Logistics information monitoring method and system based on artificial intelligence

Abstract

The application provides an artificial intelligence-based logistics information monitoring method and system, which comprises the following steps: collecting logistics data in a logistics transportation process, performing feature extraction on the logistics data to obtain logistics data features; inputting the logistics data features into a pre-trained artificial intelligence model to obtain a logistics state prediction result; obtaining an actual logistics state, performing deviation calculation on the logistics state prediction result and the actual logistics state to obtain logistics state deviation data; performing abnormal situation determination on the logistics state deviation data by using a decision tree algorithm to obtain a logistics situation determination result; and if the logistics situation determination result is abnormal, pushing abnormal information to a related personnel terminal and issuing a logistics abnormality alarm. In the application, the defect of insufficient accuracy in the current logistics detection process is overcome.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, in particular to a logistics information monitoring method and system based on artificial intelligence. BACKGROUND

[0002] With the rapid development and intelligent transformation of the logistics industry, real-time and accurate information monitoring of the logistics transportation process has become a key link to ensure logistics efficiency and service quality. Traditional logistics information monitoring methods mainly rely on manual inspection and simple rule matching systems, which have problems such as low monitoring efficiency, high labor cost, and inability to cope with complex scenarios.

[0003] In the logistics state prediction process, there is a lack of deviation calculation of the prediction result and the actual state, which cannot quantify the actual influence degree and potential risk of the deviation; and in the abnormal situation judgment link, there are often misjudgments or omissions, which are difficult to meet the high-precision requirements of the logistics industry for abnormal monitoring. SUMMARY

[0004] The main purpose of the present application is to provide a logistics information monitoring method and system based on artificial intelligence, which aims to overcome the defects of insufficient accuracy in the current logistics detection process.

[0005] To achieve the above purpose, the present application provides a logistics information monitoring method based on artificial intelligence, comprising the following steps:

[0006] Collecting logistics data in the logistics transportation process, extracting features from the logistics data to obtain logistics data features;

[0007] Inputting the logistics data features into a pre-trained artificial intelligence model to obtain a logistics state prediction result;

[0008] Obtaining the actual logistics state, calculating the deviation of the logistics state prediction result and the actual logistics state to obtain logistics state deviation data;

[0009] According to the logistics state deviation data, using a decision tree algorithm to determine abnormal situations to obtain a logistics situation determination result;

[0010] If the logistics situation determination result is abnormal, push the abnormal information to the related personnel terminal and issue a logistics abnormality alarm.

[0011] Further, the logistics data includes logistics document information, sensor collected data, and transportation vehicle GPS positioning data.

[0012] Further, the deviation calculation of the logistics state prediction result and the actual logistics state to obtain the logistics state deviation data comprises:

[0013] performing time series decomposition on the logistics state prediction result to separate time series decomposition components, including a trend item, a periodic item, and a residual item;

[0014] performing similarity calculation on the time series decomposition components of the prediction result and corresponding components of the actual logistics state based on a dynamic time warping distance measurement model to obtain a time warping similarity;

[0015] calculating a statistical distance of the logistics state prediction result and the actual logistics state in a feature space based on a Mahalanobis distance algorithm to obtain a Mahalanobis distance;

[0016] fusing the time warping similarity and the Mahalanobis distance to construct a bias confidence function and output a bias confidence;

[0017] performing clustering analysis on the bias confidence by a fuzzy C-means clustering algorithm to generate a bias heat map containing a space-time dimension;

[0018] performing morphological filtering on the bias heat map to extract a bias core region and calculate a space-time diffusion index thereof;

[0019] comprehensively taking the time warping similarity, the Mahalanobis distance, the bias confidence, and the space-time diffusion index as the logistics state bias data.

[0020] Further, according to the logistics state bias data, an abnormal situation is determined by a decision tree algorithm to obtain a logistics situation determination result, including:

[0021] constructing a training sample, the training sample including a multi-dimensional feature vector of logistics business rules, the multi-dimensional feature vector including a time warping similarity, a Mahalanobis distance, a bias confidence, and a space-time diffusion index;

[0022] performing weight distribution on the multi-dimensional feature vector by a feature importance sorting algorithm to fuse and generate a weighted feature matrix;

[0023] recursively dividing the weighted feature matrix by a CART decision tree algorithm based on Gini index optimization to construct an abnormality determination decision tree model;

[0024] performing semantic labeling on leaf nodes of the abnormality determination decision tree model by a knowledge base to generate a rule set including an abnormal type, a severity level, and a disposal priority;

[0025] obtaining a pre-constructed decision tree dynamic pruning mechanism based on reinforcement learning, and adjusting a branch structure of the abnormality determination decision tree model in real time according to a historical determination accuracy to obtain a trained decision tree model;

[0026] The logistics state deviation data is input into the trained decision tree model, and an abnormal thermodynamic map containing a time-space dimension and a treatment priority ranking result are output as a logistics condition determination result.

[0027] Further, the logistics data features are input into a pre-trained artificial intelligence model to obtain logistics state prediction results, including:

[0028] The logistics data features are subjected to spatio-temporal tensor decomposition to extract trend tensors, periodic tensors and residual tensors;

[0029] The trend tensors, periodic tensors and residual tensors are subjected to feature enhancement by a spatio-temporal graph attention network to generate a spatio-temporal correlation feature matrix;

[0030] The spatio-temporal correlation feature matrix is subjected to time series prediction based on a multi-scale causal convolution network to output a multi-step probability prediction distribution;

[0031] The multi-step probability prediction distribution is subjected to uncertainty quantification processing to obtain a confidence interval prediction result;

[0032] The confidence interval prediction result is subjected to logistics rule constraint processing to generate a logistics state prediction result conforming to business logic.

[0033] Further, the logistics data features are input into a pre-trained artificial intelligence model to obtain logistics state prediction results, including:

[0034] The logistics data features are subjected to quantum state encoding to construct a logistics feature quantum superposition state representation;

[0035] The logistics feature quantum superposition state representation is subjected to feature correlation enhancement processing by a quantum entanglement algorithm to generate a high-dimensional feature entanglement tensor;

[0036] The high-dimensional feature entanglement tensor is subjected to causal structure learning processing to identify a key causal relationship network in the logistics system;

[0037] A quantum-inspired Monte Carlo tree search algorithm is used to perform multi-step prediction path exploration based on the key causal relationship network to generate a logistics state probability tree containing a time-space dimension;

[0038] A quantum measurement simulation technique is used to collapse the logistics state probability tree to obtain a logistics state prediction distribution;

[0039] The logistics state prediction distribution is subjected to logistics rule verification to generate a logistics state prediction result conforming to business logic constraints.

[0040] The present application also provides a logistics information monitoring system based on artificial intelligence, comprising:

[0041] The collection module is configured to collect logistics data in a logistics transportation process, perform feature extraction on the logistics data, and obtain logistics data features;

[0042] The prediction module is configured to input the logistics data features into a pre-trained artificial intelligence model, and obtain a logistics state prediction result;

[0043] The calculation module is configured to obtain an actual logistics state, perform deviation calculation on the logistics state prediction result and the actual logistics state, and obtain logistics state deviation data;

[0044] The determination module is configured to perform abnormal situation determination on the logistics state deviation data by using a decision tree algorithm, and obtain a logistics situation determination result;

[0045] The alarm module is configured to push abnormal information to a terminal of a relevant person if the logistics situation determination result is abnormal, and issue a logistics abnormality alarm.

[0046] The present application also provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method according to any one of the preceding embodiments when executing the computer program.

[0047] The present application also provides a computer readable storage medium storing a computer program, wherein the computer program is executed by a processor to implement the steps of the method according to any one of the preceding embodiments.

[0048] The present application provides a logistics information monitoring method and system based on artificial intelligence, which comprises the following steps: collecting logistics data in a logistics transportation process, performing feature extraction on the logistics data, and obtaining logistics data features; inputting the logistics data features into a pre-trained artificial intelligence model, and obtaining a logistics state prediction result; obtaining an actual logistics state, performing deviation calculation on the logistics state prediction result and the actual logistics state, and obtaining logistics state deviation data; performing abnormal situation determination on the logistics state deviation data by using a decision tree algorithm, and obtaining a logistics situation determination result; and pushing abnormal information to a terminal of a relevant person if the logistics situation determination result is abnormal, and issuing a logistics abnormality alarm. In the present application, the deviation calculation on the logistics state prediction result and the actual logistics state can quantify the actual influence degree and potential risk of the deviation, and then the decision tree algorithm is used to perform abnormal situation determination, so that a precise logistics situation determination result is obtained, and the defect of insufficient accuracy in the current logistics detection process is overcome. BRIEF DESCRIPTION OF DRAWINGS

[0049] Figure 1 is a step schematic diagram of the logistics information monitoring method based on artificial intelligence in an embodiment of the present application;

[0050] Figure 2 is a structural block diagram of an artificial intelligence-based logistics information monitoring system in an embodiment of the present application;

[0051] Figure 3 is a structural schematic block diagram of a computer device in an embodiment of the present application.

[0052] The implementation, functional features and advantages of the present application will be further described with reference to the accompanying drawings in conjunction with embodiments. DETAILED DESCRIPTION

[0053] In order to make the purpose, technical solutions and advantages of the present application clearer and more apparent, the present application will be further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0054] Referring to Figure 1 , an artificial intelligence-based logistics information monitoring method is provided in an embodiment of the present application, comprising the following steps:

[0055] Step S1, collecting logistics data in the logistics transportation process, performing feature extraction on the logistics data to obtain logistics data features;

[0056] Step S2, inputting the logistics data features into a pre-trained artificial intelligence model to obtain a logistics state prediction result;

[0057] Step S3, obtaining an actual logistics state, performing deviation calculation on the logistics state prediction result and the actual logistics state to obtain logistics state deviation data;

[0058] Step S4, performing abnormal situation judgment using a decision tree algorithm according to the logistics state deviation data to obtain a logistics situation judgment result;

[0059] Step S5, if the logistics situation judgment result is abnormal, pushing abnormal information to a related personnel terminal and issuing a logistics abnormality alarm.

[0060] In the present embodiment, as described in step S1 above, first, various data collection devices and systems are used to comprehensively collect multi-source data generated in the logistics transportation process, which covers GPS positioning information of the transportation vehicle, cargo state sensor data, logistics order and scheduling information, etc. Due to the heterogeneity and redundancy of the original logistics data, direct use cannot meet the subsequent analysis requirements, so the collected logistics data needs to be processed using a specific feature extraction algorithm. This processing process removes noise and irrelevant data by mining key information in the data, extracts core features that can effectively represent the logistics transportation state, and thus obtains structured and refined logistics data features, laying a foundation for subsequent analysis.

[0061] As described in step S2 above, the logistics data features obtained in step S1 are input into an artificial intelligence model that has been trained on a large amount of historical logistics data. The model, which has been optimized and adjusted in depth, has the ability to perform complex pattern recognition and relationship analysis on logistics data features. Through the multi-layer operation and learning mechanism within the model, it can predict future or current logistics states based on the input feature data and output logistics state prediction results containing transportation timeliness, cargo loss probability, path deviation possibility, etc., providing forward-looking information for logistics operation decision-making.

[0062] As described in step S3 above, the actual state data during logistics transportation is obtained through real-time monitoring equipment and systems, which reflects the true situation of logistics activities. In order to evaluate the accuracy of the prediction results of the artificial intelligence model, the logistics state prediction results obtained in step S2 are compared with the actual logistics state data for deviation calculation. This calculation process uses professional data measurement methods to quantify the difference between the prediction results and the actual situation from multiple dimensions such as time, space, and cargo state, thereby obtaining logistics state deviation data that can fully reflect the prediction error, which provides a quantitative basis for subsequent anomaly determination.

[0063] As described in step S4 above, based on the logistics state deviation data obtained in step S3, a decision tree algorithm is used to determine abnormal situations. The decision tree algorithm constructs a tree-shaped decision model with each feature in the deviation data as a decision node, and makes layer-by-layer judgments on whether the logistics state is abnormal according to pre-set classification rules and thresholds. The decision tree algorithm can handle complex multi-dimensional data relationships and identify data feature combinations that meet abnormal patterns through deep analysis of deviation data, thereby outputting logistics situation determination results containing abnormal type, occurrence probability, etc., and achieving accurate identification of logistics transportation abnormalities.

[0064] As described in step S5 above, when the logistics situation determination result obtained in step S4 shows that there is an abnormality, the system will automatically trigger an abnormal information push mechanism. This mechanism communicates with the terminal devices (such as mobile phones, computers, smart terminals, etc.) of relevant personnel, and according to pre-set priority and notification strategies, pushes alarm information containing detailed information of the abnormality (such as abnormal occurrence time, location, type, potential impact, etc.) to relevant personnel, ensuring that they can obtain the abnormal situation in a timely manner and take targeted measures to reduce the negative impact of the abnormality on logistics transportation and ensure the stability and safety of logistics operations.

[0065] In an embodiment, the logistics data includes logistics document information, sensor collection data, and transportation vehicle GPS positioning data.

[0066] In the present embodiment, the aforementioned logistics document information includes order number, shipper information, consignee information, cargo name, specification, quantity, weight, volume, shipping address, consignee address, transportation mode, transportation plan time, etc. These information clearly shows the basic elements and expected process of logistics business, and is the core record of logistics activities. Through analysis of the logistics document information, the source, destination, transportation requirements and expected transportation time node of the goods can be mastered, thereby providing basic data for logistics resource scheduling, path planning, time efficiency prediction, etc. For example, according to the shipping address and consignee address of the order, the optimal transportation route can be planned; according to the transportation plan time, the actual transportation progress can be compared and monitored.

[0067] The aforementioned sensor collected data is obtained by various sensors installed in the goods packaging, transportation vehicles or warehouse environment, such as temperature and humidity sensors, vibration sensors, weight sensors, gas sensors, etc. These data reflect the physical environment and state changes of the goods in real time. When transporting precision instruments, the data collected by the vibration sensor can determine whether there is violent vibration during transportation to avoid damage to the goods; in cold chain logistics, the data collected by the temperature and humidity sensor can ensure that the goods are always in a suitable storage and transportation environment, and the quality of the goods is guaranteed. Sensor collected data provides direct and key information for goods state monitoring, quality assurance and abnormal warning.

[0068] The aforementioned GPS positioning data of the transportation vehicle provides real-time geographic location, driving speed, driving direction, driving trajectory, etc. of the transportation vehicle. Through processing and analysis of these data, the running state and position of the vehicle can be mastered in real time, and whether the vehicle is driving according to the predetermined route, whether there is delay, deviation, etc. can be determined. In the event of a sudden traffic situation, the transportation route can be adjusted in time according to the GPS positioning data to ensure that the goods are delivered on time; at the same time, it is also helpful for logistics enterprises to efficiently manage the vehicle fleet, optimize vehicle scheduling and improve transportation efficiency.

[0069] In an embodiment, a deviation calculation is performed on the logistics state prediction result and the actual logistics state to obtain logistics state deviation data, including:

[0070] A time series decomposition is performed on the logistics state prediction result to separate out time series decomposition components, including a trend item, a periodic item and a residual item;

[0071] Based on a dynamic time warping distance measurement model, a similarity calculation is performed on the time series decomposition components of the prediction result and the corresponding components of the actual logistics state to obtain a time warping similarity;

[0072] Based on a Mahalanobis distance algorithm, a statistical distance of the logistics state prediction result and the actual logistics state in a feature space is calculated to obtain a Mahalanobis distance;

[0073] Fusing the time regularized similarity and Mahalanobis distance, constructing a bias confidence function, and outputting a bias confidence;

[0074] Performing clustering analysis on the bias confidence through a fuzzy C-means clustering algorithm to generate a bias heat map containing a time-space dimension;

[0075] Performing morphological filtering on the bias heat map, extracting a bias core region, and calculating a time-space diffusion index thereof;

[0076] Taking the time regularized similarity, Mahalanobis distance, bias confidence, and time-space diffusion index as a whole as the logistics state bias data.

[0077] In this embodiment, during the logistics transportation process, various state indicators (such as the amount of goods transported and the speed of vehicle travel) will change dynamically over time, and the logistics state prediction result also presents complex time series characteristics. To further analyze these characteristics, a time series decomposition technique is used to process the prediction result. Specifically, an algorithm based on a statistical model or frequency domain analysis, such as the Holt-Winters seasonal decomposition method or Fourier transform, is used to decompose the original logistics state prediction time series into three core components.

[0078] Trend component: By analyzing the long-term trend of historical data, the overall change direction of the logistics state over time is extracted. For example, during the e-commerce promotion period, the logistics order volume usually presents a continuous growth trend, and the trend component can accurately capture this growth trend, providing a basis for evaluating the long-term carrying capacity and development trend of the logistics system.

[0079] Periodic component: Identifies periodically recurring patterns in the data, which may be daily, weekly, or monthly regular fluctuations. Taking urban distribution as an example, the vehicle transportation efficiency during the morning and evening peak hours presents a periodic decrease, and the periodic component can effectively extract this periodic change feature to help predict the fluctuation of the logistics state in a specific time period.

[0080] Residual component: Contains random fluctuations or abnormal disturbances that cannot be explained by the trend component and the periodic component, which may be caused by sudden traffic control, weather changes, and other occasional factors. Separating the residual component helps to more accurately locate and analyze the impact of abnormal events on the logistics state.

[0081] In logistics data, due to various uncertain factors in the transportation process (such as traffic congestion, loading and unloading delay, etc.), the predicted results and the actual state data often cannot be strictly aligned on the time axis, and the traditional Euclidean distance is difficult to accurately measure the similarity between the two. The dynamic time warping (DTW) distance measurement model constructs a two-dimensional matrix, where each element in the matrix represents the distance between the predicted sequence and the actual sequence at the corresponding time point. Then, a dynamic programming algorithm is used to search for an optimal path in the matrix, which allows the time axis to be stretched and bent, so that the two sequences are best matched overall.

[0082] In specific operation, the trend item, the periodic item and the residual item of the predicted result and the corresponding components of the actual logistics state are respectively calculated by DTW. For example, when comparing the predicted vehicle speed trend with the actual speed trend, the DTW algorithm dynamically adjusts the comparison path according to the possible time offset in the actual driving process, such as acceleration and deceleration, and finally calculates a value that comprehensively reflects the similarity between the two trend sequences, i.e. the time warping similarity. The higher the similarity value, the better the matching degree between the predicted result and the actual state in the corresponding component.

[0083] Logistics state data usually contains multiple interrelated feature dimensions, such as transportation time, cargo weight, driving route, etc. In the feature space, Euclidean distance does not consider the correlation and dimension difference between features, which may lead to misjudgment of the similarity between data points. Mahalanobis distance algorithm introduces the covariance matrix of sample data to standardize the feature space, eliminating the influence of dimension and considering the correlation between features. In the calculation process, first, the mean vector and covariance matrix of each feature dimension are calculated according to the historical logistics data. Then, for a given logistics state prediction result vector and actual state vector, the distance between them in the standardized feature space is calculated by the Mahalanobis distance formula. For example, when analyzing the logistics state of different transport vehicles, Mahalanobis distance can consider the correlation between vehicle speed, fuel consumption, load and other features, accurately measure the deviation between predicted state and actual state in the overall feature combination, and more effectively identify potential abnormal situations.

[0084] The time warping similarity reflects the similarity between the prediction and the actual situation from the perspective of pattern matching of time series, while the Mahalanobis distance measures the difference between the two from the perspective of statistical deviation of feature space. In order to more comprehensively and accurately evaluate the credibility of the deviation of the logistics state prediction, it is necessary to fuse these two indicators. When constructing the deviation confidence function, weighted summation, nonlinear function mapping (such as neural network, support vector machine) or fusion method based on evidence theory are used. For example, by weighted summation, according to the characteristics of the actual logistics scene and historical data analysis, different weight coefficients are given to the time warping similarity and the Mahalanobis distance. The setting of the weight aims to highlight the importance of different indicators in a specific scenario. Substitute the two indicators into the function for calculation, and output a deviation confidence value between 0 and 1. The closer the value is to 1, the higher the deviation confidence between the prediction result and the actual state, that is, the more significant the difference between the two, and the more likely there is an abnormal situation; on the contrary, the closer the value is to 0, the lower the deviation confidence, and the more consistent the prediction result with the actual state.

[0085] Traditional hard clustering algorithms require each data point to belong to only one cluster, but in logistics deviation data, due to the uncertainty and complexity of data, the deviation of a certain region or time point may have multiple degrees of characteristics. Therefore, it is more appropriate to use fuzzy C-means clustering algorithm. This algorithm optimizes the objective function through iteration, calculates the membership of each data point (i.e. each deviation confidence value) to different cluster centers, so that data points can belong to multiple clusters with different probabilities at the same time.

[0086] In the logistics scene, the deviation confidence data is divided according to time and space dimensions, for example, using hours as the time unit and geographical grid as the spatial unit. Cluster analysis is performed on the deviation confidence in each spatio-temporal unit to determine different cluster centers representing different levels of deviation. According to the membership of data points to each cluster center, a corresponding color value is assigned to each spatio-temporal unit to generate a deviation heat map. The darker the color, the higher the deviation confidence in the spatio-temporal unit, i.e. the greater the possibility of abnormal logistics state. Through the heat map, logistics managers can intuitively understand the distribution of deviation in the spatio-temporal dimension and quickly locate the high-risk areas and time periods.

[0087] The bias heat map may be disturbed by noise during generation, resulting in some isolated high-bias points or discontinuous areas in the image, which will affect the identification of real abnormal areas. Morphological filtering is based on the principle of mathematical morphology, and uses a predefined structural element (such as a square or circle) to perform erosion and dilation operations on the bias heat map. The erosion operation removes pixels on the image boundary that do not conform to the shape of the structural element, eliminating isolated high-bias noise points; the dilation operation adds pixels to connect adjacent high-bias areas, making the real abnormal area more complete and continuous. After morphological filtering, threshold segmentation or region growing algorithms are used to extract the bias core area, i.e., the area with high bias confidence and spatial continuity.

[0088] To further evaluate the severity and development trend of the abnormal situation, the spatiotemporal diffusion index of the bias core area is calculated. In the time dimension, the change of the core area over time is analyzed, such as calculating the growth rate of the area, the moving speed of the core position, etc.; in the spatial dimension, the expansion range and direction of the area in geographical space are measured. Through the spatiotemporal diffusion index, the range and duration of the abnormal situation that may be affected can be predicted, providing more forward-looking decision-making basis for logistics scheduling and emergency handling.

[0089] The time regularization similarity reflects the matching degree of the prediction result and the actual state in the time series pattern, the Mahalanobis distance reflects the statistical deviation of the two in the feature space, the bias confidence comprehensively evaluates the reliability of the bias, and the spatiotemporal diffusion index describes the spatiotemporal development trend of the abnormal situation. These four indicators comprehensively characterize the bias features between the logistics state prediction result and the actual state from different angles. Integrating them into logistics state bias data forms a multi-dimensional information set. Logistics managers and decision systems can not only determine whether there is a bias, but also in-depth analyze the causes, severity, and possible development trend of the bias. For example, when the time regularization similarity is low and the Mahalanobis distance is large, it indicates that there are significant differences between the prediction result and the actual state in the time series pattern and feature combination; combined with a high bias confidence and a large spatiotemporal diffusion index, it can be judged that the abnormal situation is more serious and has a further diffusion trend, prompting relevant personnel to take targeted measures in time, optimize the logistics scheduling scheme, and reduce the negative impact of the abnormal situation on logistics operation.

[0090] In an embodiment, according to the logistics state bias data, a decision tree algorithm is used to determine the abnormal situation, and a logistics situation determination result is obtained, including:

[0091] A training sample is constructed, which includes a multi-dimensional feature vector of logistics business rules, and the multi-dimensional feature vector includes time regularization similarity, Mahalanobis distance, bias confidence, and spatiotemporal diffusion index;

[0092] The application feature importance ranking algorithm is applied to weight distribution of the multi-dimensional feature vector, and a weighted feature matrix is generated by fusion;

[0093] The weighted feature matrix is recursively divided by a CART decision tree algorithm based on Gini index optimization, and an abnormality judgment decision tree model is constructed;

[0094] The leaf nodes of the abnormality judgment decision tree model are semantically annotated by a knowledge base to generate a rule set containing abnormality types, severity levels and disposal priorities;

[0095] A decision tree dynamic pruning mechanism based on reinforcement learning is obtained, and the branch structure of the abnormality judgment decision tree model is adjusted in real time according to the historical judgment accuracy to obtain a trained decision tree model;

[0096] The logistics state deviation data is input into the trained decision tree model, and an abnormality heat map containing time and space dimensions and disposal priority ranking results are output as the logistics condition judgment results.

[0097] In this embodiment, in the logistics business scenario, the judgment of abnormal conditions needs to rely on rich and representative data samples. Therefore, first, a training sample set is constructed based on logistics state deviation data. The time similarity, Mahalanobis distance, deviation confidence, and space diffusion index of multiple dimensions are integrated to form a multi-dimensional feature vector. The above features describe the deviation of the logistics state from different angles, and combined with the logistics business rules (such as transportation time limit, cargo carrying limit, etc.), the data samples that meet the rules are filtered and labeled to determine whether the samples belong to abnormal conditions and the abnormal types, thereby constructing a complete and business-oriented training sample, providing a reliable data basis for subsequent model training.

[0098] Since the contributions of each feature in the multi-dimensional feature vector to the judgment of abnormal conditions are different, in order to more accurately reflect the importance of the features, a feature importance ranking algorithm (such as the information gain rate based on decision tree method) is used to evaluate each feature. The importance of the features is ranked by calculating the contribution of each feature in distinguishing normal and abnormal samples. For example, in some logistics scenarios, the space diffusion index is important for judging whether the abnormality will spread on a large scale, and its importance may be higher than that of other features. According to the ranking results, each feature is assigned a corresponding weight, and the higher the weight, the more critical the feature is in abnormality judgment. The weight and the original feature vector are fused to generate a weighted feature matrix, so that important features can be paid more attention to in the subsequent model training process, improving the judgment accuracy and effectiveness of the model.

[0099] CART(Classification and Regression Tree, classification and regression tree) decision tree algorithm is a classic decision tree construction method. In order to better adapt to the logistics abnormality judgment scene, the optimization method based on Gini index is adopted. The Gini index is used to measure the impurity of the sample set. The smaller the Gini index, the more single the data category in the sample set. In the process of constructing the decision tree, the weighted feature matrix is taken as the input, starting from the root node, for each sample set on the node, the Gini index change after each feature division is calculated, and the feature with the largest Gini index decrease is selected as the split feature of the node. For example, if the deviation confidence is taken as the split feature, after dividing the sample set into high confidence and low confidence two groups, the Gini index decrease is calculated to be the most significant, and the deviation confidence is selected as the split basis of the node. By recursively dividing the nodes, until the preset stopping condition (such as the number of node samples is less than the threshold, the Gini index cannot be significantly reduced, etc.) is met, a decision tree model is finally constructed, which can classify and judge abnormal situations according to the feature combination of logistics state deviation data.

[0100] The constructed decision tree model only outputs the classification result. In order to make its result more business readable and operable, the leaf nodes of the decision tree are semantically annotated by using the pre-constructed logistics knowledge base. The knowledge base stores rich logistics business knowledge, industry standards and historical case experience, covering all kinds of possible abnormal situations and their corresponding feature description, processing method and other information. For each leaf node of the decision tree, according to its sample features and classification results, the knowledge base is matched to annotate the abnormal type (such as transportation delay, cargo damage, path deviation, etc.), severity level (mild, moderate, severe) and disposal priority (emergency treatment, as soon as possible, routine treatment) represented by the node. For example, if the sample features of a leaf node show that the space-time diffusion index is large and the deviation confidence is high, it is judged as a serious transportation delay anomaly that needs emergency treatment by combining the knowledge base. Through semantic annotation, the decision tree model is converted into a rule set containing clear business semantics, which is convenient for logistics managers to understand and apply.

[0101] In practical applications, the logistics business scene is constantly changing, and static decision tree models may overfit or underfit, affecting the accuracy of anomaly determination. Therefore, a dynamic pruning mechanism based on reinforcement learning is introduced. Reinforcement learning interacts with the environment through agents, and learns the optimal strategy according to the reward signals fed back by the environment. In the decision tree model, the adjustment of the branch structure of the decision tree is regarded as the action of the agent, and the historical determination accuracy is taken as the reward signal. When the model determines the new logistics state deviation data in practical applications, the accuracy of the current model is calculated by comparing the determination result with the actual situation. If the accuracy decreases, it means that the model may be overfit, and some unnecessary branches are deleted through the pruning mechanism to simplify the model structure; if the accuracy is low and there is underfitting, appropriate branches are added to enrich the expression ability of the model. Through continuous interaction with actual data, the branch structure of the decision tree is adjusted in real time according to the accuracy feedback, so that the model can adapt to the dynamic changes of logistics business, and finally a well-trained and stable decision tree model is obtained.

[0102] The real-time acquired logistics state deviation data is input into the trained and optimized decision tree model, and the model determines and classifies the data layer by layer according to the constructed rules and adjusted structure. Based on the decision result, an abnormal heat map containing time and space dimensions is generated to visually display the distribution of abnormal situations in time and space. For example, different colors and shades are used to represent the abnormal degree of different regions at different time points on the map, and the deeper the color, the more serious the abnormality. At the same time, according to the disposal priority information marked by the leaf nodes, all detected abnormal situations are prioritized to form a disposal priority sorting result. Logistics managers can quickly locate abnormal high-incidence areas and time periods from the abnormal heat map, combine the disposal priority sorting result, and prioritize the processing of high-priority abnormal situations to reasonably arrange resources, improve the efficiency and safety of logistics operations, and ultimately use the abnormal heat map and disposal priority sorting result as comprehensive logistics condition determination results to provide strong support for logistics decision-making.

[0103] In an embodiment, the logistics data features are input into a pre-trained artificial intelligence model to obtain logistics state prediction results, including:

[0104] The logistics data features are subjected to spatiotemporal tensor decomposition to extract trend tensors, periodic tensors and residual tensors;

[0105] The trend tensors, periodic tensors and residual tensors are subjected to feature enhancement by a spatiotemporal graph attention network to generate a spatiotemporal correlation feature matrix;

[0106] The spatiotemporal correlation feature matrix is subjected to time series prediction based on a multi-scale causal convolution network to output a multi-step probability prediction distribution;

[0107] The multi-step probability prediction distribution is subjected to uncertainty quantification processing to obtain a confidence interval prediction result.

[0108] The confidence interval prediction result is subjected to logistics rule constraint processing to generate a logistics state prediction result conforming to business logic.

[0109] In this embodiment, in a logistics system, data features often exhibit complex spatiotemporal dependencies. Spatiotemporal tensor decomposition techniques represent multi-dimensional logistics data features as tensors and use Tucker decomposition or CANDECOMP / PARAFAC (CP) decomposition to decompose the original tensor into three components with physical meaning.

[0110] Trend tensor: captures the changing patterns of logistics data over long time scales, such as seasonal business growth trends or long-term evolution patterns of regional logistics demand. Through low-rank tensor approximation techniques, the core components reflecting the overall change direction of the data are extracted.

[0111] Periodic tensor: identifies periodically repeating spatiotemporal patterns, such as the impact of daily traffic peak hours on delivery efficiency or the periodic fluctuations in weekly order volume. Fourier transform combined with tensor decomposition is used to separate components with specific periodic characteristics.

[0112] Residual tensor: contains unstructured disturbances caused by random fluctuations and abnormal events, such as the impact of sudden weather events or temporary traffic restrictions on logistics. The extraction of this component helps subsequent anomaly detection and uncertainty analysis.

[0113] The above decomposition methods can simultaneously preserve the dependencies of data in time and space dimensions, providing more accurate feature representations for subsequent prediction.

[0114] Logistics systems are essentially complex spatiotemporal networks, and traditional methods struggle to effectively capture the dynamic associations between nodes. Spatiotemporal graph attention networks (STGAT) enhance the features of decomposed tensors through the following mechanisms:

[0115] Spatial attention mechanism: automatically identifies key nodes and paths in the logistics network by calculating attention weights between nodes. For example, in a city delivery network, this mechanism can highlight the importance of transportation hubs or distribution centers.

[0116] Temporal attention mechanism: dynamically adjusts the weights of different time steps to capture long and short-term dependencies in time series data. For logistics prediction, this mechanism can distinguish between regular fluctuations and the impact of abnormal events.

[0117] Spacetime Fusion: Combining spatial and temporal attention mechanisms to generate a spacetime correlation feature matrix. This matrix not only preserves the decomposition characteristics of the original tensor but also enhances the dynamic interaction between nodes.

[0118] In this way, the model can adaptively focus on important spatiotemporal regions in the logistics network, improving the accuracy of the prediction.

[0119] Traditional time series prediction models struggle to handle both varying patterns at different time scales and causal relationships. The Multiscale Causal Convolutional Network (MCCN) addresses this issue through the following design:

[0120] Multiscale Convolution Structure: Parallel processing of input data using convolution kernels of different sizes, capturing both short-term fluctuations and long-term trends. For example, small convolution kernels capture subtle changes in daily delivery volumes, while large convolution kernels capture weekly / monthly demand changes.

[0121] Causal Convolution Design: By limiting the receptive field of convolution operations, the model ensures that it only uses historical information for prediction, adhering to the causal constraints of the logistics system.

[0122] Probabilistic Prediction Framework: Instead of point estimates, the model outputs probability distributions of predicted values, quantifying uncertainty through Monte Carlo Dropout or Bayesian Neural Networks. For example, the model not only predicts the arrival time of goods but also provides a probability distribution interval for the arrival time.

[0123] The above architecture effectively handles the multiscale characteristics and uncertainty in logistics data, providing more comprehensive information for decision-making.

[0124] There are various sources of uncertainty in logistics systems, such as traffic condition changes, weather impacts, etc. Uncertainty quantification is achieved through the following steps:

[0125] Model Ensemble: Train multiple models with different initialization parameters and generate multiple prediction results through ensemble methods such as Bagging or Dropout Ensemble.

[0126] Probability Distribution Estimation: Based on the ensemble prediction results, estimate the probability distribution parameters (such as mean and variance) of the predicted values.

[0127] Confidence Interval Calculation: According to the probability distribution characteristics, calculate the prediction interval at a certain confidence level. For example, a 95% confidence interval means that the true value has a 95% probability of falling within this interval.

[0128] In the logistics scenario, the uncertainty quantification results can be used to assess the risk level. For example, predicting the delay probability distribution of a certain transportation route helps dispatchers develop more robust delivery plans.

[0129] There are various hard constraints and domain knowledge in logistics business, such as vehicle load limit, delivery time window, service level agreement, etc. Rule constraint processing ensures that the prediction results conform to business logic through the following mechanisms:

[0130] Constraint encoding: convert logistics rules into differentiable constraint functions, such as representing vehicle load limit as an inequality constraint of predicted load.

[0131] Constraint optimization: use Lagrange multiplier method or projection gradient descent method to project the prediction results to the feasible solution space while maintaining the probability characteristics of the prediction model. For example, optimize the delivery path prediction under the premise of meeting all time window constraints.

[0132] Knowledge injection: inject logistics domain knowledge (such as common abnormal patterns, optimal operation processes) into the prediction model through pre-training or fine-tuning mechanisms. For example, use historical abnormal cases to train the model to identify potential risks.

[0133] In an embodiment, the logistics data features are input into a pre-trained artificial intelligence model to obtain a logistics state prediction result, including:

[0134] Quantum state encoding is performed on the logistics data features to construct a logistics feature quantum superposition state representation;

[0135] A quantum entanglement algorithm is applied to perform feature correlation enhancement processing on the logistics feature quantum superposition state representation to generate a high-dimensional feature entanglement tensor;

[0136] Causal structure learning processing is performed on the high-dimensional feature entanglement tensor to identify a key causal relationship network in the logistics system;

[0137] Through a quantum-inspired Monte Carlo tree search algorithm, a multi-step prediction path exploration is performed based on the key causal relationship network to generate a logistics state probability tree containing space-time dimensions;

[0138] Quantum measurement simulation technology is used to perform collapse processing on the logistics state probability tree to obtain a logistics state prediction distribution;

[0139] The logistics state prediction distribution is verified against logistics rules to generate a logistics state prediction result that conforms to business logic constraints.

[0140] In this embodiment, the logistics data features cover multi-dimensional information such as transportation time, cargo weight, vehicle speed, etc. In the traditional way, these features exist in the form of classical data, which can only represent one certain state at a time. Quantum state encoding technology, inspired by quantum computing principles, maps these classical logistics data features to quantum state space, giving data the "quantum superposition" property. Specifically, through a pre-set encoding rule, the originally independent multi-dimensional feature vectors are transformed into a form similar to the probability amplitude distribution of quantum bits. This allows a set of logistics data features to represent multiple possible states simultaneously, just like a particle in the quantum world can be in multiple positions at the same time. The quantum superposition state representation constructed by this encoding lays the foundation for subsequent parallel processing of complex logistics feature relationships and mining of potential laws, greatly improving the efficiency and dimensionality of data processing compared to traditional data representation forms.

[0141] In quantum entanglement, particles have non-local correlation characteristics beyond spatial distance. In the logistics data processing scenario, applying quantum entanglement algorithms is to use this characteristic to strengthen the relationship between logistics features. By designing a special algorithm process, like building a quantum circuit, the logistics features encoded as quantum superposition states are operated. The algorithm will implement specific operations on the originally independent logistics feature quantum bits, prompting them to have strong correlations and form a state similar to quantum entanglement. This operation expands the low-dimensional logistics feature space to high dimensions, making multiple factors such as transportation time, vehicle state, and weather conditions, which may only have simple linear relationships, have complex nonlinear correlations. The final generated high-dimensional feature entanglement tensor can more comprehensively and deeply express the complex relationships between various elements in the logistics system, providing rich and accurate feature information for subsequent accurate analysis of the internal logic of the logistics system.

[0142] After obtaining the high-dimensional feature entanglement tensor, a large amount of logistics features and their interrelated information are contained, but these information has not yet explicitly shown the causal relationship. The causal structure learning process aims to sort out the real causal logic relationship network in the logistics system from complex tensor data. Through a specially designed causal discovery algorithm, the process simulates the intervention and observation of feature variables in the quantum world. Specifically, the algorithm will assume to change a certain logistics feature variable, and then observe the changes of other variables. Through multiple simulations and analyses, the causal influence strength between different features is quantified. Then, using graph theory related methods, each logistics feature is regarded as a node, and the features with significant causal relationship are connected by edges to construct a complete causal relationship network. By setting a reasonable threshold, the key causal edges are filtered out to obtain the key causal relationship network. For example, through this process, it can accurately distinguish whether "road construction leads to traffic congestion, and then causes transportation delay", or simply a correlation relationship, providing a reliable structural basis for subsequent accurate prediction of logistics state changes.

[0143] Monte Carlo tree search algorithm is a classic algorithm commonly used for path search and decision-making, while quantum-inspired Monte Carlo tree search algorithm combines the unique advantages of quantum computing on this basis. Starting from the current known logistics state, the algorithm explores the possible evolution paths of future logistics state based on the key causal relationship network. It draws on the idea of quantum parallel sampling, and instead of exploring paths one by one like traditional algorithms, it can evaluate multiple possible prediction paths simultaneously, greatly improving search efficiency. At the same time, using a strategy similar to quantum amplitude amplification, it selectively enhances the search weight of high-probability paths, making the algorithm more inclined to explore those more likely to occur logistics state evolution paths. During the exploration process, each path corresponds to a possible evolution sequence of logistics state in time and space dimensions. With continuous simulation and exploration, the algorithm calculates the probability of each path occurring and integrates this information to ultimately construct a logistics state probability tree containing time and space dimensions. This probability tree not only shows the various possible situations of future logistics state, but also accurately quantifies the probability of each situation occurring and their evolution rules in time and space dimensions, providing comprehensive and forward-looking information for logistics decision-making.

[0144] In quantum mechanics, measuring a quantum system causes the wave function to collapse, causing the system to change from multiple possible states to one definite state. In logistics state prediction, the quantum measurement simulation technique is used to simulate this process to handle the logistics state probability tree. First, special "measurement operators" are designed for key variables of the logistics state, such as time, location, etc. These operators are like measurement tools in the quantum world. Then, through Monte Carlo sampling, multiple "measurements" are simulated. Each simulated measurement causes the probability tree to "collapse" to a certain path, corresponding to a specific logistics state evolution. Through a large number of simulated measurements, the results of each collapse are collected and statistically analyzed to calculate the frequency of different logistics states. Based on these statistical data, a logistics state prediction distribution is constructed. In this way, the complex multiple possibilities in the probability tree are transformed into a prediction distribution that conforms to classical probability theory, retaining the advantages of path diversity and accurate probability evaluation brought by quantum computing, and presenting in a form that is easy to understand and apply, providing clear probability information support for logistics decision-making.

[0145] In actual operation, the logistics industry has many business rules and constraints, such as vehicle load restrictions, delivery time window requirements, service level agreement provisions, etc. These rules are important guidelines to ensure the normal and efficient operation of logistics business. Therefore, after obtaining the logistics state prediction distribution, it is necessary to compare and verify these prediction results with actual business rules. First, various logistics business rules are converted into specific mathematical constraints, such as using time intervals to represent delivery time window constraints. Then, using mathematical methods such as convex optimization, the prediction distribution is mapped into a feasible region that satisfies all constraints, and the prediction results that do not meet the rules are adjusted. Finally, the probability distribution parameters are re-evaluated and calculated in the feasible region to ensure that the final generated logistics state prediction results not only conform to the probability laws of statistics, but also strictly meet the actual operation requirements of logistics business. For example, when the predicted arrival time of a delivery vehicle exceeds the service commitment time window, after rule verification and adjustment, the algorithm will correct the prediction distribution based on comprehensive consideration of traffic optimization and other factors, and give a new prediction result that meets the time window constraint, so that the prediction result truly has practical application value and can directly guide logistics operation decision-making.

[0146] In an embodiment, if the logistics situation determination result is abnormal, an abnormal information push is performed on the relevant personnel terminal, and a logistics abnormality alarm is issued, including:

[0147] Obtain the number information of each relevant personnel terminal; sequentially combine each number information to obtain a number string;

[0148] The characters in the number string are sequentially added to each node of a preset ternary tree template to obtain a numbered ternary tree;

[0149] Character attribute feature extraction is performed on the number string, and based on the extracted character attribute features, an ellipse and a triangle are constructed;

[0150] Based on the triangle, structural variation is performed on the numbered ternary tree to obtain a variant ternary tree;

[0151] The variant ternary tree is superimposed on the ellipse, based on the positional relationship between each node of the variant ternary tree and the ellipse, a target node is selected, and a string is generated by combining the characters on the target node;

[0152] The string is sent to the logistics data collection end as an encryption key for the collection end to encrypt the logistics data.

[0153] In this embodiment, when the logistics condition determination result shows that there is an anomaly and an alarm is issued, the terminal number information collection program will be started immediately. All personnel terminals related to this anomaly are automatically searched, including but not limited to logistics dispatchers, transport drivers, warehouse managers, and other terminal devices. After obtaining the unique number information of these terminals, all numbers are arranged and combined in sequence according to the preset rules (such as the importance of personnel responsibilities, the time sequence of terminal access to the system, etc.), and finally a continuous number string is formed. This string integrates the identification information of the personnel related to the anomaly handling, providing basic data for subsequent data processing.

[0154] A ternary tree template is preset, which specifies the basic structure and node distribution rules of the ternary tree. At this time, each character in the number string obtained in the previous step is sequentially assigned to each node of the ternary tree in the order from left to right. During the addition process, the construction principle of the ternary tree is followed, i.e. each node has at most three child nodes, if the current node has been assigned three child nodes, the character is added to the next layer node, until all characters in the number string are added, thereby constructing a numbered ternary tree containing number information. This process converts linear string data into hierarchical tree data, facilitating subsequent data feature extraction and structural operations.

[0155] After obtaining the numbered string, the attribute characteristics of each character in the string are analyzed and extracted. The character attribute characteristics include, but are not limited to, the ASCII code value of the character, the position sequence number of the character in the string, the frequency of the character, the type of the character, and the like. According to these extracted attribute characteristics, an ellipse and a triangle are generated through a preset mathematical algorithm and a graph construction rule. For example, the major axis and the minor axis length of the ellipse can be determined according to the statistical distribution of the character ASCII code value, and the vertex coordinates of the triangle can be determined according to the regularity of the character position sequence number.

[0156] The numbered ternary tree is structurally adjusted based on the constructed triangle. The specific operation mode is associated with the geometric characteristics of the triangle, for example, according to the three vertex coordinates of the triangle, the three different levels or regions of the ternary tree are respectively corresponded, and operations such as splitting, merging or reconnecting of the nodes in these regions are performed; or according to the length and angle relationship of the edges of the triangle, the weight or direction of the edges in the ternary tree is adjusted. Through a series of regular structural variation operations, the original numbered ternary tree is transformed into a variant ternary tree with new structural characteristics, which further excavates the potential features and relationships of the data.

[0157] The variant ternary tree and the ellipse constructed before are superimposed on the same plane, and the positional relationship between the nodes of the variant ternary tree and the ellipse is analyzed at this time. The positional relationship includes whether the node is inside, outside or on the boundary of the ellipse, and the distance from the node to the center of the ellipse and the like. According to the pre-set screening rules, the target nodes that meet the conditions are selected based on these positional relationships. For example, a plurality of nodes located inside the ellipse and closest to the center of the ellipse are selected, or nodes intersecting the boundary of the ellipse are selected. The characters carried on the selected target nodes are extracted and combined in a certain order to form a new string.

[0158] The string obtained after a series of data processing and graph operations has unique structure and information characteristics, which is used as an encryption key. The key is sent to the logistics data collection end device through a secure data transmission channel. After receiving the key, the collection end starts the data encryption program and encrypts the subsequent collected logistics data according to the key. The encrypted logistics data can only be read and analyzed by authorized devices and personnel who have the corresponding decryption key during transmission and storage, thereby effectively protecting the security and privacy of the logistics data.

[0159] In an embodiment, after the abnormal information is pushed to the terminal of the related personnel and the logistics abnormality alarm is issued, the method further includes:

[0160] Obtain the logistics document number and transport vehicle ID information triggering the abnormal alarm, combine the order flow data and transport route data in the logistics business system, extract the complete logistics business path related to the abnormality, arrange the node numbers in the complete logistics business path in business order to form an ordered number sequence;

[0161] Map each number in the ordered number sequence to four dimensions of a four-dimensional space according to its corresponding business timestamp, geographic location coordinates, cargo type code, and transport mode code, determine a plurality of coordinate points in the four-dimensional space, and connect the coordinate points to construct a four-dimensional hypercube containing space-time and business attribute information;

[0162] Calculate the length difference value of each edge of the four-dimensional hypercube, and perform topological distortion transformation on the four-dimensional hypercube by taking the length difference value as a parameter to generate a distorted hypercube;

[0163] Convert the geometric elements of the distorted hypercube into nodes, edges, and subtree structures of a binary tree according to a preset mapping rule to form an initial binary tree, and perform node splitting or merging operations on the initial binary tree according to the adjustment coefficient corresponding to the abnormal alarm to obtain a variant binary tree;

[0164] Preorder traversal is performed on the variant binary tree, the node numbers obtained by traversal are combined into a string in turn, and the string is sent to the logistics data collection end through a secure channel for encrypting the collected logistics data.

[0165] In this embodiment, when a logistics abnormality is detected and an alarm is triggered, a data association retrieval mechanism is first started. Through the timestamp and type label of the abnormal event, the relevant logistics document number (such as order number and transport number) and transport vehicle ID are located in the logistics business system. Combined with the order flow data (including order creation, sorting, warehouse out, transportation, and receipt status change records) and the transport route data (GPS track, estimated arrival time, and passing stations), a complete logistics business path graph is constructed. The path graph takes logistics nodes (warehouses, transfer stations, and distribution points) as vertices and business flow relationship as edges to form a directed acyclic graph. The unique number of each node (such as warehouse code and equipment ID) is extracted in business execution order to generate an ordered number sequence, providing a basic data structure for subsequent multi-dimensional mapping.

[0166] A four-dimensional mapping function is designed to convert the one-dimensional number sequence into a geometric structure in a four-dimensional space. The specific mapping rule can be:

[0167] X-axis: business timestamp (normalize the timestamp to the [0, 1] interval).

[0168] Y-axis, Z-axis: geographic location coordinates (convert latitude and longitude to planar rectangular coordinates).

[0169] W-axis: Combined feature vector of cargo type code and transportation mode code (processed by one-hot encoding and PCA dimension reduction).

[0170] Each logistics node number corresponds to a coordinate point in four-dimensional space, and all points form a hypercube structure. This structure not only contains the spatio-temporal position information of the nodes, but also encodes the business attributes through the W-axis, forming a multi-dimensional integrated geometric expression.

[0171] Perform geometric feature analysis on the four-dimensional hypercube, calculate the Euclidean distance of each edge, and obtain the length set. Calculate the difference value of edge length by standard deviation, which reflects the imbalance of logistics business processes (such as some links taking too long or having an abnormal geographical span). Take the length difference value as the control parameter, and apply topological distortion transformation:

[0172] The nodes corresponding to the edges with length difference values exceeding the threshold value are nonlinearly displaced in the four-dimensional space. Adopt Lorentz transformation to simulate the effect of space-time distortion, so that the hypercube undergoes morphological distortion while maintaining topological connectivity. This distortion operation converts logistics abnormal features into structural deformation in geometric space, enhancing the relevance of the key and abnormal scenarios.

[0173] Design a structure mapping algorithm from four dimensions to two dimensions:

[0174] Vertex mapping: The vertices of the hypercube are converted into the leaf nodes of the binary tree, and the node value is the original logistics node number.

[0175] Edge mapping: The edges of the hypercube are converted into the edges of the binary tree, and the weight of the edge is determined by the length of the edge.

[0176] Face structure mapping: The faces of the hypercube are converted into the sub-trees of the binary tree, and the area of the face corresponds to the depth of the sub-tree.

[0177] After generating the initial binary tree, determine the adjustment coefficient according to the severity level of the abnormal alarm (such as slight, moderate, and severe). High severity level triggers node splitting operation (splitting one node into multiple sub-nodes, increasing the branch complexity of the tree), and low severity level performs node merging (reducing redundant branches, improving the compactness of the tree). This dynamic adjustment binds the binary tree structure with the depth of abnormal features.

[0178] A pre-order traversal is performed on the mutated binary tree (root node -> left subtree -> right subtree), and the node numbers are extracted in turn and combined into a string. This string contains the complete information chain from logistics anomaly detection to topology transformation, and is unique and scenario-specific. The string is transmitted to the logistics data collection end through a quantum secure communication channel (such as quantum key distribution based on the BB84 protocol). The collection end uses the string as the key of the AES-256 encryption algorithm to encrypt the real-time collected logistics data (such as vehicle location, cargo status, environmental parameters), ensuring the security of the data during transmission and storage. The encrypted data can only be decrypted using the same key, forming a secure transmission channel strongly bound to the abnormal event.

[0179] The core of this scheme is to combine the spatiotemporal characteristics of logistics business processes with geometric topology transformation, and through the structure conversion of four-dimensional hypercubes and mutated binary trees, to realize the deep mapping of abnormal features to encryption keys. Each step is closely related to the logistics business logic, making the generated key not only random, but also containing the inherent characteristics of the abnormal event, significantly improving the security and traceability of encryption.

[0180] In an embodiment, after the abnormal information is pushed to the terminal of the relevant personnel and the logistics abnormality alarm is issued, the method further includes:

[0181] Obtaining the logistics event number associated with the abnormal information, splitting each number into a character sequence, and arranging it into a two-dimensional matrix;

[0182] Mapping the matrix elements of the two-dimensional matrix to a three-dimensional space, with the numerical value of each element determining its height on the Z-axis, forming a hyper surface with undulating characteristics; using Gaussian curvature to analyze the bending degree of each point on the hyper surface, identifying the key influence area;

[0183] Fractal iteration is performed on the hyper surface, with the level of the abnormal alarm as the iteration number control parameter;

[0184] Constructing the topology skeleton of the hyper surface, selectively pruning the skeleton according to the upstream and downstream relationships of the logistics business process, and forming a simplified topology graph;

[0185] Calculating the betweenness centrality of each node in the topology graph, mapping the betweenness centrality value to a binary sequence, and generating a dynamic key for encryption based on the binary sequence and sending it to the logistics data collection end for encryption of the collected logistics data.

[0186] In this embodiment, when the logistics anomaly alert is issued, all logistics event numbers associated with the anomaly are immediately retrieved, such as order numbers, shipping numbers, equipment identification codes, etc. Subsequently, the system splits each number by character to obtain corresponding character sequences, which contain numbers, letters, or special symbols. Based on the chronological order of the occurrence of the logistics event, or the associated logic of the number in the business process, these character sequences are arranged in sequence to construct a two-dimensional matrix. The rows and columns of the matrix represent different dimensional information organization methods, respectively. The matrix structurally integrates the originally dispersed number information, laying a foundation for subsequent multi-dimensional analysis.

[0187] After the two-dimensional matrix is constructed, each element in the matrix is mapped to a coordinate point in a three-dimensional space. The row and column positions of the element in the matrix determine its coordinates in the X and Y axes of the three-dimensional space, while the numerical value of the element itself (such as the ASCII code value corresponding to the character, the numerical value after conversion processing, etc.) determines the height of the point in the Z-axis direction. A large number of coordinate points together form a hyper-surface with undulating characteristics, which intuitively presents the distribution and variation of the logistics event number information from a geometric perspective. Then, the system uses the Gaussian curvature calculation method to calculate each point on the hyper-surface to quantify the degree of curvature of the point. The greater the Gaussian curvature, the more intense the curvature of the surface in this region, meaning that the corresponding logistics event number information changes more complex and critical. Through Gaussian curvature analysis of the entire hyper-surface, key impact areas on the hyper-surface can be accurately identified, which often contain important information closely related to logistics anomalies.

[0188] After identifying the key impact areas of the hyper-surface, fractal iteration deformation operation is started. Fractal iteration is based on fractal geometry principles, making the hyper-surface exhibit self-similar complex structures at different scales. The level of the anomaly alert will be used as a control parameter for the number of fractal iterations. The higher the alert level, the more serious the abnormal situation, the more iterations are set, and the complexity of the structure of the hyper-surface will also increase accordingly. Through fractal iteration, the original geometric shape of the hyper-surface evolves, further mining the potential complex relationships and patterns of the data on the basis of preserving the original information characteristics, enhancing the correlation between the hyper-surface and the characteristics of the logistics anomaly.

[0189] After the fractal iteration deformation is completed, topological analysis is performed on the super surface to construct a topological skeleton. The topological skeleton retains the key connection structure and topological properties of the super surface, and simplifies the complex geometric form of the super surface into a network structure composed of nodes and lines. In this network structure, the nodes represent key regions or feature points of the super surface, and the lines represent the connection relationship between the regions. Based on the upstream and downstream logic of the logistics business process, selective pruning operation is performed on the topological skeleton. For nodes and lines that are not strongly associated with the logistics business process, redundant or secondary, removal processing is performed, and the part closely related to the core process of the business is retained, thereby obtaining a simplified topological graph. The simplified topological graph focuses on the key path and core elements of the logistics business, and more clearly shows the important structural relationship related to the logistics anomaly.

[0190] After obtaining the simplified topological graph, the betweenness centrality of each node in the graph is calculated. The betweenness centrality reflects the importance of the node in connecting other nodes in the entire network. The higher the betweenness centrality value of a node, the more critical the node is in the logistics business process, and the greater the impact on information transmission and business operation. The betweenness centrality value of each node is converted and processed to map to the corresponding binary sequence. These binary sequences contain quantitative information about the importance of the nodes in the topological graph. Finally, through specific algorithms and rules, the binary sequences are integrated and transformed to generate a dynamic key for encryption. The dynamic key is sent to the logistics data acquisition end through a secure communication protocol. After receiving the key, the acquisition end uses the key to encrypt the subsequent collected logistics data, ensuring the security of the logistics data during transmission and storage, and realizing the close combination and effective linkage of logistics anomaly alarm and data encryption.

[0191] Reference Figure 2 In another embodiment of the present application, a logistics information monitoring system based on artificial intelligence is also provided, comprising:

[0192] An acquisition module is configured to acquire logistics data in a logistics transportation process, extract features from the logistics data, and obtain logistics data features.

[0193] A prediction module is configured to input the logistics data features into a pre-trained artificial intelligence model to obtain a logistics state prediction result.

[0194] A calculation module is configured to obtain an actual logistics state, calculate the deviation between the logistics state prediction result and the actual logistics state, and obtain logistics state deviation data.

[0195] A determination module is configured to determine an abnormal situation using a decision tree algorithm based on the logistics state deviation data to obtain a logistics situation determination result.

[0196] An alarm module is configured to push abnormal information to a terminal of a related personnel and issue a logistics abnormality alarm if the logistics condition determination result is abnormal.

[0197] In the embodiment, the specific implementation of each module in the system embodiment is described above in the method embodiment, and thus will not be described here.

[0198] With reference to Figure 3 In the embodiment, a computer device can be a server, and the internal structure thereof can be as shown in Figure 3 The computer device includes a processor, a memory, a display screen, an input device, a network interface and a database connected through a system bus. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium. The database of the computer device is configured to store corresponding data in the embodiment. The network interface of the computer device is configured to communicate with an external terminal through a network connection. The computer program is executed by the processor to implement the above method.

[0199] Those skilled in the art can understand Figure 3 that the structure shown in the embodiment is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied.

[0200] The embodiment of the present application also provides a computer readable storage medium having a computer program stored thereon. The computer program is executed by the processor to implement the above method. It can be understood that the computer readable storage medium in the embodiment can be a volatile readable storage medium or a non-volatile readable storage medium.

[0201] To sum up, the logistics information monitoring method and system based on artificial intelligence provided in the embodiments of the present application comprises: collecting logistics data in a logistics transportation process, performing feature extraction on the logistics data to obtain logistics data features; inputting the logistics data features into a pre-trained artificial intelligence model to obtain a logistics state prediction result; obtaining an actual logistics state, performing deviation calculation on the logistics state prediction result and the actual logistics state to obtain logistics state deviation data; performing abnormal situation determination on the logistics state deviation data by using a decision tree algorithm to obtain a logistics condition determination result; and if the logistics condition determination result is abnormal, performing abnormal information pushing on a related personnel terminal and issuing a logistics abnormality alarm. In the present application, by performing deviation calculation on the logistics state prediction result and the actual logistics state, the actual influence degree and potential risks of the deviation can be quantified, and then the decision tree algorithm is used to perform abnormal situation determination to obtain a precise logistics condition determination result, thereby overcoming the defect of insufficient accuracy in the current logistics detection process.

[0202] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware, and the computer program can be stored in a non-volatile computer readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. Any reference to memory, storage, database or other medium provided by the present application and used in the embodiments can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration but not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM) and memory bus dynamic RAM, etc.

[0203] It is to be understood that the terminology "including", "comprising", or any other variation thereof, is intended to cover a non-exclusive inclusion such that process, method, article, or apparatus that comprises a list of elements does not include only those elements but can also include other elements not expressly listed or inherent to such process, method, article, or apparatus. An element proceeded by "comprises a... " does not, without more constraints, exclude the presence of additional identical elements in the process, method, article, or apparatus that comprises the element.

[0204] The above description is merely the preferred embodiments of the present application, and is not intended to limit the patent scope of the present application. Any equivalent structure or equivalent process transformation made according to the content of the present application specification and drawings, or directly or indirectly applied to other related technical fields, is also included in the patent protection scope of the present application.

Claims

1. A logistics information monitoring method based on artificial intelligence, characterized in that, Includes the following steps: Collect logistics data during the logistics transportation process, extract features from the logistics data, and obtain logistics data features; The logistics data features are input into a pre-trained artificial intelligence model to obtain logistics status prediction results. The process involves obtaining the actual logistics status and calculating the deviation between the predicted and actual logistics status to obtain logistics status deviation data. This includes: performing a time-series decomposition on the predicted logistics status to separate time-series components, including trend, periodic, and residual components; calculating the similarity between the time-series components of the predicted result and the corresponding components of the actual logistics status based on a dynamic time warping distance metric model to obtain a time warping similarity; calculating the statistical distance between the predicted and actual logistics status in the feature space based on the Mahalanobis distance algorithm to obtain the Mahalanobis distance; fusing the time warping similarity and the Mahalanobis distance to construct a deviation confidence function and output the deviation confidence; performing cluster analysis on the deviation confidence using a fuzzy C-means clustering algorithm to generate a deviation heatmap containing spatiotemporal dimensions; performing morphological filtering on the deviation heatmap to extract the core deviation region and calculate its spatiotemporal diffusion index; and combining the time warping similarity, Mahalanobis distance, deviation confidence, and spatiotemporal diffusion index as the logistics status deviation data. Based on the logistics status deviation data, anomaly judgment is performed using a decision tree algorithm to obtain the logistics status judgment result; If the logistics situation is determined to be abnormal, an abnormality information will be pushed to the relevant personnel's terminals, and a logistics abnormality alarm will be issued.

2. The logistics information monitoring method based on artificial intelligence according to claim 1, characterized in that, The logistics data includes logistics document information, sensor data, and GPS positioning data of transport vehicles.

3. The logistics information monitoring method based on artificial intelligence according to claim 1, characterized in that, Based on the logistics status deviation data, anomaly detection is performed using a decision tree algorithm to obtain the logistics status determination result, including: Construct training samples, which include multi-dimensional feature vectors of logistics business rules. These multi-dimensional feature vectors include time warping similarity, Mahalanobis distance, bias confidence, and spatiotemporal diffusion index. The multi-dimensional feature vectors are weighted using a feature importance ranking algorithm and then fused to generate a weighted feature matrix. The weighted feature matrix is ​​recursively partitioned using the CART decision tree algorithm optimized by the Gini index to construct an anomaly detection decision tree model. Using a knowledge base, the leaf nodes of the anomaly determination decision tree model are semantically annotated to generate a rule set containing anomaly type, severity level, and handling priority. Obtain a pre-built dynamic pruning mechanism for decision trees based on reinforcement learning, and adjust the branch structure of the anomaly judgment decision tree model in real time according to the historical judgment accuracy to obtain a trained decision tree model. The logistics status deviation data is input into the trained decision tree model, and the output includes an anomaly heat map containing spatiotemporal dimensions and the disposal priority ranking result, which serves as the logistics status judgment result.

4. The logistics information monitoring method based on artificial intelligence according to claim 1, characterized in that, The logistics data features are input into a pre-trained artificial intelligence model to obtain logistics status prediction results, including: The characteristics of the logistics data are subjected to spatiotemporal tensor decomposition to extract trend tensor, periodic tensor and residual tensor; The trend tensor, period tensor, and residual tensor are enhanced by applying a spatiotemporal graph attention network to generate a spatiotemporal correlation feature matrix. The spatiotemporal correlation feature matrix is ​​used to perform temporal prediction based on a multi-scale causal convolutional network, and a multi-step probability prediction distribution is output. Uncertainty quantification is performed on the multi-step probability prediction distribution to obtain the confidence interval prediction result; The confidence interval prediction results are subjected to logistics rule constraints to generate logistics status prediction results that conform to business logic.

5. The logistics information monitoring method based on artificial intelligence according to claim 1, characterized in that, The logistics data features are input into a pre-trained artificial intelligence model to obtain logistics status prediction results, including: The logistics data features are quantum state encoded to construct a quantum superposition state representation of the logistics features; The quantum entanglement algorithm is applied to enhance the feature correlation of the quantum superposition state representation of the logistics features to generate a high-dimensional feature entanglement tensor. Causal structure learning is performed on the high-dimensional feature entanglement tensor to identify key causal relationship networks in the logistics system; Using a quantum-inspired Monte Carlo tree search algorithm, a multi-step prediction path is explored based on the key causal relationship network to generate a probability tree of logistics status containing spatiotemporal dimensions. The probability tree of the logistics state is collapsed using quantum measurement simulation technology to obtain the predicted distribution of the logistics state. The logistics status prediction distribution is validated according to logistics rules to generate logistics status prediction results that conform to business logic constraints.

6. A logistics information monitoring system based on artificial intelligence, characterized in that, include: The data acquisition module is used to collect logistics data during the logistics transportation process, and to extract features from the logistics data to obtain logistics data features. The prediction module is used to input the features of the logistics data into a pre-trained artificial intelligence model to obtain the logistics status prediction result; The calculation module is used to acquire the actual logistics status and calculate the deviation between the predicted logistics status and the actual logistics status to obtain logistics status deviation data. This includes: performing time-series decomposition on the predicted logistics status to separate time-series components, including trend, periodic, and residual components; calculating the similarity between the time-series components of the predicted result and the corresponding components of the actual logistics status based on a dynamic time warping distance metric model to obtain time warping similarity; calculating the statistical distance between the predicted logistics status and the actual logistics status in the feature space based on the Mahalanobis distance algorithm to obtain Mahalanobis distance; fusing the time warping similarity and Mahalanobis distance to construct a deviation confidence function and output the deviation confidence; performing cluster analysis on the deviation confidence using a fuzzy C-means clustering algorithm to generate a deviation heatmap containing spatiotemporal dimensions; performing morphological filtering on the deviation heatmap to extract the core deviation region and calculate its spatiotemporal diffusion index; and combining the time warping similarity, Mahalanobis distance, deviation confidence, and spatiotemporal diffusion index as the logistics status deviation data. The judgment module is used to determine abnormal situations based on the logistics status deviation data using a decision tree algorithm, and obtain the logistics status judgment result. The alarm module is used to push abnormal information to the terminals of relevant personnel and issue a logistics abnormality alarm if the logistics situation is determined to be abnormal.

7. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 5.

Images