An intelligent express sorting path optimization method based on big data analysis

Abstract

The application relates to the technical field of big data analysis, and discloses an intelligent express sorting path optimization method based on big data analysis, which comprises the following steps: acquiring multi-dimensional environment time sequence data of a distribution center warehouse area; the multi-dimensional environment time sequence data comprises illumination intensity time sequence data, humidity time sequence data and historical fault record data; acquiring historical operation data of multiple AGVs when performing sorting tasks in the warehouse area; the historical operation data comprises a two-dimensional code recognition error rate, a retry number and a path planning deviation record. According to the application, the illumination intensity, humidity and historical fault record and other multi-dimensional environment time sequence data of the distribution center warehouse area are systematically collected, and are associated with the historical operation data of the AGVs, so that the limitation of the traditional path optimization method which only depends on single operation data is broken, the path planning system has the ability of active sensing and quantitative environmental risk, and real and reliable data basis is provided for subsequent prediction and optimization.

Description

Technical Field

[0001] This invention relates to the field of big data analytics, and in particular to an intelligent express delivery sorting path optimization method based on big data analytics. Background Technology

[0002] In existing intelligent express sorting systems, AGVs typically rely on ground QR codes for navigation and positioning. Their path planning is mostly based on static grid maps or historical operating data, failing to fully consider the dynamic impact of environmental factors such as aging warehouse lighting equipment and excessive humidity during the rainy season on the QR code recognition rate. There is also a lack of means to jointly model environmental time-series data with AGV operation fault records, resulting in a significant increase in the QR code recognition error rate when lighting decays and humidity rises. AGVs frequently retry, deviate from their paths, or even stop, severely restricting sorting efficiency and system stability. Summary of the Invention

[0003] This invention provides an intelligent express sorting path optimization method based on big data analysis to solve existing technical problems, thereby addressing the issues of QR code recognition failure and AGV path reliability degradation caused by aging lighting and excessive humidity during the rainy season.

[0004] To address the aforementioned technical problems, according to one aspect of the present invention, more specifically, a smart express sorting path optimization method based on big data analysis, comprising the following steps: Step S1: Obtain multi-dimensional environmental time-series data of the distribution center warehouse area. The multi-dimensional environmental time-series data includes lighting intensity time-series data, humidity time-series data, and historical fault record data. Step S2: Obtain historical operating data of multiple AGVs performing sorting tasks in the warehouse area. The historical operating data includes QR code recognition error rate, number of retries, and path planning deviation records. Step S3: Construct an environmental identification joint prediction model, which is used to establish a dynamic correlation between multi-dimensional environmental time-series data and the historical operation data of AGV; Step S4: Input the current environmental data and weather forecast data into the environmental recognition joint prediction model to predict the downward trend of QR code recognition rate caused by lighting aging and excessive humidity in the next rainy season, and generate the predicted recognition rate loss value of each grid node. Step S5: Based on the predicted recognition rate loss value, dynamically adjust the grid map weight of the distribution center warehouse area and perform global path replanning for multiple AGVs. The global path replanning includes avoiding grid nodes whose predicted recognition rate loss value is higher than a preset threshold, or allocating redundant verification time windows for AGVs passing through the grid nodes. Step S6: During the next rainy season, collect real-time data on the actual QR code recognition error rate and path replanning correction; use the actual data as feedback to incrementally train the environmental recognition joint prediction model, and dynamically update the weight coefficients and influence functions in the calculation formula of the prediction recognition rate loss value.

[0005] Furthermore, the method for constructing the environmental identification joint prediction model includes: Step S301: Use a temporal convolutional network to extract features from the lighting intensity time series data and humidity time series data to generate a first environmental feature vector; Step S302: Use a graph neural network to extract spatial topological features from the historical fault record data and path planning deviation records to generate a second spatial feature vector; Step S303: Perform cross-modal fusion of the first environmental feature vector and the second spatial feature vector to generate a joint feature representation; Step S304: Using the QR code recognition error rate as the prediction target, train a regression predictor based on the joint feature representation to obtain the joint prediction model for environmental recognition.

[0006] Furthermore, the first environmental feature vector is a high-dimensional abstract feature representation that can characterize the dynamic changes of the reservoir area's environmental state over time after performing multi-scale convolution operations on the lighting intensity time-series data and humidity time-series data through a temporal convolutional network. It is used to reflect the combined impact of lighting aging and excessive humidity on the identified environment.

[0007] Furthermore, the second spatial feature vector is a low-dimensional embedded feature representation generated by learning the spatial topology of historical fault record data and path planning deviation records through a graph neural network. This second spatial feature vector uses the nodes of the grid graph as the basic unit and can quantify the fault distribution and deviation propagation relationship of each grid node in historical operation.

[0008] Furthermore, the formula for calculating the predicted recognition rate loss value is as follows: ; In the above formula, Represents grid nodes In the predicted time The recognition rate loss value; Indicates the reference lighting intensity; Indicates predicted lighting intensity; This indicates the predicted ambient humidity. Indicates the humidity threshold; This represents the space-time effect function based on historical fault records; Indicates the activation function; , , This represents the adjustable weighting coefficient.

[0009] Furthermore, the predicted lighting intensity is based on the current usage time and maintenance records of the lighting equipment, as well as a preset lighting attenuation model, and combined with the light factor in the weather forecast. It is used to estimate the actual lighting brightness at each grid node in the storage area at a future time point, and is used to characterize the degree of impact of lighting aging on the QR code recognition rate. The predicted environmental humidity is based on real-time collected humidity data of the reservoir area and the rainfall and relative humidity prediction results in the meteorological forecast. It is calculated by time-series extrapolation or physical model to obtain the air relative humidity value at each grid node in the reservoir area at a future time point. It is used to characterize the degree of impact of excessive humidity during the rainy season on the accuracy of the optical sensor.

[0010] Furthermore, the activation function is a nonlinear mapping function used to convert the weighted sum of the inputs into a prediction recognition rate loss value limited to a predetermined interval; the activation function includes, but is not limited to, the Sigmoid function, the ReLU function or its variants, which are used to give the prediction result probabilistic meaning or to normalize the numerical range.

[0011] Furthermore, the space-time influence function based on historical fault records is an empirical function that reflects the law of fault propagation. Its input is the spatial coordinates of the grid node and the predicted time point, and its output is the degree of influence of the node on the historical fault event at a given time. This space-time influence function is constructed by statistically analyzing the spatiotemporal distribution characteristics of historical fault data. It can characterize the cumulative effects and diffusion trends of lighting aging and humidity anomalies in different spatial locations and time windows.

[0012] Furthermore, the dynamic adjustment of the grid map weights includes: For grid nodes whose predicted recognition rate loss exceeds a preset first threshold, their passage cost in the path planning algorithm is increased to infinity, making them obstacles. For raster nodes whose predicted recognition rate loss value falls between the first and second thresholds, the passage time cost is increased proportionally to their loss value, as shown in the following formula: ; In the above formula, This indicates the additional travel time cost required. Indicates the original passage cost; This represents the penalty coefficient.

[0013] This invention provides an intelligent express sorting path optimization method based on big data analysis. Compared with existing technologies, this method achieves the following advantages: 1. This invention systematically collects multi-dimensional environmental time-series data such as lighting intensity, humidity, and historical fault records in the distribution center warehouse area, and establishes a correlation with the historical operating data of AGVs (QR code recognition error rate, number of retries, and path deviation records). This breaks the limitation of traditional path optimization methods that rely solely on single operating data, enabling the path planning system to proactively perceive and quantify environmental risks, and providing a real and reliable data foundation for subsequent prediction and optimization.

[0014] 2. This invention innovatively constructs a joint prediction model for environmental recognition that integrates temporal convolutional networks and graph neural networks. This model can simultaneously capture the evolution of environmental variables over time and the topological propagation relationship of faults in the grid space, achieving high-precision spatiotemporal prediction of the declining trend of QR code recognition rate and significantly improving the ability to predict the impact of complex environments such as aging lighting during the rainy season and excessive humidity.

[0015] 3. This invention dynamically adjusts the passage cost of the grid map based on the predicted recognition rate loss value, and adopts strategies to avoid or increase redundant verification time windows for high-risk nodes. It transforms the traditional static shortest path planning into a dynamic path replanning oriented towards reliability, effectively reducing the recognition failure rate and shutdown risk of AGV in harsh environments, and ensuring the continuity and efficiency of express sorting operations.

[0016] 4. This invention introduces a closed-loop feedback mechanism to collect QR code recognition error rate and path correction data in real time during actual operation, and to perform incremental training and dynamic updating of weight coefficients on the joint prediction model, so that the system has self-evolution and long-term adaptability, avoiding prediction failure caused by environmental drift such as equipment aging and seasonal changes, and continuously optimizing the path decision-making effect. Attached Figure Description

[0017] Figure 1 This is a flowchart of the present invention; Figure 2 In diagram a, the relationship between predicted lighting intensity and time is shown in this invention. Figure 2 Figure b is a graph showing the relationship between predicted environmental humidity and time in this invention; Figure 2 In the diagram, c represents the relationship between the spatiotemporal impact function of historical faults and time in this invention. Figure 2 In the figure, d is a graph showing the relationship between the predicted loss value at grid node (10, 10) and time in this invention; Figure 3 This is a diagram showing the relationship between the predicted recognition rate loss value and the spatial coordinates of the grid nodes in this invention. Figure 4 This is a diagram showing the relationship between the adjusted passage cost and the spatial coordinates of the grid nodes in this invention. Detailed Implementation

[0018] To make the technical solution of the present invention clearer, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] Example 1 like Figure 1 As shown, according to one aspect of the present invention, an intelligent express sorting path optimization method based on big data analysis is provided, comprising the following steps: Step S1: Obtain multi-dimensional environmental time-series data of the distribution center warehouse area. The multi-dimensional environmental time-series data includes lighting intensity time-series data, humidity time-series data, and historical fault record data. The innovation of this step lies in the fact that it is the first time that the multi-dimensional environmental time-series data (lighting intensity, humidity, and historical fault records) of the distribution center warehouse area is used as a pre-input for path optimization, breaking through the limitation of traditional methods that only rely on AGV operation data. The principle is to systematically collect environmental-related time-series data to provide a quantifiable environmental state benchmark for subsequent modeling. The advantage is that it lays the foundation for data-driven environmental perception, enabling path optimization to actively respond to environmental changes rather than passively reacting.

[0020] Step S2: Obtain historical operating data of multiple AGVs performing sorting tasks in the warehouse area. The historical operating data includes QR code recognition error rate, number of retries, and path planning deviation records. The innovation of this step lies in the synchronous collection of historical operating data of AGVs (QR code recognition error rate, number of retries, and path deviation records), and constructing a performance label set corresponding to the environmental data in Step S1. The principle is to establish a statistical correlation between AGV recognition failure and environmental factors by aligning the environment and operating data. The advantage is that it provides a real target value for supervised learning of the subsequent joint prediction model, improving the relevance and trainability of the prediction.

[0021] Step S3: Construct an environmental identification joint prediction model. This model is used to establish a dynamic correlation between multi-dimensional environmental time-series data and AGV historical operation data. The core innovation of this step lies in the construction of the "environmental identification joint prediction model," which for the first time integrates temporal convolutional networks and graph neural networks to establish a dynamic nonlinear correlation between multi-dimensional environmental time-series data and AGV operation data. Its principle is to achieve deep modeling of the environment-performance coupling mechanism through cross-modal feature extraction and fusion. The advantage is that the model can simultaneously capture the temporal evolution law and the spatial topological propagation relationship, significantly improving the prediction ability of complex environmental impacts.

[0022] Step S4: Input the current environmental data and weather forecast data into the joint prediction model for environmental identification to predict the downward trend of QR code recognition rate caused by lighting aging and excessive humidity in the next rainy season, and generate the predicted recognition rate loss value for each grid node. The innovation of this step lies in inputting the current environmental data and weather forecast into the model to predict the downward trend of QR code recognition rate caused by lighting aging and excessive humidity in the next rainy season, and output the recognition rate loss value for each grid node. The principle is to use the trained joint prediction model to perform spatiotemporal extrapolation to achieve forward-looking risk quantification. The advantage is that it enables the system to have "predictive ability", which can identify high-risk areas before problems occur, and provide a basis for decision-making for proactive scheduling.

[0023] Step S5: Based on the predicted recognition rate loss value, dynamically adjust the grid map weights of the sorting center warehouse area and perform global path replanning for multiple AGVs. The innovation of this step lies in dynamically adjusting the grid map weights based on the predicted recognition rate loss value and implementing global path replanning, including avoiding high-loss nodes or allocating redundant verification time windows for necessary nodes. The principle is to convert the predicted loss value into a dynamic passage cost in path planning and adopt a risk-weighted path optimization strategy. The advantage is that it realizes the leap from "static shortest path" to "dynamic reliable path" in path planning, which significantly reduces the risk of recognition failure and AGV stagnation during the rainy season while ensuring sorting efficiency.

[0024] Among them, global path replanning includes avoiding grid nodes whose predicted recognition rate loss value is higher than a preset threshold, or allocating redundant verification time windows for AGVs that pass through grid nodes. Step S6: During the next rainy season, collect real-time data on actual QR code recognition error rates and path replanning corrections; use this real-time data as feedback to incrementally train the joint prediction model for environmental recognition, dynamically updating the weight coefficients and influence functions in the formula for calculating the loss value of the prediction recognition rate. The innovation of this step lies in introducing a closed-loop feedback mechanism, collecting real-time data on actual recognition error rates and path corrections, incrementally training the joint prediction model for environmental recognition, and dynamically updating the weight coefficients and influence functions in the loss function. Its principle is based on online learning to continuously correct the model's prediction bias, enabling the system to adapt to long-term changes in the environment and equipment. The advantage is that the model has self-evolution capabilities; as operational data accumulates, prediction accuracy and path optimization effects continuously improve, avoiding the shortcomings of traditional methods that "train once, fail long-term."

[0025] Example 2 like Figure 1 As shown, according to one aspect of the present invention, a method for constructing an environmental identification joint prediction model includes: Step S301: Use a temporal convolutional network to extract features from the lighting intensity time-series data and humidity time-series data to generate a first environmental feature vector; wherein, the first environmental feature vector is a high-dimensional abstract feature representation that can characterize the dynamic changes of the reservoir area's environmental state over time after the lighting intensity time-series data and humidity time-series data are subjected to multi-scale convolution operations by the temporal convolutional network, and is used to reflect the combined impact of lighting aging and excessive humidity on the identified environment.

[0026] The innovation of this step lies in the first application of Temporal Convolutional Network (TCN) to extract time-series data features of lighting intensity and humidity in the express sorting environment, rather than traditional time-series statistical methods. Its principle is to capture environmental change patterns over different time spans through multi-scale convolutional kernels, generating a first environmental feature vector that can reflect the combined effects of lighting aging and excessive humidity. Compared with RNN or LSTM, TCN has advantages such as strong parallel computing capabilities, stable gradients, and flexible control of the receptive field, ensuring that the model can effectively capture the long-term dependence and abrupt changes in the environmental state of the storage area.

[0027] Step S302: Use a graph neural network to extract spatial topological features from historical fault record data and path planning deviation records to generate a second spatial feature vector. The second spatial feature vector is a low-dimensional embedded feature representation generated by learning the spatial topological structure of historical fault record data and path planning deviation records through a graph neural network. The second spatial feature vector uses the nodes of the grid graph as basic units and can quantify the fault distribution and deviation propagation relationship of each grid node in historical operation.

[0028] The innovation of this step lies in using a graph neural network (GNN) to perform spatial topology learning on historical fault records and path planning deviation records, breaking through the limitation of traditional methods that ignore the spatial propagation effect of faults. The principle is to regard grid map nodes as graph vertices and use the message passing mechanism of GNN to quantify the fault distribution and deviation propagation relationship of each node in historical operation, generating a second spatial feature vector. The advantage is that it can explicitly model the mutual influence of faults between adjacent grid nodes, enabling the model to have spatial generalization ability. Even nodes that have not directly experienced faults can infer potential risks through the graph structure.

[0029] Step S303: Perform cross-modal fusion of the first environmental feature vector and the second spatial feature vector to generate a joint feature representation. The innovation of this step lies in the first cross-modal fusion of temporal features (first environmental feature vector) and spatial features (second spatial feature vector) to form a joint feature representation. The principle is to use an attention mechanism or a combination of splicing and linear transformation to enable the model to automatically learn the interaction and coupling relationship between the two while retaining the independent information of time and space. The advantage is that the joint features can simultaneously reflect the "deterioration trend of the environment over time" and the "spreading law of the fault in space", providing a more comprehensive and robust input representation for subsequent predictions and significantly improving the expressive power of the model.

[0030] Step S304: Using the QR code recognition error rate as the prediction target, train a regression predictor based on the joint feature representation to obtain the joint prediction model for environmental recognition. The innovation of this step lies in using the QR code recognition error rate as the direct prediction target and training a regression predictor based on the joint feature representation to construct a complete joint prediction model for environmental recognition. Its principle is to use the learned joint features as input and establish a nonlinear mapping from environmental-spatial features to the recognition error rate through a regressor (such as a fully connected network or XGBoost). The advantage is that it achieves end-to-end prediction from multi-source heterogeneous data to quantifiable risk values, enabling the model not only to describe the past but also to accurately estimate the recognition performance at different time-space nodes in the future, providing a reliable basis for subsequent dynamic path planning.

[0031] Example 3 like Figure 2 As shown, according to one aspect of the present invention, the predicted lighting intensity is a value estimated based on the current usage time of the lighting equipment, maintenance records, and a preset lighting attenuation model, combined with the illumination factor in the weather forecast, to estimate the actual lighting brightness at each grid node in the reservoir area at a future time point, used to characterize the degree of influence of lighting aging on the QR code recognition rate; the predicted environmental humidity is a value calculated based on real-time collected humidity data of the reservoir area and the rainfall and relative humidity prediction results in the weather forecast, through time-series extrapolation or a physical model, to determine the relative humidity value at each grid node in the reservoir area at a future time point, used to characterize the degree of influence of excessive humidity during the rainy season on the accuracy of the optical sensor.

[0032] In this embodiment, the activation function is a nonlinear mapping function used to convert the weighted sum of the inputs into a prediction recognition rate loss value limited to a predetermined interval. The spatial-temporal influence function based on historical fault records is an empirical function reflecting the fault propagation pattern. Its inputs are the spatial coordinates of the grid node and the prediction time point, and its output is the degree to which the node is affected by historical fault events at a given time.

[0033] In intelligent express sorting scenarios, AGVs rely on ground QR codes for navigation and positioning. QR code recognition rates are affected by a variety of environmental factors. Based on historical data analysis and physical mechanisms, the main reasons for the decline in recognition rates include: the decay of warehouse lighting over time, resulting in insufficient illumination on the QR code surface; and increased humidity during the rainy season causing condensation on the QR code surface and fogging of the optical sensor lens, introducing recognition errors.

[0034] in, This indicates the design baseline lighting intensity for the storage area (unit: Lux), which is the minimum illuminance required to ensure proper QR code recognition. Indicates the predicted time Below is the actual lighting intensity (unit: Lux) of the current lighting equipment after aging and degradation. Figure 2 As shown in 'a'; In the predicted time The ambient relative humidity (unit: %). This indicates the upper humidity threshold for the normal operation of a QR code recognition system (typically 80%–90%, depending on the specific device model). Figure 2 As shown in b; Represents the spatial coordinates of a grid node; This indicates a predicted time point (usually a specific moment during the next rainy season, such as a certain day and time).

[0035] The specific process of obtaining the above formula is as follows: I. The effect of lighting intensity on recognition rate is generally positively correlated: the closer the illuminance is to or exceeds the reference value, the higher the recognition rate; when the illuminance is below the reference value, the recognition rate decreases. Define the lighting loss factor. for: ; when ,otherwise The value of this factor ranges from [0, 1], with a larger value indicating more severe lighting insufficiency. Because the lighting aging process is non-linear (e.g., the light decay of lamps accelerates at the end of their lifespan), a coefficient is introduced. Weighting, This data was obtained by fitting historical data and reflects the sensitivity of recognition rate loss to decreased illumination.

[0036] II. Humidity has a "threshold effect" on optical sensors and QR code surfaces: when the humidity is below a certain level... The effect of humidity is negligible; however, once the threshold is exceeded, the recognition rate drops rapidly with increasing humidity. Define a humidity loss factor. for: ; when ,otherwise Similarly, weighting coefficients are introduced. The intensity of the effect of humidity regulation. It can be obtained through experimental calibration, that is, by testing the curve of QR code recognition rate as a function of humidity in a humidity control box and fitting it.

[0037] Third, the recognition problem of some nodes is not only determined by the current environment, but also related to past failure modes. For example, if a node's QR code is damaged due to long-term water accumulation, its recognition rate will still be low even if the current humidity is not high. This spatiotemporal correlation requires a function... To express. Definition as follows: ; In the above formula, This represents the total number of historical failure events. Indicates the first Spatial coordinates and time of the secondary failure event; Indicates the spatial influence radius (determined based on the correlation between adjacent grids, for example, the side length of 2 to 5 grids); This represents the time-dependent decay constant (reflecting the rate at which the impact of a fault decays over time, e.g., by day or week). Indicates the first The weight of each failure event can be set according to the severity of the event (such as the number of consecutive failures or the duration of AGV downtime).

[0038] The physical meaning of this function is that each historical failure event contributes to the additional recognition rate loss of nodes within a certain spatial range around it and over a subsequent period of time. The contribution value decays exponentially with the increase of spatial distance and time interval. The range of values ​​is controlled within the interval [0, 1] through normalization, and a coefficient is introduced. Adjust its overall influence, such as Figure 2 As shown in c in the figure.

[0039] IV. The three factors mentioned above (lighting, humidity, and historical faults) are independent of each other but work together to affect the recognition rate loss. Under the assumption of no interactive coupling, the comprehensive loss can be achieved through linear superposition or nonlinear combination. Considering the synergistic effect among the three factors in practice (e.g., the loss under low illumination and high humidity is much greater than the sum of their individual effects), this patent adopts a nonlinear combination with an activation function to enhance the model's expressive power. The initial weighted sum ∑ is defined as: ; The first two items have already been defined in the first step (note that when...). and (The corresponding term is 0). Since ∑ may exceed the interval [0, 1], and the actual recognition rate loss should not increase indefinitely (the upper limit is 1, representing complete failure to recognize), an activation function is introduced to map it to [0, 1]. Commonly used activation functions include: ; when The output value is between 0.5 and 1. In this embodiment, the Sigmoid function is preferred to obtain smooth saturation characteristics, which physically avoids the loss value from exceeding 1.

[0040] like Figure 2 As shown in d, the final formula for calculating the prediction recognition rate loss is: ; In the above formula, Represents grid nodes In the predicted time The recognition rate loss value; Indicates the reference lighting intensity; Indicates predicted lighting intensity; This indicates the predicted ambient humidity. Indicates the humidity threshold; This represents the space-time effect function based on historical fault records; Indicates the activation function; , , This represents the adjustable weighting coefficient.

[0041] Example 4 like Figure 3 , Figure 4 As shown, according to one aspect of the present invention, dynamically adjusting the grid map weights includes: For grid nodes whose predicted recognition rate loss exceeds a preset first threshold, their passage cost in the path planning algorithm is increased to infinity, making them obstacles. For raster nodes whose predicted recognition rate loss value falls between the first and second thresholds, the passage time cost is increased proportionally to their loss value, as shown in the following formula: ; In the above formula, This indicates the additional travel time cost required. Indicates the original passage cost; This represents the penalty coefficient. The specific process for obtaining the above formula is as follows: I. In intelligent express sorting and distribution centers, AGV path planning is based on a grid map. Each grid node... It has a theoretical passage cost in its original state. This cost typically consists of the physical distance to the node, the historical average travel time, or the preset path cost.

[0042] When predicted in time At that time, node A decrease in QR code recognition rate due to aging lighting and excessive humidity will result in a predicted recognition rate loss value. The larger the loss value, the higher the risk of the AGV failing to recognize the QR code, retrying, or deviating from the path at that node. Therefore, it is necessary to dynamically increase the passage cost of that node to guide the path planning algorithm to prioritize more reliable nodes.

[0043] II. The increment of the passage cost should satisfy the following: The larger the value, the greater the cost increment; the cost can only increase or remain unchanged, not decrease (because increasing risk does not reduce the toll cost); and there exists a control coefficient to balance the weights of the original cost and the risk penalty. To maintain the stability of the path planning algorithm, linear or affine transformations are usually used. Therefore, the simplest form can be set as: ; In the above formula, It is a non-negative monotonically increasing function, and satisfies (No additional cost is incurred when there is no predicted loss).

[0044] III. Choosing the Linear Form The degree of risk is proportional to the increase in cost, which facilitates engineering implementation and manual calibration; and The penalty coefficient (unit: dimensionless or time / distance conversion factor) reflects the decision-maker's sensitivity to recognition rate risk. The larger the value, the more the system tends to avoid high-risk areas. Therefore: ; ; ; The formula is derived based on the fundamental principles of risk-weighted path planning, quantifying the predicted risk of identification failure as additional travel time or distance cost, and using a linear penalty factor as the decision variable. Final form While maintaining the complexity of path calculation, it actively avoids problems such as rainy season and aging lighting, effectively reducing the actual failure rate of AGV when performing sorting tasks.

[0045] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for optimizing intelligent express sorting routes based on big data analysis, characterized in that, Includes the following steps: Step S1: Obtain multi-dimensional environmental time-series data of the distribution center warehouse area. The multi-dimensional environmental time-series data includes lighting intensity time-series data, humidity time-series data, and historical fault record data. Step S2: Obtain historical operating data of multiple AGVs performing sorting tasks in the warehouse area. The historical operating data includes QR code recognition error rate, number of retries, and path planning deviation records. Step S3: Construct an environmental identification joint prediction model, which is used to establish a dynamic correlation between multi-dimensional environmental time-series data and the historical operation data of AGV; Step S4: Input the current environmental data and weather forecast data into the environmental recognition joint prediction model to predict the downward trend of QR code recognition rate caused by lighting aging and excessive humidity in the next rainy season, and generate the predicted recognition rate loss value of each grid node. Step S5: Based on the predicted recognition rate loss value, dynamically adjust the grid map weight of the distribution center warehouse area and perform global path replanning for multiple AGVs. The global path replanning includes avoiding grid nodes whose predicted recognition rate loss value is higher than a preset threshold, or allocating redundant verification time windows for AGVs passing through the grid nodes. Step S6: During the next rainy season, collect real-time data on the actual QR code recognition error rate and path replanning correction; use the actual data as feedback to incrementally train the environmental recognition joint prediction model, and dynamically update the weight coefficients and influence functions in the calculation formula of the prediction recognition rate loss value.

2. The intelligent express sorting path optimization method based on big data analysis according to claim 1, characterized in that: The method for constructing the joint prediction model for environmental identification includes: Step S301: Use a temporal convolutional network to extract features from the lighting intensity time series data and humidity time series data to generate a first environmental feature vector; Step S302: Use a graph neural network to extract spatial topological features from the historical fault record data and path planning deviation records to generate a second spatial feature vector; Step S303: Perform cross-modal fusion of the first environmental feature vector and the second spatial feature vector to generate a joint feature representation; Step S304: Using the QR code recognition error rate as the prediction target, train a regression predictor based on the joint feature representation to obtain the joint prediction model for environmental recognition.

3. The intelligent express sorting path optimization method based on big data analysis according to claim 2, characterized in that: The first environmental feature vector is a high-dimensional abstract feature representation that can characterize the dynamic changes of the reservoir area's environmental state over time after performing multi-scale convolution operations on the lighting intensity time-series data and humidity time-series data through a time convolutional network. It is used to reflect the combined impact of lighting aging and excessive humidity on the identified environment.

4. The intelligent express sorting path optimization method based on big data analysis according to claim 2, characterized in that: The second spatial feature vector is a low-dimensional embedded feature representation generated by learning the spatial topology of historical fault record data and path planning deviation records through a graph neural network. This second spatial feature vector uses the nodes of the grid graph as the basic unit and can quantify the fault distribution and deviation propagation relationship of each grid node in historical operation.

5. The intelligent express sorting path optimization method based on big data analysis according to claim 1, characterized in that: The formula for calculating the predicted recognition rate loss value is as follows: ； In the above formula, Represents grid nodes In the predicted time The recognition rate loss value; Indicates the reference lighting intensity; Indicates predicted lighting intensity; This indicates the predicted ambient humidity. Indicates the humidity threshold; This represents the space-time effect function based on historical fault records; Indicates the activation function; , , This represents the adjustable weighting coefficient.

6. The intelligent express sorting path optimization method based on big data analysis according to claim 5, characterized in that: The predicted lighting intensity is a value estimated based on the current usage time and maintenance records of the lighting equipment, as well as a preset lighting attenuation model, and combined with the light factor in the weather forecast. It is used to characterize the degree of impact of lighting aging on the QR code recognition rate. The predicted environmental humidity is based on real-time collected humidity data of the reservoir area and the rainfall and relative humidity prediction results in the meteorological forecast. It is calculated by time-series extrapolation or physical model to obtain the air relative humidity value at each grid node in the reservoir area at a future time point. It is used to characterize the degree of impact of excessive humidity during the rainy season on the accuracy of the optical sensor.

7. The intelligent express sorting path optimization method based on big data analysis according to claim 5, characterized in that: The activation function is a non-linear mapping function used to convert the weighted sum of the inputs into a prediction recognition rate loss value limited to a predetermined interval.

8. The intelligent express sorting path optimization method based on big data analysis according to claim 5, characterized in that: The space-time influence function based on historical fault records is an empirical function that reflects the law of fault propagation. Its input is the spatial coordinates of the grid node and the predicted time point, and its output is the degree of influence of the node on the historical fault event at a given time.

9. The intelligent express sorting path optimization method based on big data analysis according to claim 5, characterized in that: The dynamic adjustment of raster map weights includes: For grid nodes whose predicted recognition rate loss exceeds a preset first threshold, their passage cost in the path planning algorithm is increased to infinity, making them obstacles. For raster nodes whose predicted recognition rate loss value falls between the first and second thresholds, the passage time cost is increased proportionally to their loss value, as shown in the following formula: ； In the above formula, This indicates the additional travel time cost required. Indicates the original passage cost; This represents the penalty coefficient.

Images