An efficiency prediction system for port industry and port logistics collaboration

Abstract

The application discloses a kind of efficiency prediction systems for facing port industry and port logistics coordination, it is related to port logistics efficiency prediction technical field, including following steps: collection port industry goods in port's retention time, and collection truck in port's turnover time of transporting port industry goods, obtain goods retention time data and truck turnover time data;And carry out hierarchical calculation processing, obtain average retention data and average turnover data;And respectively carry out analysis correction processing, obtain reference retention turnover data;Efficiency prediction model is constructed based on reference retention turnover data, and average retention time of goods is predicted;The present application is used to solve the problem that the existing port logistics efficiency prediction technology cannot establish a prediction model according to the historical retention time of goods in the port when analyzing and predicting the retention time of goods in the port of port industry, objectively and stably predict the retention time of goods in the port.

Description

Technical Field

[0001] This invention relates to the field of port logistics efficiency prediction technology, specifically an efficiency prediction system for the collaboration between port-related industries and port logistics. Background Technology

[0002] Port logistics efficiency prediction technology refers to a comprehensive technical system that uses multi-source operational data from the entire port logistics process to predict the changing trends of core port logistics efficiency indicators in a specific future period through data preprocessing, model building, and training. This provides data support for port scheduling optimization, port-related industry collaboration, and logistics resource allocation.

[0003] Accurate prediction of the synergy efficiency between port-adjacent industries and port logistics is crucial throughout the entire chain of port operation optimization, cost reduction and efficiency improvement for port-adjacent industries, and enhancement of regional economic competitiveness. For ports, accurate prediction of changes in key indicators corresponding to synergy efficiency allows for targeted adjustments to maximize the smoothness of logistics operations. For port-adjacent industries, reliable synergy efficiency prediction can guide their industrial production plans and mitigate supply chain disruption risks. Accurate synergy efficiency prediction can also provide a basis for macro-level decisions such as port spatial layout and the improvement of multimodal transport systems. However, existing port logistics efficiency prediction technologies often use Data Envelopment Analysis (DEA) to analyze and predict the dwell time of goods in port-adjacent industries. DEA is highly sensitive to data quality, has poor anti-interference capabilities, and poor prediction stability. Furthermore, using DEA often requires collecting multiple data sources, and the selection of input and output indicators is subjective, affecting the objectivity of the results. Therefore, existing port logistics efficiency prediction technologies cannot establish predictive models based on historical dwell time of goods in port-adjacent industries to provide objective and stable predictions of the dwell time of goods in port. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in the prior art. It obtains cargo dwell time data and truck turnaround time data by collecting data on the port dwell time of goods from port-related industries and the turnaround time of trucks transporting these goods. These data are then processed through hierarchical calculations to obtain average dwell time data and average turnaround time data. Further analysis and correction are performed to obtain reference dwell time and turnaround time data. An efficiency prediction model is then constructed based on this reference dwell time and turnaround time data to predict the average dwell time of goods. This addresses the problem that existing port logistics efficiency prediction technologies cannot establish a prediction model based on historical dwell time of goods in the port, thus failing to objectively and stably predict the dwell time of goods in the port.

[0005] To achieve the above objectives, this application provides an efficiency prediction system for the collaboration between port-adjacent industries and port logistics, including a data collection module, a hierarchical statistics module, an analysis and correction module, and an efficiency prediction module.

[0006] The data collection module is used to collect the dwell time of goods in port-adjacent industries and the turnaround time of trucks transporting goods in port, thereby obtaining cargo dwell time data and truck turnaround time data.

[0007] The hierarchical statistics module includes a first processing unit and a second processing unit. The first processing unit is used to perform hierarchical calculation and processing on cargo dwell time data to obtain average dwell time data; the second processing unit is used to perform hierarchical calculation and processing on truck turnaround time data to obtain average turnaround data.

[0008] The analysis and correction module performs analysis and correction processing based on the average retention data and the average turnover data to obtain reference retention and turnover data.

[0009] The efficiency prediction module includes a construction unit and a prediction unit. The construction unit builds an efficiency prediction model based on reference retention and turnover data, and the prediction unit is used to predict the average retention time of goods.

[0010] Furthermore, the data collection module is configured with a data collection strategy, which includes:

[0011] Let any port to be predicted be referred to as the first port, any port-adjacent industry corresponding to the first port be referred to as the first port-adjacent industry, and the goods that the first port-adjacent industry needs to transport through the first port be referred to as the first goods; let the container carrying the first goods be referred to as the first container; and let the truck transporting the first container be referred to as the first truck.

[0012] The time taken from the arrival of any first container at the first port to its departure from the first port is recorded as the corresponding first container's dwell time at the port.

[0013] The time taken for any first truck to enter the first port and leave the first port is recorded as the turnaround time of the corresponding first truck at the port.

[0014] Furthermore, data collection strategies also include:

[0015] Set the data collection time period to T1. Any data collection time period is recorded as the first time period. Within the first time period, the dwell time of each container leaving the first port is collected and recorded as the cargo dwell time information of the first time period.

[0016] Within the first time period, the turnaround time of each truck at the port is collected and recorded as the truck turnaround time information for the first time period.

[0017] The information on cargo dwell time and truck turnaround time is collected periodically and repeatedly for each collection period, and then arranged in chronological order and recorded as cargo dwell time data and truck turnaround time data, respectively.

[0018] Furthermore, the first processing unit is configured with a first processing strategy, which includes:

[0019] Arrange the cargo dwell time information of the first time period in ascending order and denote it as the first dwell time sequence. Calculate the mean and standard deviation of the first dwell time sequence and denote them as AP and AB in order. Remove the dwell time in the first dwell time sequence that is not located in [AP-k1*AB, AP+k1*AB] and obtain the second dwell time sequence. Here, k1 is the set proportional coefficient.

[0020] Set the interval duration to t2, and set multiple consecutive duration intervals based on t2, which are denoted as duration interval 1 to duration interval n1, where n1 is the total number of duration intervals set; calculate the standard deviation of the second retention sequence, denoted as CB;

[0021] Based on the second retention sequence, count the number of retention times contained in each time interval from time interval 1 to time interval 2, and denote the time interval with the most and second most retention times as the first interval and the second interval, respectively.

[0022] Furthermore, the first processing strategy also includes:

[0023] Expand the first interval to both sides by a range of CB to obtain the first dense interval, and expand the second interval to both sides by a range of CB to obtain the second dense interval; take the union of the first dense interval and the second dense interval, and denote it as the effective dense interval; denote the dwell time in the second dwell sequence that is located in the effective dense interval as the normal dwell time, and count the total number of normal dwell times, and denote it as AF;

[0024] The effective dense interval is evenly divided into n2 sub-intervals, which are denoted as dense sub-interval 1 to dense sub-interval n2 respectively; and the number of normal dwell times contained in each dense sub-interval is counted, where n2 is the set number.

[0025] Let any normal dwell time be denoted as the first dwell time LT, and let the number of normal dwell times contained in the dense subinterval containing the first dwell time be denoted as BF; let BF / AF be denoted as the weight of the first dwell time, and let (BF / AF)*LT be denoted as the weighted dwell time of the first dwell time.

[0026] Repeatedly calculate the weights of all normal dwell times and the corresponding weighted dwell times, and sum them separately, recording them as AQ and QT respectively in order; calculate QT / AQ, and record it as the representative dwell time of the first time period; repeatedly calculate the representative dwell time of each collection time period, and arrange them in chronological order, recording them as the dwell time series, and label them as the average dwell data.

[0027] Furthermore, the second processing unit is configured with a second processing strategy, which includes:

[0028] Arrange the truck turnaround time information of the first time period in ascending order and denote it as the first turnaround sequence. Calculate the mean and standard deviation of the first turnaround sequence and denote them as DP and DB in order. Remove the dwell time in the first turnaround sequence that is not located in [DP-k2*DB, DP+k2*DB] and obtain the second turnaround sequence. Here, k2 is the set proportional coefficient.

[0029] Plot a kernel density curve for the second rotation sequence and obtain the lowest point between the two peaks on the kernel density curve, which is denoted as the valley point. Use the valley point as a threshold to divide the second rotation sequence into two parts, which are denoted as the first rotation sequence and the second rotation sequence, respectively.

[0030] For the first week's rotor sequence, calculate the median and coefficient of variation of the first week's rotor sequence, and denote them as CE and CV respectively in order. Calculate the corrected representative value corresponding to the first week's rotor sequence, denoted as XT1, where XT1 = CE * (1 - k3 * CV), and k3 is the set scaling factor. Repeat the calculation of the corrected representative value corresponding to the second week's rotor sequence, denoted as XT2.

[0031] Obtain the number of turnover times contained in the first and second rotor sequences, respectively, and denote them as WF1 and WF2 in order; calculate [WF1 / (WF1+WF2)]*XT1+[WF2 / (WF1+WF2)]*XT2, and denote it as the representative turnover time of the first time period; repeat the calculation of the representative turnover time for each acquisition time period, arrange them in chronological order, denote them as the turnover time series, and mark them as the average turnover data.

[0032] Furthermore, the analysis and correction module is configured with analysis and correction strategies, which include:

[0033] For a time series, any data point in the time series is denoted as TL(i), where i represents the position number; k4 data points are taken before and after TL(i) and denoted as the adjacent data of TL(i), and the mean and standard deviation of the adjacent data are calculated and denoted as GP(i) and GB(i) in order, where k4 is the number of sets;

[0034] If the absolute difference between TL(i) and GP(i) is greater than k5*GB(i), then TL(i) is marked as a suspected outlier; otherwise, it is marked as a normal data point; where k5 is the set scaling factor.

[0035] If TL(i) is a suspected outlier, then perform linear fitting from TL(i-k6) to TL(i) to obtain the slope of the fitting, denoted as E1; and perform linear fitting from TL(i) to TL(i+k6) to obtain the slope of the fitting, denoted as E2.

[0036] Furthermore, the analysis and correction strategies also include:

[0037] Let |E1-E2| be denoted as the slope difference CE corresponding to TL(i); repeatedly calculate the slope difference corresponding to all data in the residence time series; and take the slope difference of the first k6 data of TL(i) as the set of slope differences of the first neighboring region; and take the slope difference of the last k6 data of TL(i) as the set of slope differences of the last neighboring region; where k6 is the set number;

[0038] Calculate the mean and standard deviation of the set of slope differences in the previous neighborhood, and denote them as HP1 and HB1 in order; calculate the mean and standard deviation of the set of slope differences in the subsequent neighborhood, and denote them as HP2 and HB2 in order.

[0039] If the absolute difference between CE and HP1 is greater than k7*HB1 and the absolute difference between CE and HP2 is greater than k7*HB2, then mark TL(i) as an event anomaly point; otherwise, mark TL(i) as a normal data point, and repeat the acquisition of all event anomaly points in the retention time series.

[0040] Furthermore, the analysis and correction strategies also include:

[0041] If TL(i) is an event outlier, then the adjacent data of TL(i) are arranged in order of size, and the median is obtained and recorded as the neighborhood median. TL(i) is replaced with the neighborhood median. The replacement is repeated for all event outliers. After completion, the corrected sequence corresponding to the retention time series is obtained and recorded as the reference retention sequence.

[0042] Repeatedly acquire the corrected sequence corresponding to the turnover time series, and record it as the reference turnover sequence; record the reference retention sequence and the reference turnover sequence as the reference retention turnover data of the first port-adjacent industry; repeatedly acquire the reference retention turnover data of multiple port-adjacent industries.

[0043] Furthermore, the efficiency prediction module is configured with an efficiency prediction strategy, which includes:

[0044] An initial prediction model is constructed based on the BiLSTM model. The initial prediction model includes an input layer, a bidirectional LSTM layer, a Dropout layer, a fully connected layer, and an output layer. The model input is set as the reference dwell time sequence and the corresponding reference turnover sequence, and the model output is the representative dwell time for each future collection time period.

[0045] The initial prediction model was trained using reference retention and turnover data to obtain an efficiency prediction model. Then, using reference retention and turnover data of the first port-adjacent industry, the representative retention time of the first port-adjacent industry in each future collection time period was predicted to obtain the efficiency prediction results corresponding to the collaboration between the first port-adjacent industry and port logistics.

[0046] The beneficial effects of this invention are as follows: This invention collects the dwell time of goods in port-adjacent industries and the turnaround time of trucks transporting these goods, obtaining cargo dwell time data and truck turnaround time data; it then performs hierarchical calculations on the cargo dwell time data and truck turnaround time data to obtain average dwell time data and average turnaround data; based on the average dwell time data and average turnaround data, it performs analysis and correction processing to obtain reference dwell time and turnaround data; based on the reference dwell time and turnaround data, it constructs an efficiency prediction model to predict the average dwell time of goods; when analyzing and predicting the dwell time of goods in port-adjacent industries, a prediction model can be established based on the historical dwell time of goods in the port to provide an objective and stable prediction of the dwell time of goods in the port.

[0047] This invention first removes extreme outliers, then statistically analyzes dwell time by interval to find effective dense intervals, and further subdivides these intervals into sub-intervals. Each dwell time is weighted according to its sample proportion within its sub-interval, resulting in a representative dwell time for that period. Compared to a simple arithmetic mean, this weighting assigns greater weight to common dwell times, making the output representative value more reflective of normal dwell times rather than noise or extreme events. Turnover time is divided into subsequences using the valleys of the kernel density curve, and each subsequence is corrected using the median and coefficient of variation. This allows for convergence of the representative value even with a low coefficient of variation, avoiding overestimation or unstable estimation of the representative value due to dispersion. Furthermore, the slope difference is used to distinguish between sudden event anomalies and normal fluctuations, differentiating between sudden event anomalies and genuine trend fluctuations, preventing the correction of genuine trend changes as anomalies, and reducing the risk of incorrect corrections. Attached Figure Description

[0048] Figure 1 This is a schematic diagram of the system of the present invention;

[0049] Figure 2 This is a flowchart of the steps of the method of the present invention;

[0050] Figure 3This is a flowchart illustrating the process of obtaining the representative residence time of the present invention;

[0051] Figure 4 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Example 1, please refer to Figure 1 As shown, this application provides an efficiency prediction system for the collaboration between port-adjacent industries and port logistics, including a data collection module, a hierarchical statistics module, an analysis and correction module, and an efficiency prediction module.

[0054] The data collection module is used to collect the dwell time of goods in port-adjacent industries and the turnaround time of trucks transporting goods in port, thus obtaining cargo dwell time data and truck turnaround time data.

[0055] The data collection module is configured with a data collection strategy, which includes: designating any port to be predicted as the first port, designating any port-adjacent industry corresponding to the first port as the first port-adjacent industry, designating the goods that the first port-adjacent industry needs to transport through the first port as the first goods; designating the container carrying the first goods as the first container; and designating the truck transporting the first container as the first truck.

[0056] The time taken from the arrival of any first container at the first port to its departure from the first port is recorded as the corresponding first container's dwell time at the port. The shorter the dwell time of the first container at the port, that is, the shorter the dwell time of the goods, the higher the efficiency of port logistics in the customs clearance, loading and unloading, and transportation of goods for port-adjacent industries. This is the core indicator of the efficiency of port logistics collaboration between port-adjacent industries and port logistics.

[0057] The time taken for any first truck to enter the first port and leave the first port is recorded as the turnaround time of the first truck in the port. The turnaround time of the first truck in the port reflects the operation of the entire process from the truck entering the port to picking up and delivering goods and leaving the port. The shorter the time, the smoother the connection between port gate passage, yard scheduling and loading and unloading operations, the higher the efficiency of matching trucks and goods, and it also shows that the port logistics and port-adjacent enterprises are closely coordinated.

[0058] The data collection period is set to T1. Any data collection period is recorded as the first time period. Within the first time period, the dwell time of each container leaving the first port is collected and recorded as the cargo dwell time information of the first time period. T1 can be flexibly set according to the actual application scenario. In this embodiment, T1 = 7 days, i.e., one week.

[0059] Within the first time period, the turnaround time of each truck at the port is collected and recorded as the truck turnaround time information for the first time period.

[0060] The information on cargo dwell time and truck turnaround time is collected periodically and repeatedly for each collection period, and then arranged in chronological order and recorded as cargo dwell time data and truck turnaround time data, respectively.

[0061] In the specific implementation process, truck turnaround time is the core variable that directly affects cargo dwell time. It is also easy to collect, and there is an instant transmission relationship between the two, which cannot be replaced by other variables, such as weather and policies. When predicting cargo dwell time in the future, adding only the variable representing truck turnaround time can simplify the structure of the subsequent prediction model to the greatest extent, reduce the cost of data collection and processing, and improve the accuracy and stability of subsequent predictions without increasing the difficulty and complexity.

[0062] The hierarchical statistics module includes a first processing unit and a second processing unit. The first processing unit is used to perform hierarchical calculation and processing on cargo dwell time data to obtain average dwell time data; the second processing unit is used to perform hierarchical calculation and processing on truck turnaround time data to obtain average turnaround data.

[0063] The first processing unit is configured with a first processing strategy, which includes: (see details) Figure 3 As shown, the cargo dwell time information for the first time period is arranged in ascending order and denoted as the first dwell time sequence. The mean and standard deviation of the first dwell time sequence are calculated and distributed in order as AP and AB. Dwell times in the first dwell time sequence that are not located in [AP-k1*AB, AP+k1*AB] are removed. After this, the second dwell time sequence is obtained, where k1 is a set proportional coefficient. In this embodiment, k1=3, which can be adjusted flexibly. Removing dwell times that are not located in [AP-k1*AB, AP+k1*AB] can effectively remove measurement errors, input anomalies, or extreme cases, and reduce the impact of a few erroneous data on subsequent calculations and processing.

[0064] The interval duration is set to t2, meaning the size of each duration interval is t2. Multiple consecutive duration intervals are set based on t2, denoted as duration interval 1 to duration interval n1, where n1 is the total number of duration intervals. The standard deviation of the second retention sequence is calculated and denoted as CB. In this embodiment, t2 = 0.5 hours, which can be set flexibly but should not be too large, generally not exceeding 2 hours. Setting duration intervals can discretize the continuous retention time space, making it easier to count the number of samples in each interval, which is convenient for subsequent processing.

[0065] Based on the second delay sequence, count the number of delay times contained in each time interval from time interval 1 to time interval 2. The time interval with the most and second most delay times is denoted as the first interval and the second interval, respectively. By counting, find the two intervals with the densest distribution to discover the main behavioral patterns of the data. Because the distribution characteristics of the delay times corresponding to goods shipped from the port and goods received at the port are not the same, it is necessary to find the two intervals with the densest distribution.

[0066] The first interval is expanded to both sides by one CB to obtain the first dense interval, and the second interval is expanded to both sides by one CB to obtain the second dense interval. The union of the first and second dense intervals is denoted as the effective dense interval. The residence time of the second residence sequence located in the effective dense interval is denoted as the normal residence time, and the total number of normal residence times is denoted as AF. Simply taking the interval boundary may be too strict. Expanding the CB based on the standard deviation can include nearby and reasonable samples, avoiding ignoring normal samples outside the boundary. The effective dense interval can cover the intervals corresponding to the two main patterns.

[0067] The effective dense interval is evenly divided into n2 sub-intervals, which are denoted as dense sub-interval 1 to dense sub-interval n2. The number of normal dwell times contained in each dense sub-interval is counted, where n2 is the set number. In this embodiment, n2=20, and n2 can be flexibly set according to the actual application scenario. Subdividing the effective dense interval into sub-intervals can more accurately estimate the sample proportion of a specific duration.

[0068] Let any normal dwell time be denoted as the first dwell time LT, and let the number of normal dwell times contained in the dense subinterval containing the first dwell time be denoted as BF; let BF / AF be denoted as the weight of the first dwell time, and let (BF / AF)*LT be denoted as the weighted dwell time of the first dwell time.

[0069] Repeatedly calculate the weights of all normal dwell times and their corresponding weighted dwell times, sum them separately, and record them as AQ and QT respectively in order; calculate QT / AQ, and record it as the representative dwell time of the first time period. Repeatedly calculate the representative dwell time of each collection time period, arrange them in chronological order, and record them as the dwell time series, which is marked as the average dwell data; if there are more samples in a sub-interval of a certain duration, it means that the duration is more typical and has a higher weight, so the representative value is closer to the normal rather than a rare extreme; avoid the pull of a few but extreme normal or marginal samples on the representative value, and improve representativeness and robustness.

[0070] The second processing unit is configured with a second processing strategy, which includes: arranging the truck turnaround time information of the first time period in ascending order, denoted as the first turnaround sequence; calculating the mean and standard deviation of the first turnaround sequence, and distributing them sequentially as DP and DB; removing the dwell time in the first turnaround sequence that is not located in [DP-k2*DB, DP+k2*DB]; and obtaining the second turnaround sequence after completion, where k2 is a set proportional coefficient; in this embodiment, k2=3, which can be flexibly adjusted; and initially removing extreme outliers to obtain relatively clean data for processing more stable input.

[0071] Plot a kernel density curve for the second turnover sequence and obtain the lowest point between the two peaks on the kernel density curve, which is denoted as the valley point. Use the valley point as a threshold to divide the second turnover sequence into two parts, which are denoted as the first week rotor sequence and the second week rotor sequence, respectively. Truck turnover time data often shows a bimodal distribution because the turnover time of pickup trucks and delivery trucks is different, so two peaks will be formed.

[0072] For the first week's rotor sequence, calculate the median and coefficient of variation of the first week's rotor sequence, and denote them as CE and CV respectively in order. Calculate the corrected representative value corresponding to the first week's rotor sequence, denoted as XT1, where XT1 = CE * (1 - k3 * CV), and k3 is the set scaling factor. Repeat the calculation of the corrected representative value corresponding to the second week's rotor sequence, denoted as XT2. In this embodiment, k3 = 0.5, which can be flexibly set.

[0073] Obtain the number of turnaround times contained in the first and second rotor sequences, respectively, and denote them as WF1 and WF2 in order. Calculate [WF1 / (WF1+WF2)]*XT1 + [WF2 / (WF1+WF2)]*XT2, and denote it as the representative turnaround time of the first time period. Repeat the calculation of the representative turnaround time for each acquisition time period, arrange them in chronological order, and denote them as the turnaround time series, labeled as the average turnaround data. If the data volume of a subsequence is larger, its representative value has a greater impact on the final representative turnaround time, making the representative turnaround time more reasonable.

[0074] In practice, the smaller the coefficient of variation, the more stable the corresponding weekly rotor sequence data, and the smaller the correction value; conversely, the correction value needs to be used to adjust the median. If the weekly rotor sequence has a large dispersion, that is, a large coefficient of variation, it means that the mode is unstable or fluctuates greatly. The median is corrected downward by the coefficient of variation, that is, the representative value is reduced, so as to reduce the impact of high fluctuation group on the overall representativeness; avoid directly regarding the median of a high fluctuation group as a stable representative.

[0075] The analysis and correction module performs analysis and correction processing based on the average retention data and the average turnover data to obtain reference retention and turnover data.

[0076] The analysis and correction module is configured with an analysis and correction strategy, which includes: for a time series, any data point in the time series is denoted as TL(i), where i represents the position number; k4 data points are taken before and after TL(i) and denoted as the adjacent data of TL(i), and the mean and standard deviation of the adjacent data are calculated, excluding TL(i), and denoted as GP(i) and GB(i) in order, where k4 is the set number; in this embodiment, k4=3, which can be flexibly set; unlike using the global mean and standard deviation of the entire series, the local mean and standard deviation can reflect the normal level and fluctuation range near the data point, thereby avoiding the judgment of reasonable fluctuations as abnormal;

[0077] If the absolute difference between TL(i) and GP(i) is greater than k5*GB(i), then TL(i) is marked as a suspected outlier; otherwise, it is marked as a normal data point. Here, k5 is a set scaling factor. In this embodiment, k5=3, which can be set flexibly. Points that deviate significantly from the baseline are picked out to reduce the amount of subsequent calculations.

[0078] If TL(i) is a suspected outlier, then perform linear fitting from TL(i-k6) to TL(i) to obtain the slope of the fitting, denoted as E1; and perform linear fitting from TL(i) to TL(i+k6) to obtain the slope of the fitting, denoted as E2.

[0079] Let |E1-E2| be denoted as the slope difference CE corresponding to TL(i); repeatedly calculate the slope difference corresponding to all data in the residence time series; and take the slope difference of the first k6 data of TL(i) as the set of slope differences of the first neighboring region; and take the slope difference of the last k6 data of TL(i) as the set of slope differences of the last neighboring region; where k6 is the set number; in this embodiment, k6=3, which can be set flexibly; a single point anomaly will cause a huge difference in the slope between the front and back sides; while the fitting slope of the true trend breakpoint or continuous change will have a certain difference in the front and back segments; the slope difference can quantify the severity of the sudden change, thereby screening out the true anomaly points;

[0080] Calculate the mean and standard deviation of the set of slope differences in the previous neighborhood, and denote them as HP1 and HB1 in order; calculate the mean and standard deviation of the set of slope differences in the subsequent neighborhood, and denote them as HP2 and HB2 in order; by comparing the slope difference of CE with the slope difference of surrounding points, it can be determined whether the CE exceeds the normal slope difference level for that period.

[0081] If the absolute difference between CE and HP1 is greater than k7*HB1 and the absolute difference between CE and HP2 is greater than k7*HB2, then TL(i) is marked as an event anomaly point; otherwise, TL(i) is marked as a normal data point. All event anomaly points in the dwell time series are repeatedly obtained. In this embodiment, k7=2.5, which can be flexibly set. It is required that CE is significantly different from the normal levels on both sides before and after, which can avoid misjudging a point that is part of the trend on one side as an anomaly. For example, a point in a continuously rising trend may be different from the previous segment, but similar to the next segment, and should not be regarded as an anomaly.

[0082] If TL(i) is an event outlier, then the adjacent data of TL(i) are arranged in order of size, and the median is obtained and denoted as the neighborhood median. TL(i) is replaced with the neighborhood median. This replacement is repeated for all event outliers. After completion, the corrected sequence corresponding to the retention time series is obtained and denoted as the reference retention sequence. Replacing with the neighborhood median instead of the mean or interpolation can better resist the influence of outliers that may also exist in adjacent positions. The replacement can smooth out sudden event anomalies without changing the overall trend.

[0083] Repeatedly obtain the corrected sequence corresponding to the turnover time series, which is consistent with the process of obtaining the reference retention sequence; record it as the reference turnover sequence; record the reference retention sequence and the reference turnover sequence as the reference retention turnover data of the first port-adjacent industry; repeatedly obtain the reference retention turnover data of multiple port-adjacent industries;

[0084] In practice, the data at event anomalies are not erroneous data, but rather data that deviates from normal levels due to interference from special events. For example, if temporary port maintenance or extreme weather causes a sudden increase or decrease in representative dwell time during a certain week, such anomalies can seriously mislead subsequent model learning and need to be corrected.

[0085] The efficiency prediction module includes a construction unit and a prediction unit. The construction unit builds an efficiency prediction model based on reference retention and turnover data, and the prediction unit is used to predict the average retention time of goods.

[0086] The efficiency prediction module is configured with an efficiency prediction strategy, which includes: constructing an initial prediction model based on a BiLSTM model. The initial prediction model includes an input layer, a bidirectional LSTM layer, a Dropout layer, a fully connected layer, and an output layer. The model input is set as the reference dwell time sequence and the corresponding reference turnover sequence, and the model output is the representative dwell time for each future collection time period. The bidirectional LSTM layer can learn time dependencies within each time window. The bidirectional structure allows the network to capture temporal correlations from the past to the present and from the present to the past simultaneously within the same window, making the prediction more accurate. The Dropout layer can randomly discard a portion of neuron connections, reducing overfitting and enhancing the model's generalization ability.

[0087] The initial prediction model was trained using reference retention and turnover data, resulting in an efficiency prediction model. Then, using reference retention and turnover data of the first port-adjacent industry, the representative retention time of the first port-adjacent industry in each future collection time period was predicted, and the efficiency prediction results corresponding to the collaboration between the first port-adjacent industry and port logistics were obtained.

[0088] In practice, other machine learning models or algorithms can be selected for prediction based on the actual application scenario, such as the CNN+LSTM hybrid model, the GRU model, and the LSTM model.

[0089] Example 2, please refer to Figure 2 As shown, this application provides an efficiency prediction method for the collaboration between port-adjacent industries and port logistics, including the following steps:

[0090] Step S1 involves collecting the port dwell time of goods from port-adjacent industries and the turnaround time of trucks transporting these goods, thus obtaining cargo dwell time data and truck turnaround time data. Step S1 includes the following sub-steps:

[0091] Step S101: Any port to be predicted is designated as the first port; any port-adjacent industry corresponding to the first port is designated as the first port-adjacent industry; the goods that the first port-adjacent industry needs to transport through the first port are designated as the first goods; the container carrying the first goods is designated as the first container; and the truck transporting the first container is designated as the first truck.

[0092] Step S102: Record the time taken from the arrival of any first container at the first port to its departure from the first port as the corresponding first container's dwell time at the port.

[0093] Step S103: Record the time taken for any first truck to enter the first port and leave the first port as the turnaround time of the corresponding first truck at the port.

[0094] Step S104: Set the collection time period to T1. Any collection time period is recorded as the first time period. Within the first time period, collect the dwell time corresponding to the first container leaving the first port and record it as the cargo dwell time information of the first time period.

[0095] Step S105, and within the first time period, collect the turnaround time of each first truck at the port, and record it as the truck turnaround time information of the first time period;

[0096] Step S106: Periodically collect cargo dwell time information and truck turnaround time information for each collection time period, and arrange them in chronological order, recording them as cargo dwell time data and truck turnaround time data respectively.

[0097] Step S2 involves performing stratified calculations on the cargo dwell time data and truck turnaround time data to obtain average dwell time data and average turnaround data. Step S2 includes the following sub-steps:

[0098] Step S201: Arrange the cargo dwell time information of the first time period in ascending order and record it as the first dwell time sequence. Calculate the mean and standard deviation of the first dwell time sequence and record them as AP and AB in order. Remove the dwell time in the first dwell time sequence that is not located in [AP-k1*AB, AP+k1*AB]. After completion, the second dwell time sequence is obtained, where k1 is the set proportional coefficient.

[0099] Step S202: Set the interval duration to t2, and set multiple consecutive duration intervals based on t2, which are denoted as duration interval 1 to duration interval n1, where n1 is the total number of duration intervals set; calculate the standard deviation of the second retention sequence, denoted as CB;

[0100] Step S203: According to the second retention sequence, count the number of retention times contained in each time interval from time interval 1 to time interval 2, and record the time interval with the most retention times and the second most retention times as the first interval and the second interval, respectively.

[0101] Step S204: Expand the first interval to both sides by a range of CB to obtain the first dense interval; expand the second interval to both sides by a range of CB to obtain the second dense interval; take the union of the first dense interval and the second dense interval, and denote it as the effective dense interval; denote the dwell time in the second dwell sequence that is located in the effective dense interval as the normal dwell time; count the total number of normal dwell times, and denote it as AF.

[0102] Step S205: Divide the effective dense interval evenly into n2 sub-intervals, which are denoted as dense sub-interval 1 to dense sub-interval n2; and count the number of normal dwell times contained in each dense sub-interval, where n2 is the set number.

[0103] Step S206: Denote any normal dwell time as the first dwell time LT, and denote the number of normal dwell times contained in the dense sub-interval where the first dwell time is located as BF; denote BF / AF as the weight of the first dwell time, and denote (BF / AF)*LT as the weighted dwell time of the first dwell time.

[0104] Step S207: Repeatedly calculate the weights of all normal dwell times and the corresponding weighted dwell times, and sum them separately, recording them as AQ and QT respectively in order; calculate QT / AQ, and record it as the representative dwell time of the first time period; repeat the calculation of the representative dwell time of each collection time period, and arrange them in chronological order, recording them as the dwell time series, and label them as the average dwell data.

[0105] Step S208: Arrange the truck turnaround time information of the first time period in ascending order and record it as the first turnaround sequence. Calculate the mean and standard deviation of the first turnaround sequence and record them as DP and DB in order. Remove the dwell time in the first turnaround sequence that is not located in [DP-k2*DB, DP+k2*DB]. After completion, the second turnaround sequence is obtained, where k2 is the set proportional coefficient.

[0106] Step S209: Plot a kernel density curve for the second rotation sequence and obtain the lowest point between the two peaks on the kernel density curve, which is denoted as the valley point. Use the valley point as a threshold to divide the second rotation sequence into two parts, which are denoted as the first rotation rotor sequence and the second rotation rotor sequence, respectively.

[0107] Step S210: For the first week rotor sequence, calculate the median and coefficient of variation of the first week rotor sequence, and denot them as CE and CV respectively in order. Calculate the corrected representative value corresponding to the first week rotor sequence, denoted as XT1, where XT1 = CE * (1 - k3 * CV), and k3 is the set scaling factor. Repeat the calculation of the corrected representative value corresponding to the second week rotor sequence, denoted as XT2.

[0108] Step S211: Obtain the number of turnover times contained in the first and second rotor sequences, respectively, and denote them as WF1 and WF2 in order; calculate [WF1 / (WF1+WF2)]*XT1+[WF2 / (WF1+WF2)]*XT2, and denote it as the representative turnover time of the first time period; repeat the calculation of the representative turnover time of each acquisition time period, arrange them in chronological order, denote them as the turnover time series, and mark them as the average turnover data.

[0109] Step S3 involves analyzing and correcting the average retention data and average turnover data to obtain reference retention and turnover data. Step S3 includes the following sub-steps:

[0110] Step S301: For the retention time series, any data point in the retention time series is denoted as TL(i), where i represents the position number; k4 data points are taken before and after TL(i) and denoted as the adjacent data of TL(i), and the mean and standard deviation of the adjacent data are calculated and denoted as GP(i) and GB(i) in order, where k4 is the number of sets;

[0111] Step S302: If the absolute difference between TL(i) and GP(i) is greater than k5*GB(i), then TL(i) is marked as a suspected outlier; otherwise, it is marked as a normal data point; where k5 is the set scaling factor.

[0112] In step S303, if TL(i) is a suspected outlier, then take TL(i-k6) to TL(i) for linear fitting and obtain the slope of the fitting, denoted as E1; and take TL(i) to TL(i+k6) for linear fitting and obtain the slope of the fitting, denoted as E2.

[0113] Step S304: Denote |E1-E2| as the slope difference CE corresponding to TL(i); Repeat the calculation of the slope difference corresponding to all data in the residence time series; Take the slope difference of the first k6 data of TL(i) and denote it as the set of slope differences of the first neighboring region; Take the slope difference of the last k6 data of TL(i) and denote it as the set of slope differences of the last neighboring region; where k6 is the set number.

[0114] Step S305: Calculate the mean and standard deviation of the set of slope differences in the previous neighborhood, and denote them as HP1 and HB1 in order; calculate the mean and standard deviation of the set of slope differences in the subsequent neighborhood, and denote them as HP2 and HB2 in order.

[0115] Step S306: If the absolute difference between CE and HP1 is greater than k7*HB1 and the absolute difference between CE and HP2 is greater than k7*HB2, then mark TL(i) as an event anomaly point; otherwise, mark TL(i) as a normal data point. Repeat the process of obtaining all event anomaly points in the retention time series.

[0116] Step S307: If TL(i) is an event outlier, then arrange the adjacent data of TL(i) in order of size and obtain the median, which is recorded as the neighborhood median data. Replace TL(i) with the neighborhood median data. Repeat the replacement for all event outliers. After completion, the corrected sequence corresponding to the retention time series is obtained, which is recorded as the reference retention sequence.

[0117] Step S308: Repeatedly obtain the corrected sequence corresponding to the turnover time series and record it as the reference turnover sequence; record the reference retention sequence and the reference turnover sequence as the reference retention turnover data of the first port-adjacent industry; repeatedly obtain the reference retention turnover data of multiple port-adjacent industries.

[0118] Step S4 involves constructing an efficiency prediction model based on reference transit data and predicting the average transit time for goods. Step S4 includes the following sub-steps:

[0119] Step S401: Construct an initial prediction model based on the BiLSTM model. The initial prediction model includes an input layer, a bidirectional LSTM layer, a Dropout layer, a fully connected layer, and an output layer. Set the model input as the reference dwell time sequence and the corresponding reference turnover sequence, and set the model output as the representative dwell time for each future acquisition time period.

[0120] Step S402: The initial prediction model is trained using reference retention and turnover data. After completion, an efficiency prediction model is obtained. Then, the representative retention time of the first port-adjacent industry is predicted for each future collection time period using reference retention and turnover data. The efficiency prediction results corresponding to the collaboration between the first port-adjacent industry and port logistics are obtained.

[0121] Example 3, please refer to Figure 4 As shown, Figure 4 A schematic diagram of an electronic device is provided, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The memory stores computer-readable instructions, and the processor can call these instructions. When the processor executes a computer-readable instruction, it performs steps as described in an efficiency prediction method for the collaboration between port-adjacent industries and port logistics, to achieve the following functions: collecting the dwell time of goods from port-adjacent industries in the port and collecting the turnaround time of trucks transporting these goods in the port, obtaining cargo dwell time data and truck turnaround time data; performing hierarchical calculations on the cargo dwell time data and truck turnaround time data respectively, obtaining average dwell time data and average turnaround data; performing analysis and correction processing based on the average dwell time data and average turnaround data respectively, obtaining reference dwell time and turnaround data; constructing an efficiency prediction model based on the reference dwell time and turnaround data, and predicting the average dwell time of goods.

[0122] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0123] Example 4: This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the steps of an efficiency prediction method for the collaboration between port-adjacent industries and port logistics, to achieve the following functions: collecting the dwell time of goods in the port and the turnaround time of trucks transporting goods in the port, obtaining cargo dwell time data and truck turnaround time data; performing hierarchical calculation processing on the cargo dwell time data and truck turnaround time data respectively, obtaining average dwell time data and average turnaround data; performing analysis and correction processing on the average dwell time data and average turnaround data respectively, obtaining reference dwell time and turnaround data; constructing an efficiency prediction model based on the reference dwell time and turnaround data, and predicting the average dwell time of goods.

[0124] Based on the above description of the embodiments, the embodiments of the present invention can be provided as methods, systems, or computer program products. Based on this understanding, the above technical solutions, in essence or in terms of their contribution to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or certain parts of the embodiments.

[0125] In the embodiments provided in this application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces. The indirect coupling or communication connection between systems, modules, and units may be electrical, mechanical, or other forms.

[0126] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. An efficiency prediction system for the collaboration between port-adjacent industries and port logistics, characterized in that, It includes a data collection module, a hierarchical statistics module, an analysis and correction module, and an efficiency prediction module; The data collection module is used to collect the dwell time of goods in port-adjacent industries and the turnaround time of trucks transporting goods in port, thereby obtaining cargo dwell time data and truck turnaround time data. The hierarchical statistics module includes a first processing unit and a second processing unit. The first processing unit is used to perform hierarchical calculation and processing on cargo dwell time data to obtain average dwell time data; the second processing unit is used to perform hierarchical calculation and processing on truck turnaround time data to obtain average turnaround data. The analysis and correction module performs analysis and correction processing based on the average retention data and the average turnover data to obtain reference retention and turnover data. The efficiency prediction module constructs an efficiency prediction model based on reference retention and turnover data, and predicts the average retention time of goods. The analysis and correction module is configured with analysis and correction strategies, which include: For a time series, any data point in the time series is denoted as TL(i), where i represents the position number; k4 data points are taken before and after TL(i) and denoted as the adjacent data of TL(i), and the mean and standard deviation of the adjacent data are calculated and denoted as GP(i) and GB(i) in order, where k4 is the number of sets; If the absolute difference between TL(i) and GP(i) is greater than k5*GB(i), then TL(i) is marked as a suspected outlier; otherwise, it is marked as a normal data point; where k5 is the set scaling factor. If TL(i) is a suspected outlier, then perform linear fitting from TL(i-k6) to TL(i) to obtain the slope of the fitting, denoted as E1; and perform linear fitting from TL(i) to TL(i+k6) to obtain the slope of the fitting, denoted as E2. Let |E1-E2| be denoted as the slope difference CE corresponding to TL(i); repeatedly calculate the slope difference corresponding to all data in the residence time series; and take the slope difference of the first k6 data of TL(i) as the set of slope differences of the first neighboring region; and take the slope difference of the last k6 data of TL(i) as the set of slope differences of the last neighboring region; where k6 is the set number; Calculate the mean and standard deviation of the set of slope differences in the previous neighborhood, and denote them as HP1 and HB1 in order; calculate the mean and standard deviation of the set of slope differences in the subsequent neighborhood, and denote them as HP2 and HB2 in order. If the absolute difference between CE and HP1 is greater than k7*HB1 and the absolute difference between CE and HP2 is greater than k7*HB2, then mark TL(i) as an event anomaly point; otherwise, mark TL(i) as a normal data point, and repeat the acquisition of all event anomaly points in the retention time series.

2. The efficiency prediction system for port-related industries and port logistics collaboration according to claim 1, characterized in that, The data collection module is configured with data collection strategies, which include: Let any port to be predicted be referred to as the first port, any port-adjacent industry corresponding to the first port be referred to as the first port-adjacent industry, and the goods that the first port-adjacent industry needs to transport through the first port be referred to as the first goods; let the container carrying the first goods be referred to as the first container; and let the truck transporting the first container be referred to as the first truck. The time taken from the arrival of any first container at the first port to its departure from the first port is recorded as the corresponding first container's dwell time at the port. The time taken for any first truck to enter the first port and leave the first port is recorded as the turnaround time of the corresponding first truck at the port.

3. The efficiency prediction system for port-related industries and port logistics collaboration according to claim 2, characterized in that, Data collection strategies also include: Set the data collection time period to T1. Any data collection time period is recorded as the first time period. Within the first time period, the dwell time of each container leaving the first port is collected and recorded as the cargo dwell time information of the first time period. Within the first time period, the turnaround time of each truck at the port is collected and recorded as the truck turnaround time information for the first time period. The information on cargo dwell time and truck turnaround time is collected periodically and repeatedly for each collection period, and then arranged in chronological order and recorded as cargo dwell time data and truck turnaround time data, respectively.

4. The efficiency prediction system for the collaboration between port-adjacent industries and port logistics as described in claim 3, characterized in that, The first processing unit is configured with a first processing strategy, which includes: Arrange the cargo dwell time information of the first time period in ascending order and denote it as the first dwell time sequence. Calculate the mean and standard deviation of the first dwell time sequence and denote them as AP and AB in order. Remove the dwell time in the first dwell time sequence that is not located in [AP-k1*AB, AP+k1*AB] and obtain the second dwell time sequence. Here, k1 is the set proportional coefficient. Set the interval duration to t2, and set multiple consecutive duration intervals based on t2, which are denoted as duration interval 1 to duration interval n1, where n1 is the total number of duration intervals set; calculate the standard deviation of the second retention sequence, denoted as CB; Based on the second retention sequence, count the number of retention times contained in each time interval from time interval 1 to time interval 2, and denote the time interval with the most and second most retention times as the first interval and the second interval, respectively.

5. The efficiency prediction system for the collaboration between port-adjacent industries and port logistics as described in claim 4, characterized in that, The first processing strategy also includes: Expand the first interval to both sides by a range of CB to obtain the first dense interval, and expand the second interval to both sides by a range of CB to obtain the second dense interval; take the union of the first dense interval and the second dense interval, and denote it as the effective dense interval; denote the dwell time in the second dwell sequence that is located in the effective dense interval as the normal dwell time, and count the total number of normal dwell times, and denote it as AF; The effective dense interval is evenly divided into n2 sub-intervals, which are denoted as dense sub-interval 1 to dense sub-interval n2 respectively; and the number of normal dwell times contained in each dense sub-interval is counted, where n2 is the set number. Let any normal dwell time be denoted as the first dwell time LT, and let the number of normal dwell times contained in the dense subinterval containing the first dwell time be denoted as BF; let BF / AF be denoted as the weight of the first dwell time, and let (BF / AF)*LT be denoted as the weighted dwell time of the first dwell time. Repeatedly calculate the weights of all normal dwell times and the corresponding weighted dwell times, and sum them separately, recording them as AQ and QT respectively in order; calculate QT / AQ, and record it as the representative dwell time of the first time period; repeatedly calculate the representative dwell time of each collection time period, and arrange them in chronological order, recording them as the dwell time series, and label them as the average dwell data.

6. The efficiency prediction system for the collaboration between port-adjacent industries and port logistics as described in claim 5, characterized in that, The second processing unit is configured with a second processing strategy, which includes: Arrange the truck turnaround time information of the first time period in ascending order and denote it as the first turnaround sequence. Calculate the mean and standard deviation of the first turnaround sequence and denote them as DP and DB in order. Remove the dwell time in the first turnaround sequence that is not located in [DP-k2*DB, DP+k2*DB] and obtain the second turnaround sequence. Here, k2 is the set proportional coefficient. Plot a kernel density curve for the second rotation sequence and obtain the lowest point between the two peaks on the kernel density curve, which is denoted as the valley point. Use the valley point as a threshold to divide the second rotation sequence into two parts, which are denoted as the first rotation sequence and the second rotation sequence, respectively. For the first week's rotor sequence, calculate the median and coefficient of variation of the first week's rotor sequence, and denote them as CE and CV respectively in order. Calculate the corrected representative value corresponding to the first week's rotor sequence, denoted as XT1, where XT1 = CE * (1 - k3 * CV), and k3 is the set scaling factor. Repeat the calculation of the corrected representative value corresponding to the second week's rotor sequence, denoted as XT2. Obtain the number of turnover times contained in the first and second rotor sequences, respectively, and denote them as WF1 and WF2 in order; calculate [WF1 / (WF1+WF2)]*XT1+[WF2 / (WF1+WF2)]*XT2, and denote it as the representative turnover time of the first time period; repeat the calculation of the representative turnover time for each acquisition time period, arrange them in chronological order, denote them as the turnover time series, and mark them as the average turnover data.

7. The efficiency prediction system for the collaboration between port-adjacent industries and port logistics as described in claim 6, characterized in that, Analysis and correction strategies also include: If TL(i) is an event outlier, then the adjacent data of TL(i) are arranged in order of size, and the median is obtained and recorded as the neighborhood median. TL(i) is replaced with the neighborhood median. The replacement is repeated for all event outliers. After completion, the corrected sequence corresponding to the retention time series is obtained and recorded as the reference retention sequence. Repeatedly acquire the corrected sequence corresponding to the turnover time series, and record it as the reference turnover sequence; record the reference retention sequence and the reference turnover sequence as the reference retention turnover data of the first port-adjacent industry; repeatedly acquire the reference retention turnover data of multiple port-adjacent industries.

8. The efficiency prediction system for the collaboration between port-adjacent industries and port logistics according to claim 7, characterized in that, The efficiency prediction module is configured with efficiency prediction strategies, which include: An initial prediction model is constructed based on the BiLSTM model. The initial prediction model includes an input layer, a bidirectional LSTM layer, a Dropout layer, a fully connected layer, and an output layer. The model input is set as the reference dwell time sequence and the corresponding reference turnover sequence, and the model output is the representative dwell time for each future collection time period. The initial prediction model was trained using reference retention and turnover data to obtain an efficiency prediction model. Then, using reference retention and turnover data of the first port-adjacent industry, the representative retention time of the first port-adjacent industry in each future collection time period was predicted to obtain the efficiency prediction results corresponding to the collaboration between the first port-adjacent industry and port logistics.

Images