A rule and data double-driven port operation planning method and system

Abstract

The application relates to the technical field of port planning, and discloses a port operation planning method and system based on rule and data double driving, which comprises the following steps: collecting multi-source prediction data of a port; judging the stationarity of the multi-source prediction data to obtain a stationarity judgment result; predicting based on the multi-source prediction data by using a prediction method corresponding to the stationarity judgment result to obtain a target prediction result of the port; performing reliability verification on the prediction method corresponding to the target prediction result; and when the reliability verification is passed, constructing a target function based on the target prediction result under the constraint of port business rules and solving the target function to obtain an operation scheme of the port. The application effectively improves the throughput prediction accuracy by taking into account the flexibility of data driving and the constraint rigidity of rule driving, thereby improving the operation planning efficiency, and providing strong support for efficient operation of the port.

Description

Technical Field

[0001] This invention relates to the field of port planning technology, specifically to a port operation planning method and system based on rules and data. Background Technology

[0002] As a key node in the logistics network, the operation planning of ports directly affects ship turnaround, customs clearance, and the stability of regional supply chains. Therefore, scientific and efficient operation planning for ports is of great significance.

[0003] Currently, port data is typically collected in real time via the Internet of Things (IoT), and operational planning is based on machine learning models. However, port throughput is significantly affected by unforeseen factors and is a strongly non-stationary sequence. When the non-stationarity of the data exceeds a preset threshold, existing models often accumulate errors due to repeated differencing, leading to inaccurate throughput predictions and consequently low operational planning efficiency. Furthermore, existing technologies use a single dimension in data collection, further amplifying the discrepancy between predictions and actual conditions, resulting in a disconnect between port operational planning and actual needs. Summary of the Invention

[0004] This invention provides a port operation planning method and system based on rules and data to solve the problems of inaccurate throughput prediction in existing technologies, which leads to low operation planning efficiency and a disconnect between port operation planning and actual needs.

[0005] In a first aspect, the present invention provides a port operation planning method based on both rules and data, the method comprising: Collect multi-source prediction data for ports; The stationarity of the multi-source prediction data is assessed, and the stationarity assessment results are obtained. The prediction method corresponding to the stationarity judgment result is adopted, and prediction is made based on multi-source prediction data to obtain the target prediction result of the port. Verify the reliability of the prediction method corresponding to the target prediction result; During the reliability verification, under the constraints of port business rules, an objective function is constructed and solved based on the target prediction results to obtain the port's operation plan.

[0006] This invention compensates for the lack of information in single-source data by collecting multi-source forecast data from ports, providing a comprehensive and continuous data source for operational planning. The multi-source forecast data is assessed for stationarity, and the corresponding forecasting method is used to predict the target throughput, outputting the target forecast result. This avoids the problems of insufficient adaptability of single forecasting methods to data with different stationarities and the susceptibility to forecasting errors, effectively improving the accuracy and precision of throughput forecasting. The forecasting method corresponding to the target forecast result undergoes reliability verification, further improving the accuracy and precision of throughput forecasting and ensuring high reliability of the forecast data used for subsequent operational planning. During reliability verification, under the constraints of port business rules, an objective function is constructed and solved based on the target forecast result to obtain the port's operational plan. By balancing the flexibility of data-driven approaches with the rigidity of rule-driven constraints, the accuracy of throughput forecasting is effectively improved, thereby increasing operational planning efficiency and providing strong support for efficient port operations.

[0007] In one alternative implementation, the multi-source prediction data includes historical throughput data; The stationarity of multi-source prediction data is assessed, and the stationarity assessment results are obtained, including: Based on historical throughput data from multi-source prediction data, a historical throughput sequence is constructed. A regression equation is constructed based on the historical throughput sequence; Regression estimation is performed on the regression equation to obtain the regression coefficients and the standard errors of the regression coefficients; The ratio of the regression coefficient to the standard error is calculated as an indicator of stationarity. When the stationarity index is less than the preset critical value, the stationarity judgment result is determined to be stationary; When the stationarity index is not less than the preset critical value, the stationarity judgment result is determined to be non-stationary.

[0008] This invention constructs a historical throughput sequence and a regression equation based on the sequence. By calculating a stationarity index to test stationarity, it achieves a statistical significance test for the stationarity of the sequence. This allows for the adoption of appropriate prediction methods based on the stationarity judgment results, thereby improving prediction accuracy.

[0009] In one optional implementation, a prediction method corresponding to the stationarity assessment result is used to perform prediction based on multi-source prediction data to obtain the target prediction result for the port, including: When the stationarity assessment result is stationary, a pre-trained long short-term memory network is used to make predictions based on multi-source prediction data to obtain the target prediction result.

[0010] This invention improves prediction efficiency by directly using a pre-trained long short-term memory network to make predictions based on multi-source prediction data when the stationarity judgment result is stationary.

[0011] In one optional implementation, a prediction method corresponding to the stationarity judgment result is used to make a prediction based on multi-source prediction data to obtain the target prediction result of the port, which further includes: When the stationarity judgment result is non-stationary, an attention mechanism is used to assign attention weights to each time step in the multi-source prediction data. The data at each time step is weighted by its corresponding attention weight to obtain the weighted multi-source prediction data. A pre-trained long short-term memory network is used to make predictions based on weighted multi-source prediction data to obtain initial prediction results; Residual modeling is performed based on historical port forecasts to obtain correction amounts; A time-varying weighted fusion algorithm is used to fuse the initial prediction results and the correction values ​​to obtain the target prediction results.

[0012] This invention introduces an attention mechanism to dynamically weight sudden factors when the data is non-stationary, and uses historical port prediction data to model residuals to obtain correction amounts, thereby correcting the prediction results of the LSTM network. This effectively avoids the accumulation of prediction errors in non-stationary scenarios and ensures accurate throughput prediction even in complex and ever-changing environments.

[0013] In one alternative implementation, residual modeling is performed based on historical port forecasts to obtain correction amounts, including: Obtain the port's first historical actual throughput and the corresponding historical throughput forecast within a preset historical time window; Calculate the residual sequence between the first historical actual throughput and the historical throughput prediction; Construct an autoregressive integral moving average model of the residual sequence; The correction amount is obtained by making predictions based on the autoregressive integral moving average model.

[0014] This invention obtains the first historical actual throughput and the corresponding historical throughput prediction value within a preset historical time window, constructs an autoregressive integral moving average model based on the residual sequence between the two, and uses this model to make predictions to obtain the corresponding correction amount. This can provide targeted compensation and correction for the initial prediction results, effectively reduce prediction bias, and improve the accuracy of the prediction results.

[0015] In one optional implementation, the reliability of the prediction method corresponding to the target prediction result is verified, including: Obtain the port's second historical actual throughput within a preset verification period; The prediction method corresponding to the target prediction result is adopted, and prediction is performed based on multi-source prediction data corresponding to the preset verification period to obtain historical prediction results; Error index is calculated based on the second historical actual throughput and historical prediction results; If the error index is not greater than the preset business tolerance threshold, the reliability verification is deemed to have passed. If the error index exceeds the preset business tolerance threshold, the reliability verification is deemed to have failed.

[0016] This invention verifies the reliability of the prediction method by predicting and calculating an error index based on the second historical actual throughput, and comparing it with a preset service tolerance threshold. This objectively quantifies and evaluates the prediction accuracy, which helps improve the reliability of the prediction results.

[0017] In one alternative implementation, the method further includes: If the reliability verification fails, the parameters of the prediction method corresponding to the target prediction result are adaptively adjusted, and the prediction method corresponding to the stationarity judgment result is used to make predictions based on multi-source prediction data to obtain the target prediction result of the port. This process continues until the reliability verification passes.

[0018] This invention ensures the reliability of prediction results by adaptively adjusting the parameters of the prediction method when the reliability verification fails, and then re-predicting and verifying until the reliability verification passes.

[0019] In one optional implementation, upon passing the reliability verification, under the constraints of port business rules, an objective function is constructed and solved based on the target prediction results to obtain the port's operation plan, including: When passing the reliability verification, a set of constraints is constructed based on the port business rules; Construct a multi-objective function, which represents minimizing the total ship demurrage time and maximizing the throughput plan completion rate; Under the constraints of the set of constraints, the multi-objective function is solved based on the target prediction results to obtain the actual amount of work completed by each ship in the port at each time step. Based on the actual amount of work completed by each vessel in the port at each time step, multiple decision variables are determined to form an operation plan.

[0020] This invention ensures that subsequent optimization schemes conform to the actual operational boundaries of the port by constructing a set of constraints based on port business rules after the prediction results have passed reliability verification. A multi-objective function is constructed that balances minimizing total vessel waiting time with maximizing throughput plan completion rate. Under these constraints, the multi-objective function is solved based on the objective prediction results to obtain the optimal actual workload completed by each vessel at each time step. Finally, based on this workload, multi-dimensional decision variables are determined to form a complete operational plan. This transforms the abstract optimization results into directly executable scheduling instructions, effectively improving the rationality of port operation organization, reducing vessel waiting costs, ensuring the stable achievement of throughput plans, and achieving efficient integration of data prediction and intelligent scheduling.

[0021] Secondly, the present invention provides a port operation planning system based on both rules and data, the system comprising: The data acquisition and preprocessing module is used to collect multi-source forecast data for the port. The prediction module is used to determine the stationarity of multi-source prediction data, obtain the stationarity determination result, use the prediction method corresponding to the stationarity determination result, make predictions based on the multi-source prediction data, obtain the target prediction result of the port, and perform reliability verification on the prediction method corresponding to the target prediction result. The optimization decision module is used to construct and solve the objective function based on the target prediction results when the reliability verification is passed, under the constraints of the port business rules, to obtain the port's operation plan.

[0022] Thirdly, the present invention provides an electronic device, comprising: a memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the computer instructions to perform the rule-based and data-driven port operation planning method described in the first aspect or any corresponding embodiment thereof.

[0023] Fourthly, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the rule-based and data-driven port operation planning method described in the first aspect or any corresponding embodiment thereof. Attached Figure Description

[0024] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0025] Figure 1This is a schematic diagram of a port operation planning system based on rules and data dual-drive according to an embodiment of the present invention; Figure 2 This is a flowchart of a port operation planning method based on rules and data dual-drive according to an embodiment of the present invention; Figure 3 This is a flowchart illustrating the determination of target prediction results according to an embodiment of the present invention; Figure 4 This is a schematic diagram of the hardware structure of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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, 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.

[0027] It is understood that before using the technical solutions disclosed in the various embodiments of the present invention, users should be informed of the types, scope of use, and usage scenarios of the personal information involved in the present invention and their authorization should be obtained in accordance with relevant laws and regulations through appropriate means.

[0028] The terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this invention, "a plurality of" means two or more, unless otherwise explicitly specified.

[0029] Figure 1 This is a schematic diagram of a port operation planning system based on rule-based and data-driven approaches according to an embodiment of the present invention, such as... Figure 1 As shown, the system includes: a data acquisition and preprocessing module for acquiring multi-source forecast data of the port; a forecasting module for judging the stationarity of the multi-source forecast data, obtaining the stationarity judgment result, using the forecasting method corresponding to the stationarity judgment result, making a forecast based on the multi-source forecast data to obtain the target forecast result of the port, and performing reliability verification on the forecasting method corresponding to the target forecast result; and an optimization decision module for constructing and solving an objective function based on the target forecast result under the constraints of the port's business rules when the reliability verification is passed, to obtain the port's operation plan.

[0030] Specifically, the data acquisition and preprocessing module collects and preprocesses multi-source forecast data from the port in real time, providing high-quality data for subsequent stages. The forecasting module assesses the stationarity of the multi-source forecast data, employs corresponding forecasting methods based on the stationarity results, and outputs the target forecast result. Furthermore, it performs reliability verification on the target forecast result to improve the accuracy of throughput forecasting. Upon passing the reliability verification, the optimization decision-making module, constrained by port business rules, constructs and solves an objective function based on the target forecast result to obtain the port's operational plan. Through the collaborative operation of these modules, accurate throughput forecasting is ensured even in complex and ever-changing environments, resulting in precise and efficient operational plans. This significantly improves port operational efficiency, effectively alleviates port congestion, reduces operating costs, mitigates congestion risks, and increases the monthly throughput plan completion rate.

[0031] According to an embodiment of the present invention, a port operation planning method based on rules and data dual-drive is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0032] This embodiment provides a port operation planning method based on both rules and data, which can be used in the aforementioned port operation planning system based on both rules and data. Figure 2 This is a flowchart of a port operation planning method based on rule-based and data-driven approaches according to an embodiment of the present invention, such as... Figure 2 As shown, the process includes the following steps: Step S201: Collect multi-source prediction data for the port.

[0033] Specifically, for the port to be planned, a one-hour time step is used, meaning one time step corresponds to a continuous short period of time. Multi-source prediction data for 30 consecutive time steps are collected. During data collection, the current time step is used as the baseline, i.e., as the prediction time step. Data from 29 consecutive time steps are collected back to historical data, and combined with the real-time data of the current time step, resulting in 30 time steps of multi-source prediction data. The data collection dimensions cover historical throughput data, ship and cargo demand data, yard and equipment status data, meteorological and environmental data, and time and calendar data. The specific data collected at each time step is as follows: (1) Historical throughput data is used to capture the long-term trend and short-term fluctuations of port operations, including: the actual throughput of the previous time step, the sliding average throughput of the past 24 time steps, the sliding average throughput of the past 7 days, the throughput of the same time step last week (used to capture the weekly periodicity of operations), the import unpacking volume of the previous time step, and the export loading volume of the previous time step. (2) Ship and cargo demand data are used to quantify current and future operational pressure, including: the number of ships berthed in port at this time step, the number of ships waiting at anchorage, the number of ships expected to arrive in the next 24 time steps (based on the advance notice list issued by the shipping company, inputting real-time location features, historical channel traffic efficiency, current sea condition influencing factors and anchorage congestion level as feature variables, and using a pre-trained multilayer perceptron algorithm to calculate the arrival probability of each ship on the way, automatically removing interference items with arrival probabilities lower than the preset threshold, and adding the ships already queuing at the anchorage outside the port to summarize the accurate arrival number in the next 24 time steps), the average tonnage of ships in port (distinguishing between branch line ships and mother ships), the total inventory of imported heavy containers to be unloaded, the total inventory of exported heavy containers to be loaded, the proportion of dangerous goods / special containers (affecting operational efficiency), and the occupancy rate of refrigerated container sockets; (3) The status data of the yard and equipment is used to reflect the upper limit of the port's real-time operation capacity, including: the utilization rate (%) of the import full container yard, the utilization rate (%) of the export full container yard, the utilization rate (%) of the empty container yard, the number of available quay cranes, the number of available yard cranes, and the average turnaround time of the gate trucks (reflecting the smoothness of the port's collection and distribution). (4) Meteorological environmental data is the main source for judging stability and directly affects the start and stop of operations, including: instantaneous wind speed at this time step (m / s, the quay crane needs to stop operating when it exceeds the preset threshold), visibility (km, affecting navigation and operations in foggy weather), rainfall (mm / h, heavy rain reduces visibility and road surface friction, affecting operational safety and efficiency), and temperature (°C, extreme high / low temperatures affect the performance of port machinery and equipment). (5) Time and calendar data are used to help identify the social cyclical patterns of workload, including: time index (0-23, using one-hot encoding or sine transformation processing), weekday index (1-7, distinguishing between weekdays and weekends), holiday identifier (0 / 1 binary variable, such as a sharp decrease in workload during the Spring Festival) and work shift identifier (day shift / night shift).

[0034] The data collected at 30 time steps are integrated into multi-source prediction data. By collecting data from multiple dimensions, the missing information from single data points is compensated for, providing a comprehensive and continuous data source for operation planning.

[0035] Optionally, after collecting the above data, data standardization and integrity processing can be performed. During standardization, the maximum-minimum standardization method is used to transform the collected data, and the transformed data is checked to see if it falls within the [0,1] range. If the deviation exceeds 5%, reprocessing is triggered. During integrity processing, it is ensured that the data at 30 time steps is continuous and unbroken, and that all dimensions of data are complete. If data loss is detected, imputation is performed promptly. For example, linear interpolation can be used to accurately imput short-term missing data, ensuring the integrity and validity of the data.

[0036] Step S202: Perform a stationarity assessment on the multi-source prediction data to obtain the stationarity assessment result.

[0037] Specifically, because port throughput is easily affected by various unforeseen factors such as extreme weather, the corresponding multi-source forecast data tends to exhibit strong non-stationary characteristics. If a unified model is directly used for processing, it is easy to cause error accumulation. Therefore, it is necessary to judge the stationarity of the multi-source forecast data in order to determine whether the data is stationary.

[0038] Step S203: Using the prediction method corresponding to the stationarity judgment result, prediction is made based on multi-source prediction data to obtain the target prediction result of the port.

[0039] Specifically, to avoid the problem that a single forecasting method is not adaptable enough to data with different stationarities and is prone to forecasting errors, a corresponding forecasting method is adopted based on the stationarity judgment results of multi-source forecasting data to make forecasts, and finally outputs the target forecast result of port throughput, which effectively improves the accuracy and precision of throughput forecasting.

[0040] Step S204: Verify the reliability of the prediction method corresponding to the target prediction result.

[0041] Specifically, the recent performance of the prediction method is verified. If the recent performance is good, it means that the prediction method is currently reliable and the prediction results obtained by using the prediction method can be used to ensure the accuracy and precision of throughput prediction, and ensure that the prediction data used for subsequent operation planning has high reliability.

[0042] Step S205: Upon passing the reliability verification, under the constraints of the port business rules, an objective function is constructed and solved based on the target prediction results to obtain the port's operation plan.

[0043] Specifically, port business rules are the rigid constraints and operational guidelines for port operations, covering rules such as yard capacity limits, ship loading and unloading priorities, and equipment operating capabilities. When the target prediction result passes reliability verification, meaning the port throughput prediction result has high accuracy and reliability, a multi-objective function is constructed and solved under the rigid constraints of the port business rules. This accurately yields the optimal port operation plan that meets throughput requirements, adapts to the actual port operation scenario, satisfies rule constraints, and takes into account operational efficiency. This ensures the scientific nature and feasibility of the operation plan and avoids a disconnect between the planned scheme and actual operations.

[0044] This invention compensates for the lack of information in single-source data by collecting multi-source forecast data from ports, providing a comprehensive and continuous data source for operational planning. The multi-source forecast data is assessed for stationarity, and the corresponding forecasting method is used to predict the target throughput, outputting the target forecast result. This avoids the problems of insufficient adaptability of single forecasting methods to data with different stationarities and the susceptibility to forecasting errors, effectively improving the accuracy and precision of throughput forecasting. The forecasting method corresponding to the target forecast result undergoes reliability verification, further improving the accuracy and precision of throughput forecasting and ensuring high reliability of the forecast data used for subsequent operational planning. During reliability verification, under the constraints of port business rules, an objective function is constructed and solved based on the target forecast result to obtain the port's operational plan. By balancing the flexibility of data-driven approaches with the rigidity of rule-driven constraints, the accuracy of throughput forecasting is effectively improved, thereby increasing operational planning efficiency and providing strong support for efficient port operations.

[0045] This embodiment provides a port operation planning method based on both rules and data, which can be used in the aforementioned port operation planning system based on both rules and data. The method specifically includes the following steps: Step S301: Collect multi-source prediction data for the port. For details, please refer to [link / reference]. Figure 1 Step S201 of the illustrated embodiment will not be described again here.

[0046] Step S302: Perform a stationarity assessment on the multi-source prediction data to obtain the stationarity assessment result. The multi-source prediction data includes historical throughput data.

[0047] Specifically, step S302 includes: Step S3021: Construct a historical throughput sequence based on historical throughput data from the multi-source prediction data.

[0048] Specifically, the actual throughput of the previous time step is selected from the historical throughput data and arranged in order from first to last time step to form a historical throughput sequence.

[0049] Step S3022: Construct a regression equation based on the historical throughput sequence.

[0050] Specifically, an enhanced regression equation (1) is constructed based on the historical throughput sequence to eliminate the effects of autocorrelation: (1) In the formula, Represents the number of times in the historical throughput sequence The actual throughput of the previous time step in the previous time step; Indicates the first The first-order difference corresponding to the actual throughput of the previous time step is obtained by... calculate; This represents the constant term, i.e., the intercept of the regression equation; Indicates the trend coefficient; This represents the time trend term, used to capture the linear changing trend of historical throughput sequences; Represents the regression coefficient, indicating the degree of influence of the previous throughput on the current throughput; Indicates the lag order, which is automatically selected according to the AIC criterion (Akaike Information Criterion); Indicates the lag difference coefficient; This represents the lagged difference term, used to eliminate higher-order autocorrelation in the residuals; This represents the preset white noise residual coefficient, used to represent random fluctuations that cannot be explained by the regression equation.

[0051] Step S3023: Perform regression estimation on the regression equation to obtain the regression coefficients and the standard errors of the regression coefficients.

[0052] Specifically, the constant term, trend coefficient, regression coefficient, and lag difference coefficient in the above regression equation are all unknowns. By performing least squares regression estimation on the regression equation, the estimated values ​​and standard deviations of each coefficient can be obtained. Since the core of stationarity judgment is to determine whether there is a unit root in the historical throughput series, and a regression coefficient of 0 indicates the existence of a unit root, while a non-zero value indicates the absence of a unit root, the estimated values ​​and standard errors of the regression coefficients can be extracted from them.

[0053] Step S3024: Calculate the ratio of the regression coefficient to the standard error, which serves as an indicator of stationarity.

[0054] Specifically, since regression coefficients are estimated values, it is impossible to determine whether their zero or non-zero values ​​are due to actual existence or error. Therefore, directly using regression coefficients to determine stationarity is inaccurate. Thus, this embodiment of the invention uses the ratio of the regression coefficient to its standard error as a stationarity indicator to quantify the degree of deviation of the regression coefficient from zero, thereby achieving a statistical significance test for the existence of a unit root in the sequence and improving the accuracy of stationarity determination.

[0055] Step S3025: When the stationarity index is less than the preset critical value, the stationarity judgment result is determined to be stationary.

[0056] Specifically, the preset critical value is -2.86, the critical value at the 5% confidence level in the ADF (Augmented Dickey-Fuller) test. If the stationarity index is less than -2.86, it can be statistically determined that there is no unit root in the historical throughput series, and that it has mean regression characteristics, thus the multi-source prediction data is considered to be stationary data.

[0057] Step S3026: When the stationarity index is not less than the preset critical value, the stationarity judgment result is determined to be non-stationary.

[0058] Specifically, if the stationarity index is not less than -2.86, it can be statistically determined that the historical throughput sequence has a unit root and has random walk characteristics. Therefore, the multi-source prediction data is considered to be non-stationary data and is severely affected by sudden factors.

[0059] By constructing a historical throughput sequence and then using the regression equation to perform an ADF stationarity test, we can accurately determine whether the historical throughput sequence has a unit root, thus achieving a statistical significance test for the stationarity of the sequence and ensuring the reliability of the stationarity judgment.

[0060] Step S303: Using the prediction method corresponding to the stationarity judgment result, prediction is made based on multi-source prediction data to obtain the target prediction result of the port.

[0061] Specifically, when the stationarity assessment result is stationary, the above step S303 includes: Step S3031: When the stationarity judgment result is stationary, a pre-trained long short-term memory network is used to make a prediction based on multi-source prediction data to obtain the target prediction result.

[0062] Specifically, when the stationarity judgment result is stationary, it can adapt to the learning requirements of Long Short-Term Memory (LSTM) networks for time series patterns. Therefore, a pre-trained LSTM network is directly used to make predictions with multi-source prediction data as input, and the target prediction result for prediction time step a is output, including the throughput prediction value of each time step in a+72 time steps and the ship operation requirements (such as operation load requirements, berth occupancy time requirements, mechanical configuration requirements, etc.).

[0063] Optionally, the LSTM network includes an input layer, three stacked LSTM layers, a Dropout regularization layer, a fully connected layer, and an output layer. Historical data covering all four seasons of the previous year is collected from the port via IoT devices, according to the data dimensions of the multi-source prediction data, to improve the model's generalization ability. This data is cleaned and divided into training and validation sets in a 7:3 ratio. The LSTM network is pre-trained using the training set to learn the inherent patterns, periodic characteristics, fluctuation trends, and long-term dependencies between different influencing factors influencing port throughput over time. This includes seasonal variations, differences between weekdays and holidays, ship operation patterns, berth occupancy characteristics, stable correlation patterns between machinery configuration requirements and throughput, and the model's prediction accuracy is verified using the validation set. Pre-training is completed when the model's prediction accuracy reaches a preset value, resulting in a pre-trained LSTM network with stable prediction capabilities and strong generalization ability.

[0064] When the stationarity assessment result is non-stationary, step S303 above includes: Step S3032: When the stationarity judgment result is non-stationary, an attention mechanism is used to assign attention weights to each time step in the multi-source prediction data.

[0065] Specifically, when the stationarity assessment result is non-stationary, it indicates that the network is susceptible to sudden disturbances such as extreme weather, sudden changes in throughput, and fluctuations in ship traffic, making it difficult for the LSTM network to accurately capture these patterns. Therefore, an attention mechanism is introduced to assign corresponding attention weights to each time step of the multi-source prediction data. The multi-source prediction data comprises data from 30 consecutive time steps. Based on whether each time step contains sudden features such as extreme weather, sudden changes in ship traffic, or drastic fluctuations in operational load, the attention mechanism automatically identifies key time steps affected by these sudden factors, assigning higher attention weights to these key time steps and lower attention weights to other stationary time steps. This strengthens the impact of sudden disturbance time steps on the prediction results and improves prediction accuracy.

[0066] Step S3033: Weight the data corresponding to each time step with its corresponding attention weight to obtain the weighted multi-source prediction data.

[0067] Specifically, a time-step weighting method is adopted, in which the attention weight corresponding to each time step is weighted and calculated with all the data in that time step to obtain attention-weighted multi-source prediction data.

[0068] Step S3034: A pre-trained long short-term memory network is used to make predictions based on weighted multi-source prediction data to obtain initial prediction results.

[0069] Specifically, since the weighted multi-source prediction data has enhanced the characteristics of sudden disturbance time steps, it can adapt to the learning requirements of LSTM networks for non-stationary data. Therefore, the pre-trained LSTM network is directly used to predict the weighted multi-source prediction data as input and output the initial prediction results, including the throughput prediction values ​​and ship operation requirements of each time step in a+72 time steps.

[0070] Step S3035: Based on the historical forecast of the port, perform residual modeling to obtain the correction amount.

[0071] Specifically, step S3035 includes: Step a1: Obtain the port's first historical actual throughput and the corresponding historical throughput prediction value within a preset historical time window.

[0072] Specifically, for each time step within a preset historical time window, the actual throughput of that time step and the predicted throughput value of the earliest time step in the prediction results output when the previous time step is the prediction time are obtained, and these are respectively used as the first historical actual throughput and the historical throughput prediction value.

[0073] Assume the preset historical time window is from 2:00 AM to 5:00 AM on January 1st, including time steps: 2:00-3:00 AM, 3:00-4:00 AM, and 4:00-5:00 AM. For the 3:00-4:00 AM time step, since it is a historical period, the actual throughput is known, and it is taken as the first historical actual throughput. Because the prediction results include the throughput prediction values ​​of each of the a+72 time steps, but not the throughput prediction value of the predicted time step itself, that is, the prediction results for 3:00-4:00 AM do not include the throughput prediction value of that time step itself. Therefore, the first throughput prediction value in the prediction results for 2:00-3:00 AM is taken as the historical throughput prediction value. It should be noted that the prediction results for multiple time steps before 3:00-4:00 AM all include the throughput prediction value for 3:00-4:00 AM. This embodiment of the invention only uses the prediction result of the previous time step as an example for explanation and is not intended to be limiting.

[0074] Step a2: Calculate the residual sequence between the first historical actual throughput and the historical throughput prediction value.

[0075] Specifically, for each time step within a preset historical time window, the difference between the first historical actual throughput and the predicted historical throughput for that time step is calculated as the residual for that time step. The residuals of each time step within the preset historical time window are arranged sequentially in chronological order to form a residual sequence.

[0076] Step a3: Construct an autoregressive integral moving average model of the residual sequence.

[0077] Specifically, the Autoregressive Integrated Moving Average (ARIMA) model of equation (2) is constructed as follows: (2) In the formula, Indicates the lag operator; This represents the order of the highest-order term in an autoregressive coefficient polynomial. Automatically selected based on AIC criteria; Represents the difference operator; The order of difference is represented by the number of differences. The residual sequence is processed by first-order difference, second-order difference, etc. After each difference, the ADF test is performed on the differenced sequence until the test result is stationary. The number of differences in the residual sequence is then taken as the order of difference. This indicates the time step index within the preset historical time window; Represents the nth element in the residual sequence The difference between each time step; This represents the order of the highest-order term in a moving average coefficient polynomial. Automatically selected using the BIC (Bayesian Information Criterion); This indicates the preset white noise.

[0078] The autoregressive coefficient polynomial in equation (2) above contains The autoregressive coefficients and the moving average coefficient polynomial contained therein Each of the moving average coefficients is an unknown quantity. The ARIMA model is fitted using the residual sequence, allowing the model to learn the linear variation of the residuals, thereby obtaining the optimal values ​​for each unknown coefficient, and finally obtaining a fully fitted ARIMA model.

[0079] Step a4: Make a prediction based on the autoregressive integral moving average model to obtain the correction amount.

[0080] Specifically, using the fitted ARIMA model, time-series extrapolation predictions are performed on a+72 time steps. The resulting residual predictions for the next a+72 time steps are defined as correction values, which quantify the degree of linear deviation between the initial prediction results identified by the ARIMA model and the actual values.

[0081] Step S3036: Use a time-varying weighted fusion algorithm to fuse the initial prediction result and the correction amount to obtain the target prediction result.

[0082] Specifically, given that the ARIMA model has high accuracy in the short term but is prone to divergence in the long term, while the LSTM network excels at capturing long-term trends, a decaying weight function is designed to introduce strong corrections only in the first 6 time steps of the prediction, and then gradually reduce the correction strength. The target prediction result is obtained through the following equation (3): (3) In the formula, Indicates the target prediction result; This indicates the initial prediction result; This represents the decay weighting function; Indicates the correction amount.

[0083] Specifically, the decaying weight function takes a value of 1 during the strong correction period (a+1 to a+6 time steps), at which time the correction amount obtained from the ARIMA model is fully superimposed, leveraging its ability to capture short-term sudden fluctuations to significantly improve recent prediction accuracy; during the transition period (a+7 to a+12 time steps), the value of the decaying weight function is determined by... The calculation shows that the value of z ranges from [7,12]. It can be seen that the function value gradually decays linearly to 0 over time, achieving a smooth transition and avoiding a cliff in the prediction curve. During the trend-dominant period (a+13 to a+72 time steps), the decay weight function is 0. At this time, it relies entirely on the initial prediction results of the LSTM network, avoiding the cumulative error interference of the long-term prediction of the ARIMA model, and accurately achieving throughput prediction for all time steps.

[0084] Optionally, for processing non-stationary data, in addition to using a combination of attention mechanism and ARIMA correction, the Variational Mode Decomposition (VMD) algorithm can also be used to decompose the multi-source prediction data into multiple relatively stationary mode components, then perform LSTM prediction on each component, and finally reconstruct the prediction results of each component to obtain the target prediction result.

[0085] By introducing an attention mechanism to dynamically weight sudden factors when the data is non-stationary, and combining it with the ARIMA model to correct the prediction results of the LSTM network, the accumulation of prediction errors in non-stationary scenarios is effectively avoided, ensuring accurate throughput prediction even in complex and ever-changing environments.

[0086] Step S304: Verify the reliability of the prediction method corresponding to the target prediction result.

[0087] Specifically, step S304 includes: Step S3041: Obtain the second historical actual throughput of the port within the preset verification period.

[0088] Specifically, the actual throughput of the port at each time step within the preset verification period closest to the current time is obtained as its corresponding second actual throughput.

[0089] Step S3042: Using the prediction method corresponding to the target prediction result, prediction is performed based on multi-source prediction data corresponding to the preset verification period to obtain historical prediction results.

[0090] Specifically, the prediction method corresponding to the target prediction result is adopted, and prediction is performed based on the multi-source prediction data collected at each time step of the preset verification period to obtain the prediction result at each time step, which is used as the historical prediction result.

[0091] Step S3043: Calculate the error index based on the second historical actual throughput and historical prediction results.

[0092] Specifically, the mean absolute percentage error (MAPE) between the second historical actual throughput and the historical forecast results is calculated using the following formula (4) as an error indicator.

[0093] (4) In the formula, This represents the number of time steps in the preset verification period, which is 24 in this embodiment of the invention. Indicates the first The second historical actual throughput at each time step; Indicates the first The predicted throughput value for each time step is determined with reference to step a1.

[0094] Step S3044: If the error index is not greater than the preset service tolerance threshold, the reliability verification is determined to be passed.

[0095] Specifically, the preset business tolerance threshold refers to the upper limit of the prediction error that is set in advance based on the actual operational needs of the port and historical operating experience, and can be taken as 8%. If the error index is not greater than the preset business tolerance threshold, it means that the model in the prediction method used is performing well and the prediction deviation is within a controllable range. It is judged to have passed the reliability verification and provides reliable data for subsequent operation planning.

[0096] Step S3045: When the error index is greater than the preset service tolerance threshold, it is determined that the reliability verification has failed.

[0097] Specifically, if the error index exceeds the preset business tolerance threshold, it indicates that the model in the prediction method used has failed to adapt to recent changes in data characteristics, is susceptible to sudden events that cause pattern deviations, and the prediction accuracy cannot meet business requirements, thus failing the reliability verification.

[0098] By predicting and calculating error indicators based on the second historical actual throughput, and comparing them with preset business tolerance thresholds, the reliability of the prediction method can be verified. This allows for an objective quantitative assessment of prediction accuracy and helps improve the reliability of prediction results.

[0099] Step S305: If the reliability verification fails, the parameters of the prediction method corresponding to the target prediction result are adaptively adjusted, and the prediction method corresponding to the stationarity judgment result is used to make predictions based on multi-source prediction data to obtain the target prediction result of the port, until the reliability verification passes.

[0100] Specifically, if the reliability verification fails, the key parameters of the model involved in the prediction method are adaptively adjusted, such as adjusting the learning rate from 0.001 to 0.0005. Then, the process returns to step S303 to re-predict based on the updated parameters, obtaining a new target prediction result, and performing the reliability verification again. Through iterative optimization and cyclic verification, the prediction accuracy is continuously improved until the verification passes, ultimately outputting a reliable throughput prediction result, which helps improve the accuracy of subsequent task planning.

[0101] In some alternative implementations, Figure 3 This is a flowchart illustrating the determination of target prediction results according to an embodiment of the present invention, such as... Figure 3As shown, multi-source prediction data for the port is collected, and the data is assessed for stationarity. If the stationarity assessment result is stationary, an LSTM network is directly used for prediction to obtain the target prediction result. If the stationarity assessment result is non-stationary, an attention mechanism is first used for weighting, followed by prediction using an LSTM network to obtain an initial prediction result. Then, ARIMA residual modeling is performed to obtain a correction. The initial prediction result and the correction are fused to obtain the target prediction result. The reliability of the prediction method corresponding to the target prediction result is verified. If it passes, the target prediction result is output; if it fails, the parameters of the prediction method are adaptively adjusted, and the adjusted prediction method is used to predict again. This process is repeated until the reliability verification passes, at which point the target prediction result is output.

[0102] Step S306: When the reliability verification is passed, under the constraints of the port business rules, an objective function is constructed and solved based on the target prediction results to obtain the port's operation plan.

[0103] Specifically, step S306 includes: Step S3061: When passing the reliability verification, construct a set of constraint conditions based on the port business rules.

[0104] Specifically, the system also includes a rule extraction and transformation module. During reliability verification, this module reads port business rules from the business rule base, specifically including two types of rules: hard constraints on physical resources and soft / hard rules for business operations. This module uses a rule parser to transform the hard constraint on physical resources into the following constraint condition for yard capacity based on flow balance, as shown in equation (5), to ensure that yard utilization does not exceed the limit at any time step and that inventory conforms to the dynamic conservation law: (5) In the formula, Indicates the first The actual container inventory at the end of each time step; Indicates the first The predicted container volume at the [number] time step is the target prediction result for the [number]th time step. The predicted throughput for each time step; Indicates the first The outbound (loading / pickup) container volume of the planned time step; This represents the yard warning threshold coefficient, such as 0.85, which is used to reserve buffer space to prevent warehouse congestion caused by prediction errors. This indicates the maximum design capacity of the storage yard.

[0105] The rule parser transforms the soft / hard rules of business operations into priority-based operation time windows of the following equation (6), which apply only to vessels marked as high priority, and is used to concentrate resources to support high-priority vessels within a limited time: (6) In the formula, Indicates a ship Arrival time; Indicates the vessels set according to the priority rules. The latest completion time; Indicates a ship In the The operational capacity per time step depends on the number of quay cranes allocated; Indicates a ship The remaining total amount of loading and unloading operations.

[0106] The two constraints mentioned above are integrated into a sparse matrix form of constraint set, which serves as the mathematical constraint boundary for subsequent optimization solutions.

[0107] Step S3062: Construct a multi-objective function, which represents minimizing the total port dwell time of ships and maximizing the throughput plan completion rate.

[0108] Specifically, the multi-objective function is shown in equation (7) below: (7) In the formula, Represents a multi-objective function; This represents the weighting coefficient for ship demurrage time. This indicates the assembly of vessels awaiting operation; Indicates a ship The actual completion time; Indicates a ship The actual arrival time at the port; Indicates a ship The length of time spent in port; This represents the weighting coefficient for the completion rate of the throughput plan, and its sum with the weighting coefficient for the time the ship spends in port is 1. This represents the planned monthly throughput value for the port. Indicates a ship In the The actual amount of work completed at each time step; This indicates the percentage of planned tasks that were not completed.

[0109] In the above formula (7) This is used to minimize two sub-objectives, which are the two parts added together in the parentheses of the above formula (7). The first sub-objective represents the total time of all ships waiting in port. By minimizing the first sub-objective, the port's operational efficiency and service level are directly reflected. The shorter the waiting time, the lower the cost of waiting for ships and the higher the utilization rate of port equipment, which can effectively alleviate the port congestion problem. Minimizing the second sub-objective is to minimize the proportion of unfinished planned operations, which is actually to maximize the completion rate of the monthly throughput plan. This directly aligns with the port's business operation goals and ensures that the operation plan revolves around the port's monthly operating goals, taking into account both operational efficiency and operating results.

[0110] Step S3063: Under the constraints of the constraint set, solve the multi-objective function based on the target prediction results to obtain the actual amount of work completed by each ship in the port at each time step.

[0111] Specifically, in the above formula (7) As the decision variable, a mixed integer programming algorithm is adopted, with the set of constraints as the boundary of the feasible region and the target prediction result as the input. The multi-objective function is solved using commercial solvers (such as Gurobi or CPLEX). By iteratively optimizing within the constraint boundary, the optimal value of the actual completed work of each ship in the port at each time step is finally obtained accurately.

[0112] Alternatively, in addition to mixed integer programming algorithms, genetic algorithms can also be used to solve the problem. By simulating the selection, crossover, and mutation operations in the biological genetic evolution process, the optimal decision variables are searched in the solution space.

[0113] Step S3064: Based on the actual amount of work completed by each vessel in the port at each time step, determine multiple decision variables and form an operation plan.

[0114] Specifically, in order to transform the optimal value of the ship time step operation quantity into an operational plan that can be implemented in the port, a full-dimensional decision variable covering binary variables, continuous variables, and integer variables is derived, corresponding to different scheduling dimensions of port operations.

[0115] The integer variable represents the number of loading and unloading equipment at each time step. First, the product of the actual completed workload and the preset workload imbalance coefficient at each time step is calculated. Then, the product of the actual hourly efficiency of a single piece of equipment (the amount of work completed by a single piece of equipment per unit time, not the rated efficiency) and the effective working time of a single piece of equipment is calculated. The ratio of the first product to the second product is the number of loading and unloading equipment at that time step.

[0116] The binary variable is used to determine the time step of a vessel's operations. When a vessel commences operations at a certain time step (i.e., when the actual amount of work completed at that time step is not zero), the variable takes a value of 1; otherwise, it takes a value of 0. For each vessel, by traversing all its time steps, a sequence of binary variable values ​​for all time steps is obtained. The first time step to take a value of 1 is identified as the vessel's start time, and the last time step to take a value of 1 is identified as the completion time, thus defining the vessel's operation time window. Based on the operation time windows of all vessels, the order of loading and unloading operations for all vessels in the port is determined according to their start times and priorities. The operation time windows and order of all vessels are integrated to generate a Gantt chart of vessel operation sequence, visually displaying the operation time steps and their sequential relationships for each vessel, providing direct timing instructions for on-site scheduling.

[0117] To translate the actual workload completed by each vessel at each time step into a specific spatial allocation plan, a vessel-yard spatial flow matrix is ​​constructed. Continuous variables are defined. , indicating the first At this time step, the ship With storage yard blocks The actual container flow allocated between them. This continuous variable must satisfy the following equation (8) showing the quay-yard flow conservation mapping and the following equation (9) showing the yard block-level capacity and safety constraints: (8) (9) In the formula, Indicates the total number of storage yard blocks; Indicates a ship In the The actual amount of work completed at each time step; Represents a storage yard block In the The real-time inventory at each time step, and referring to the above formula (5), must meet the following requirements: , Represents a storage yard block The maximum design capacity.

[0118] Equation (8) above indicates that for any vessel at any time step, the total amount allocated to all yard blocks must be equal to the actual amount of work completed by the vessel at that time step, thus clarifying the logical basis for decomposing the macro-level work volume at the shore to the micro-level blocks within the yard. Equation (9) above specifies the constraint conditions in step S3061 to the block dimension, indicating that for any yard block, the real-time inventory of that yard block is updated according to the vessel's operational attributes ("+" for unloading into the yard and "-" for loading out of the yard).

[0119] In the process of solving the multi-objective function, the solver needs to adaptively calculate the optimal solution matrix of continuous variables within the feasible region that satisfies the above equations (8) and (9). The system can directly extract the continuous variables corresponding to each storage yard block at each time step and fill them into the planning and scheduling table, thereby guiding the on-site trucks to distribute the boxes that need to be stored to the two blocks in proportion, avoiding the micro-congestion bottleneck caused by the allocation overload of a single block.

[0120] The target prediction results, the actual amount of work completed by each ship at each time step, integer variables and binary variables are also filled into the plan scheduling table, so that the throughput, ship operation requirements, the actual amount of work completed by each ship, the number of loading and unloading equipment in the port, whether the port is carrying out operations, and the actual container flow allocated between ships and yard blocks are clearly defined in the a+72 time steps, forming a complete operation plan.

[0121] In some optional implementations, the system also includes an execution and feedback module, which distributes the operation plan to the port operation control system to control the loading and unloading equipment and the yard management system to perform the operations. Real-time data on actual operations at each time step (such as actual ship loading and unloading time, actual inventory in the yard, etc.) is collected and compared with the target prediction results and the multi-objective function to calculate the relative deviation rate between the actual operation value and the predicted value. If the relative deviation rate is less than a preset tolerance threshold, it is considered normal fluctuation, and the system maintains the current model parameters and rule configuration unchanged. If the relative deviation rate exceeds the preset tolerance threshold for a consecutive preset number of time steps (e.g., the relative deviation rate is greater than 10% for three consecutive time steps), a dual-channel feedback adjustment mechanism is triggered.

[0122] The dual-channel feedback adjustment mechanism comprises two channels. Channel one is the online incremental learning of the LSTM network: data collected at time steps with a relative bias rate greater than a preset value, including actual throughput, current weather, and event markers, are automatically labeled as high-value difficult examples and added to the training set used during pre-training. The network is then fine-tuned online using the updated training set, keeping the weights of the lower-level feature extraction layers frozen while updating the weight parameters of the top-level fully connected layers. This allows the model to quickly adapt to new data distribution characteristics (such as adapting to new seasonal fluctuations or policy impacts), thereby reducing future prediction errors.

[0123] Channel 2 is for dynamic calibration of thresholds in the business rule base: In response to deviations in operational plans (such as yard congestion or idle machinery), the system automatically adjusts key rule parameters in the business rule base. If frequent alarms are triggered by the actual inventory level in the yard (i.e.,...) , (This represents the actual inventory level at the storage yard under real-time monitoring), and the system automatically adjusts it downwards. ,like , indicating that Reduce the threshold by 5% to proactively tighten the alarm limit and reserve more buffer space to cope with uncertainties.

[0124] This invention compensates for the lack of information in single-source data by collecting multi-source forecast data from ports, providing a comprehensive and continuous data source for operational planning. The multi-source forecast data is assessed for stationarity, and the corresponding forecasting method is used to predict the target throughput, outputting the target forecast result. This avoids the problems of insufficient adaptability of single forecasting methods to data with different stationarities and the susceptibility to forecasting errors, effectively improving the accuracy and precision of throughput forecasting. The forecasting method corresponding to the target forecast result undergoes reliability verification, further improving the accuracy and precision of throughput forecasting and ensuring high reliability of the forecast data used for subsequent operational planning. During reliability verification, under the constraints of port business rules, an objective function is constructed and solved based on the target forecast result to obtain the port's operational plan. By balancing the flexibility of data-driven approaches with the rigidity of rule-driven constraints, the accuracy of throughput forecasting is effectively improved, thereby increasing operational planning efficiency and providing strong support for efficient port operations.

[0125] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention.

[0126] The following is a detailed reference. Figure 4 The diagram illustrates a structural schematic suitable for implementing an electronic device according to embodiments of the present invention. The electronic device may include a processor (e.g., a central processing unit, graphics processor, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from memory 408 into random access memory (RAM) 403. RAM 403 also stores various programs and data required for the operation of the electronic device. The processor 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0127] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; memory devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic devices to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 Electronic devices with various devices are shown, but it should be understood that it is not required to implement or have all of the devices shown, and more or fewer devices may be implemented or have instead.

[0128] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a memory 408, or installed from a ROM 402. When the computer program is executed by the processor 401, it performs the functions defined in the rule-based and data-driven port operation planning method of the embodiments of the present invention.

[0129] Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0130] This invention also provides a computer-readable storage medium. The methods described above according to embodiments of the invention can be implemented in hardware or firmware, or implemented as computer code that can be recorded on a storage medium, or implemented as computer code downloaded via a network and originally stored on a remote storage medium or a non-transitory machine-readable storage medium and then stored on a local storage medium. Thus, the methods described herein can be processed by software stored on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. The storage medium can be a magnetic disk, optical disk, read-only memory, random access memory, flash memory, hard disk, or solid-state drive, etc.; further, the storage medium can also include combinations of the above types of memory. It is understood that computers, processors, microprocessor controllers, or programmable hardware include storage components capable of storing or receiving software or computer code. When the software or computer code is accessed and executed by the computer, processor, or hardware, the rule-based and data-driven port operation planning method shown in the above embodiments is implemented.

[0131] Although embodiments of the invention have been described in conjunction with the accompanying drawings, those skilled in the art can make various modifications and variations without departing from the spirit and scope of the invention, and such modifications and variations all fall within the scope defined by the appended claims.

Claims

1. A port operation planning method based on both rules and data, characterized in that, The method includes: Collect multi-source prediction data for ports; The stationarity of the multi-source prediction data is assessed to obtain the stationarity assessment result; Using the prediction method corresponding to the stationarity judgment result, prediction is performed based on the multi-source prediction data to obtain the target prediction result of the port. The reliability of the prediction method corresponding to the target prediction result is verified. During the reliability verification, under the constraints of port business rules, an objective function is constructed and solved based on the target prediction results to obtain the port's operation plan.

2. The method according to claim 1, characterized in that, The multi-source prediction data includes historical throughput data; The step of performing a stationarity assessment on the multi-source prediction data to obtain a stationarity assessment result includes: Based on the historical throughput data in the multi-source prediction data, a historical throughput sequence is constructed; Based on the aforementioned historical throughput sequence, a regression equation is constructed; The regression equation is regressed to obtain the regression coefficients and the standard errors of the regression coefficients; The ratio of the regression coefficient to the standard error is calculated as an indicator of stationarity. When the stability index is less than a preset critical value, the stability judgment result is determined to be stable; When the stationarity index is not less than a preset critical value, the stationarity judgment result is determined to be non-stationary.

3. The method according to claim 2, characterized in that, The prediction method corresponding to the stationarity judgment result, based on the multi-source prediction data, to obtain the target prediction result of the port includes: When the stationarity judgment result is stationary, a pre-trained long short-term memory network is used to make a prediction based on the multi-source prediction data to obtain the target prediction result.

4. The method according to claim 2, characterized in that, The prediction method corresponding to the stationarity judgment result, which performs prediction based on the multi-source prediction data to obtain the target prediction result of the port, further includes: When the stationarity judgment result is non-stationary, an attention mechanism is used to assign attention weights to each time step in the multi-source prediction data; The data at each time step is weighted by its corresponding attention weight to obtain the weighted multi-source prediction data. A pre-trained long short-term memory network is used to make predictions based on weighted multi-source prediction data to obtain initial prediction results; Based on the historical forecasts of the port, residual modeling is performed to obtain the correction amount; The target prediction result is obtained by fusing the initial prediction result and the correction amount using a time-varying weighted fusion algorithm.

5. The method according to claim 4, characterized in that, The residual modeling based on the historical forecasts of the port yields the correction amount, including: Obtain the port’s first historical actual throughput and the corresponding historical throughput prediction value within a preset historical time window; Calculate the residual sequence between the first historical actual throughput and the historical throughput prediction value; Construct an autoregressive integral moving average model for the residual sequence; The correction amount is obtained by making predictions based on the autoregressive integral moving average model.

6. The method according to claim 1, characterized in that, The reliability verification of the prediction method corresponding to the target prediction result includes: Obtain the second historical actual throughput of the port within a preset verification period; Using the prediction method corresponding to the target prediction result, prediction is performed based on the multi-source prediction data corresponding to the preset verification period to obtain historical prediction results; An error index is calculated based on the second historical actual throughput and the historical prediction results; When the error index is not greater than the preset service tolerance threshold, the reliability verification is determined to be passed. If the error index exceeds the preset service tolerance threshold, the reliability verification is deemed to have failed.

7. The method according to claim 6, characterized in that, The method further includes: If the reliability verification fails, the parameters of the prediction method corresponding to the target prediction result are adaptively adjusted, and the prediction method corresponding to the stationarity judgment result is used to make a prediction based on the multi-source prediction data to obtain the target prediction result of the port, until the reliability verification passes.

8. The method according to claim 1, characterized in that, During the reliability verification, under the constraints of port business rules, an objective function is constructed and solved based on the target prediction results to obtain the port's operation plan, including: When passing the reliability verification, a set of constraints is constructed based on the port business rules; Construct a multi-objective function, which represents minimizing the total ship demurrage time and maximizing the throughput plan completion rate; Under the constraints of the set of constraints, the multi-objective function is solved based on the target prediction results to obtain the actual amount of work completed by each ship in the port at each time step; Based on the actual amount of work completed by each vessel in the port at each time step, multiple decision variables are determined to form the operation plan.

9. A port operation planning system based on both rules and data, characterized in that, The system includes: The data acquisition and preprocessing module is used to collect multi-source forecast data for the port. The prediction module is used to determine the stationarity of the multi-source prediction data, obtain the stationarity determination result, use the prediction method corresponding to the stationarity determination result to make a prediction based on the multi-source prediction data, obtain the target prediction result of the port, and perform reliability verification on the prediction method corresponding to the target prediction result. The optimization decision module is used to construct and solve an objective function based on the target prediction result when the reliability verification is passed, under the constraints of the port business rules, to obtain the port's operation plan.

10. An electronic device, characterized in that, include: A memory and a processor are interconnected, the memory storing computer instructions, and the processor executing the computer instructions to perform the port operation planning method based on rules and data as described in any one of claims 1 to 8.

Images