In-transit cargo arrival time prediction method and system based on real-time environmental data
By constructing a joint traffic state vector and using a Bayesian update method, the uncertainty of prediction is quantified, which solves the problem of insufficient accuracy and reliability in the prediction of the arrival time of goods in transit in existing technologies, and realizes accurate prediction in complex traffic environments.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- HUBEI MAI RUIDA SUPPLY CHAIN CO LTD
- Filing Date
- 2026-05-08
- Publication Date
- 2026-06-09
AI Technical Summary
Existing technologies struggle to quantify the uncertainty of arrival times for goods en route, neglect real-time data quality and vehicle status, and fail to effectively capture the spatiotemporal correlation of road network traffic flow, resulting in insufficient accuracy and reliability of prediction results in complex traffic environments.
By acquiring historical and real-time traffic data of the road network, a joint traffic state vector is constructed, the basic transition probability matrix is adjusted, and the credibility of observation evidence is calculated by combining real-time traffic conditions and vehicle load conditions. The arrival time is then iteratively predicted using a Bayesian update method to quantify the prediction uncertainty and assess the credibility of the observation evidence.
It achieves accurate and reliable arrival time prediction in complex traffic environments, quantifies prediction uncertainty, closely reflects actual conditions, avoids interference from poor observation data, and improves the accuracy and stability of prediction.
Smart Images
Figure CN122175490A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the technical field of logistics management, specifically relating to a method and system for predicting the arrival time of goods in transit based on real-time environmental data. Background Technology
[0002] In modern logistics and supply chain management, accurate prediction of the arrival time of goods in transit is a core element. Traditional forecasting methods mainly rely on static route planning and simple estimations of historical average speeds, failing to fully consider the dynamic changes in road traffic conditions. Therefore, when faced with urban traffic congestion, accidents, or unforeseen circumstances common in long-distance transportation, the prediction results of these methods often deviate significantly from the actual situation.
[0003] To improve prediction accuracy, existing technologies have begun to integrate real-time traffic data, such as dynamically adjusting estimated arrival times by acquiring real-time GPS locations of vehicles and real-time traffic flow information of the road network. However, these improved methods still have significant limitations: on the one hand, they typically only output a single point-in-time prediction value, failing to quantify the uncertainty of the prediction itself, making it difficult for decision-makers to assess the reliability of the results and to conduct effective risk management and resource scheduling; on the other hand, most models ignore the impact of vehicle characteristics (such as load status) and fluctuations in the quality of observation data on prediction accuracy.
[0004] Furthermore, modeling the spatiotemporal correlation between traffic conditions on upstream and downstream road segments is crucial for improving prediction accuracy. Current mainstream methods largely rely on historical data to construct static or quasi-static transition models. When real-time traffic conditions significantly deviate from historical baselines, or when individualized factors such as vehicle load need to be incorporated, the model's adaptability and accuracy decrease significantly. Simultaneously, the credibility of observational evidence is often not assessed during the fusion of real-time vehicle observation data. For example, GPS signals are prone to momentary drift in scenarios such as tunnels and densely populated high-rise areas. Using such noisy data indiscriminately for model updates can actually interfere with the stability of predictions.
[0005] Therefore, there is an urgent need for a new method for predicting the arrival time of goods en route that can comprehensively quantify and predict uncertainties, dynamically capture the spatiotemporal correlation characteristics of road network traffic flow, and intelligently adjust information fusion strategies based on the quality of observation data, in order to cope with complex and ever-changing actual transportation scenarios. Summary of the Invention
[0006] This invention provides a method and system for predicting the arrival time of goods in transit based on real-time environmental data, in order to solve the technical problems of existing methods that are difficult to quantify uncertainty and ignore the quality of real-time data and vehicle status.
[0007] In a first aspect, the present invention provides a method for predicting the arrival time of goods in transit based on real-time environmental data, comprising: S1: Obtain historical and real-time traffic data of the road network and divide the road network into multiple road segments; initialize the probability distribution of arrival time as the initial prior probability distribution; in each iteration, calculate the Shannon entropy based on the prior probability distribution of arrival time at the current moment, and use the Shannon entropy as a quantitative indicator of prediction uncertainty. S2. Based on the prediction uncertainty quantification index, determine the number of upstream road segments N and downstream road segments M. Combine the traffic states of the vehicle's current target road segment, N upstream road segments, and M downstream road segments into a joint traffic state vector. Based on historical data, construct the basic transition probability matrix between the joint traffic state vectors. Combine the deviation of the real-time traffic state from the historical baseline and the vehicle load status to adjust the basic transition probability matrix and obtain the spatiotemporally related transition probability matrix. S3 uses the vehicle's real-time position and instantaneous kinematic parameters as multidimensional observation evidence at the current moment, calculates the degree of fluctuation of the instantaneous kinematic parameters within a preset time window, and obtains the credibility of the observation evidence; and generates evidence update weights by combining the quantitative index of prediction uncertainty and the credibility of the observation evidence. S4. Based on the spatiotemporal correlation transition probability matrix, multidimensional observation evidence, and evidence update weights, the posterior probability distribution of the arrival time at the current moment is obtained through Bayesian update, and this is used as the prior probability distribution for the next iteration. The above iterative process is repeated until the transportation ends, and the predicted arrival time is determined based on the final posterior probability distribution of the arrival time.
[0008] Further, in step S1, the Shannon entropy is calculated based on the prior probability distribution of the arrival time at the current moment, and this Shannon entropy is used as a quantification index of prediction uncertainty, including: Discretize the prediction time range as The time slice, for the first For each time slice, the prior probability of the arrival time is... ; Through formula Calculate the Shannon entropy of the prior probability distribution of arrival time. and the calculated Shannon entropy This serves as a quantitative indicator of the prediction uncertainty.
[0009] Further, in step S2, the number of upstream road segments N and the number of downstream road segments M are determined based on the prediction uncertainty quantification index, including: Preset low uncertainty threshold and high uncertainty threshold ; When the prediction uncertainty quantification index is less than the low uncertainty threshold When the number of upstream road segments N is The number of downstream road sections M is When the quantification index of prediction uncertainty is between and When the upstream road segment N is between, The number of downstream road sections M is When the quantification index of prediction uncertainty is greater than When the number of upstream road segments N is The number of downstream road sections M is ; in, , .
[0010] Furthermore, It is 2. It is 4. It is 6. It is 4. It is 8. It is 12.
[0011] Further, in step S2, a basic transition probability matrix between joint traffic state vectors is constructed based on historical data, and the basic transition probability matrix is adjusted by combining the deviation of real-time traffic conditions from historical baselines and vehicle load conditions, including: The difference between the real-time average speed and the historical average speed of each road segment in the joint traffic state vector is calculated to obtain the speed deviation value. A load adjustment coefficient is set: the load adjustment coefficient is less than 1 under full load, equal to 1 under half load, and greater than 1 under empty load. Each element in the basic transition probability matrix is multiplied by the speed adjustment factor positively correlated with the speed deviation value and the load adjustment coefficient to obtain an intermediate matrix. Each row of the intermediate matrix is normalized so that the sum of the elements in each row is 1, thus obtaining the adjusted spatiotemporal correlation transition probability matrix.
[0012] Further, in step S3, the degree of fluctuation of instantaneous kinematic parameters within a preset time window is calculated to obtain the credibility of the observational evidence, including: Collect the instantaneous speed and instantaneous acceleration of vehicles within a preset time window; Calculate the standard deviation of instantaneous velocity within the preset time window. and the standard deviation of instantaneous acceleration ; Through formula The confidence level of the observational evidence was calculated. ,in, and These are the preset weights for positive constants.
[0013] Further, in step S3, the evidence update weights are generated by comprehensively considering the quantitative index of prediction uncertainty and the credibility of observational evidence, including: The quantitative indicators of forecast uncertainty are normalized to obtain normalized uncertainty. ; Through formula Calculate evidence update weights ,in, To determine the credibility of the observed evidence, The baseline impact factor is between 0 and 1, and the evidence update weights are used for this purpose. Used to adjust the influence of observational evidence during the Bayesian update process.
[0014] Further, in step S4, based on the spatiotemporal correlation transition probability matrix, multidimensional observation evidence, and evidence update weights, the posterior probability distribution of the arrival time at the current moment is obtained through Bayesian update, including: The particle filter algorithm is used to perform Bayesian updates, where the transition probability matrix is used as the state transition model and multidimensional observation evidence is used as the observation model. When updating the weights of each particle, the observation likelihood function is modulated by updating the weights with evidence, and the influence of multidimensional observation evidence on the posterior probability distribution of arrival time is adjusted, thereby obtaining an updated particle set, which is used to represent the posterior probability distribution of arrival time.
[0015] Further, in step S4, the predicted arrival time is determined based on the posterior probability distribution of the final arrival time, including: In the posterior probability distribution of arrival time, the value at each discrete time point is... With the corresponding posterior probability of arrival time Multiply them and sum all the products to get the expected arrival time. The calculation formula is: ; The calculated expected value As the predicted arrival time.
[0016] Secondly, the present invention provides an in-transit cargo arrival time prediction system based on real-time environmental data, including a memory and a processor. The memory stores computer program instructions, and when the computer program instructions are executed by the processor, the above-mentioned in-transit cargo arrival time prediction method based on real-time environmental data is implemented.
[0017] The beneficial effects are as follows: This invention proposes a more accurate and reliable method for predicting the arrival time of goods en route. It not only provides prediction results in the form of probability distributions, quantifying the uncertainty of the prediction, but also identifies the spatiotemporal correlation characteristics of road network traffic flow by constructing traffic state combinations between upstream and downstream road segments and building a basic transition probability matrix based on historical data. Furthermore, it can correct the basic transition probability matrix based on the deviation of real-time traffic from historical baselines and key individual factors such as vehicle load status, obtaining a spatiotemporally correlated transition probability matrix, making the prediction model closely fit the actual transportation situation. By evaluating the credibility of real-time vehicle observation data and combining it with prediction uncertainty quantification indicators to jointly determine the evidence update weights, it avoids interference from poor observation data on the prediction results, achieving a prudent integration of model predictions and measured evidence. Through an iterative Bayesian update process, it can continuously absorb the latest information, improving the accuracy of arrival time prediction in complex traffic environments. Attached Figure Description
[0018] Figure 1 This is a flowchart of a method for predicting the arrival time of goods in transit based on real-time environmental data. Figure 2 To update the particle distribution comparison chart before and after arrival time; Figure 3 This is a schematic diagram of the posterior probability distribution and expected value of arrival time; Figure 4 This is a schematic diagram of a system for predicting the arrival time of goods in transit based on real-time environmental data. Detailed Implementation
[0019] An embodiment of the in-transit cargo arrival time prediction method based on real-time environmental data provided by this invention: like Figure 1 As shown, the method for predicting the arrival time of goods in transit based on real-time environmental data includes: S1: Obtain historical and real-time traffic data of the road network and divide the road network into multiple road segments; initialize the probability distribution of arrival time as the initial prior probability distribution; in each iteration, calculate the Shannon entropy based on the prior probability distribution of arrival time at the current moment, and use the Shannon entropy as a quantitative indicator of prediction uncertainty.
[0020] Specifically, the road network topology of a city or region is obtained in batches through the map service provider's application programming interface (API), and roads are divided into uniquely identified segments using intersections as nodes. Historical traffic data, such as the average traffic speed of road segments at different times of the week, is extracted from the data warehouse. Real-time traffic data, such as the current average speed and congestion level of each road segment, is obtained every minute by calling the real-time traffic information API. At the start of a transportation task, an initial arrival time is estimated based on the total mileage and the historical average speed of the entire road network. , build a The arrival time is determined by using a normal or uniform distribution centered at the origin, which serves as the probability distribution for arrival time; this is the initial prior probability distribution. This probability distribution covers... The time range before and after is relatively large and discretized into time slices with five-minute intervals, and each time slice corresponds to an initial probability.
[0021] Let the prior probability distribution of the arrival time at the current moment be... = ,..., ,in The arrival time falls on the [date]. The probability of each time slice. If the calculated Shannon entropy value is high, for example, greater than 4.5, it indicates that the probability distribution is flat and the uncertainty is high; if the Shannon entropy value is low, for example, less than 2.0, it indicates that the probability is concentrated on a few time slices and the uncertainty is low.
[0022] In an optional embodiment, in step S1, the Shannon entropy is calculated based on the prior probability distribution of the arrival time at the current moment, and this Shannon entropy is used as a quantification index of prediction uncertainty, including: Discretize the prediction time range as The time slice, for the first For each time slice, the prior probability of the arrival time is... ; Through formula Calculate the Shannon entropy of the prior probability distribution of arrival time. and the calculated Shannon entropy This serves as a quantitative indicator of the prediction uncertainty.
[0023] Specifically, the first step is to divide the continuous forecast time range. For example, if the vehicle is predicted to arrive between 10:00 AM and 11:00 AM, this hour is divided into 60 time slices, each lasting one minute. =60, where the first time slice is from 10:00 to 10:01, the second time slice is from 10:01 to 10:02, and so on.
[0024] The second step is to obtain the arrival probability for each time slot. Based on the prior model, a probability distribution is given. For example, the probability P of arriving in the 10:15 to 10:16 time slot is... The probability P(0.1) of arriving in the time slot between 10:16 and 10:17 is 0.1. The probability is 0.15, and the sum of the probabilities of all 60 time slices is 1.
[0025] The third step is to calculate the uncertainty using the Shannon entropy formula, substituting the probability values of all time slices into the formula, for example, calculating... If the probability is highly concentrated in a few adjacent time slices, for example, P( If the probability is 0.8 and the other probability values are very small, then the calculated Shannon entropy H will be very low, such as 1.5, indicating that the prediction result is very certain. Conversely, if the probability is evenly distributed across multiple time slices, for example, each of the 20 consecutive time slices has a probability of 0.05, then the calculated Shannon entropy H will be very high, such as 4.3, indicating that the prediction result has high uncertainty. The calculated Shannon entropy H reflects the reliability of the current prediction.
[0026] S2. Based on the prediction uncertainty quantification index, determine the number of upstream road segments N and downstream road segments M. Combine the traffic states of the vehicle's current target road segment, N upstream road segments, and M downstream road segments into a joint traffic state vector. Based on historical data, construct the basic transition probability matrix between the joint traffic state vectors. Combine the deviation of the real-time traffic state from the historical baseline and the vehicle load state to adjust the basic transition probability matrix and obtain the spatiotemporally correlated transition probability matrix.
[0027] Specifically, a preset entropy threshold is used, such as a high uncertainty threshold. and low uncertainty threshold When the calculated Shannon entropy is greater than When the value is less than 1, it indicates that the prediction is highly uncertain and requires consideration of a broader road network impact, necessitating the setting of higher N and M values. When the Shannon entropy is relatively high, the prediction is relatively certain, and a lower N and M are set to reduce the amount of computation. When the Shannon entropy is between the two, a moderate N and M are set.
[0028] Based on the ratio of real-time speed to free-flow speed, traffic conditions are divided into three levels: smooth, slow, and congested. Assuming the traffic conditions of the vehicle's current target road segment, the three upstream road segments, and the five downstream road segments are, in order, smooth, smooth, slow, congested, smooth, slow, slow, smooth, smooth, then the resulting joint traffic state vector is the state sequence.
[0029] By analyzing historical traffic data, a basic transition probability matrix is constructed based on the frequency with which a vehicle's current target road segment transitions to one of three states (smooth, slow, or congested) in the next time unit after a certain joint traffic state vector appears. During adjustment, the difference between the current real-time speed and the historical average speed for each road segment in the joint traffic state vector is calculated to form a speed adjustment factor. If the current speed is significantly lower than the historical value, the probability of transitioning to a congested state is increased.
[0030] Obtain the vehicle's load status, such as empty, half-loaded, or fully loaded, and set a load adjustment coefficient. If the vehicle is fully loaded, reduce the probability of the vehicle transitioning from a congested or slow-moving state to a smooth state, because heavily loaded vehicles accelerate more slowly. Apply the speed adjustment factor and load adjustment coefficient to the corresponding elements of the basic transition probability matrix using multiplication or weighted summation, and then renormalize to obtain the spatiotemporally correlated transition probability matrix.
[0031] In an optional embodiment, in step S2, determining the number of upstream road segments N and the number of downstream road segments M based on the prediction uncertainty quantification index includes: Preset low uncertainty threshold and high uncertainty threshold ; When the prediction uncertainty quantification index is less than the low uncertainty threshold When the number of upstream road segments N is The number of downstream road sections M is When the quantification index of prediction uncertainty is between and When the upstream road segment N is between, The number of downstream road sections M is When the quantification index of prediction uncertainty is greater than When the number of upstream road segments N is The number of downstream road sections M is ; in, , .
[0032] For example, It is 2. It is 4. It is 6. It is 4. It is 8. It is 12.
[0033] Specifically, two key entropy thresholds need to be set based on historical data and experience; for example, a low uncertainty threshold. Set to 2.0, high uncertainty threshold. The threshold is set to 4.5. These two thresholds categorize the degree of prediction uncertainty into three levels: low, medium, and high. The real-time calculated quantitative index of prediction uncertainty is compared with these two thresholds to adjust the spatial range of traffic condition perception.
[0034] For example, at a certain moment, the current predictive uncertainty quantification index is calculated to be 1.8. Since 1.8 is less than... The current prediction is deemed highly reliable, likely due to stable traffic conditions and smooth vehicle movement. In this case, it's assumed that only the local traffic environment closest to the vehicle's location needs to be considered; therefore, the number of upstream road segments N is set to 2, and the number of downstream road segments M is set to 4. The model will only analyze traffic data from the six road segments before and after the vehicle. If the prediction uncertainty quantification index calculated at another time is 3.5, given... and A value between 5.0 and 8 indicates a moderate level of uncertainty in the forecast. In this case, the analysis scope needs to be expanded, with N set to 4 and M set to 8. If the forecast uncertainty quantification index is as high as 5.0, exceeding 8... If the prediction is highly uncertain, and there may be an accident or severe congestion ahead, the broadest analysis range will be used. In this case, N is set to 6 and M is set to 12 to detect long-distance traffic events that may affect arrival time.
[0035] In an optional embodiment, in step S2, a basic transition probability matrix between joint traffic state vectors is constructed based on historical data, and the basic transition probability matrix is adjusted by combining the deviation of real-time traffic conditions from historical baselines and vehicle load conditions, including: The difference between the real-time average speed and the historical average speed of each road segment in the joint traffic state vector is calculated to obtain the speed deviation value. A load adjustment coefficient is set: the load adjustment coefficient is less than 1 under full load, equal to 1 under half load, and greater than 1 under empty load. Each element in the basic transition probability matrix is multiplied by the speed adjustment factor positively correlated with the speed deviation value and the load adjustment coefficient to obtain an intermediate matrix. Each row of the intermediate matrix is normalized so that the sum of the elements in each row is 1, thus obtaining the adjusted spatiotemporal correlation transition probability matrix.
[0036] Specifically, the first step is to assess real-time traffic conditions: assuming the base transition probability matrix is constructed based on historical average data, the historical average speed for road segment A at 9:00 AM on Tuesday is 50 km / h. However, real-time data shows that the average speed on this section of road is currently only 30 km / h. The speed deviation is -20. A negative deviation indicates that traffic is more congested than usual, and a speed adjustment factor less than 1 will be generated, for example, 0.7. Simultaneously, it is determined that the truck is currently fully loaded, and according to preset rules, the load adjustment coefficient corresponding to full load is set to 0.9, because loaded vehicles accelerate more slowly.
[0037] The second step is to adjust the basic transition probability matrix: Assuming the probability of transitioning from road segment A to road segment B in the basic transition probability matrix is 0.6, to reflect the current situation, this probability value is multiplied sequentially by the speed adjustment factor and the load adjustment coefficient (i.e., 0.6 × 0.7 × 0.9), resulting in a temporary unnormalized value of 0.378. This adjustment process is performed on each probability element in the basic transition probability matrix.
[0038] The third step is normalization: After performing the above adjustments on all probability elements in a row of the basic transition probability matrix, the sum of all elements in that row may no longer equal 1. For example, a row that was originally 0.6 and 0.4 may become 0.378 and 0.32 after adjustment. To make the adjusted elements a reliable probability distribution, the sum of the new row vectors is calculated, resulting in 0.698. Dividing each element in that row by the sum yields adjusted spatiotemporal correlation transition probabilities of approximately 0.54 and 0.46.
[0039] S3 uses the vehicle's real-time position and instantaneous kinematic parameters as multidimensional observation evidence at the current moment, calculates the degree of fluctuation of the instantaneous kinematic parameters within a preset time window, and obtains the credibility of the observation evidence; and generates evidence update weights by combining the quantitative index of prediction uncertainty and the credibility of the observation evidence.
[0040] Specifically, the vehicle's latitude and longitude, instantaneous speed, and instantaneous acceleration are acquired every second via the vehicle's GPS and Controller Area Network (CLAN) bus. These data collectively constitute multidimensional observational evidence. To calculate reliability, instantaneous speed data from the past 30 seconds are selected, and their standard deviation is calculated. If the standard deviation exceeds a preset threshold, the GPS signal is considered to be drifting due to obstruction by tunnels or tall buildings, indicating poor data quality and low reliability. The reliability score can be set as the reciprocal of the standard deviation and normalized.
[0041] The calculated quantitative index of prediction uncertainty and the credibility of observational evidence are normalized so that their values are both between 0 and 1. The evidence update weight is determined by both; when the quantitative index of prediction uncertainty is high or the credibility of observational evidence is high, the generated evidence update weight is correspondingly high, and vice versa.
[0042] In an optional embodiment, in step S3, the fluctuation degree of instantaneous kinematic parameters within a preset time window is calculated to obtain the credibility of the observational evidence, including: Collect the instantaneous speed and instantaneous acceleration of vehicles within a preset time window; Calculate the standard deviation of instantaneous velocity within the preset time window. and the standard deviation of instantaneous acceleration ; Through formula The confidence level of the observational evidence was calculated. ,in, and These are the preset weights for positive constants.
[0043] Specifically, the first step is data acquisition: A time window is set, for example, the past 30 seconds. During this period, instantaneous speed and instantaneous acceleration are continuously collected from the vehicle's sensors (such as GPS or CAN bus), forming two time-series datasets. For example, the instantaneous speed dataset might contain 60.1, 60.5, 59.8, 60.3, ... The instantaneous acceleration dataset includes values of 0.1, -0.2, 0.3, -0.1, ... .
[0044] The second step is to calculate the volatility: calculate the standard deviation for both the instantaneous velocity dataset and the instantaneous acceleration dataset. If a vehicle is cruising at a constant speed on a highway, the speed readings will be very close, and the calculated standard deviation of the instantaneous speed will be... It might only be 0.2 Standard deviation of instantaneous acceleration It is also close to 0. Conversely, if a vehicle frequently starts and stops in congested urban areas, the speed and acceleration readings will change drastically, leading to a more accurate calculation. Possibly as high as 10 , It could be 1.5 .
[0045] The third step is to calculate the confidence level: substitute the two standard deviations obtained into the formula, where and These are preset weighting coefficients, which can be calibrated according to vehicle type, driving scenario, or system robustness requirements. They are typically set to small positive values to avoid over-sensitivity to short-term fluctuations. For example... =0.05, =0.2. For vehicles traveling at a constant speed, the confidence level of observational evidence is... A value very close to 1 indicates high-quality and reliable current observation data. For vehicles that frequently start and stop, due to... and The value within the parentheses is a large positive number, which increases the credibility of the observational evidence. A value close to 0 indicates that the observed data fluctuates wildly and has low reliability.
[0046] In an optional embodiment, in step S3, the evidence update weight is generated by combining the prediction uncertainty quantification index and the credibility of observational evidence, including: Quantifying the uncertainty of prediction Normalization is performed to obtain the normalized uncertainty. ; Through formula Calculate evidence update weights ,in, To determine the credibility of the observed evidence, The baseline impact factor is between 0 and 1, and the evidence update weights are used for this purpose. Used to adjust the influence of observational evidence during the Bayesian update process.
[0047] Specifically, the first step is to normalize the uncertainty index: Assume the system calculates the Shannon entropy H to be 4.0, while the theoretical maximum and minimum values of the Shannon entropy H are known to be 8.0 and 0, respectively. Then, the normalized uncertainty... The calculation result is =0.5. Scaling the uncertainty to the range of 0 to 1 facilitates subsequent calculations.
[0048] The second step is to obtain the confidence level C of the observational evidence and set a preset baseline impact factor β: Assuming the confidence level C of the observational evidence calculated based on vehicle driving stability is 0.9, it indicates that the current observational data quality is very high. The preset baseline impact factor β is 0.1, which ensures that even when the prediction is very certain, new observational evidence still has a basic influence.
[0049] The third step is to calculate the evidence update weight. Substituting the above values into the formula yields a value of 0.495. Evidence Update Weight This is used to adjust the role of new observation data in updating the prediction model. If the prediction was originally highly uncertain, this normalized uncertainty... If the evidence confidence C is close to 1, then the evidence update weights are applied. The new evidence will contribute more to the revised predictions.
[0050] S4. Based on the spatiotemporal correlation transition probability matrix, multidimensional observation evidence, and evidence update weights, the posterior probability distribution of the arrival time at the current moment is obtained through Bayesian update, and this is used as the prior probability distribution for the next iteration. The above iterative process is repeated until the transportation ends, and the predicted arrival time is determined based on the final posterior probability distribution of the arrival time.
[0051] In an optional embodiment, in step S4, based on the spatiotemporal correlation transition probability matrix, multidimensional observation evidence, and evidence update weights, the posterior probability distribution of the arrival time at the current moment is obtained through Bayesian update, including: The particle filter algorithm is used to perform Bayesian updates, where the transition probability matrix is used as the state transition model and multidimensional observation evidence is used as the observation model. When updating the weights of each particle, the observation likelihood function is modulated by updating the weights with evidence, and the influence of multidimensional observation evidence on the posterior probability distribution of arrival time is adjusted, thereby obtaining an updated particle set, which is used to represent the posterior probability distribution of arrival time.
[0052] Specifically, the first step is to initialize and propagate the particles: each particle represents a possible arrival time assumption. For example, particle 1 represents an arrival time of 10:30, and particle 2 represents 10:32. At each time step, a transition probability matrix is used to predict the next state of each particle, i.e., the new location the vehicle might reach, which is equivalent to letting the particle set evolve one step forward according to the traffic model.
[0053] The second step is to calculate and modulate the observation likelihood: when new observational evidence is received, such as the vehicle's GPS location being on road segment X at a speed of 50... The likelihood is calculated by evaluating how well each particle represents its state and matches the new evidence. For example, a particle predicting a vehicle is on road segment Y will have a low likelihood, while a particle predicting a vehicle is on road segment X at a speed close to 50 will have a high likelihood. Before application, the calculated likelihood is multiplied by the previously calculated evidence update weight W. If the evidence update weight W is 0.5, then the likelihood of all particles will be multiplied by 0.5, which will either decrease or increase the overall impact of the new evidence.
[0054] The third step is resampling: the weight of each particle is updated based on the modulated likelihood value. High-weighted particles represent hypotheses that better match observational evidence and are more likely to be replicated and retained in the next resampling process; while low-weighted particles may be discarded. For example... Figure 2 As shown, through iterative cycles of prediction, weight update, and resampling, the particle ensemble gradually gathers into the state that best explains the observed data, thus forming a posterior probability distribution of arrival time.
[0055] In an optional embodiment, in step S4, determining the predicted arrival time based on the final posterior probability distribution of the arrival time includes: In the posterior probability distribution of arrival time, the value at each discrete time point is... With the corresponding posterior probability of arrival time Multiply them and sum all the products to get the expected arrival time. The calculation formula is: ; The calculated expected value As the predicted arrival time.
[0056] Specifically, the first step is to obtain the posterior probability distribution of arrival time: after Bayesian update processes such as particle filtering, a set of discrete probability distributions representing arrival time is obtained. For example, if the predicted time range is divided into multiple 1-minute time slices, the posterior probability distribution may show that the probability of arrival time at 10:45 is 0.1, at 10:46 is 0.4, at 10:47 is 0.3, and at 10:48 is 0.2.
[0057] The second step is to calculate the expected value: To provide a single, definite predicted time, rather than a probability distribution, we calculate the expected value of the posterior probability distribution of the arrival time, i.e., the weighted average. During the calculation, the time points need to be converted into calculable numerical values. For example, 10:45 is considered 645 minutes, 10:46 is considered 646 minutes, 10:47 is considered 647 minutes, and 10:48 is considered 648 minutes. The calculation result is 645 × 0.1 + 646 × 0.4 + 647 × 0.3 + 648 × 0.2 = 646.6 minutes.
[0058] The third step is to output the result: convert the calculated 646.6 minutes back to standard time format, i.e., 10:46.6. The result can be rounded or simply rounded down. The predicted arrival time output is 10:47. Figure 3 As shown, the time obtained is the most reasonable estimate after considering all possibilities and probabilities.
[0059] An embodiment of the in-transit cargo arrival time prediction system based on real-time environmental data provided by this invention: like Figure 4 As shown, the in-transit cargo arrival time prediction system based on real-time environmental data includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the above-mentioned in-transit cargo arrival time prediction method based on real-time environmental data.
[0060] The on-the-way cargo arrival time prediction system based on real-time environmental data also includes other components well known to those skilled in the art, such as communication interfaces. Their settings and functions are known in the art and will not be described in detail here.
[0061] In addition, in the description of this specification, "multiple" means at least two, such as two, three or more, etc., unless otherwise expressly and specifically defined.
Claims
1. A method for predicting the arrival time of goods in transit based on real-time environmental data, characterized in that, include: S1: Obtain historical and real-time traffic data of the road network, divide the road network into multiple road segments; initialize the probability distribution of arrival time as the initial prior probability distribution; In each iteration, the Shannon entropy is calculated based on the prior probability distribution of arrival time at the current moment, and this Shannon entropy is used as a quantitative indicator of prediction uncertainty. S2. Based on the prediction uncertainty quantification index, determine the number of upstream road segments N and downstream road segments M. Combine the traffic states of the vehicle's current target road segment, N upstream road segments, and M downstream road segments into a joint traffic state vector. Based on historical data, construct the basic transition probability matrix between the joint traffic state vectors. Combine the deviation of the real-time traffic state from the historical baseline and the vehicle load status to adjust the basic transition probability matrix and obtain the spatiotemporally related transition probability matrix. S3 uses the vehicle's real-time position and instantaneous kinematic parameters as multidimensional observation evidence at the current moment, calculates the degree of fluctuation of the instantaneous kinematic parameters within a preset time window, and obtains the credibility of the observation evidence; and generates evidence update weights by combining the quantitative index of prediction uncertainty and the credibility of the observation evidence. S4. Based on the spatiotemporal correlation transition probability matrix, multidimensional observation evidence, and evidence update weights, the posterior probability distribution of the arrival time at the current moment is obtained through Bayesian update, and this is used as the prior probability distribution for the next iteration. The above iterative process is repeated until the transportation ends, and the predicted arrival time is determined based on the final posterior probability distribution of the arrival time.
2. The method for predicting the arrival time of goods in transit based on real-time environmental data according to claim 1, characterized in that, In step S1, the Shannon entropy is calculated based on the prior probability distribution of the arrival time at the current moment, and this Shannon entropy is used as a quantification index of prediction uncertainty, including: Discretize the prediction time range as The time slice, for the first For each time slice, the prior probability of the arrival time is... ; Through formula Calculate the Shannon entropy of the prior probability distribution of arrival time. and the calculated Shannon entropy This serves as a quantitative indicator of the prediction uncertainty.
3. The method for predicting the arrival time of goods in transit based on real-time environmental data according to claim 1, characterized in that, In step S2, the number of upstream road segments N and the number of downstream road segments M are determined based on the prediction uncertainty quantification index, including: Preset low uncertainty threshold and high uncertainty threshold ; When the prediction uncertainty quantification index is less than the low uncertainty threshold When the number of upstream road segments N is The number of downstream road sections M is When the quantification index of prediction uncertainty is between and When the upstream road segment N is between, The number of downstream road sections M is When the quantification index of prediction uncertainty is greater than When the number of upstream road segments N is The number of downstream road sections M is ; in, , .
4. The method for predicting the arrival time of goods in transit based on real-time environmental data according to claim 3, characterized in that, It is 2. It is 4. It is 6. It is 4. It is 8. It is 12.
5. The method for predicting the arrival time of goods in transit based on real-time environmental data according to claim 3, characterized in that, In step S2, a basic transition probability matrix between joint traffic state vectors is constructed based on historical data. This matrix is then adjusted by considering the deviation of real-time traffic conditions from historical baselines and vehicle load conditions. This adjustment includes: The difference between the real-time average speed and the historical average speed of each road segment in the joint traffic state vector is calculated to obtain the speed deviation value. A load adjustment coefficient is set: the load adjustment coefficient is less than 1 under full load, equal to 1 under half load, and greater than 1 under empty load. Each element in the basic transition probability matrix is multiplied by the speed adjustment factor positively correlated with the speed deviation value and the load adjustment coefficient to obtain an intermediate matrix. Each row of the intermediate matrix is normalized so that the sum of the elements in each row is 1, thus obtaining the adjusted spatiotemporal correlation transition probability matrix.
6. The method for predicting the arrival time of goods in transit based on real-time environmental data according to claim 1, characterized in that, In step S3, the fluctuation of instantaneous kinematic parameters within a preset time window is calculated to obtain the credibility of the observational evidence, including: Collect the instantaneous speed and instantaneous acceleration of vehicles within a preset time window; Calculate the standard deviation of instantaneous velocity within the preset time window. and the standard deviation of instantaneous acceleration ; Through formula The confidence level of the observational evidence was calculated. ,in, and These are the preset weights for positive constants.
7. The method for predicting the arrival time of goods in transit based on real-time environmental data according to claim 6, characterized in that, In step S3, the evidence update weights are generated by combining the quantitative index of prediction uncertainty and the credibility of observational evidence, including: The quantitative indicators of forecast uncertainty are normalized to obtain normalized uncertainty. ; Through formula Calculate evidence update weights ,in, To determine the credibility of the observed evidence, The baseline impact factor is between 0 and 1, and the evidence update weights are used for this purpose. Used to adjust the influence of observational evidence during the Bayesian update process.
8. The method for predicting the arrival time of goods in transit based on real-time environmental data according to any one of claims 1-7, characterized in that, In step S4, based on the spatiotemporal correlation transition probability matrix, multidimensional observation evidence, and evidence update weights, the posterior probability distribution of arrival time at the current moment is obtained through Bayesian update, including: The particle filter algorithm is used to perform Bayesian updates, where the transition probability matrix is used as the state transition model and multidimensional observation evidence is used as the observation model. When updating the weights of each particle, the observation likelihood function is modulated by updating the weights with evidence, and the influence of multidimensional observation evidence on the posterior probability distribution of arrival time is adjusted, thereby obtaining an updated particle set, which is used to represent the posterior probability distribution of arrival time.
9. The method for predicting the arrival time of goods in transit based on real-time environmental data according to claim 8, characterized in that, In step S4, the predicted arrival time is determined based on the final posterior probability distribution of the arrival time, including: In the posterior probability distribution of arrival time, the value at each discrete time point is... With the corresponding posterior probability of arrival time Multiply them and sum all the products to get the expected arrival time. The calculation formula is: ; The calculated expected value As the predicted arrival time.
10. A system for predicting the arrival time of goods in transit based on real-time environmental data, characterized in that, It includes a memory and a processor, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the in-transit cargo arrival time prediction method based on real-time environmental data as described in any one of claims 1-9 is implemented.