Deep learning-based medicine cold chain packing material scheduling method and system

Abstract

The invention discloses a medicine cold chain packing material scheduling method and system based on deep learning. The method comprises the following steps: S1, monitoring key parameters in a cold chain logistics process in real time; s2, inputting the monitoring data into a deep learning model for analysis; s3, dynamically adjusting a packing material scheduling strategy according to an analysis result; s4, predicting future demands and transportation environments by using a deep learning model, collecting historical sales data, market trends, seasonal changes and special event information, calculating scores of various packaging material configuration schemes through a weighted calculation mode, and taking the packaging material configuration scheme with the highest score as an optimal scheduling method; according to the method, the key parameters in the cold-chain logistics process are monitored in real time, the deep learning model is trained in combination with historical data, and accurate prediction of environment changes and dynamic optimization of the packing material scheduling strategy are achieved.

Description

Technical Field

[0001] The present invention relates to the field of material scheduling, and more specifically, to a deep learning-based method for scheduling pharmaceutical cold chain packaging materials. The present invention also relates to a deep learning-based system for scheduling pharmaceutical cold chain packaging materials. Background Art

[0002] With the development of the global pharmaceutical industry, the transportation and storage requirements of drugs have become increasingly stringent, especially in the field of cold chain logistics. Pharmaceutical products are extremely sensitive to temperature and humidity environmental parameters. Any slight environmental change may cause the quality of drugs to deteriorate or even become ineffective. Therefore, it is crucial to ensure the stability and safety of pharmaceutical products during transportation and storage. Traditional cold chain logistics management mainly relies on manual monitoring and experience judgment. This method has many shortcomings: poor real-time performance: manual monitoring cannot achieve 24-hour uninterrupted monitoring, and the response speed is slow; low accuracy: judgments based on experience are easily affected by subjective factors, resulting in inaccurate decisions; low efficiency: manual adjustment of packaging materials and scheduling strategies is time-consuming and labor-intensive, and it is difficult to cope with large-scale and complex logistics needs; insufficient data utilization: a large amount of historical data is not fully utilized and cannot provide strong support for decision-making;

[0003] To address these issues, a number of cold chain monitoring systems based on the Internet of Things (IoT) have emerged in recent years. These systems install sensors at key locations in refrigerated trucks and warehouses to enable real-time collection and monitoring of environmental parameters. However, these systems still face the following challenges: Limited data analysis capabilities: While they can collect data in real time, they lack effective data analysis methods, making it difficult to extract valuable information from massive amounts of data; Insufficient predictive capabilities: Most existing monitoring systems can only perform real-time monitoring and lack the ability to predict future demand, making it impossible to prepare for scheduling in advance; Low intelligence: The system's automation and intelligence levels are low, requiring more manual intervention, which affects overall efficiency;

[0004] Therefore, we propose a pharmaceutical cold chain packaging material scheduling method and system based on deep learning to effectively solve the above problems. Summary of the Invention

[0005] The purpose of this invention is to address the shortcomings of the existing technology and propose a pharmaceutical cold chain packaging material scheduling method and system based on deep learning. By real-time monitoring of key parameters in the cold chain logistics process and combining historical data to train a deep learning model, it can achieve accurate prediction of environmental changes and dynamic optimization of packaging material scheduling strategies.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] The pharmaceutical cold chain packaging material scheduling method based on deep learning includes the following steps:

[0008] S1. Real-time monitoring of key parameters in the cold chain logistics process. Temperature, humidity, pressure, and vibration sensors are installed in refrigerated trucks, refrigerated warehouses, and packaging boxes. Environmental data is regularly collected and transmitted to a central monitoring system via wireless communication.

[0009] S2. Input monitoring data into the deep learning model for analysis. The central monitoring system establishes a deep learning model based on the received data and uses historical data to train the deep learning model so that it can accurately predict the impact of environmental changes on pharmaceutical product quality.

[0010] S3. Dynamically adjust the packaging material scheduling strategy based on the analysis results. Input the real-time monitoring data into the trained deep learning model for real-time analysis and evaluation. Based on the analysis results of the deep learning model, automatically generate packaging material scheduling instructions.

[0011] S4. Use deep learning models to predict future demand and transportation environment, collect historical sales data, market trends, seasonal changes, special event information, and predict market demand and transportation environment in the future. Calculate the scores of various packaging material configuration plans through weighted calculations, and use the one with the highest score as the optimal scheduling method.

[0012] Preferably, in step S1, after receiving the real-time data from each sensor, the central monitoring system performs preliminary processing and cleaning, including data verification, outlier detection, and data format conversion, to ensure the quality of the data input into the deep learning model;

[0013] The processed data is stored in a central database to facilitate subsequent data analysis and query. The data storage system is highly reliable and scalable, supporting real-time access and historical data backtracking.

[0014] Preferably, in step S2, the monitored temperature, humidity, pressure and vibration environment data are first normalized to meet the input requirements of the model. The data collected by n sensors, each sensor collects data with a dimension of d, and the input data matrix expression is:

[0015] X∈R n×d， Where X represents the data matrix and R represents the data set;

[0016] In the model, the convolution kernel size of layer l is k l ×k l , the number of convolution kernels is m l , then the output feature size of this layer is:

[0017]

[0018] Where s lis the step size, n and d are the height and width of the input feature map respectively, k l is the size of the convolution kernel.

[0019] Preferably, the central monitoring system establishes a deep learning model based on the received data, which is expressed as:

[0020]

[0021] Where, is the impact score of environmental changes on pharmaceutical product quality predicted by the model, X is the input environmental monitoring data, θ is the model parameter, and f represents the mapping relationship of the model;

[0022] The model parameters θ are adjusted to minimize the loss function L, which is expressed as:

[0023]

[0024] Where y is the impact score of actual environmental changes on the quality of pharmaceutical products, θ * is the optimal set of model parameters, and L(f(X; θ), y) represents the difference between the predicted result f(X; θ) using the model parameters θ and the input data X and the actual impact score y of environmental changes on the quality of pharmaceutical products.

[0025] Preferably, step S3 is specifically as follows:

[0026] S31. Generate packaging material scheduling instructions based on the analysis results of the deep learning model. The instructions may include adjusting the type, quantity, or delivery time of packaging materials to adapt to current and future environmental conditions to ensure that the quality of pharmaceutical products is not affected;

[0027] S32. The generated packaging material dispatch instructions are sent to the relevant execution departments, including the warehouse management department. The warehouse management department performs corresponding operations according to the instructions, including adjusting inventory, preparing shipments, or changing delivery routes.

[0028] S33. During the execution process, the system will continuously collect feedback information, including actual environmental data, packaging material usage, and the quality status of pharmaceutical products. This information will be used to further optimize the training of deep learning models and the formulation of scheduling strategies to improve the intelligence level and efficiency of the entire system.

[0029] Preferably, the step S31 further includes manually reviewing the generated scheduling instructions, specifically:

[0030] Preliminary assessment and preparation. Before executing the instructions, the warehouse management department needs to conduct preliminary assessment and preparation, including checking the existing inventory, confirming the availability of the required packaging materials, and preparing logistics resources;

[0031] Implement manual fine-tuning. Based on the actual work situation, the warehouse management department will conduct manual fine-tuning. When it is found that the actual demand for certain packaging materials is different from the instructions, adjustments will be made based on the actual situation to ensure that the scheduling strategy is more scientific and reasonable.

[0032] Preferably, in step S4, a deep learning model is used to predict future demand and transportation environment, collect historical sales data, market trends, seasonal changes, special event information, and predict market demand and transportation environment in the future, specifically:

[0033] Collect market feedback data, including historical sales volume, historical sales revenue, customer feedback, market research reports, and special event reports; special event reports include data on holidays, promotions, and natural disasters;

[0034] A long short-term memory network model is established and trained based on the data. The training divides the collected historical data into a training set and a test set, with the training set accounting for 70%-80% of the total data and the test set accounting for 20%-30%;

[0035] The currently known market data and features are input into the trained model. Based on the input data, the model calculates the predicted values ​​of market demand and transportation environment in the future.

[0036] Preferably, the scores of various packaging material configuration schemes are calculated by weighted calculation in step S4, and the one with the highest score is used as the optimal scheduling method, specifically:

[0037] The weighted items include product shelf life, with a weight of ω SL ; Storage temperature, weight is ω ST : The distance between the cold chain warehouse and the destination, with a weight of ω D : The complexity of the transportation route, with a weight of ω RC ;

[0038] For each packaging material configuration scheme i, define a comprehensive score S i , the expression is:

[0039] S i =w SL ·SL i +w ST ·ST i +w D (1-D i )+w RC ·(1-RC i );

[0040] Where S Li , S Ti , Di , RC i Respectively represent the scores of the i-th packaging material configuration scheme in terms of product shelf life, storage temperature, distance and route complexity. The scores are standardized values ​​ranging from 0 to 1, where 1 represents the best, and ω SL +ω ST +ω D +ω RC =1.

[0041] Preferably, the score value is normalized by taking any variable x, i.e. the actual value, and its minimum value is x min , the maximum value is x max , the minimum-maximum normalized value x′ is expressed as:

[0042]

[0043] By linearly converting the original data to between 0 and 1, the influence of different dimensions and magnitudes is eliminated, making the data comparable.

[0044] A pharmaceutical cold chain packaging material scheduling system based on deep learning, which is used to implement the above method, includes:

[0045] The data collection layer installs temperature, humidity, pressure, and vibration sensors at key locations in refrigerated trucks, refrigerated warehouses, and packaging boxes, and transmits sensor data to the central monitoring system in real time via wireless communication technology;

[0046] The data processing and storage layer performs preliminary processing and cleaning on the received data, including data verification, outlier detection, and data format conversion. The processed data is stored in a highly reliable and scalable central database, supporting real-time access and historical data backtracking.

[0047] The model training and analysis layer builds a deep learning model based on monitoring data, uses historical data for training, inputs real-time monitoring data into the trained deep learning model, performs real-time analysis and evaluation, uses the deep learning model to predict future demand and transportation environment, and collects and analyzes relevant data;

[0048] The scheduling strategy generation layer automatically generates packaging material scheduling instructions based on the analysis results of the deep learning model, including adjusting the type, quantity or delivery time of packaging materials. It calculates the scores of various packaging material configuration plans through weighted calculation and selects the one with the highest score as the optimal scheduling method.

[0049] The execution and feedback layer sends the generated packaging material scheduling instructions to the execution department for corresponding operations. During the execution process, feedback information is continuously collected, including actual environmental data, packaging material usage, and the quality status of pharmaceutical products. The collected feedback information is used to further optimize the training of deep learning models and the formulation of scheduling strategies.

[0050] Technical effects and advantages of the present invention: The pharmaceutical cold chain packaging material scheduling method and system based on deep learning provided by the present invention has the following advantages compared with traditional technologies:

[0051] By real-time monitoring and predicting environmental changes, packaging materials and scheduling strategies can be adjusted promptly to ensure the stability and safety of pharmaceutical products during transportation and storage, reduce the risk of drug quality degradation or failure due to environmental changes, and improve the overall quality level of drugs. At the same time, automated packaging material scheduling instruction generation and execution reduces the need for manual intervention, improves operational efficiency and accuracy, and dynamically adjusts the type, quantity, or delivery time of packaging materials, optimizes the allocation of logistics resources, and reduces logistics costs.

[0052] Secondly, the use of deep learning models for data analysis and prediction provides a scientific basis for decision-making, avoiding the subjective errors caused by relying on experience and judgment. Through the training of historical data and the input of real-time data, the model can continuously learn and optimize, improving the accuracy and reliability of predictions. The system integrates multiple links of data collection, processing, analysis and execution feedback, forming a closed-loop intelligent management system, continuously collecting feedback information and using it for model optimization, so that the system can continuously improve itself and improve the overall intelligence level.

[0053] Finally, the deep learning model has strong generalization capabilities and can adapt to different transportation environments and conditions, improving the adaptability and robustness of the system. By predicting future demand and transportation environment, it can make scheduling preparations in advance, effectively respond to market fluctuations and emergencies, make full use of historical data and real-time data, and explore potential correlations and patterns. It provides strong support for the formulation of packaging material scheduling strategies, selects the optimal packaging material configuration plan through weighted calculation, and achieves optimal resource allocation. BRIEF DESCRIPTION OF THE DRAWINGS

[0054] Figure 1 This is a flow chart of the pharmaceutical cold chain packaging material scheduling method based on deep learning of the present invention;

[0055] Figure 2 This is the architecture diagram of the pharmaceutical cold chain packaging material scheduling system based on deep learning in the present invention. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solutions and advantages of the present invention more clear, the present invention is further described in detail below with reference to specific embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0057] The present invention provides a pharmaceutical cold chain packaging material scheduling method and system based on deep learning. By real-time monitoring of key parameters in the cold chain logistics process and combining historical data to train a deep learning model, accurate prediction of environmental changes and dynamic optimization of packaging material scheduling strategies can be achieved.

[0058] like Figure 1 As shown in the figure, the pharmaceutical cold chain packaging material scheduling method based on deep learning includes the following steps:

[0059] S1. Real-time monitoring of key parameters in the cold chain logistics process. Temperature, humidity, pressure, and vibration sensors are installed in refrigerated trucks, refrigerated warehouses, and packaging boxes. Environmental data is regularly collected and transmitted to a central monitoring system via wireless communication. In step S1, after receiving the real-time data from each sensor, the central monitoring system performs preliminary processing and cleaning, including data verification, outlier detection, and data format conversion, to ensure the data quality of the input deep learning model.

[0060] The processed data is stored in a central database to facilitate subsequent data analysis and query. The data storage system is highly reliable and scalable, supporting real-time access and historical data backtracking.

[0061] S2. Input the monitoring data into the deep learning model for analysis. The central monitoring system establishes a deep learning model based on the received data and uses historical data to train the deep learning model so that it can accurately predict the impact of environmental changes on the quality of pharmaceutical products. In step S2, the monitored temperature, humidity, pressure, and vibration environmental data are first normalized to meet the input requirements of the model. The data collected by n sensors has a dimension of d for each sensor. The input data matrix expression is:

[0062] X∈R n×d， Where X represents the data matrix and R represents the data set;

[0063] In the model, the convolution kernel size of layer l is k l ×k l , the number of convolution kernels is m l , then the output feature size of this layer is:

[0064]

[0065] Where s l is the step size, n and d are the height and width of the input feature map respectively, k l is the size of the convolution kernel; the central monitoring system establishes a deep learning model based on the received data, which is expressed as:

[0066] Where, is the impact score of environmental changes on pharmaceutical product quality predicted by the model, X is the input environmental monitoring data, θ is the model parameter, and f represents the mapping relationship of the model;

[0067] The model parameters θ are adjusted to minimize the loss function L, which is expressed as:

[0068]

[0069] Where y is the impact score of actual environmental changes on the quality of pharmaceutical products, θ * is the optimal set of model parameters, and L(f(X; θ), y) represents the difference between the predicted result f(X; θ) using the model parameters θ and the input data X and the actual impact score y of environmental changes on the quality of pharmaceutical products.

[0070] S3. Dynamically adjust the packaging material scheduling strategy based on the analysis results. Input the real-time monitoring data into the trained deep learning model for real-time analysis and evaluation. Automatically generate packaging material scheduling instructions based on the analysis results of the deep learning model. Step S3 is specifically as follows:

[0071] S31. Generate packaging material dispatch instructions based on the analysis results of the deep learning model. The instructions may include adjusting the type, quantity, or delivery time of packaging materials to adapt to current and future environmental conditions to ensure that the quality of pharmaceutical products is not affected. Step S31 also includes manual review of the generated dispatch instructions, specifically: preliminary assessment and preparation. Before executing the instructions, the warehouse management department must conduct preliminary assessment and preparation, including checking existing inventory, confirming the availability of required packaging materials, and preparing logistics resources.

[0072] Implement manual fine-tuning. Based on the actual working situation, the warehouse management department will conduct manual fine-tuning. When it is found that the actual demand for certain packaging materials is different from the instructions, adjustments will be made based on the actual situation to ensure that the scheduling strategy is more scientific and reasonable.

[0073] S32. The generated packaging material dispatch instructions are sent to the relevant execution departments, including the warehouse management department. The warehouse management department performs corresponding operations according to the instructions, including adjusting inventory, preparing shipments, or changing delivery routes.

[0074] S33. During the execution process, the system will continuously collect feedback information, including actual environmental data, packaging material usage, and the quality status of pharmaceutical products. This information will be used to further optimize the training of deep learning models and the formulation of scheduling strategies to improve the intelligence level and efficiency of the entire system.

[0075] S4. Use deep learning models to predict future demand and transportation conditions. Collect historical sales data, market trends, seasonal changes, and special event information to forecast market demand and transportation conditions for a period of time. Calculate scores for various packaging material configuration options using a weighted calculation method, and select the one with the highest score as the optimal scheduling method.

[0076] In step S4, a deep learning model is used to predict future demand and transportation environment. Historical sales data, market trends, seasonal changes, and special event information are collected to predict market demand and transportation environment for a period of time in the future. Specifically,

[0077] Collect market feedback data, including historical sales volume, historical sales revenue, customer feedback, market research reports, and special event reports; special event reports include data on holidays, promotions, and natural disasters;

[0078] Build and train a long short-term memory network model based on the data. The collected historical data is divided into a training set and a test set. The training set accounts for 70%-80% of the total data, and the test set accounts for 20%-30%;

[0079] Input the currently known market data and features into the trained model. Based on the input data, the model calculates the predicted market demand and transportation environment for a period of time in the future.

[0080] In step S4, the scores of various packaging material configuration schemes are calculated by weighted calculation, and the one with the highest score is taken as the optimal scheduling method. Specifically, the weighted items include the shelf life of the product, with a weight of ω SL ; Storage temperature, weight is ω ST : The distance between the cold chain warehouse and the destination, with a weight of ω D : The complexity of the transportation route, with a weight of ω RC ; For each packaging material configuration scheme i, define a comprehensive score S i , the expression is:

[0081] S i =w SL ·SL i +w ST ·ST i +w D (1-D i )+w RC ·(1-RCi );

[0082] Where S Li , S Ti , D i , RC i Respectively represent the scores of the i-th packaging material configuration scheme in terms of product shelf life, storage temperature, distance and route complexity. The scores are standardized values ​​ranging from 0 to 1, where 1 represents the best, and ω SL +ω ST +ω D +ω RC =1; Score standardization is to standardize any variable x, that is, the actual value, to its minimum value x min , the maximum value is x max , the minimum-maximum normalized value x′ is expressed as:

[0083]

[0084] By linearly converting the original data to between 0 and 1, the influence of different dimensions and magnitudes is eliminated, making the data comparable;

[0085] For example, based on the above, there are four groups of solutions: A, B, C, and D:

[0086] plan Product shelf life rating Storage temperature score Distance score Route complexity rating A 0.8 0.9 0.7 0.6 B 0.7 0.85 0.8 0.7 C 0.9 0.75 0.6 0.8 D 0.6 0.65 0.9 0.9

[0087] ω SL =0.3;ω ST =0.25:ω D =0.2:ω RC =0.25;

[0088] According to the above calculation:

[0089] Score A =0.3×0.8+0.25×0.9+0.2×0.7+0.25×0.6=0.755;

[0090] Score B =0.3×0.7+0.25×0.85+0.2×0.8+0.25×0.7=0.7575;

[0091] Score C =0.3×0.9+0.25×0.75+0.2×0.6+0.25×0.8=0.7775;

[0092] Score D =0.3×0.6+0.25×0.65+0.2×0.9+0.25×0.9=0.7475;

[0093] According to the above calculations, Plan C has the highest comprehensive score, so Plan C is the optimal scheduling method.

[0094] This embodiment also proposes a pharmaceutical cold chain packaging material scheduling system based on deep learning, such as Figure 2 , the system is used to implement the above method, including:

[0095] The data collection layer installs temperature, humidity, pressure, and vibration sensors at key locations in refrigerated trucks, refrigerated warehouses, and packaging boxes, and transmits sensor data to the central monitoring system in real time via wireless communication technology.

[0096] The data processing and storage layer performs preliminary processing and cleaning on the received data, including data verification, outlier detection, and data format conversion. The processed data is stored in a highly reliable and scalable central database, supporting real-time access and historical data backtracking.

[0097] The model training and analysis layer builds a deep learning model based on monitoring data, uses historical data for training, inputs real-time monitoring data into the trained deep learning model, performs real-time analysis and evaluation, uses the deep learning model to predict future demand and transportation environment, and collects and analyzes relevant data.

[0098] The scheduling strategy generation layer automatically generates packaging material scheduling instructions based on the analysis results of the deep learning model, including adjusting the type, quantity or delivery time of packaging materials. It calculates the scores of various packaging material configuration plans through weighted calculation and selects the one with the highest score as the optimal scheduling method.

[0099] The execution and feedback layer sends the generated packaging material scheduling instructions to the execution department for corresponding operations. During the execution process, feedback information is continuously collected, including actual environmental data, packaging material usage, and the quality status of pharmaceutical products. The collected feedback information is used to further optimize the training of deep learning models and the formulation of scheduling strategies.

[0100] In summary: compared with traditional technologies, the present invention has the following advantages:

[0101] By real-time monitoring and predicting environmental changes, packaging materials and scheduling strategies can be adjusted promptly to ensure the stability and safety of pharmaceutical products during transportation and storage, reduce the risk of drug quality degradation or failure due to environmental changes, and improve the overall quality level of drugs. At the same time, automated packaging material scheduling instruction generation and execution reduces the need for manual intervention, improves operational efficiency and accuracy, and dynamically adjusts the type, quantity, or delivery time of packaging materials, optimizes the allocation of logistics resources, and reduces logistics costs.

[0102] Secondly, the use of deep learning models for data analysis and prediction provides a scientific basis for decision-making, avoiding the subjective errors caused by relying on experience and judgment. Through the training of historical data and the input of real-time data, the model can continuously learn and optimize, improving the accuracy and reliability of predictions. The system integrates multiple links of data collection, processing, analysis and execution feedback, forming a closed-loop intelligent management system, continuously collecting feedback information and using it for model optimization, so that the system can continuously improve itself and improve the overall intelligence level.

[0103] Finally, the deep learning model has strong generalization capabilities and can adapt to different transportation environments and conditions, improving the adaptability and robustness of the system. By predicting future demand and transportation environment, it can make scheduling preparations in advance, effectively respond to market fluctuations and emergencies, make full use of historical data and real-time data, and explore potential correlations and patterns. It provides strong support for the formulation of packaging material scheduling strategies, selects the optimal packaging material configuration plan through weighted calculation, and achieves optimal resource allocation.

[0104] Finally, it should be noted that the above is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or make equivalent substitutions for some of the technical features therein. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A pharmaceutical cold chain packaging material scheduling method based on deep learning, characterized in that: The steps include: S1. Real-time monitoring of key parameters in the cold chain logistics process. Temperature, humidity, pressure, and vibration sensors are installed in refrigerated trucks, refrigerated warehouses, and packaging boxes. Environmental data is collected regularly and transmitted to the central monitoring system via wireless communication. S2. Input the monitoring data into the deep learning model for analysis. The central monitoring system establishes a deep learning model based on the received data and uses historical data to train the deep learning model so that it can accurately predict the impact of environmental changes on the quality of pharmaceutical products. S3. Dynamically adjust the packaging material scheduling strategy based on the analysis results, input the real-time monitoring data into the trained deep learning model, conduct real-time analysis and evaluation, and automatically generate packaging material scheduling instructions based on the analysis results of the deep learning model; S4. Use deep learning models to predict future demand and transportation environment, collect historical sales data, market trends, seasonal changes, special event information, and predict market demand and transportation environment in the future. Calculate the scores of various packaging material configuration schemes through weighted calculation, and use the one with the highest score as the optimal scheduling method.

2. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 1 is characterized in that: In step S1, after receiving the real-time data from each sensor, the central monitoring system performs preliminary processing and cleaning, including data verification, outlier detection, and data format conversion, to ensure the data quality of the input deep learning model; The processed data is stored in a central database to facilitate subsequent data analysis and query. The data storage system has high reliability and scalability, and supports real-time access and historical data backtracking.

3. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 1 is characterized in that: In step S2, the monitored temperature, humidity, pressure and vibration environment data are first normalized to meet the input requirements of the model. The data collected by n sensors, each sensor collects data with a dimension of d, and the input data matrix expression is: X∈R n×d , where X represents the data matrix and R represents the data set; In the model, the convolution kernel size of the lth layer is k l ×k l , the number of convolution kernels is m l , then the output feature size of this layer is: In the formula, s l is the step size, n and d are the height and width of the input feature map respectively, k l is the size of the convolution kernel.

4. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 3 is characterized in that: The central monitoring system establishes a deep learning model based on the received data, and the expression is: In the formula, is the impact score of environmental changes on the quality of pharmaceutical products predicted by the model, X is the input environmental monitoring data, θ is the model parameter, and f represents the mapping relationship of the model; The model parameters θ are adjusted to minimize the loss function L, which is expressed as: Where y is the impact score of actual environmental changes on the quality of pharmaceutical products, θ * is the optimal set of model parameters, and L(f(X; θ), y) represents the difference between the predicted result f(X; θ) using the model parameters θ and the input data X and the actual impact score y of environmental changes on the quality of pharmaceutical products.

5. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 1 is characterized in that: Step S3 is specifically as follows: S31. Generate packaging material scheduling instructions based on the analysis results of the deep learning model, including adjusting the type, quantity or delivery time of packaging materials to adapt to current and future environmental conditions to ensure that the quality of pharmaceutical products is not affected; S32. The generated packaging material dispatching instruction will be sent to the relevant execution departments, including the warehouse management department. The warehouse management department will perform corresponding operations according to the instruction, including adjusting inventory, preparing shipment or changing the delivery route; S33. During the execution process, the system will continuously collect feedback information, including actual environmental data, packaging material usage, and the quality status of pharmaceutical products. The information will be used to further optimize the training of deep learning models and the formulation of scheduling strategies to improve the intelligence level and efficiency of the entire system.

6. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 5 is characterized in that: The step S31 also includes manually reviewing the generated dispatch instruction, specifically: Preliminary assessment and preparation. Before executing the instruction, the warehouse management department needs to conduct preliminary assessment and preparation, including checking the existing inventory, confirming the availability of the required packaging materials, and preparing logistics resources; Implement manual fine-tuning. According to the actual work situation, the warehouse management department will conduct manual fine-tuning. When it is found that the actual demand for certain packaging materials differs from the instructions, adjustments will be made based on the actual situation to ensure that the scheduling strategy is more scientific and reasonable.

7. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 1 is characterized in that: In step S4, the deep learning model is used to predict future demand and transportation environment, collect historical sales data, market trends, seasonal changes, special event information, and predict market demand and transportation environment in the future, specifically: Collect market feedback data, including historical sales volume, historical sales, customer feedback, market research reports, and special event reports; special event reports include data on holidays, promotional activities, and natural disasters; A long short-term memory network model is established and trained according to the data. The training divides the collected historical data into a training set and a test set, wherein the training set accounts for 70%-80% of the total data and the test set accounts for 20%-30%; The currently known market data and features are input into the trained model, and the model calculates the predicted values ​​of market demand and transportation environment in the future based on the input data.

8. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 7 is characterized in that: The scores of various packaging material configuration schemes are calculated by weighted calculation in step S4, and the one with the highest score is taken as the optimal scheduling method, specifically: The weighted items include the product shelf life, with a weight of ω SL ; Storage temperature, weight is ω ST : The distance between the cold chain warehouse and the destination, with a weight of ω D : The complexity of the transportation route, with a weight of ω RC ; For each packaging material configuration scheme i, define a comprehensive score S i , the expression is: S i =w SL ·SL i +w ST ·ST i +w D ·(1-D i )+w RC ·(1-RC i ); In the formula, S Li , S Ti , D i , R.C. i Respectively represent the scores of the i-th packaging material configuration scheme in terms of product shelf life, storage temperature, distance and route complexity. The scores are standardized values ​​ranging from 0 to 1, where 1 represents the best, and ω SL +ω ST +ω D +ω RC =1.

9. The pharmaceutical cold chain packaging material scheduling method based on deep learning according to claim 8, characterized in that: Rating standardization is to standardize any variable x, that is, the actual value, to its minimum value x min , the maximum value is x max , the minimum-maximum normalized value x′, expressed as: By linearly converting the original data to between 0 and 1, the influence of different dimensions and magnitudes is eliminated, making the data comparable.

10. A pharmaceutical cold chain packaging material scheduling system based on deep learning, the system being used to implement any of the methods of claims 1-9, characterized in that: include: Data collection layer: temperature, humidity, pressure, and vibration sensors are installed at key locations of refrigerated trucks, refrigerated warehouses, and packaging boxes, and the sensor data is transmitted to the central monitoring system in real time through wireless communication technology; Data processing and storage layer: The received data is initially processed and cleaned, including data verification, outlier detection, and data format conversion. The processed data is stored in a highly reliable and scalable central database, supporting real-time access and historical data backtracking. Model training and analysis layer: build deep learning models based on monitoring data, use historical data for training, input real-time monitoring data into the trained deep learning models, conduct real-time analysis and evaluation, use deep learning models to predict future demand and transportation environment, collect relevant data and analyze them; The scheduling strategy generation layer automatically generates packaging material scheduling instructions based on the analysis results of the deep learning model, including adjusting the type, quantity or delivery time of packaging materials, calculating the scores of various packaging material configuration plans through weighted calculation, and selecting the one with the highest score as the optimal scheduling method. The execution and feedback layer sends the generated packaging material scheduling instructions to the execution department to perform corresponding operations and continuously collects feedback information during the execution process, including actual environmental data, packaging material usage, and the quality status of pharmaceutical products: the collected feedback information is used to further optimize the training of deep learning models and the formulation of scheduling strategies.

Images