Real-time dynamic route induction method for ships in inland river route network

Abstract

The invention discloses a real-time dynamic ship route induction method for a ship in an inland river route network. The method comprises the following steps: collecting information; calculating the real-time dynamic speed of the ship and the interpenetration clearance of the ship; establishing a ship route network weighting topological graph; establishing a generalized path resistance function; establishing an induction path optimization model and solving an optimal route; and outputting a real-time dynamic ship induction route. The real-time dynamic induction route of the ship is calculated by establishing the ship route induction model on the basis of the real-time dynamic ship traffic flow prediction result in combination with the ship route network topological structure in the inland river route network, so that the method has the characteristics of high universality and wide coverage; the LSTM network is adopted to predict the ship navigational speed, the prediction precision is high, convergence is fast, and the workload is small; according to the invention, accurate real-time dynamic induction route information of the ship can be provided, and the reliability is good; due to the fact that the ant colony optimization algorithm of operational research is used for achieving real-time dynamic route planning of the ships in the inland river route network, economical efficiency is good.

Description

Technical Field

[0001] The invention relates to a ship route induction method, in particular to a real-time dynamic route induction method for ships in an inland river route network. Background Art

[0002] With the rapid development of international trade and the increasing prosperity of water transport, the traffic volume of water transport is also growing rapidly, the shipping traffic volume is growing rapidly, the waterway traffic is becoming more and more busy, and the difficulty of water transport traffic organization is gradually increasing. At present, many countries in the world, including developed countries, generally adopt intelligent traffic control systems, and the core application of intelligent transportation systems in water transport is the real-time dynamic guidance of ship routes.

[0003] Regardless of whether the ship is sailing on the inland waterway network or the ocean route network, the traditional ship route induction method is only based on radar, Automatic Identification System (AIS system), VTS and GPS to guide the ship to operate along a fixed route. This route induction method belongs to static navigation, which does not take into account the changes in the traffic flow of each route in different time periods within the navigation area. The changes in the road resistance of each route are caused by changes. It has a certain degree of blindness, and thus cannot guarantee the effective reduction of the operating costs of shipping companies, and also brings certain difficulties to the management of the maritime department. Summary of the invention

[0004] In order to solve the above problems existing in the prior art, the present invention proposes a real-time dynamic induction method for ship routes in an inland waterway network based on real-time dynamic AIS data and a dynamic route network topology structure, which can improve the macro-management level of the maritime administration over the ships in the navigation area, and also reduce the operating costs of related shipping companies and auxiliary companies.

[0005] In order to achieve the above object, the technical solution of the present invention is as follows: A real-time dynamic route induction method for ships in an inland waterway network comprises the following steps:

[0006] A. Collect information

[0007] Obtain traffic flow information through the ship information network, and extract speed, statistical flow, ship type, ship length, ship width, time, travel direction, ship speed, track direction, longitude, latitude and wind speed from it according to the characteristics of the original ship AIS system data;

[0008] B. Calculate the real-time dynamic speed of ships and the gap between ships

[0009] The long short-term memory network, i.e., LSTM network, is used to predict the real-time dynamic speed of the ship. The real-time dynamic time series data S(t) of the ship at time t is used as the input of the LSTM network, and the average value data of the real-time dynamic speed time series of the ship at time t+1, t+2, ..., t+5 is used as the output. The specific steps are as follows:

[0010] B1. Normalize the original data set

[0011] After repairing the data of the original ship automatic identification system, i.e., the AIS system, the time series data with equal intervals of time interval K is obtained to form the original data set X. The formula for normalizing the original data set X is as follows:

[0012]

[0013] Where: X raw is the data value in the original data set X; min(X) is the minimum value in the original data set X; max(X) is the maximum value in the original data set X; x trans are the normalized data values, all ranging from 0 to 1, dimensionless;

[0014] B2. LSTM network neuron input data weight initialization

[0015] The Xavier method is used to initialize the weights of the LSTM network neuron input data, and the grid search method is used to optimize the LSTM network time step, hidden layer nodes, learning rate, and batch size parameters. The LSTM network parameters are selected as follows: 4 input layer nodes, 100 hidden layer nodes, 2 output layer nodes, the initial learning rate is set to 0.001, and the batch size is 20.

[0016] B3. Training LSTM network

[0017] The input of the LSTM network takes time order into consideration. When considering the ship information at multiple consecutive moments, the vectors of multiple moments are no longer concatenated into one vector, but a first-order time step is added. The format of the input data is in matrix form.

[0018] The training samples are represented as follows:

[0019] {xt:[S(t),S(t-1),L,S(t-n+1)], Yt:[L(t+1),L(t+2),L(t+3),L(t+4),L(t+5)]}

[0020] Where n is the time step, and the adam optimization algorithm is used to update the weights of the input data of the upper layer of neurons in the LSTM network, and the final weights of the network are determined according to the set error rate.

[0021] Construct a training data set: use the sequence data S(t-105), S(t-104), …, S(t-1) of 105 consecutive moments as the input of the training LSTM network, and use the sequence data S(t-100), S(t-99), …, S(t) of 101 moments that are 5 moments later than the input data as the output of the training LSTM network. Save the parameters of the LSTM network after training.

[0022] B4. Prediction using LSTM network

[0023] After training is completed, the test samples are input into the LSTM network for prediction. At the same time, the predicted results are denormalized so that the predicted track data has real physical meaning.

[0024] Verify the prediction results of the LSTM network: Take the ship's dynamic speed sequence data S(t-35), S(t-34), ..., S(t-1) at 35 consecutive moments as the initial input of the test LSTM network, and predict the corresponding ship's dynamic speed sequence data S(t-30), S(t-29), ..., S(t) from t-30 to t, so as to obtain the predicted value of the ship's dynamic speed. Since the predicted output ship's dynamic speed data is in groups of 5, and the required data is the average value of each group of predicted data. Therefore, when drawing the actual and predicted comparison chart, take each moment and the 4 data before it as a group, calculate its average value as the actual data, and the output data of each group of 5 is the predicted data.

[0025] B5. Calculation of ship insertion clearance

[0026] Since the main channel of inland waterways is often dredged, many construction ships often cross the main channel to carry out sand dumping operations. Therefore, the trajectory of the construction ship should also be taken into account in the optimization of the inland waterway network trajectory, and the real-time dynamic ship interlacing gap needs to be calculated. The specific technical solution is as follows:

[0027] Collect waterway hydrological, meteorological and traffic environment data;

[0028] Calculate the navigation time of ships on the waterway;

[0029] Establish a waterway ship arrival model;

[0030] Calculate the channel ship crossing gap and establish the ship interspersed gap model as follows:

[0031] z t0 =min(z i ,z j ) (2)

[0032] z i=max(z qi -h qi ,z hi -h hi ) (3)

[0033] z j =max(z qj -h qj ,z hj -h hj ) (4)

[0034] Where:

[0035] Z i ——The maximum distance maintained by the target ship when it passes through the bow or stern of ship i, excluding the safety distance;

[0036] h hi ——The target ship passes through the stern of ship i, and the horizontal safe distance between the target ship and the target ship;

[0037] h qi ——The target ship passes through the bow of ship i, and the horizontal safe distance between the target ship and the ship;

[0038] Z j ——The maximum distance that the target ship can maintain when passing through the bow or stern of ship j, excluding the safety distance;

[0039] h hj ——The target ship passes through the stern of ship j and the horizontal safe distance between the target ship and the ship;

[0040] h qj ——The target ship passes through the bow of ship j and is at a safe lateral distance from the ship.

[0041] C. Establish a weighted topological map of the ship route network

[0042] The inland waterway network is divided into an inland waterway network formed by the main channel plus a grid-type branch channel and an inland waterway network formed by the main channel plus a branch channel of a tree-branch type. The nodes are interrupted at the points where ships pass through the main channel of the tree-branch type network and the intersections of routes in the grid type network to establish a weighted topological map of the ship route network.

[0043] D. Establishing a generalized road resistance function

[0044] When different routes intersect, there is a potential traffic conflict between ships, which will cause delays. This is also one of the factors that cause delays in ships passing through the induction area. Therefore, the ship travel time is used as the road resistance function, that is: T = t + d s +θd z , where d s The average conflict delay time of ships on the main route; d zis the average delay time of delayed ships when different routes intersect; when calculating the delay in the intersection area of ​​different routes, judge the size between the ship flow on the left route and the merging or passing capacity. When there is an intersection area of ​​different routes and the arrival rate of ships on the left route is greater than the merging or passing capacity, the traffic on the left route will be blocked. θ is the discrimination coefficient, θ for the left route is 1, and θ for the right route is 0.

[0045] The ship travel time includes the sailing time prediction result and the ship delay time, which is used as one of the decision variables of the objective function in the dynamic route induction process.

[0046] E. Establish an induced path optimization model and solve the best route

[0047] Based on the research on the induced path of the target ship and the crossing clearance of the target ship, the induced path optimization model is obtained by combining them as follows:

[0048]

[0049] The constraints are:

[0050]

[0051] v 0 >0 (7)

[0052] Y (i,j) >0 (8)

[0053] t a +(m-1)t g ≤t ij ≤t c +(m-1)t g (9)

[0054] t h >t z 1,2 (10)

[0055] Where:

[0056] L ij Indicates the construction operation segment with starting point i and end point j;

[0057] x ij Indicates whether the ship has passed through segment L ij , 1 means passing, 0 means not passing;

[0058] T ij Indicates that the construction operation segment L is passed ij Total sailing time of the vessel;

[0059] t ij Indicates that the construction operation segment L is passedij The interval time between the two ships;

[0060] M represents the set of route nodes, M = {A, 1, 2, ..., B};

[0061] Y (i,j) Indicates construction operation segment L ij Mileage;

[0062] t a It indicates the time required for the target ship to sail from the departure position to the intersection;

[0063] t c It represents the minimum time required for the target ship to complete a crossing;

[0064] t g It indicates the time distance between two consecutive target ships when multiple target ships are crossing.

[0065] m represents the number of target ships in the queue ready to be inserted;

[0066] t h Indicates the clearance between two ships during normal navigation;

[0067] t z 1 Indicates the critical gap for ships to merge into;

[0068] t z 2 Indicates the critical gap through which the ship passes;

[0069] t s It indicates the time between two ships merging or passing in the delayed direction;

[0070] Constraint (6) indicates that if a ship passes through the construction section L ij If yes, it is 1, otherwise it is 0;

[0071] Constraint (7) indicates that the ship's speed is always a positive real number at any time, that is, the ship is always in a sailing state;

[0072] Constraint (8) indicates that the mileage of a route segment is a positive real number;

[0073] Constraint (9) ensures that ships on the main route do not experience conflict delays caused by the target ship intersecting.

[0074] Constraint (10) ensures that ships on the main route do not experience conflicts and delays due to intersections between different routes.

[0075] F. Output real-time dynamic ship guidance route to guide ship navigation

[0076] Output the real-time dynamic optimal guidance results of the ship's route to provide decision-making information for ship drivers.

[0077] Furthermore, the method for repairing the data of the original ship automatic identification system (AIS system) described in step B comprises the following steps:

[0078] B11. Outlier Removal

[0079] Use the binning method to divide the data into a series of "boxes" of equal width. If the value of a data point on a certain attribute is outside 3 times the standard deviation of the statistical mean of all data points in the "box" on this attribute, this point is considered an outlier and needs to be removed.

[0080] B12. Time alignment and dynamic data completion

[0081] For the outliers removed in step B11 and the missing values ​​existing in the AIS data itself, the cubic spline interpolation method is used to repair them to obtain track data with equal time intervals.

[0082] B13. Static data completion

[0083] In the event of long-term loss of basic information on ship type and size, the largest ship information in the waterway section is extracted to replace it.

[0084] Furthermore, the time interval K in step B is 3-7 min.

[0085] Compared with the prior art, the present invention has the following beneficial effects:

[0086] 1. Since the present invention is based on the results of real-time dynamic ship traffic flow prediction, combined with the topological structure of the ship route network in the inland river route network, by establishing a ship route induction model, the real-time dynamic induction route of the ship is calculated, which is different from the existing static induction route planning of ships, and has the characteristics of strong versatility and wide coverage;

[0087] 2. The LSTM network used in the present invention predicts the ship speed with high prediction accuracy, fast convergence and small workload;

[0088] 3. Since the present invention can provide accurate real-time dynamic guided route information of ships, it has strong scientificity, high calculation accuracy and good reliability;

[0089] 4. Since the present invention utilizes the ant colony optimization algorithm of operations research to solve the real-time dynamic route planning of ships in the inland waterway network, it can effectively reduce the operating costs of shipping companies and has good economic efficiency.

[0090] 5. By adopting the ship route dynamic induction method proposed in the present invention, the management department can dynamically plan the ship routes within the jurisdiction, thereby achieving macro-management of the ships in the entire jurisdiction. BRIEF DESCRIPTION OF THE DRAWINGS

[0091] Figure 1 It is a schematic diagram of the process of the present invention.

[0092] Figure 2 It is a map of a place somewhere at the mouth of the Yangtze River.

[0093] Figure 3 This is a schematic diagram of the gap where ships pass through.

[0094] Figure 4 This is the LSTM network prediction flow chart.

[0095] Figure 5 This is the LSTM network prediction result diagram.

[0096] Figure 6 It is a schematic diagram of the waterway.

[0097] Figure 7 It is a curve chart showing the relationship between the speed of ships on the delayed main route and time.

[0098] Figure 8 It is a curve chart showing the relationship between the displacement of delayed ships on the main route and time.

[0099] Fig. 9 It is the flow chart of the ship induced path algorithm.

[0100] Fig.10 This is a schematic diagram of the induced path planning under case 1.

[0101] Fig.11 This is a schematic diagram of the induced path planning under case 2.

[0102] Fig.12 This is a schematic diagram of the induced path planning under situation 3.

[0103] Fig.13 This is a schematic diagram of the induced path planning under situation 4.

[0104] Fig.14 This is a schematic diagram of the target ship's induced route in situation 1.

[0105] Fig.15 This is a schematic diagram of the target ship's induced route in situation 2.

[0106] Fig.16 This is a schematic diagram of the target ship's induced route in case 2 where the crossing angle is changed.

[0107] Fig.17This is a schematic diagram of the target ship's induced route under situation 3.

[0108] Fig.18 This is a schematic diagram of the target ship's induced route under situation 4.

[0109] Fig.19 It is a schematic diagram of the target ship's induced route in case 4 of changing the crossing angle.

[0110] Fig. 20 This is a schematic diagram of the target ship induced path program running results. DETAILED DESCRIPTION

[0111] The present invention will be further described below in conjunction with the accompanying drawings. Figure 1 As shown, the real-time dynamic route induction method for ships in an inland waterway network includes the following steps:

[0112] 1. Information Data Collection

[0113] 1. Inland waterway ship information data collection

[0114] The present invention monitors a place near the Yangtze River estuary. Figure 2 At the bold black line mark shown, the AIS system collects ship information on the waterway at this section, including ship traffic, ship speed, ship size, current wind force and other related information. The collection time is four days from December 22 to 25, 2017, from 9:30 to 17:00 every morning. The collection results of a certain period are shown in Table 1. The collected information is mainly used for subsequent simulation of ships in the channel, using random distribution to generate initial ship positions, sizes and speeds, and using the generated ship data to study the induced path of the ship.

[0115] Table 1 Ship information statistics 1

[0116]

[0117] from Figure 2 It can be seen that there are planned directions in the waterway, and most ships sail in the specified waterway according to the waterway and its specified direction. While collecting ship information in the waterway, a certain number of ships are selected to count their dynamic information during navigation. Statistics are performed every short period of time to obtain their continuous dynamic navigation information within a certain navigation time. The data of a ship in a certain period of time is shown in Table 2. The dynamic navigation data of these ships combined with the flow data collected in the previous time period can be used for ship speed prediction training in the subsequent LSTM network.

[0118] Table 2 Ship dynamic information statistics 2

[0119]

[0120]

[0121] 2. Related data processing

[0122] Because the actual AIS system data contains a large amount of erroneous data and missing data, in addition, since the transmission time intervals of the AIS system data are different at different ship speeds, the dynamic navigation data sampling intervals may be unequal. Some of the AIS system data used in the present invention are dynamic information. As can be seen from Table 1, in the process of real-time data collection, there will be missing ship types, lengths and widths of ships, and the data needs to be repaired. Therefore, the relevant data needs to be repaired, and the repair method is as follows:

[0123] (1) Elimination of outliers. Use the binning method to divide the data into a series of "boxes" of equal width. If the value of a data point on a certain attribute is outside the three standard deviations of the statistical mean of all data points in the "box" on this attribute, this point is considered an outlier and needs to be eliminated.

[0124] (2) Time alignment and dynamic data completion: For the outliers removed in step (1) and the missing values ​​in the AIS system data itself, the cubic spline interpolation method is used to repair them to obtain track data with equal time intervals.

[0125] (3) Static data completion. In the event of long-term loss of basic information such as ship type and size, the largest ship information in the waterway section can be extracted to replace it. This method can fully consider issues such as the distance between ships and ensure the safety of navigation and route guidance planning.

[0126] According to steps (1), (2) and (3), the continuous AIS system data can be processed to obtain a relatively complete corresponding ship navigation dynamic information record after processing, which is the AIS system time series data. Table 3 is a part of the intercepted and processed time series data of a ship's navigation dynamics. Not only the abnormal data is removed and repaired, but also the coordinates are transformed from longitude and latitude to Mercator projection. After the Mercator coordinate values ​​are properly rotated according to the angle of the midline of the channel, the prediction that can be used for subsequent crossing gaps and induced routes can be obtained. The MMSI of the ship is 440287000, the date is December 23, 2017, and the time interval between adjacent data is 1 minute. The data with gray shading in Table 3 are the data filled by cubic linear interpolation and the previously missing data.

[0127] Table 3 Ship information repair

[0128]

[0129] 2. Real-time dynamic ship traffic flow information prediction

[0130] In the ship traffic flow information, according to the characteristics of the original ship AIS system data, traffic flow information such as speed, statistical flow, ship type, ship length, ship width, time, driving direction, ship speed, track direction, longitude, latitude and wind force are extracted to establish the predicted data. In addition, wind force also has a certain degree of influence on ship traffic. For example, in windy weather, the ship speed will slow down or even stop sailing to avoid collision. Here, additional wind force information is added to assist in the prediction, and the wind force is classified according to the degree, and added as an influencing factor for the change of ship traffic flow information. In addition, it is also necessary to obtain the geographical location information (latitude and longitude) of the ship and the information on the size of the ship, but this is not the data used for prediction, but for the calculation of the interlaced gap later. Combined with other influencing factors, a systematic analysis of the data is carried out to find out the regularity, and the corresponding prediction methods and models are established to predict the changes in traffic flow in the next few periods.

[0131] 1. Ship insertion gap theory

[0132] Compared with the ocean route network, the network topology of the inland waterway network is slightly simpler, and its longitudinal main channel is relatively fixed. The transverse route presents a corresponding transverse network with the entrance and exit positions of the branch river. This network structure has the situation that the ship crosses the longitudinal channel during navigation, and the interlaced gaps that may appear in the longitudinal channel lead to the difference of the real-time dynamic induced route trajectory of the ship. However, since the ship crosses the longitudinal route, it is easy to cause collision accidents between ships. Therefore, it is necessary to predict the crossing gap during the navigation of the ship. At present, when a large number of ships arrive in the domestic Yangtze River waterway, the Maritime Safety Administration usually adopts a fleet of 10 ships to guide the ships through the Yangtze River waterway, which leads to a large number of ship delays. The present invention is used to predict the waiting time for the safe crossing gap. If the target ship finds the time and place for the safe crossing gap with no waiting time or the least waiting time during navigation, and crosses by changing the course during navigation at the time and place where it can be crossed, the original waiting time can be saved, the overall travel time of the target ship is greatly reduced, and the operation efficiency is improved. Therefore, by predicting the changes in the positions of the ships in the channel and the target ship itself when the target ship travels for a certain period of time, the waiting time for the safe crossing gap to appear after each travel period can be predicted, and the crossing gap that allows the target ship to cross the channel with no waiting time or minimum waiting time can be found.

[0133] Figure 3 (a) and (b) are at t 0 At that moment, the target ship needs to cross half of the channel; Figure 3(c) and (d) are at t c1 At this moment, after the target ship has crossed half of the channel, it readjusts its angle and prepares to cross the second half of the channel.

[0134] 2. Vessel traffic flow information prediction

[0135] According to the characteristics of the original ship AIS system data, factors such as speed, statistical flow, location (longitude, latitude), and ship size are extracted from it. The wind factor is also added. Wind, flow, and speed are used as elements to construct the ship route network traffic flow time series data, and the implementation process of the ship flow and speed prediction model based on the LSTM network is analyzed.

[0136] Based on the Windows platform, the Python version of the LSTM network for ship navigation dynamics prediction was developed using machine learning libraries such as Keras and TensorFlow. The ship navigation dynamics time series data S(t) at time t is used as the input of the LSTM network, and the average value of the ship speed state time series at time t+1, t+2, ..., t+5 is used as the output. The experimental process is shown in Figure 4 , the prediction step is as shown in step B of the invention content.

[0137] The comparison between the predicted results and the actual data is shown in Table 4 and Appendix Figure 5 As shown in the figure, due to the slow and time-intensive changes in ship speed, the predicted results are basically around the actual results.

[0138] Table 4 Ship speed prediction

[0139]

[0140]

[0141] 3. Establishing a weighted topological map of the ship route network

[0142] Since China's inland waterway network is composed of the Grand Canal, the Yangtze River system and the Pearl River system, it is a "one vertical, two horizontal and two networks", forming a tree-fork type + grid type inland waterway network. Among them: the inland waterway network centered on Jining and the Yangtze River Delta (including Shanghai, Nanjing, Hangzhou, and Hefei) is basically dominated by a grid-type route network, while the inland waterway network centered on Wuhan, Changsha, Nanchang, the Pearl River Delta of Chongqing and Nanning is basically dominated by a tree-fork type route network. Therefore, for engineering ships sailing in the tree-fork type inland waterway network for channel dredging, they often cross the main channel to carry out sand throwing operations. In addition, there are also some passenger ferries sailing in this area that need to cross the main channel. For ships sailing in the grid-type inland waterway network, real-time dynamic heading induction is required based on the shortest ship travel time. Therefore, it is necessary to perform node interruption processing at the intersection of ships in the main channel of the tree-fork type inland waterway network and the intersection of routes in the grid-type inland waterway network to establish a route network topology structure. Because of the Maritime Administration's control mechanism for the release of ships in the waterway, the distribution of ships in the waterway is not dense, and their navigation conditions can be regarded as those of free-flowing ships. In the absence of special circumstances, they will basically travel at their maximum speed and will not change their speed.

[0143] 4. Establishing route resistance function

[0144] The route resistance of the present invention is not composed of a single indicator, but takes into account the route resistance value of the minimum ship travel time as one of the decision variables of the objective function in the dynamic route induction process.

[0145] 1. Delay time due to conflict between different routes

[0146] When different routes intersect, there is a potential traffic conflict between ships, which will cause delays. This is also one of the factors causing delays in ships passing through the induction area. Calculating the delay time of ships at the intersection of different routes can be regarded as a ship queuing system. According to the relevant knowledge of queuing theory, the average delay time d of delayed ships is z Equal to its system service rate μ z Subtract the ship arrival rate λ l The reciprocal of is calculated according to the following formula:

[0147]

[0148] Where: l The ship arrival rate per unit time on the main route, μ z System service rate for ship queuing system.

[0149] 2. Delay time of ships on the main route

[0150] Since the minimum safe time interval must be maintained between ships, the actual crossing gap that can be accepted by the target ship is t-τ, so the probability P(t) of the target ship's acceptable crossing gap t is:

[0151] P(t)=λe -λ(t-τ) (12)

[0152] Among them, λ is the ship arrival rate per unit time on the main route, which is calculated as follows:

[0153]

[0154] Among them, q' represents the ship traffic volume per unit time. There is a flow conversion here. The predicted ship traffic volume is at intervals of one hour or half an hour. Therefore, the predicted ship traffic volume needs to be converted into the ship traffic volume with the same time unit as the bow time distance.

[0155] According to the relevant theoretical knowledge of queuing theory, in a single server queue system (M / M / 1), the probability p that the number of target ships ready to be inserted is m is m for:

[0156] p m =(1-ρ)ρ m (14)

[0157]

[0158] Among them, ρ represents the service intensity, which reflects the utilization of service agencies in queuing theory and expresses traffic saturation in traffic engineering. g It is expressed as the number of target ships that need to intersperse the main route ship flow per unit time; c g It indicates the number of target ships that need to be interspersed and sail through the intersection with the minimum safe distance between target ships in a unit time. That is to say, when as many target ships as possible are needed to intersperse, and the follow-up time between target ships only meets the minimum safe time of ship navigation, the number of target ships that can be sent out in a unit time.

[0159] also, Figure 7 and Figure 8 The figure shows the relationship between the navigation speed, displacement and the acceptable gap between the two ships after the ship on the main route is delayed by the target ship crossing. Figure 7It can be seen that the delayed ship on the main route starts to slow down before reaching the place where it conflicts with the target ship. At the conflict point, the ship's sailing speed is reduced to the minimum. After the ship on the main route completely passes the conflict area, it starts to accelerate again until it returns to its normal sailing speed. Assuming that the acceleration and deceleration of the delayed ship on the main route are the same during the acceleration and deceleration process, the delay time of the ship on the main route during the acceleration and deceleration process is also the same.

[0160] according to Figure 8 It can be seen that when the ship affected and delayed on the main route begins to slow down a distance before reaching the conflict zone, the time distance between the ship and the previous ship when it sails to the conflict zone is exactly the minimum time t required for the target ship to complete an insertion. c , then the delay time of the ship during deceleration is t c Subtract the original gap t between the two ships, that is, t c -t. Due to the acceleration process, the total delay time of the ships on the main route during the process of a target ship completing the insertion is 2(t c -t). Combining the previously analyzed conditions that cause ship delays on the main route, we can get the functional relationship of the ship delay time on the main route when a target ship intersects:

[0161]

[0162] Combined with the target ship's departure interval, it is the target ship's follow-up time t g , it can be concluded that when there are m target ships that need to be interspersed in the route, the functional relationship of the ship delay time on the main route is:

[0163]

[0164] Through the probability formula (12) of the target ship's acceptable insertion gap, the probability formula (14) of m target ships crossing, and the functional relationship (17) of the delay time of ships on the main route, the calculation formula (18) for calculating the delay time can be used to calculate the average conflict delay time of ships on the main route affected by the insertion of target ships without absolute priority navigation, as shown in formula (19):

[0165]

[0166] 5. Build a model and find the best route

[0167] The shortest path problem in a dynamic time-varying route network. The road resistance of each route segment in the route network is a function of the traffic flow in different time periods. Through LSTM network prediction, the predicted results of ship traffic in the route network divided by time periods in a certain period of time in the future are stored in the database. Based on the complete route segment ship traffic flow data based on time series, the ship flow in a specific time period determines the road resistance value of each route segment at the current moment. Finally, the ant colony algorithm is used to solve the objective function of each route segment to obtain the route result with the smallest generalized road resistance value in the time-varying route network. See step E of the invention content for details.

[0168] In view of the route induction problem, the present invention adopts the ant colony algorithm with better effect to solve the induction model, that is, the ship's departure point is taken as the starting point o of the ant path, the ship's target point is taken as the end point of the ant path search, and the node set selected by the ant represents the feasible solution of the ship's path. Assume that the initial time T = T 0 , then the basic process of solving the problem using the ant colony algorithm is as follows:

[0169] (1) The total number of ants m and the pheromone influence factor ε 1 , heuristic function influence factor ε 2 、Information volatility factor ε 3 , number of iterations u, maximum number of iterations u max Set the parameters such as , let the node x = o where m ants are located at the initial moment;

[0170] (2) Based on the generated construction area and the coordinates of each node, calculate the relative size and direction of the driving direction between each node and the meteorological and hydrological factors such as wind, waves and currents;

[0171] (3) Calculate the free speed V of the ship on each section based on meteorological and hydrological conditions and the ship's own conditions f , and accordingly calculate the corresponding wind and wave influence coefficient;

[0172] (4) The ants calculate the travel time and cost from node x to the feasible nodes according to the ship flow and wind and wave influence coefficient predicted by the feasible solution node connected to node x at time T, and then select the next node according to probability according to formula (20). Where: Assume that the total number of ants in the ant colony is m, the number of channel nodes is n, and the distance between adjacent nodes o and w is d ow , whose travel time is T ow , the pheromone concentration between two nodes of section ow at time t is τ ow (t). At the initial moment, each road section is set with a common initial pheromone concentration τ 0 Ant k (k = 1, 2, 3 ... m) on a node o according to the pheromone concentration and heuristic function between the nodes connected to it with probability P ow k(t) Select the next node of the path, P ow k (t) represents the probability of ant k transferring from node o to node w at time t, and its formula is as follows:

[0173]

[0174] (5) Let T = T + T xr (r is the next node selected, T xr is the travel time of the flight segment xr), let x = r, return to step (4) and continue to select the next node until reaching the target point;

[0175] (6) When ant k reaches the target point, it records the path according to the actual time taken, completes the cycle of all m ants in turn, records the optimal path and time taken in this iteration, and updates the pheromone concentration of the flight segment according to the following formula:

[0176]

[0177] where Δτ ij k Indicates that the kth ant is in segment L ij The pheromone released when walking, Δτ ij Indicates that all m ants in the ant colony are in segment L ij The total amount of pheromones released during the period.

[0178] (7) Let t = t + 1, x = o, T = T 0 , if t≤t max Go to step (4) and proceed to the next iteration, otherwise go to step (8);

[0179] (8) Output the optimal solution.

[0180] Its flow chart is as follows Fig. 9 shown.

[0181] 6. Output guided routes to guide ships

[0182] Output the best induction result of the ship route to provide decision-making information for ship driving. Generally speaking, the riverbed is usually deep in the middle and shallow on both sides. The water depth in the middle makes it difficult to slow down the water flow, so the flow rate in the middle of the river will be greater than on both sides of the river. Therefore, when establishing the induction network, the water speed in the central part of the channel will be slightly greater than on both sides. Considering efficiency, in order to increase the speed of the ship as much as possible and reduce the travel time, when the target ship is in the same direction as the water flow speed, the target ship is first crossed to the middle part of the channel, and the sailing distance in the middle part of the channel is as long as possible, and then consider a second crossing; on the contrary, when the target ship is in the opposite direction of the water flow speed, it is hoped that the target ship can travel as close to both sides of the channel as possible. In this case, it can directly cross to the other side of the channel at the right time.

[0183] Because of the strong supervision of the Maritime Safety Administration, ships in the channel usually sail on one side of the channel according to their own course, rather than sailing in the channel in a scattered manner. Therefore, for safety reasons, the target ship should sail in the same direction as other ships in the channel as much as possible. If the direction from the departure point to the destination of the target ship is the same as the sailing direction of the ships sailing on the departure point side, there is no rush to cross to the other side of the channel; on the contrary, if the direction from the departure point to the destination of the target ship is opposite to the sailing direction of the ships sailing on the departure point side, it is necessary to arrange the crossing as soon as possible.

[0184] Based on the above considerations, we have summarized the preliminary guidance route planning for four situations for the tree-branch type inland river route network, including the following four situations:

[0185] For different crossing situations of the target ship, in situation 1 (the direction from the starting point to the destination of the target ship is the same as the water speed and the heading of the ship on the same side, such as Fig.10 In the simulation shown in Figure 2, the following results are obtained by running the C language: Fig.14As shown in , the navigation time indicates the predicted total time from the start of the prediction to the end of the navigation, the waiting time indicates the waiting time before the crossing, and the lateral navigation distance and longitudinal navigation distance indicate the changes in the distances in each direction in space from the previous navigation time to the next navigation time. If only the lateral navigation distance changes between the two navigation times, it means that the target ship only needs to sail along the channel at the current position; if both the lateral navigation distance and the longitudinal navigation distance change between the two navigation times, it means that the target ship has crossed and can cross according to the set prediction angle. In the current simulated waterway, the route induction is carried out under situation 1, and the induction results are: sailing at the set speed of 1.5m / s, sailing horizontally to 132m, the sailing time is about 1 minute, and a crossing without waiting can be carried out; crossing is carried out according to the 90° set in the prediction, sailing longitudinally for 550m, and drifting horizontally for 260m due to the water flow, and the first crossing is completed in 7 minutes; then continue to sail horizontally in the waterway for about 1793m, at this time it has been sailing for about 20 minutes, and prepare for the second crossing; the second crossing does not need to wait, and can be carried out directly, the longitudinal distance of the crossing is 950m, and the horizontal drift is about 687m, at this time it arrives near the destination.

[0186] like Fig.15 As shown in the figure, in the current simulated channel, in case 2 (the direction from the target ship's departure point to the destination is opposite to the water speed and the heading of the ship on the same side (such as Fig.11 As shown in the figure, the graphic explanation is Fig.10 The result of the induction is as follows: the ship sails at the set speed of 1.5m / s, sails horizontally to 48m, and the sailing time is about 1 minute, which allows for a no-waiting crossing; the ship crosses at the 90° set during the prediction, sails 1500m longitudinally, and drifts 733m laterally due to the water flow, completing the crossing in 17 minutes, and the drift at this time is a reverse drift; the ship then continues to sail horizontally in the channel for about 3685m, and it takes about 94 minutes to sail to the vicinity of the destination. Because the water flow speed is opposite to the ship speed, the lateral drift direction of the target ship when crossing at 90° is opposite to the direction from the starting point to the end point, so the total sailing time is greatly increased. At this time, the crossing angle is adjusted to 120°, and the results are as shown in the attached figure. Fig.16 As shown, it can be seen that the required navigation time is shortened to 78 minutes, while the lateral displacement during the crossing is greatly reduced.

[0187] like Fig.17 As shown in , in the current simulated channel, in case 3 (the direction from the target ship’s departure point to the destination is the same as the water speed and opposite to the direction of the ship on the same side (such as Fig.12 As shown in the figure, the graphic explanation is Fig.10The route induction was carried out under the same conditions, and the results of the induction were: the ship sailed at the set speed of 1.5m / s and could directly cross without waiting; the ship crossed at the 90° set during the prediction, sailed 950m longitudinally, and drifted 687m laterally due to the water flow, completing the first crossing in 10 minutes; then the ship continued to sail laterally in the waterway for about 1925m, at which time it had sailed for about 24 minutes and was ready for the second crossing; the second crossing did not require waiting and could be carried out directly, with a longitudinal distance of 550m and a lateral drift of about 260m, arriving near the destination.

[0188] like Fig.18 As shown in the figure, in the current simulated channel, in case 4 (the direction from the target ship's departure point to the destination is opposite to the water speed and the same as the heading of the ship on the same side (such as Fig.13 As shown in the figure, the graphic explanation is Fig.10 The result of the induction is: sailing at the set speed of 1.5m / s, sailing horizontally to 3696m, sailing time is about 77 minutes, and no waiting can be done; crossing according to the 90° set in the prediction, sailing 1500m longitudinally, drifting 734m horizontally due to the water flow, and completing the crossing in 93 minutes. The drift at this time is reverse drift; then continue to sail horizontally in the channel for about 38m. At this time, it takes about 94 minutes to sail and arrive near the destination. Because the water flow speed is opposite to the ship speed, the lateral drift direction of the target ship when crossing at 90° is opposite to the direction from the starting point to the end point, so the total sailing time is greatly increased. At this time, the crossing angle is adjusted to 60°, and the results are as shown in the attached figure. Fig.19 As shown, it can be seen that the required navigation time is shortened to 78 minutes, and the lateral displacement during the crossing is greatly reduced.

[0189] In order to verify the safety of the path, the data changes are further visualized through programming, and the following is obtained: Fig.17 The route is displayed dynamically step by step, and only two more obvious pictures are captured for display. Among them, the hexagon represents the current position of the target ship, the thick black solid line represents the induced path of the target ship, the black dot represents the current position of the ship sailing in the channel, and the thin black solid line represents the predicted route of the ship sailing in the channel. The upper left corner shows the guidance information for the target ship in the induced path, including the sailing time, waiting time and sailing distance in each direction, such as Fig. 20 (c).

[0190] like Fig. 20 (a), when the target ship completes its first crossing, the intersection and ratio between the route of the sailing ship and the crossing path show that the target ship passes by the stern of ship i at a safe distance. Fig. 20In (b), the second crossing is completed, and there is a new ship on the right, and the target ship safely crosses from its bow. The result of the induced path is ideal.

[0191] For the attached Figure 6 The grid-type inland waterway route network shown in (b) and (d) can directly calculate the travel time of ships on each route of the grid-type inland waterway route network through the aforementioned step D, and then use the induction model of step E to calculate the real-time dynamic induced route of the ship using the ant colony algorithm in the fifth step of the aforementioned implementation method.

[0192] In summary, the present invention predicts the travel time of ships based on the real-time dynamic prediction of ship traffic flow, and establishes a generalized resistance function including ship travel time, sailing distance, and sailing fuel consumption as targets in combination with the weighted topological map of ship routes. A multi-objective induction model of ship routes is established based on the generalized resistance function combined with the time-varying route network in the induction area, and finally, the A* and other related optimization algorithms are used to solve it, and the dynamic induction routes of ships in different time periods are determined, which has the characteristics of wide coverage and strong applicability. The real-time dynamic route induction of ships will target the reasonable planning of ship routes in complex sea areas. Whether the real-time dynamic route planning of ships is reasonable will affect the arrival time of ships, the order of entry and exit, and the berth assignment, and will also affect the reasonable formulation of the port rear yard loading and unloading operation plan in the port production scheduling. Different from the static navigation of ship navigation, the present invention designs and develops a real-time dynamic route induction model for ships, which has less workload, good economy, and strong versatility. The present invention only needs to make real-time dynamic predictions of the traffic flow of the route network, combine it with the weighted topological map of ship routes in navigable waters, and establish a multi-objective induction model for ship routes based on a generalized road resistance function that includes ship travel time, sailing distance, and sailing fuel consumption as targets. Then, the real-time dynamic route induction of ships in different time periods can be calculated based on relevant optimization algorithms, thereby realizing real-time dynamic macro-management of ships in navigable waters.

Claims

1. A real-time dynamic route guidance method for ships in an inland waterway network, characterized by: The following steps are involved: A. Collect information Obtain traffic flow information through the ship information network, and extract speed, statistical flow, ship type, ship length, ship width, time, travel direction, ship speed, track direction, longitude, latitude and wind speed from it according to the characteristics of the original ship AIS system data; B. Calculate the real-time dynamic speed of ships and the gap between ships The long short-term memory network, i.e., LSTM network, is used to predict the real-time dynamic speed of the ship. The real-time dynamic time series data S(t) of the ship at time t is used as the input of the LSTM network, and the average value data of the real-time dynamic speed time series of the ship at time t+1, t+2, ..., t+5 is used as the output. The specific steps are as follows: B1. Normalize the original data set After repairing the data of the original ship automatic identification system, i.e., the AIS system, the time series data with equal intervals of time interval K is obtained to form the original data set X. The formula for normalizing the original data set X is as follows: Where: X raw is the data value in the original data set X; min(X) is the minimum value in the original data set X; max(X) is the maximum value in the original data set X; x trans are the normalized data values, all ranging from 0 to 1, dimensionless; B2. LSTM network neuron input data weight initialization The Xavier method is used to initialize the weights of the LSTM network neuron input data, and the grid search method is used to optimize the LSTM network time step, hidden layer nodes, learning rate, and batch size parameters; the LSTM network parameters are selected as follows: the input layer nodes are 4, the hidden layer nodes are 100, the output layer nodes are 2, the initial learning rate is set to 0.001, and the batch size is 20; B3. Training LSTM network The input of the LSTM network takes time order into consideration. When considering the ship information at multiple consecutive moments, the vectors of multiple moments are no longer concatenated into one vector, but a first-order time step is added. The format of the input data is in matrix form. The training samples are represented as follows: {xt:[S(t),S(t-1),L,S(t-n+1)], Yt:[L(t+1),L(t+2),L(t+3),L(t+4),L(t+5)]} Where n is the time step, and the adam optimization algorithm is used to update the weights of the input data of the upper layer of neurons in the LSTM network, and the final weights of the network are determined according to the set error rate; Construct a training data set: use the sequence data S(t-105), S(t-104), …, S(t-1) of 105 consecutive moments as the input of the LSTM network, and use the sequence data S(t-100), S(t-99), …, S(t) of 101 moments delayed by 5 moments than the input data as the output of the LSTM network. Save the parameters of the LSTM network after training. B4. Prediction using LSTM network After the training is completed, the test samples are input into the LSTM network for prediction. At the same time, the prediction results are denormalized so that the predicted track data has real physical meaning. Verify the prediction results of the LSTM network: Take the ship's dynamic speed sequence data S(t-35), S(t-34), ..., S(t-1) at 35 consecutive moments as the initial input of the test LSTM network, and predict the corresponding ship's dynamic speed sequence data S(t-30), S(t-29), ..., S(t) from t-30 to t, so as to obtain the predicted value of the ship's dynamic speed; since the predicted output ship's dynamic speed data is in groups of 5, and the required data is the average value of each group of predicted data; therefore, when drawing the comparison chart of the actual and predicted values, take each moment and the 4 data before it as a group, calculate its average value as the actual data, and the output data of each group of 5 is the predicted data; B5. Calculation of ship insertion clearance Since the main channel of inland waterways is often dredged, many construction ships often cross the main channel to carry out sand dumping operations; therefore, the trajectory of the construction ship should also be taken into account in the optimization of the inland waterway network trajectory, and the real-time dynamic ship interlacing gap needs to be calculated. The specific technical solution is as follows: Collect waterway hydrological, meteorological and traffic environment data; Calculate the navigation time of ships on the waterway; Establish a ship arrival model for the waterway; Calculate the channel ship crossing gap and establish the ship interspersed gap model as follows: With t0 =min(of i ,With j ) (2) z i =max(z qi -h qi ,z hi -h hi ) (3) z j =max(z qj -h qj ,z hj -h hj ) (4) Where: Z i ——The maximum distance maintained by the target ship when it passes through the bow or stern of ship i, excluding the safety distance; h hi ——The target ship passes through the stern of ship i, and the horizontal safe distance between the target ship and the target ship; h qi ——The target ship passes through the bow of ship i, and the horizontal safe distance between the target ship and the ship; Z j ——The maximum distance that the target ship can maintain when passing through the bow or stern of ship j, excluding the safety distance; h hj ——The target ship passes through the stern of ship j and the horizontal safe distance between the target ship and the ship; h qj ——The target ship passes through the bow of ship j, and the horizontal safe distance between the target ship and the target ship; C. Establish a weighted topological map of the ship route network The inland waterway network is divided into an inland waterway network formed by the main channel plus the grid-type branch channel and an inland waterway network formed by the main channel plus the branch channel. The nodes at the intersection of the main channel of the branch-type network and the intersection of the routes in the grid-type network are interrupted to establish a weighted topological diagram of the ship route network. D. Establishing a generalized road resistance function When different routes intersect, there is a potential traffic conflict between ships, which will cause delays. This is also one of the factors that cause delays in ships passing through the induction area. Therefore, the ship travel time is used as the road resistance function, that is, T = t + d s +θd z , where d s The average conflict delay time of ships on the main routes; d z is the average delay time of delayed ships when different routes intersect; when calculating the delay in the intersection area of ​​different routes, judge the size between the ship flow on the left route and the merging or crossing capacity; when there is an intersection area of ​​different routes and the arrival rate of ships on the left route is greater than the merging or crossing capacity, the traffic on the left route will be blocked; θ is the discrimination coefficient, θ for the left route is 1, and θ for the right route is 0; The ship travel time includes the sailing time prediction result and the ship delay time, which is used as one of the decision variables of the objective function in the dynamic route induction process; E. Establish an induced path optimization model and solve the best route Based on the research on the induced path of the target ship and the crossing clearance of the target ship, the induced path optimization model is obtained by combining them as follows: The constraints are: v0＞0 (7) Y (i,j) ＞0 (8) t a +(m-1)t g ≤t ij ≤t c +(m-1)t g (9) t h >t z 1,2 (10) Where: L ij Indicates the construction operation segment with starting point i and end point j; x ij Indicates whether the ship has passed through segment L ij , 1 means passing, 0 means not passing; T ij Indicates that the construction operation segment L is passed ij Total sailing time of the vessel; t ij Indicates that the construction operation segment L is passed ij The interval time between the two ships; M represents the set of route nodes, M = {A, 1, 2, ..., B}; Y (i,j) Indicates construction operation segment L ij Mileage; t a It indicates the time required for the target ship to sail from the departure position to the intersection; t c It represents the minimum time required for the target ship to complete a crossing; t g It indicates the time distance between two consecutive target ships when multiple target ships are crossing. m represents the number of target ships in the queue ready to be inserted; t h Indicates the clearance between two ships during normal navigation; t z 1 Indicates the critical gap for ships to merge into; t z 2 Indicates the critical gap through which the ship passes; t s It indicates the time between two ships merging or passing in the delayed direction; Constraint (6) indicates that if a ship passes through the construction section L ij If yes, it is 1, otherwise it is 0; Constraint (7) indicates that the ship's speed is always a positive real number at any time, that is, the ship is always in a sailing state; Constraint (8) indicates that the mileage of a route segment is a positive real number; Constraint (9) ensures that ships on the main route do not cause conflict delays caused by target ship interspersed; Constraint (10) ensures that ships on the main route do not experience conflicts and delays due to intersections between different routes; F. Output real-time dynamic ship guidance route to guide ship navigation Output the real-time dynamic optimal guidance results of the ship's route to provide decision-making information for ship drivers.

2. According to claim 1, a real-time dynamic route guidance method for ships in an inland waterway network is characterized by: The method for repairing the data of the original ship automatic identification system (AIS system) described in step B comprises the following steps: B11. Outlier Removal Use binning to divide the data into a series of "bins" of equal width. If the value of a data point on a certain attribute is outside the three standard deviations of the statistical mean value of all data points in the "bin" on this attribute, this point is considered an outlier and needs to be removed. B12. Time alignment and dynamic data completion For the outliers removed in step B11 and the missing values ​​existing in the AIS data itself, the cubic spline interpolation method is used to repair them to obtain track data with equal time intervals; B13. Static data completion In the event of long-term loss of basic information on ship type and size, the largest ship information in the waterway section is extracted to replace it.

3. According to claim 1, a real-time dynamic route guidance method for ships in an inland waterway network is characterized by: The time interval K in step B is 3-7 min.

Images