corresponding humidity value using an internal calibration curve. This yields humidity data in %RH units, and the conversion process accurately reflects humidity changes within the 0-100%RH range.

[0077] A photoresistor in a light sensor is a light-sensitive semiconductor element whose resistance changes significantly with light intensity. In a storage environment, when light intensity increases, the photoresistor absorbs more photons, increasing the internal carrier concentration and enhancing conductivity, thus decreasing resistance. Conversely, when light intensity decreases, the number of absorbed photons decreases, reducing carrier concentration and weakening conductivity, leading to an increase in resistance. The light sensor circuit provides a stable operating voltage to the photoresistor and indirectly reflects the resistance change by measuring the current flowing through it. Based on a pre-established table of resistance and light intensity, the resistance change is converted into a specific light intensity value, ultimately yielding light intensity data in lux, accurately capturing changes from weak to strong light.

[0078] The drug status sensor is equipped with a near-infrared spectroscopy scanning module. During operation, it emits a specific range of near-infrared light towards the coenzyme Q10 soft capsules on the storage rack. Under near-infrared light irradiation, the active ingredient in the coenzyme Q10 soft capsules exhibits characteristic absorption of specific wavelengths of light, forming a unique absorption spectrum. The sensor's spectral detector captures the reflected or transmitted spectral signal, converts it into an electrical signal, and transmits it to the internal data processing unit. This unit preprocesses the spectral signal, extracting characteristic peaks related to the coenzyme Q10 active ingredient content and the integrity of the soft capsule's appearance, and converts this characteristic information into quantifiable drug status data. For example, when the active ingredient content of the soft capsule decreases, the absorption intensity of the corresponding characteristic peak changes systematically. The sensor can identify this change to generate corresponding drug status data, enabling real-time monitoring of the drug's status.

[0079] In this embodiment, after receiving preprocessed environmental parameter data such as temperature, humidity, and light, as well as drug status data, the data is divided into sliding window sequences according to a preset fixed length, such as 30 minutes per time step, with a sliding window length of 10 time steps. During the division process, the windows slide sequentially at intervals of each time step to ensure reasonable data overlap between adjacent windows, thereby preserving the temporal continuity of the data and ultimately obtaining time series data containing time dimension information. Subsequently, this submodule inputs the generated time series data into an LSTM model, which contains two interconnected LSTM branch networks, responsible for feature extraction and prediction related to environmental parameters and simulation and prediction related to drug activity degradation, respectively.

[0080] The first LSTM branch network begins processing environmental parameters from the input time-series data. This branch network consists of two LSTM layers, each with 128 hidden units. During data processing, the LSTM layers dynamically update the hidden state based on the current input environmental parameters and the previous hidden state, through dynamic adjustments to the input gate, forget gate, and output gate. This process gradually extracts the time-dependent features of environmental parameters at different time steps, such as the periodic variation of temperature throughout the day and the lag characteristics of humidity changes with temperature. A Bahdanau attention mechanism is added to the output of the LSTM layers, first processing the input data... The hidden state of the last layer is input into a fully connected layer. After a linear transformation, an attention energy vector of the same length as the time step is generated. This vector is used to measure the importance of the hidden state at each time step to the current prediction. Then, the energy vector is normalized into attention weights by the softmax function, so that the sum of the weights is 1. After that, the attention weights are used to perform a weighted summation of all the hidden states of the LSTM layer to obtain a feature vector containing key time information. After the dimension of this feature vector is adjusted by passing it through a fully connected layer, it is concatenated with the hidden state of the current time step to finally generate a context vector that integrates global time features and features of the current time step.

[0081] After receiving the context vector generated by the weighted summation submodule, it is input into the fully connected output layer. The fully connected output layer contains multiple neurons, the number of which corresponds to the number of time steps within the future time window to be predicted. For example, to predict the environmental parameter change trends in the next 1 hour, 2 hours, and 4 hours, 3 output neurons are set accordingly. This layer performs nonlinear transformation and integration on the context vector, mapping the extracted time-dependent features to the prediction space of environmental parameters. During the output process, the physical meaning of the environmental parameters is combined to constrain the output results. For example, the temperature prediction value must be within a reasonable physical range. Finally, the specific change trends of environmental parameters such as temperature, humidity, and light in different future time windows are output. For example, the temperature will rise from 23℃ to 24.5℃ in the next 1 hour, and the humidity will drop from 50%RH to 45%RH in the next 2 hours.

[0082] The first LSTM branch network outputs time-dependent features of environmental parameters and drug state data from the time-series data, such as the content of coenzyme Q10 active ingredient and the appearance of soft capsules. These two types of data are then fused to form comprehensive data that includes both environmental factors and the drug's own state. This comprehensive data is then input into another LSTM branch network, which processes the data through an internal gating mechanism and introduces an attention mechanism to model the nonlinear relationship between environmental parameters and active degradation. During processing, this submodule dynamically adjusts the attention weights based on changes in the drug state data. When the drug's state changes slightly, the attention weight of the environmental parameters at the corresponding time step is increased; when the drug's state is stable, the weights of each time step are evenly distributed. Through this dynamic adjustment, the key environmental factors and timing of action affecting the drug's activity degradation are captured more accurately, thereby simulating the degradation pathway of coenzyme Q10 activity. Finally, the probability distribution of activity degradation within future time windows, such as 1 day, 3 days, and 7 days, is output through a fully connected layer and a softmax function. For example, the probability of the activity degradation rate being 0-1% within the next 3 days is 80%, and the probability of 1%-2% is 15%, etc., which is used as the final prediction result.

[0083] In this embodiment, predicted values ​​of environmental parameters such as temperature, humidity, and light intensity within a future time window are extracted from the output of the LSTM model. Simultaneously, the system's preset target ranges for environmental parameters are invoked. When determining these target ranges, the storage characteristics of the coenzyme Q10 soft capsules and historical data optimization results are considered, and differentiated standards are established for different parameters: the target temperature range is set at 20-25℃, within which the coenzyme Q10 degradation rate is less than 0.1% / day; the target humidity range is set at 40-60%RH to avoid excessive humidity causing the soft capsules to absorb moisture and stick together, or excessively low humidity causing them to stick together. To prevent dryness and cracking, the target light range is set to <50 lux to reduce the damage of strong light to the Coenzyme Q10 molecular structure. After obtaining the predicted values ​​of temperature, humidity, and light for the next 1, 2, and 4 hours from the LSTM model output, the deviation of each parameter prediction value from the corresponding target range is calculated. If the predicted temperature is 27℃, the deviation from the upper limit of the target of 25℃ is +2℃; if the predicted humidity is 35%RH, the deviation from the lower limit of the target of 40%RH is -5%RH. Simultaneously, the deviation change rate is calculated, which is the ratio of the difference between two consecutive prediction deviations to the time interval, as shown above. The temperature deviation is +1℃, the current temperature is +2℃, and the interval is 1 hour, so the rate of change is +1℃ / hour. Subsequently, the submodule initiates a fuzzy control algorithm, using the deviation and its rate of change as input variables. It converts these into fuzzy quantities such as temperature deviation "small positive," "medium positive," and "large positive" using membership functions such as triangular membership functions. Then, it calls the built-in fuzzy rule library, which includes preset rules such as "if the temperature deviation is large positive and the rate of change is small positive, then adjust the cooling power to the maximum" and "if the humidity deviation is large negative and the rate of change is small negative, then adjust the humidification power to the maximum," to infer the fuzzy quantities and output the result. Fuzzy control quantities, such as "high" cooling power and "medium" humidification time, are used. Finally, through defuzzification, the fuzzy control quantities are converted into specific quantitative control parameters, such as 80% cooling power and 10 minutes humidification time, generating a control strategy that includes equipment operation type, operating parameters, and duration. After the strategy is generated, the submodule automatically converts it into control instructions that the actuator can recognize, such as "start cooling, power 80%" for the air conditioner and "start, run for 10 minutes" for the humidifier, and sends them to the corresponding devices through the PLC controller's communication interface, while recording the instruction sending time and content.

[0084] The trigger submodule maintains real-time communication with temperature, humidity, and light sensors in the storage environment. After the actuator responds to control commands, it collects current environmental parameters every 3 minutes. After collection, the submodule compares the actual parameters with preset thresholds (i.e., the upper and lower limits of the target range for environmental parameters): if the temperature is stable at 22-24℃, humidity at 45-55%RH, and light intensity <40 lux, it is determined that the threshold has been reached, and the next cycle of monitoring begins; if any parameters are not up to standard, such as the temperature remaining at 26℃, exceeding the 25℃ upper limit, a trigger signal is immediately generated, and the cycle restarts. The newly activated generation submodule generates a control strategy. When triggered, the submodule simultaneously extracts the current value of the parameter that did not meet the standard and the predicted value before the control command is executed. It calculates the deviation of the actual control effect. For example, if the expected temperature drop after cooling is 24℃, but the actual temperature is 26℃, the deviation is +2℃. This deviation is then passed to the generation submodule as an additional parameter to assist in optimizing the new round of control strategy, such as appropriately increasing the cooling power. At the same time, the submodule records the number of triggers and the reasons for failure. If the same parameter fails to meet the standard after three consecutive triggers, it automatically sends a device maintenance prompt to the system management terminal to avoid ineffective control cycles.

[0085] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. An AI-based dynamic storage environment control system for coenzyme Q10 soft capsules, characterized in that, The system includes: The data acquisition module is used to collect temperature data, humidity data, light intensity data, and drug status data; The data preprocessing module is used to clean the collected data, remove outliers and noisy data, and normalize the data to obtain preprocessed data. The model prediction module is used to input preprocessed data into the LSTM model, predict the changing trends of environmental parameters within the future time window, simulate the degradation pathway of coenzyme Q10 activity, and output the prediction results. The environmental control module is used to generate control strategies based on the prediction results and to dynamically adjust environmental parameters through actuators.

2. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 1, characterized in that, The data acquisition module includes: The initialization submodule is used to initialize the temperature sensor, humidity sensor, light sensor and drug status monitoring sensor. The temperature sensor, humidity sensor and light sensor are evenly distributed in different positions in the storage area, and the drug status monitoring sensor is installed at the detection point of the drug storage rack. The data acquisition submodule is used to generate acquisition signals at a preset sampling frequency, trigger each sensor to start data acquisition, and obtain temperature data, humidity data, light intensity data, and drug status data. The transmitting submodule is used to convert the collected data into digital signals and transmit them wirelessly to the data preprocessing module.

3. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 2, characterized in that, The acquisition submodule includes: Temperature sensors detect changes in ambient temperature via thermocouples and convert them into electrical signals to obtain temperature data. Humidity sensors detect humidity by measuring changes in capacitance, thus obtaining humidity data. A light sensor obtains light intensity data by measuring the change in the resistance of a photoresistor. The drug status sensor acquires drug status data through near-infrared spectral scanning.

4. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 1, characterized in that, The data preprocessing module includes: The judgment submodule is used to calculate the mean and standard deviation of the collected data using the 3σ criterion algorithm, mark data that are out of range as outliers, determine the sensor failure for consecutive outliers and issue a prompt, and remove isolated outliers. The noise reduction submodule is used to filter noise from the data after removing outliers using adaptive wavelet denoising, so as to obtain the noise-reduced data. The normalization submodule is used to normalize the denoised data using the min-max normalization algorithm to obtain the processed data.

5. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 4, characterized in that, The noise reduction submodule includes: The Daubechies wavelet basis function is used to perform multi-scale decomposition on the outlier-removed data to separate high-frequency noise from low-frequency effective signals. The wavelet coefficients with concentrated noise energy are located by modulus maxima detection. The high-frequency noise is compressed using a soft threshold function. The inverse transform of the denoised wavelet coefficients is then performed to obtain the denoised data.

6. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 1, characterized in that, The model prediction module includes: The input submodule is used to divide the preprocessed data into a fixed-length sliding window sequence to obtain time series data, and input the time series data into the LSTM model, which contains two LSTM branch networks. The weighted summation submodule is used by the first LSTM branch network to progressively extract time-dependent features from environmental parameters in time series data. It uses attention weights to perform weighted summation on the hidden states of the LSTM layer, and then concatenates the hidden state of the current time step with the fully connected layer to generate a context vector. The prediction submodule is used to predict the changing trends of environmental parameters within a future time window based on context vectors through a fully connected output layer. The dynamic adjustment submodule combines the output of the first LSTM branch network with the drug status data in the time series data and inputs it into another LSTM branch network. It models the nonlinear relationship between environmental parameters and active degradation through the attention mechanism, dynamically adjusts the attention weights, simulates the degradation path of coenzyme Q10 activity, and outputs the probability distribution of active degradation within the future time window to obtain the prediction results.

7. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 6, characterized in that, The first LSTM branch network contains two LSTM layers, each containing 128 hidden units. The LSTM layers extract the time-dependent features of environmental parameters by dynamically updating the hidden states. An attention mechanism is added to the output of the LSTM layer. The hidden state of the last layer of the LSTM layer is input into a fully connected layer to generate an attention energy vector of the same length as the time step. The energy vector is normalized into attention weights by the softmax function. The attention mechanism adopts Bahdanau attention.

8. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 1, characterized in that, The environmental control module includes: The generation submodule is used to obtain the prediction results output by the LSTM model, generate control strategies using fuzzy control algorithms, convert the control strategies into control instructions for the actuators, and send them to the corresponding devices through the PLC controller. The trigger submodule is used to collect environmental parameters in real time after execution. If the environmental parameters do not reach the preset threshold, the control strategy generation will be retried.

9. The AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 8, characterized in that, The generation submodule includes: The target difference of environmental parameters is determined, and the deviation and rate of change of the deviation between the prediction results output by the LSTM model and the target difference are used as inputs. The control quantity is output through the fuzzy rule base to generate a control strategy.

10. A method for implementing an AI-based dynamic storage environment control system for coenzyme Q10 soft capsules as described in claim 1, characterized in that, The method includes the following steps: Collect temperature data, humidity data, light intensity data, and drug status data; The collected data is cleaned to remove outliers and noisy data, and then normalized to obtain preprocessed data. The preprocessed data is input into the LSTM model to predict the changing trends of environmental parameters within the future time window, and to simulate the degradation pathway of coenzyme Q10 activity, and output the prediction results. Control strategies are generated based on the prediction results, and environmental parameters are dynamically adjusted through actuators.

Images