Intelligent logistics system based on optimal scheduling

Abstract

The invention discloses an intelligent logistics system based on optimal scheduling, and relates to the field of logistics management, and the system comprises the steps: arranging environment sensors at key nodes of a logistics network, and transmitting collected real-time data to edge computing equipment for preprocessing; analyzing the preprocessed data by using a deep learning model, and predicting an environment change trend in a non-period of time; dynamically adjusting a logistics scheduling strategy based on the prediction result; after receiving the adjusted scheduling strategy, defining a state space, an action space and a reward function of the logistics system; training a reinforcement learning algorithm model, and continuously optimizing a decision strategy through interaction with the environment; and calling the reinforcement learning model through an API interface to make a real-time decision, and generating an optimal scheduling scheme. According to the invention, by integrating environment perception, deep learning prediction, adaptive scheduling and reinforcement learning, the defects of an existing intelligent logistics system in the aspects of environment perception, prediction and dynamic scheduling optimization are overcome, and the intelligent level and operation efficiency of logistics management are significantly improved.

Description

Technical Field

[0001] The present invention relates to the field of logistics management, and in particular to an intelligent logistics system based on optimized scheduling. Background Art

[0002] By integrating advanced information technology, communication technology, and automation technology, the smart logistics system realizes the efficient configuration and management of logistics resources, thereby improving the quality and efficiency of logistics services. In particular, with the support of big data and cloud computing, the smart logistics system can process massive amounts of data in real time, providing important information support for logistics decision-making. In addition, the application of edge computing technology makes data processing closer to the data source, effectively reducing data transmission delays and improving the system's response speed.

[0003] Existing smart logistics systems still face challenges in environmental perception and prediction, dynamic scheduling optimization, etc. The ability to predict environmental changes is limited. Traditional statistical methods have difficulty capturing complex nonlinear relationships, while simple machine learning models often have low prediction accuracy due to insufficient data or improper feature selection. Dynamic scheduling strategies lack flexibility and adaptability. Most systems use static or semi-static scheduling modes and cannot respond quickly to real-time environmental changes, resulting in low scheduling efficiency and increased logistics costs. Summary of the invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides a smart logistics system based on optimized scheduling to solve the problems of the existing smart logistics systems in terms of environmental perception, prediction and dynamic scheduling optimization.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, an embodiment of the present invention provides a smart logistics system based on optimized scheduling, which includes:

[0008] Data collection and processing module: Environmental sensors are deployed at key nodes in the logistics network, and the collected real-time data is transmitted to edge computing devices for pre-processing;

[0009] Environmental change prediction module: Use deep learning models to analyze preprocessed data and predict environmental change trends over a period of time;

[0010] Adaptive scheduling module: dynamically adjusts logistics scheduling strategies based on prediction results;

[0011] Environmental modeling module: After receiving the adjusted scheduling strategy, it defines the state space, action space and reward function of the logistics system;

[0012] Model reinforcement learning module: trains reinforcement learning algorithm models and continuously optimizes decision-making strategies through interaction with the environment;

[0013] Intelligent decision support module: calls the reinforcement learning model through the API interface to make real-time decisions and generate the optimal scheduling plan.

[0014] As a preferred solution of the file encryption method of the present invention, wherein: the environmental sensor includes a meteorological sensor and a traffic sensor;

[0015] The meteorological sensor monitors temperature, humidity and air pressure;

[0016] The traffic sensor monitors the traffic flow and traffic light status at the location;

[0017] The logistics scheduling strategy includes delivery routes and delivery times.

[0018] As a preferred solution of the file encryption method of the present invention, the environmental sensors are arranged at the key nodes of the logistics network, and the collected real-time data is transmitted to the edge computing device for preprocessing. The specific steps are as follows:

[0019] Deploy environmental sensors in warehouses, distribution centers, and major traffic intersections in the logistics network

[0020] Connect the sensor to the edge computing device through the LoRa module, set the sampling frequency of the sensor, and transmit the data detected by the sensor to the edge computing device in real time through the LoRa module;

[0021] The edge computing device receives real-time data and uses a sliding window filtering algorithm to clean the data and remove invalid values ​​and outliers.

[0022] Take 5 data points as a window, calculate the average value of the data points in the window, and when the difference between a data point and the window average value exceeds 3 times the standard deviation, replace it with the window average value;

[0023] The cleaned data is uniformly converted into JSON format, and the Z-score value of each data point is calculated. When the Z-score value is not between [-3, 3], the data point is considered an outlier and is removed.

[0024] As a preferred solution of the file encryption method of the present invention, the deep learning model is used to analyze the pre-processed data to predict the environmental change trend over a period of time. The specific steps are as follows:

[0025] Extract the data values ​​of each sensor from the JSON data, calculate the maximum and minimum values ​​of each sensor data, and use the maximum and minimum normalization method to normalize the preprocessed data;

[0026] The normalized data is randomly divided into training set and test set;

[0027] Use the Keras library to build an LSTM model, import the training set and test set into the Keras library, and build a prediction model with two layers of LSTM, 128 neurons in each layer, and an output layer;

[0028] Convert the test set and training set into a three-dimensional array with the shape of: number of samples, time steps, number of features;

[0029] Use the training set data to train the prediction model, set the batch size to 32, and the number of training rounds to 50. At the same time, use the test set data to calculate the mean square error of the output results and evaluate the performance of the prediction model;

[0030] The time step data is extracted from the most recent sensor data, and the extracted time is predicted through the prediction model to generate the environmental change trend in the future.

[0031] As a preferred solution of the file encryption method of the present invention, the logistics scheduling strategy is dynamically adjusted based on the prediction results. The specific steps are as follows:

[0032] Based on the predicted environmental data, the impact on logistics and distribution is evaluated, and the transportation cost, delivery time, and delivery vehicle delay are scored according to different environmental changes;

[0033] Based on the evaluation results, and referring to the delivery time and number of delays of the delivery route and the current delivery time, the Scikit-Learn decision model is used to determine whether the logistics scheduling strategy needs to be adjusted.

[0034] As a preferred solution of the file encryption method of the present invention, wherein: the three aspects of transportation cost, delivery time, and delivery vehicle delay are scored according to different environmental changes. The specific steps are as follows:

[0035] Weather condition scoring rules: if the temperature is below 0 degrees Celsius or above 35 degrees Celsius, the score is 2; if the humidity is above 80%, the score is 2; if the air pressure drops significantly, the score is 3; if there is a forecast of thunderstorm or heavy rainfall, the score is 4;

[0036] Traffic condition scoring rules: If the traffic volume increases by more than 20% compared to normal days, the score is 2; if there is an accident or construction on the road, the score is 3; if the traffic light fails or traffic is controlled, the score is 4;

[0037] Add up all the individual scores; the higher the total score, the greater the potential risk. The impact level is divided according to the total score, with 0-5 points for slight impact, 6-10 points for moderate impact, 11-15 points for severe impact, and 16 points or above for extremely severe impact.

[0038] As a preferred solution of the file encryption method of the present invention, the Scikit-Learn decision model is used to determine whether the logistics scheduling strategy needs to be adjusted. The specific steps are as follows:

[0039] When the comprehensive score is greater than 11, the historical performance score of the current delivery route is less than 2.5, and the traffic condition score is greater than 3, the delivery route will be replanned, giving priority to routes with better traffic conditions;

[0040] When the comprehensive score is greater than 11, the current delivery time is before midnight, and the weather condition score is greater than 3, a time period with less traffic flow will be selected for delivery based on traffic flow forecasts.

[0041] As a preferred solution of the file encryption method of the present invention, after receiving the adjusted scheduling strategy, the state space, action space and reward function of the logistics system are defined. The specific steps are as follows:

[0042] State space refers to all factors that affect logistics scheduling strategies, including static factors and dynamic factors. The static factors include the location of the distribution point, warehouse capacity, type and number of distribution vehicles, and the dynamic factors include weather conditions, traffic conditions, and distribution task status.

[0043] Action space refers to the decision of scheduling adjustment strategy, including,adjusting delivery route and adjusting delivery time;

[0044] The reward function gives positive rewards for successfully delivering goods on time, gives additional positive rewards for avoiding delivery delays due to bad weather or traffic conditions, gives positive rewards for reducing unnecessary delivery costs (such as fuel costs and labor costs), and gives negative rewards for delivery delays or cost overruns, reflecting the effectiveness of logistics scheduling decisions.

[0045] As a preferred solution of the file encryption method of the present invention, the reinforcement learning algorithm is used to train the model, and the decision strategy is continuously optimized through interaction with the environment. The specific steps are as follows:

[0046] Build a reinforcement learning environment based on the defined state space and action space, receive actions and return new states and immediate rewards;

[0047] Initialize the learning rate, initial exploration rate, and discount factor of the reinforcement learning algorithm;

[0048] Use a multi-layer convolutional neural network as a reinforcement learning model to interact with the environment and store the generated experience tuples in the experience replay pool;

[0049] A batch of experience tuples are randomly selected from the experience replay pool. For each experience tuple, the target Q value is calculated. The expression is:

[0050]

[0051] Among them, y i is the i-th target Q value, r i is the reward obtained from the environment immediately after executing the i-th action, γ is the discount factor, and s i+1 is the next state, a' is the next state s i+1 All possible actions of θ - is the target network parameter;

[0052] Use gradient descent to minimize the loss function and update the model parameters;

[0053] By monitoring the behavior of the reinforcement learning model, evaluating its performance in the actual environment, adding positive rewards and negative rewards, and dynamically adjusting the reward function;

[0054] Through continuous interaction between the reinforcement learning model and the environment, accumulation of experience tuples, and continuous trial and error, the reinforcement learning model is gradually optimized and training is completed.

[0055] As a preferred solution of the file encryption method of the present invention, the specific steps of calling the reinforcement learning model through the API interface to make real-time decisions and generate the optimal scheduling plan are as follows:

[0056] The trained reinforcement learning model is deployed on the edge computing device and integrated with the logistics scheduling system. The edge computing device is responsible for collecting the latest environmental data and calling the model to obtain the best scheduling strategy based on the current status.

[0057] After the reinforcement learning model is launched, we continue to collect interaction data in real environments and further optimize the reinforcement learning model through online learning.

[0058] The beneficial effects of the present invention are as follows: the present invention uses a deep learning model to analyze the preprocessed data, predicts the environmental change trend in the future, and realizes the scientific prediction of future environmental changes. By dynamically adjusting the logistics scheduling strategy based on the prediction results, the effective allocation and utilization of logistics resources are realized. After receiving the adjusted scheduling strategy, the state space, action space and reward function of the logistics system are defined, and the refined management of the logistics scheduling process is realized. By using the reinforcement learning algorithm to train the model, the decision-making strategy is continuously optimized through interaction with the environment, and the continuous improvement of the logistics scheduling strategy is realized. The reinforcement learning model is called through the API interface for real-time decision-making, and the optimal scheduling plan is generated, thereby realizing the intelligent and automated logistics scheduling. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0060] Figure 1 This is a flow chart of the intelligent logistics system based on optimized scheduling in Example 1.

[0061] Figure 2 This is a module diagram of the intelligent logistics system based on optimized scheduling in Example 1. DETAILED DESCRIPTION

[0062] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the accompanying drawings.

[0063] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0064] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0065] Example 1, reference Figure 1 and Figure 2 , which is the first embodiment of the present invention, and provides a smart logistics system based on optimized scheduling, comprising the following steps:

[0066] S1. Environmental sensors are deployed at key nodes in the logistics network and the collected real-time data is transmitted to edge computing devices for preprocessing.

[0067] Furthermore, environmental sensors, including meteorological sensors and traffic sensors, are deployed in warehouses, distribution centers, and major traffic intersections in the logistics network;

[0068] The meteorological sensor monitors temperature, humidity and air pressure;

[0069] The traffic sensor monitors the traffic flow and traffic light status at the location;

[0070] Connect the sensor to the edge computing device through the LoRa module, set the sampling frequency of the sensor, and transmit the data detected by the sensor to the edge computing device in real time through the LoRa module;

[0071] Furthermore, the edge computing device receives real-time data and uses a sliding window filtering algorithm to clean the data and remove invalid and outliers;

[0072] Take 5 data points as a window, calculate the average value of the data points in the window, and when the difference between a data point and the window average value exceeds 3 times the standard deviation, replace it with the window average value;

[0073] The cleaned data is uniformly converted into JSON format, and the Z-score value of each data point is calculated. When the Z-score value is not between [-3, 3], the data point is considered an outlier and is removed.

[0074] S2. Use deep learning models to analyze preprocessed data and predict environmental change trends over a period of time.

[0075] Furthermore, the data values ​​of each sensor are extracted from the JSON data, the maximum and minimum values ​​of each sensor data are calculated respectively, and the preprocessed data is normalized using the maximum and minimum normalization method. The expression is as follows:

[0076]

[0077] Among them, x ′ is the normalized data value, x is the data value, min(x) is the minimum data value, and max(x) is the maximum data value;

[0078] Furthermore, the normalized data is randomly divided into training and test sets;

[0079] Use the Keras library to build an LSTM model, import the training set and test set into the Keras library, and build a prediction model with two layers of LSTM, 128 neurons in each layer, and an output layer;

[0080] Convert the test set and training set into a three-dimensional array with the shape of: number of samples, time steps, number of features;

[0081] Use the training set data to train the prediction model, set the batch size to 32, and the number of training rounds to 50. At the same time, use the test set data to calculate the mean square error of the output results and evaluate the performance of the prediction model;

[0082] The time step data is extracted from the most recent sensor data, and the extracted time is predicted through the prediction model to generate the environmental change trend in the future.

[0083] S3. Dynamically adjust logistics scheduling strategies based on prediction results.

[0084] Furthermore, based on the predicted environmental data, the impact on logistics and distribution is evaluated, and the transportation cost, delivery time, and delivery vehicle delay are scored for different environmental changes;

[0085] Weather condition scoring rules: if the temperature is below 0 degrees Celsius or above 35 degrees Celsius, the score is 2; if the humidity is above 80%, the score is 2; if the air pressure drops significantly, the score is 3; if there is a forecast of thunderstorm or heavy rainfall, the score is 4;

[0086] Traffic condition scoring rules: If the traffic volume increases by more than 20% compared to normal days, the score is 2; if there is an accident or construction on the road, the score is 3; if the traffic light fails or traffic is controlled, the score is 4;

[0087] All individual scores are added together. The higher the total score, the greater the potential risk. The impact level is divided according to the total score: 0-5 points for slight impact, 6-10 points for moderate impact, 11-15 points for severe impact, and 16 points or above for extremely severe impact.

[0088] Furthermore, based on the evaluation results, and referring to the delivery time and number of delays of the delivery route and the current delivery time, the Scikit-Learn decision model is used to determine whether the logistics scheduling strategy needs to be adjusted;

[0089] The logistics scheduling strategy includes delivery routes and delivery times;

[0090] When the comprehensive score is greater than 11, it means that environmental changes have a serious impact on delivery. When the historical performance score of the current delivery route is less than 2.5, it means that the historical performance of the current route is poor. When the traffic condition score is greater than 3, it means that the traffic condition is poor. Use the path optimization algorithm (genetic algorithm) to re-plan the delivery route and give priority to routes with better traffic conditions.

[0091] When the comprehensive score is greater than 11, it means that environmental changes have a serious impact on delivery. When the current delivery time is before midnight and the weather condition score is greater than 3, it means that the weather conditions are not good. According to the traffic flow forecast, the time period with less traffic flow is selected for delivery.

[0092] S4. After receiving the adjusted scheduling strategy, define the state space, action space and reward function of the logistics system.

[0093] Furthermore, the state space refers to all factors that affect the logistics scheduling strategy, including static factors and dynamic factors. The static factors include the location of the distribution point, warehouse capacity, type and number of distribution vehicles, and the dynamic factors include weather conditions, traffic conditions, and distribution task status.

[0094] Action space refers to the decision of scheduling adjustment strategy, including,adjusting delivery route and adjusting delivery time;

[0095] Furthermore, the reward function gives positive rewards by successfully delivering goods on time, gives additional positive rewards for avoiding delivery delays due to bad weather or traffic conditions, gives positive rewards for reducing unnecessary delivery costs (such as fuel costs and labor costs), and gives negative rewards for delivery delays or cost overruns, reflecting the effectiveness of logistics scheduling decisions.

[0096] S5. Train the reinforcement learning algorithm model and continuously optimize the decision-making strategy through interaction with the environment.

[0097] Build a reinforcement learning environment based on the defined state space and action space, receive actions and return new states and immediate rewards;

[0098] Furthermore, the learning rate of the reinforcement learning algorithm was initially set to 0.001, the initial exploration rate was 1.0, and it was gradually reduced to 0.01 with training, and the discount factor was 0.99;

[0099] Use a multi-layer convolutional neural network as a reinforcement learning model to interact with the environment and store the generated experience tuples in the experience replay pool, which helps to break the correlation between data and improve training efficiency;

[0100] A batch of experience tuples are randomly selected from the experience replay pool. For each experience tuple, the target Q value is calculated. The expression is:

[0101] y i =r i +γmax a′ Q(s i+1 ,a′;θ - )

[0102] Among them, y i is the i-th target Q value, r i is the reward obtained from the environment immediately after executing the i-th action, γ is the discount factor, and s i+1 is the next state, a′ is the next state s i+1 All possible actions of θ - is the target network parameter;

[0103] Use gradient descent to minimize the loss function and update the model parameters;

[0104] By monitoring the behavior of the reinforcement learning model, evaluating its performance in the actual environment, adding positive rewards and negative rewards, and dynamically adjusting the reward function, we can ensure that the model learns more effective strategies and avoids falling into local optimal solutions.

[0105] Through continuous interaction between the reinforcement learning model and the environment, accumulation of experience tuples, and continuous trial and error, the reinforcement learning model is gradually optimized and training is completed.

[0106] S6. Call the reinforcement learning model through the API interface to make real-time decisions and generate the optimal scheduling plan.

[0107] Furthermore, the trained reinforcement learning model is deployed on the edge computing device and integrated with the logistics scheduling system. The edge computing device is responsible for collecting the latest environmental data and calling the model to obtain the best scheduling strategy based on the current status;

[0108] After the reinforcement learning model is launched, we continue to collect interaction data in real environments and further optimize the reinforcement learning model through online learning.

[0109] In summary, the present invention achieves scientific prediction of future environmental changes by: using a deep learning model to analyze preprocessed data and predicting environmental change trends in the future; achieving effective allocation and utilization of logistics resources by dynamically adjusting the logistics scheduling strategy based on the prediction results; defining the state space, action space and reward function of the logistics system after receiving the adjusted scheduling strategy; achieving refined management of the logistics scheduling process; training the model with a reinforcement learning algorithm and continuously optimizing the decision-making strategy through interaction with the environment; achieving continuous improvement of the logistics scheduling strategy; calling the reinforcement learning model through an API interface for real-time decision-making, generating the optimal scheduling plan, and achieving intelligent and automated logistics scheduling.

[0110] Example 2, referring to Table 1, is the second example of the present invention. In order to further verify the technical solution of the present invention, experimental simulation data of a smart logistics system based on optimized scheduling are provided.

[0111] In order to verify the effectiveness of the intelligent logistics system based on optimized scheduling (hereinafter referred to as this system), a typical logistics distribution scenario was selected in this experiment.

[0112] First, environmental sensors are installed at key nodes in the logistics network, such as warehouses, distribution centers, and major traffic intersections. These sensors include meteorological sensors for monitoring temperature, humidity, and air pressure, as well as traffic sensors for monitoring vehicle flow and traffic light status. All sensors are connected to edge computing devices via LoRa modules, and the sampling frequency of the sensors is set to once per minute to ensure that environmental change information can be captured in a timely manner.

[0113] During the data collection phase, the system preprocessed the collected raw data, cleaned the data using a sliding window filtering algorithm, and removed invalid values ​​and outliers. Subsequently, the preprocessed data was normalized using the maximum and minimum normalization method to ensure the data quality of the input deep learning model. In this experiment, a prediction model consisting of two layers of LSTM units, each containing 128 neurons, and an output layer was constructed to predict environmental change trends in the future.

[0114] After the prediction model is trained, the system dynamically adjusts the logistics scheduling strategy based on the prediction results. By quantifying the impact of transportation costs, delivery time, and delivery vehicle delays, the system can more accurately identify situations that require scheduling strategy adjustments. For example, when it is predicted that a certain area will encounter bad weather or traffic congestion, the system will adjust the delivery route or time in advance to reduce potential risks.

[0115] In order to evaluate the actual effect of this system, this experiment also set up a control group, that is, the traditional logistics scheduling mode, without using any prediction or adaptive scheduling technology. The two groups ran the same logistics distribution task for a week and collected data on multiple indicators including delivery on-time rate, transportation cost, customer satisfaction, etc.

[0116] The details are shown in Table 1 below:

[0117] Table 1 Comparison of logistics scheduling effects between this system and traditional mode

[0118]

[0119] Through the comparative analysis of the above experimental data, the adaptive scheduling module of this system can dynamically adjust the delivery plan according to the predicted results of environmental changes, avoiding delivery delays caused by external factors (such as bad weather or traffic congestion). This intelligent scheduling method improves the overall delivery efficiency and reliability.

[0120] The cost reduction is mainly due to the fact that the system reduces unnecessary driving distance and waiting time by optimizing the delivery route and time. Especially in the face of traffic congestion or bad weather, the strategy can be flexibly adjusted to avoid unnecessary cost expenditure.

[0121] The improvement of customer satisfaction directly reflects the improvement of delivery service quality and efficiency, and also means that customers' expectations for delivery time are better met. This is of great significance for improving brand image and service competitiveness.

[0122] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A smart logistics system based on optimized scheduling, characterized by: include, Data collection and processing module: Environmental sensors are deployed at key nodes in the logistics network, and the collected real-time data is transmitted to edge computing devices for pre-processing; Environmental change prediction module: Use deep learning models to analyze preprocessed data and predict environmental change trends over a period of time; Adaptive scheduling module: dynamically adjusts logistics scheduling strategies based on prediction results; Environmental modeling module: After receiving the adjusted scheduling strategy, it defines the state space, action space and reward function of the logistics system; Model reinforcement learning module: trains reinforcement learning algorithm models and continuously optimizes decision-making strategies through interaction with the environment; Intelligent decision support module: calls the reinforcement learning model through the API interface to make real-time decisions and generate the optimal scheduling plan.

2. The intelligent logistics system based on optimized scheduling as claimed in claim 1, characterized in that: The environmental sensors include meteorological sensors and traffic sensors; The meteorological sensor monitors temperature, humidity and air pressure; The traffic sensor monitors the traffic flow and traffic light status at the location; The logistics scheduling strategy includes delivery routes and delivery times.

3. The intelligent logistics system based on optimized scheduling as claimed in claim 2, characterized in that: The environmental sensors are arranged at the key nodes of the logistics network, and the collected real-time data is transmitted to the edge computing device for preprocessing. The specific steps are as follows: Deploy environmental sensors in warehouses, distribution centers, and major traffic intersections in the logistics network Connect the sensor to the edge computing device through the LoRa module, set the sampling frequency of the sensor, and transmit the data detected by the sensor to the edge computing device in real time through the LoRa module; The edge computing device receives real-time data and uses a sliding window filtering algorithm to clean the data and remove invalid values ​​and outliers. Take 5 data points as a window, calculate the average value of the data points in the window, and when the difference between a data point and the window average value exceeds 3 times the standard deviation, replace it with the window average value; The cleaned data is uniformly converted into JSON format, and the Z-score value of each data point is calculated. When the Z-score value is not between [-3, 3], the data point is considered an outlier and is removed.

4. The intelligent logistics system based on optimized scheduling as claimed in claim 3, characterized in that: The deep learning model is used to analyze the preprocessed data and predict the environmental change trend over a period of time. The specific steps are as follows: Extract the data values ​​of each sensor from the JSON data, calculate the maximum and minimum values ​​of each sensor data, and use the maximum and minimum normalization method to normalize the preprocessed data; The normalized data is randomly divided into training set and test set; Use the Keras library to build an LSTM model, import the training set and test set into the Keras library, and build a prediction model with two layers of LSTM, 128 neurons in each layer, and an output layer; Convert the test set and training set into a three-dimensional array with the shape of: number of samples, time steps, number of features; Use the training set data to train the prediction model, set the batch size to 32, and the number of training rounds to 50. At the same time, use the test set data to calculate the mean square error of the output results and evaluate the performance of the prediction model; The time step data is extracted from the most recent sensor data, and the extracted time is predicted through the prediction model to generate the environmental change trend in the future.

5. The intelligent logistics system based on optimized scheduling as claimed in claim 4, characterized in that: The specific steps of dynamically adjusting the logistics scheduling strategy based on the prediction results are as follows: Based on the predicted environmental data, the impact on logistics and distribution is evaluated, and the transportation cost, delivery time, and delivery vehicle delay are scored according to different environmental changes; Based on the evaluation results, and referring to the delivery time and number of delays of the delivery route and the current delivery time, the Scikit-Learn decision model is used to determine whether the logistics scheduling strategy needs to be adjusted.

6. The intelligent logistics system based on optimized scheduling as claimed in claim 5, characterized in that: According to different environmental changes, the three aspects of transportation cost, delivery time and delivery vehicle delay are scored. The specific steps are as follows: Weather condition scoring rules: if the temperature is below 0 degrees Celsius or above 35 degrees Celsius, the score is 2; if the humidity is above 80%, the score is 2; if the air pressure drops significantly, the score is 3; if there is a forecast of thunderstorm or heavy rainfall, the score is 4; Traffic condition scoring rules: If the traffic volume increases by more than 20% compared to normal days, the score is 2; if there is an accident or construction on the road, the score is 3; if the traffic light fails or traffic is controlled, the score is 4; Add up all the individual scores; the higher the total score, the greater the potential risk. The impact level is divided according to the total score, with 0-5 points for slight impact, 6-10 points for moderate impact, 11-15 points for severe impact, and 16 points or above for extremely severe impact.

7. The intelligent logistics system based on optimized scheduling as claimed in claim 6, characterized in that: The Scikit-Learn decision model is used to determine whether the logistics scheduling strategy needs to be adjusted. The specific steps are as follows: When the comprehensive score is greater than 11, the historical performance score of the current delivery route is less than 2.5, and the traffic condition score is greater than 3, the delivery route will be replanned, giving priority to routes with better traffic conditions; When the comprehensive score is greater than 11, the current delivery time is before midnight, and the weather condition score is greater than 3, a time period with less traffic flow will be selected for delivery based on traffic flow forecasts.

8. The intelligent logistics system based on optimized scheduling as claimed in claim 7, characterized in that: After receiving the adjusted scheduling strategy, the state space, action space and reward function of the logistics system are defined. The specific steps are as follows: The state space refers to all factors that affect the logistics scheduling strategy. Including static factors and dynamic factors, the static factors, Including, the location of the distribution point, warehouse capacity, type and number of distribution vehicles, the dynamic factors, Including weather conditions, traffic conditions, and delivery task status; Action space refers to the decision of scheduling adjustment strategy, including,adjusting delivery route and adjusting delivery time; The reward function gives positive rewards for successfully delivering goods on time, gives additional positive rewards for avoiding delivery delays due to bad weather or traffic conditions, gives positive rewards for reducing unnecessary delivery costs (such as fuel costs and labor costs), and gives negative rewards for delivery delays or cost overruns, reflecting the effectiveness of logistics scheduling decisions.

9. The intelligent logistics system based on optimized scheduling as claimed in claim 8, characterized in that: The reinforcement learning algorithm is used to train the model and continuously optimize the decision-making strategy through interaction with the environment. The specific steps are as follows: Build a reinforcement learning environment based on the defined state space and action space, receive actions and return new states and immediate rewards; Initialize the learning rate, initial exploration rate, and discount factor of the reinforcement learning algorithm; Use a multi-layer convolutional neural network as a reinforcement learning model to interact with the environment and store the generated experience tuples in the experience replay pool; A batch of experience tuples are randomly selected from the experience replay pool. For each experience tuple, the target Q value is calculated. The expression is: y i =r i +γmax a′ Q(s i+1 ,a′;θ - ) Among them, y i is the i-th target Q value, r i is the reward obtained from the environment immediately after executing the i-th action, γ is the discount factor, and s i+1 is the next state, a′ is the next state s i+1 All possible actions of θ - is the target network parameter; Use gradient descent to minimize the loss function and update the model parameters; By monitoring the behavior of the reinforcement learning model, evaluating its performance in the actual environment, adding positive rewards and negative rewards, and dynamically adjusting the reward function; Through continuous interaction between the reinforcement learning model and the environment, accumulation of experience tuples, and continuous trial and error, the reinforcement learning model is gradually optimized and training is completed.

10. The intelligent logistics system based on optimized scheduling as claimed in claim 9, characterized in that: The API interface is used to call the reinforcement learning model to make real-time decisions and generate the optimal scheduling plan. The specific steps are as follows: The trained reinforcement learning model is deployed on the edge computing device and integrated with the logistics scheduling system. The edge computing device is responsible for collecting the latest environmental data and calling the model to obtain the best scheduling strategy based on the current status. After the reinforcement learning model is launched, we continue to collect interaction data in real environments and further optimize the reinforcement learning model through online learning.

Images