A port load prediction method and algorithm

Abstract

The present application relates to a kind of port load prediction method and algorithm, including algorithm part and actual prediction method part, algorithm part includes selection and division time interval, data acquisition, data processing, the training and test of convolutional neural network, finally obtains qualified convolutional neural network, thus is applied to actual: through the present application can reach the beneficial effect of accurately predicting port load.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of waterway traffic management, in particular to a port load prediction method and algorithm. BACKGROUND

[0002] With the continuous advancement of port electrification and the increasing demand of the shipping industry for carbon emissions, the use of clean energy such as solar, wind, and hydrogen energy for power generation to supply power to the port has received more and more attention. However, there is a great uncertainty in the energy consumption of the port, which brings certain challenges to the management of the energy-consuming side. In order to better coordinate the port energy consumption and supply, it is necessary to predict the energy consumption of the port. SUMMARY

[0003] The present application aims to overcome the problem of energy consumption evaluation and provides a port load prediction method and algorithm.

[0004] The present application provides an algorithm for obtaining a qualified port load prediction neural network, and the technical solution is as follows:

[0005] An algorithm for obtaining a qualified port load prediction neural network, characterized by comprising the following steps:

[0006] Step 1, selecting an initial time t o and an ending time t e , and dividing the time interval [t o , t e ] into m equal parts to obtain the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], where t o =t0<t1<t2<...<t m =t e ;

[0007] Step 2, obtaining the climate data, day type data, season type data, work volume data, port-berthing ship data, power of each hydrogen energy device used, and power of each diesel device used in each interval of the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ];

[0008] Step 3, taking the column vector i-1 composed of the climate data, day type data, season type data, work volume data, port-berthing ship data, power of each hydrogen energy device used, and power of each diesel device used in the interval [t i , t m ] as the i-th input vector, forming an input data set A, where 1≤i≤m,

[0009] A = {X1, X2, ..., X} m};

[0010] Step 4: Obtain the interval sequence [t0, t1], [t1, t2], ..., [t... m-1 , t m The data on electricity, hydrogen, and diesel consumption for each interval in the [t] section will then be processed by the interval [t]. i-1 , t i The column vector containing the electricity consumption, hydrogen consumption, and diesel consumption data is shown in the image. As the i-th output vector, it forms the output dataset B, where 1 ≤ i ≤ m.

[0011] B = {Y1, Y1, ..., Y1} m};

[0012] Step 5: Construct the Cartesian product (X) of the i-th input vector and the i-th output vector. i Y i Let be a set U of elements, where 1 ≤ i ≤ m, i.e., let

[0013] U=={(X1,Y1),…,(X m Y m )}

[0014] The set U is taken as the dataset, and the dataset is divided into a training set C and a test set D, wherein,

[0015] C = {X1, Y1), ..., X s Y s )}

[0016] D = {X} s+1 Y s+1 ), ..., (X m Y m )};

[0017] Step Six: After preprocessing the dataset, a convolutional neural network is selected to process each element (X) in the training set C. i Y i The first component X) i As the input vector of the convolutional neural network, each element (X) in the dataset i Y i The second component Y) i The convolutional neural network is trained using the output vector of the convolutional neural network to obtain a trained convolutional neural network.

[0018] Step seven, using test set D to test the trained convolutional neural network, and obtaining a qualified convolutional neural network.

[0019] Compared with the prior art, the technical scheme provided in the present application can bring at least the following beneficial effects: the technical scheme takes into account various complex factors affecting the port load, and the trained qualified convolutional neural network can more accurately predict the port load. Since it is difficult to establish an accurate mathematical model between the influencing factors of the port load and the corresponding port load, the convolutional neural network can learn the prediction model by collecting historical values of the port load influencing factors and the corresponding historical load data, without the need to establish an accurate mathematical model; the convolutional neural network used can well capture the characteristics between various influencing factors, improve the prediction accuracy, and the pooling process can improve the learning efficiency.

[0020] As a preferred, the climate data in step two includes visibility data, wind data, rain data and thunder data; in this way, the climate factors affecting the port load are considered in detail.

[0021] As a preferred, in step two, the visibility data of the climate data includes the following steps:

[0022] Step A1, using fuzzy logic to divide the visibility into three categories of no influence, weak influence and strong influence;

[0023] Step A2, selecting three data α 11 , α 12 , α 13 , where α 11 < α 12 < α 13 , let α 11 represent the membership of no influence of the visibility, α 12 represent the membership of weak influence of the visibility, and α 13 represent the membership of strong influence of the visibility, and according to the fuzzy logic, the visibility membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ] is determined, if a certain interval belongs to no influence, the data α 11 representing that the visibility is no influence is taken as the visibility data of the corresponding interval, if a certain interval belongs to weak influence, the data α 12 representing that the visibility is weak influence is taken as the visibility data of the corresponding interval, and if a certain interval belongs to strong influence, the data α 13 representing that the visibility is strong influence is taken as the visibility data of the corresponding interval.

[0024] Step two, obtaining the wind data from the climate data, includes the following steps:

[0025] Step B1: Using fuzzy logic, classify wind into three categories: no impact, weak impact, and strong impact.

[0026] Step B2: Select three data α 21 α 22 α 23 , where α 21 <α 22 <α 23 Let α 21 The wind's membership has no effect, let α 22 Let α represent the weak influence of the wind. 23 The strong influence representing the wind membership is determined based on fuzzy logic, and the interval sequence [t0, t1], [t1, t2], ..., [t...] is judged. m-1 , t m The wind attribute of each interval in the data is defined as follows: if an interval belongs to the "no influence" category, then the data α representing "no influence" for wind is assigned. 21 For wind data within a corresponding interval, if a certain interval belongs to a weak influence, then the data α representing wind as a weak influence will be used. 22 For wind data within a corresponding interval, if a certain interval belongs to a strong influence, then the data α representing wind as a strong influence will be used. 23 As wind data for the corresponding interval;

[0027] Step two, obtaining the rainfall data from the climate data, includes the following steps:

[0028] Step C1: Using fuzzy logic, divide the rain into three categories: no impact, weak impact, and strong impact.

[0029] Step C2: Select three data α 31 α 32 α 33 , where α 31 <α 32 <α 33 Let α 31 The term α represents the unaffected element belonging to the rain element. 32 The table describes the weak influence of rain, α. 33 The table shows the strong influence of the rain membership, and the interval sequences [t0, t1], [t1, t2], ..., [t] are determined based on fuzzy logic. m-1 , t m The rain attribute of each interval in the data is defined as follows: if an interval belongs to the "no influence" category, then the data representing rain as having no influence will be α. 31 For rainfall data within a corresponding interval, if a certain interval belongs to a weak influence, then the data α representing rainfall as having a weak influence will be used.32 As the rain data of the corresponding interval, if a certain interval belongs to strong influence, data a 33 As the rain data of the corresponding interval;

[0030] In the step two, the thunder data of the climate data includes the following steps:

[0031] Step D1, using fuzzy logic, divide the thunder into three categories of no influence, weak influence and strong influence;

[0032] Step D2, select three data a 41 , a 42 , a 43 , where a 41 < a 42 < a 43 , let a 41 represent the thunder belonging to no influence, a 42 represent the thunder belonging to weak influence, and a 43 represent the thunder belonging to strong influence, and according to fuzzy logic to judge the rain belonging of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to no influence, data a 41 will be represented as the thunder data of the corresponding interval, if a certain interval belongs to weak influence, data a 42 will be represented as the thunder data of the corresponding interval, if a certain interval belongs to strong influence, data a 43 will be represented as the thunder data of the corresponding interval;

[0033] Fuzzy logic can quantify and materialize the qualitative fuzzy influence factors, which is conducive to the construction of neural network data input.

[0034] As preferred, in the step two, the acquisition of the day type data includes the following steps:

[0035] Step E1, divide the day into holiday, double holiday and working day, select three data a 51 , a 52 , a 53 , where a 51 < a 52 < a 53 , let a 51 represent the day belonging to holiday, a 52 represent the day belonging to double holiday, and a 53 represent the day belonging to working day;

[0036] Step E2, judging the day affiliation of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval is affiliated to a holiday, then data α 51 representing that the day is a holiday is taken as the day data of the corresponding interval, if a certain interval is affiliated to a double holiday, then data α 52 representing that the day is a double holiday is taken as the day data of the corresponding interval, if a certain interval is affiliated to a working day, then data α 53 representing that the day is a working day is taken as the day data of the corresponding interval;

[0037] In this way, the day type is quantified, and the differences between holidays, working days and double holidays are fully considered.

[0038] As a preferred embodiment, in the step two, the acquisition of the season type data comprises the following steps:

[0039] Step F1, dividing the seasons into four seasons, spring, summer, autumn and winter, and selecting four data β1, β2, β3, β4, wherein β1<β2<β3<β4, β1 represents that the season is spring, β2 represents that the season is summer, β3 represents that the season is autumn, and β4 represents that the season is winter;

[0040] Step F2, judging the season affiliation of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval is affiliated to spring, then data β1 representing that the season is spring is taken as the season data of the corresponding interval, if a certain interval is affiliated to summer, then data β2 representing that the season is summer is taken as the season data of the corresponding interval, if a certain interval is affiliated to autumn, then data β3 representing that the season is autumn is taken as the season data of the corresponding interval, and if a certain interval is affiliated to winter, then data β4 representing that the season is winter is taken as the season data of the corresponding interval;

[0041] In this way, the quantification of the season is realized while the influence of the season factor is considered.

[0042] As a preferred embodiment, the work load data acquired in the step two comprises the total work load of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], wherein the influence of the work load is considered.

[0043] As a preferred embodiment, in the step two, the data of the berthing ship comprises type data, shore power connection state data, maximum sustainable power of the auxiliary engine of the berthing ship and berthing time data;

[0044] The step two includes the following steps of acquiring the data of the berthing ship:

[0045] In step G1, the type of the berthing ship is divided into bulk carrier, container ship, oil tanker and other, wherein, the data a represents the type of the berthing ship is bulk carrier, the data b represents the type of the berthing ship is container ship, the data c represents the type of the berthing ship is oil tanker, and the data d represents the type of the berthing ship is other, at the same time, if there is no berthing ship shore power connection in a certain interval, 0 is taken as the shore power connection state data of the corresponding interval, if there is berthing ship shore power connection in a certain interval, 1 is taken as the shore power connection state data of the corresponding interval;

[0046] In step G2, the shore power connection state data of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ] and the type data of the berthing ship, the maximum sustainable power of the auxiliary machinery of the berthing ship and the berthing time data are acquired.

[0047] Such an evaluation mode provides input data and prepares for energy saving at the same time.

[0048] Preferably, the ratio of the data amount of the training set C to the data amount of the test set is 7:3, which reflects the scientific nature of the test and the training.

[0049] Preferably, the convolutional neural network includes an input layer, a convolutional layer C1, a pooling layer S1, a convolutional layer C2, a pooling layer S2, a full connection layer and an output layer, and the input layer, the convolutional layer C1, the pooling layer S1, the convolutional layer C2, the pooling layer S2, the full connection layer and the output layer are sequentially arranged according to the running order; the convolutional neural network can learn a prediction model by collecting historical values of port load influencing factors and corresponding historical load data, thereby without the need to establish an accurate mathematical model.

[0050] The port load prediction method provided by the application has the technical scheme as follows.

[0051] The port load prediction method adopts a port load qualified prediction neural network algorithm based on the port load qualified prediction neural network obtained by the above method, and includes the following steps:

[0052] In step 10.1, a time period is selected as a prediction time period.

[0053] Step 10.2, obtaining the climate data, day type data, season type data, operation amount data, ported ship data, power of each hydrogen energy equipment used and power of each diesel equipment used of the predicted time period as prediction input data;

[0054] Step 10.3, inputting the prediction input data into the obtained port load qualified prediction neural network to obtain output power, hydrogen consumption and diesel consumption data.

[0055] Compared with the prior art, the technical scheme provided by the present application can at least bring the following beneficial effects: by using the trained qualified port load qualified prediction neural network, only the relevant data collected needs to be input, and the power consumption, hydrogen consumption and diesel consumption data can be predicted, so that the problem of accurate port load prediction is solved by a simple method. BRIEF DESCRIPTION OF DRAWINGS

[0056] Figure 1 is a program structure diagram in the embodiment of the present application;

[0057] Figure 2 is a convolutional neural network structure diagram of the present application. DETAILED DESCRIPTION

[0058] The technical scheme in the embodiment of the present application will be described clearly and completely below in combination with the drawings in the embodiment of the present application.

[0059] In combination with Figures 1-2 , the present embodiment first introduces a technical scheme of an algorithm of an obtained port load qualified prediction neural network in detail.

[0060] The algorithm of the obtained port load qualified prediction neural network provided by the embodiment of the present application has the following characteristics:

[0061] Step one, selecting an initial time t o and an ending time t e , and equally dividing the time interval [t o , t e ] into m parts to obtain interval sequences [t0, t1], [t1, t2],..., [t m-1 , t m ], wherein t o =t0<t1<t2<...<t m =t e ;

[0062] Step two, obtaining the interval sequences [t0, t1], [t1, t2],..., [t m-1 , tm the climate data, the day type data, the season type data, the operation amount data, the port- calling ship data, the power of each hydrogen energy device used, and the power of each diesel device used in each interval.

[0063] In particular, the climate data in this embodiment includes visibility data, wind data, rain data, and thunder data.

[0064] The visibility data of the climate data is obtained by the following steps:

[0065] Step A1, using fuzzy logic, the visibility is divided into three categories of no influence, weak influence and strong influence;

[0066] Step A2, three data, α 11 , α 12 , and α 13 are selected, where α 11 < α 12 < α 13 , α 11 represents the membership of no influence of the visibility, α 12 represents the membership of weak influence of the visibility, and α 13 represents the membership of strong influence of the visibility, and the membership of the visibility in each interval of the interval sequence [t0, t1], [t1, t2],..., [t m-1 t m ] is determined according to the fuzzy logic, if a certain interval belongs to no influence, the data α 11 representing that the visibility is no influence is taken as the visibility data of the corresponding interval, if a certain interval belongs to weak influence, the data α 12 representing that the visibility is weak influence is taken as the visibility data of the corresponding interval, and if a certain interval belongs to strong influence, the data α 13 representing that the visibility is strong influence is taken as the visibility data of the corresponding interval.

[0067] The wind data of the climate data is obtained by the following steps:

[0068] Step B1, using fuzzy logic, the wind is divided into three categories of no influence, weak influence and strong influence;

[0069] Step B2, three data, α 21 , α 22 , and α 23 are selected, where α 21 < α 22 < α 23 , α 21 represents the membership of no influence of the wind, α 22 represents the membership of weak influence of the wind, and α 23representing the wind belonging to the strong influence, and according to the fuzzy logic, judging the wind belonging to each interval of the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to the no influence, taking the data α 21 representing the wind being the no influence as the wind data of the corresponding interval, if a certain interval belongs to the weak influence, taking the data α 22 representing the wind being the weak influence as the wind data of the corresponding interval, if a certain interval belongs to the strong influence, taking the data α 23 representing the wind being the strong influence as the wind data of the corresponding interval.

[0070] The wind data of the climate data includes the following steps:

[0071] Step C1, using the fuzzy logic, dividing the wind into three categories of no influence, weak influence and strong influence;

[0072] Step C2, selecting three data α 31 , α 32 , α 33 , wherein α 31 < α 32 < α 33 , letting α 31 represent the wind belonging to the no influence, α 32 represent the wind belonging to the weak influence, and α 33 represent the wind belonging to the strong influence, and according to the fuzzy logic, judging the wind belonging to each interval of the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to the no influence, taking the data α 31 representing the wind being the no influence as the wind data of the corresponding interval, if a certain interval belongs to the weak influence, taking the data α 32 representing the wind being the weak influence as the wind data of the corresponding interval, if a certain interval belongs to the strong influence, taking the data α 33 representing the wind being the strong influence as the wind data of the corresponding interval.

[0073] The wind data of the climate data includes the following steps:

[0074] Step D1, using the fuzzy logic, dividing the wind into three categories of no influence, weak influence and strong influence;

[0075] Step D2, selecting three data α 41 , α 42 , α 43 , wherein α 41 < α 42 < α 43 , letting α41 representing no influence of thunder, a 42 representing weak influence of thunder, a 43 representing strong influence of thunder, and according to the rain membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to no influence, the data a 41 representing no influence of thunder is taken as the thunder data of the corresponding interval, if a certain interval belongs to weak influence, the data a 42 representing weak influence of thunder is taken as the thunder data of the corresponding interval, if a certain interval belongs to strong influence, the data a 43 representing strong influence of thunder is taken as the thunder data of the corresponding interval.

[0076] For example, the data corresponding to the climate information can be described by the following table.

[0077] No impact Weak impact Strong impact Visibility 11 12 13 Wind 21 22 23 Rain 31 32 33 Thunder 41 42 43

[0078] In the above table, the data representing no influence of visibility is 11, the data representing weak influence of visibility is 12, the data representing strong influence of visibility is 13, and so on.

[0079] In step two, obtaining the day type data comprises the following steps:

[0080] Step E1, dividing the day into a holiday, a double holiday and a working day, and selecting three data a 51 , a 52 , a 53 , wherein a 51 < a 52 < a 53 , a 51 represents the holiday membership of the day, a 52 represents the double holiday membership of the day, and a 53 represents the working day membership of the day.

[0081] Step E2, judging the day membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to a holiday, the data a 51 representing the holiday of the day is taken as the day data of the corresponding interval, if a certain interval belongs to a double holiday, the data a 52 representing the double holiday of the day is taken as the day data of the corresponding interval, if a certain interval belongs to a working day, the data a 53 representing the working day of the day is taken as the day data of the corresponding interval.

[0082] For example, the following table data can be used to represent the membership of different days.

[0083] Day type Weekday Weekend Holiday Data volume 3 2 1

[0084] In step two, the seasonal type data is obtained by the following steps:

[0085] In step F1, the season is divided into four seasons: spring, summer, autumn and winter. Select β1, β2, β3 and β4, where β1<β2<β3<β4, β1 represents the season is spring, β2 represents the season is summer, β3 represents the season is autumn, and β4 represents the season is winter.

[0086] In step F2, the seasonal membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ] is determined. If a certain interval belongs to spring, the data β1 representing the season is spring is taken as the seasonal data of the corresponding interval. If a certain interval belongs to summer, the data β2 representing the season is summer is taken as the seasonal data of the corresponding interval. If a certain interval belongs to autumn, the data β3 representing the season is autumn is taken as the seasonal data of the corresponding interval. If a certain interval belongs to winter, the data β4 representing the season is winter is taken as the seasonal data of the corresponding interval.

[0087] For example, spring, summer, autumn and winter can be represented by 4, 5, 6 and 7 respectively.

[0088] In this embodiment, the obtained work quantity data includes the total work quantity of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ].

[0089] For bulk cargo terminals, the work quantity is the mass of the cargo. For container terminals, the work quantity is the standard box quantity.

[0090] In step two, the data of the port-berthing ship includes type data, shore power connection state data, maximum sustainable power of the port-berthing ship auxiliary machine and port-berthing time data.

[0091] In step two, the data of the port-berthing ship is obtained by the following steps:

[0092] Step G1, divide the type of the port-berthing ship into bulk carrier, container ship, oil tanker and other, use data a to represent the type of the port-berthing ship is bulk carrier, data b to represent the type of the port-berthing ship is container ship, data c to represent the type of the port-berthing ship is oil tanker, data d to represent the type of the port-berthing ship is other, at the same time, if there is no port-berthing ship shore power connection in a certain interval, 0 is taken as the shore power connection state data of the corresponding interval, if there is port-berthing ship shore power connection in a certain interval, 1 is taken as the shore power connection state data of the corresponding interval;

[0093] Step G2, obtain the port-berthing ship shore power connection state data and the type data of the port-berthing ship, the maximum sustainable power of the auxiliary machinery of the port-berthing ship, and the port-berthing time data of each interval in the interval sequence [t0, t1], [t1, t2], …, [t m-1 , t m ].

[0094] The values of a, b, c and d can be equal to 1, 2, 3 and 4, and the power of each hydrogen energy device used and the power of each diesel device used can be obtained from the database.

[0095] Step three, the column vector i-1 , t i ] composed of the climate data, the type of day data, the type of season data, the operation amount data, the port-berthing ship data, the power of each hydrogen energy device used and the power of each diesel device used in the interval [t is taken as the i-th input vector, forming the input data set A, where 1≤i≤m,

[0096] A={X1, X2, …, X m}.

[0097] Step four, obtain the power consumption, hydrogen consumption and diesel consumption data of each interval in the interval sequence [t0, t1], [t1, t2], …, [t m-1 , t m ], and then the column vector i-1 , t i ] composed of the power consumption, hydrogen consumption and diesel consumption data in the interval [t is taken as the i-th output vector, forming the output data set B, where 1≤i≤m,

[0098] B={Y1, Y1, …, Y m}

[0099] Step five, establish a set U with the Cartesian product (X i , Y i ) of the i-th input vector and the i-th output vector as an element, where 1≤i≤m, that is,

[0100] U == {(X1, Y1),..., (X m , Y m )}

[0101] , and dividing the data set into a training set C and a test set D, wherein

[0102] C == {(X1, Y1),..., (X s , Y s )}

[0103] D == {(X s+1 , Y s+1 ),..., (X m , Y m )}.

[0104] Step six, after preprocessing the data set, a convolutional neural network is selected, the first component X i of each element (X i , Y i ) in the training set C is taken as the input vector of the convolutional neural network, the second component Y i of each element (X i , Y i ) in the data set is taken as the ideal output vector of the convolutional neural network, the convolutional neural network is trained, and a trained convolutional neural network is obtained.

[0105] Step seven, the test set D is used to test the trained convolutional neural network, and a qualified convolutional neural network is obtained.

[0106] The ratio of the data quantity of the training set C to the data quantity of the test set in the embodiment is 7:3. The convolutional neural network includes an input layer, a convolutional layer C1, a pooling layer S1, a convolutional layer C2, a pooling layer S2, a fully connected layer with three hidden layers, and an output layer, and the input layer, the convolutional layer C1, the pooling layer S1, the convolutional layer C2, the pooling layer S2, the fully connected layer, and the output layer are sequentially arranged along the running direction of the convolutional neural network.

[0107] In the embodiment, the convolutional kernels of the convolutional layer C1 and the convolutional layer C2 are both 3x1 matrices, the operation step length is 1, the number is 8, the filtering region of the pooling layer S1 and the pooling layer S2 is a 2x1 filtering unit, the step length is 2, and the maximum pooling is used for pooling operation. The activation function of the fully connected layer adopts a logarithmic loss function, a logarithmic likelihood cost objective function, and a normalized exponential function.

[0108] The embodiment also provides a port load prediction method, which uses a port load qualified prediction neural network obtained based on a port load qualified prediction neural network algorithm, and includes the following steps:

[0109] Step 10.1, selecting a time period as a prediction time period;

[0110] Step 10.2, obtaining the climate data, day type data, season type data, operation amount data, ported ship data, power of each hydrogen energy equipment used and power of each diesel equipment used in the prediction time period as prediction input data;

[0111] Step 10.3, inputting the prediction input data into the obtained port load qualification prediction neural network to obtain output electric quantity, hydrogen consumption and diesel consumption data.

[0112] Of course, before entering the neural network training, a preprocessing can be performed, which can be performed by a normalization manner, and at the same time, some invalid data or unreliable data need to be removed. In summary, although the embodiments of the present application have been shown and described, it can be understood by those skilled in the art that various changes, modifications, replacements and modifications can be made to the embodiments without departing from the principles and spirits of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A method of obtaining a qualified port load forecasting neural network, characterized in that: Comprising the following steps: Step one, select initial time t o and end time t e , and divide the time interval [t o , t e ] into m equal parts to obtain the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], wherein t o = t0 < t1 < t2 <... < t m = t e ; Step 2: Obtain the interval sequence [t0, t1], [t1, t2], ..., [t... m-1 , t m The data includes climate data, daily data, seasonal data, operational data, berthing vessel data, power of each hydrogen energy unit, and power of each diesel energy unit used in each interval. Step 3: The interval [t] i-1 , t i The column vector within contains the climate data, daily data, seasonal data, operational data, berthing vessel data, power output of each hydrogen generator, and power output of each diesel generator. The i-th input vector forms the input dataset A, where 1 ≤ i ≤ m. A = {X1, X2,..., X m}; Step four, obtaining the electricity consumption, hydrogen consumption and diesel consumption data of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , m t i-1 , i ] and then forming a column vector as the i-th output vector, forming an output data set B, where 1≤i≤m, B = {Y1, Y1,..., Y m}; Step five, establish a set U with the Cartesian product (X i , Y i ) of the ith input vector and the ith output vector as an element, where 1≤i≤m, that is, let U = {(X1, Y1),..., (X m , Y m )} , the set U is taken as a data set, and the data set is divided into a training set C and a test set D, wherein, C = {(X1, Y1),..., (X s , Y s )} D = {(X s+1 , Y s+1 ),..., (X m , Y m )}; Step six, after preprocessing the data set, selecting a convolutional neural network, taking the first component X of each element (X i ,Y i ) in the training set C as the input vector of the convolutional neural network, taking the second component Y of each element (X i , Y i ) in the data set as the ideal output vector of the convolutional neural network, training the convolutional neural network, and obtaining a trained convolutional neural network; i , i Step seven, using the test set D to test the trained convolutional neural network, and obtaining a qualified convolutional neural network; The climate data in step two includes visibility data, wind data, rain data and thunder data, and in step two, the visibility, wind, rain and thunder are divided into influence forms according to fuzzy logic respectively, and each influence form is valued to obtain the visibility data, wind data, rain data and thunder data.

2. The method of obtaining a qualified port load forecasting neural network according to claim 1, characterized in that: In the step two, the visibility data of the climate data includes the following steps: Step A1, using fuzzy logic, the visibility is divided into three categories: no influence, weak influence and strong influence; Step A2, select three data α 11 , α 12 , α 13 , where α 11 < α 12 < α 13 , α 11 represents the visibility membership of no influence, α 12 represents the visibility membership of weak influence, and α 13 represents the visibility membership of strong influence, and according to fuzzy logic, judge the visibility membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval is a membership of no influence, then the data α 11 representing the visibility of no influence is taken as the visibility data of the corresponding interval, if a certain interval is a membership of weak influence, then the data α 12 representing the visibility of weak influence is taken as the visibility data of the corresponding interval, and if a certain interval is a membership of strong influence, then the data α 13 representing the visibility of strong influence is taken as the visibility data of the corresponding interval; In the step two, the wind data of the climate data includes the following steps: Step B1, using fuzzy logic, the wind is divided into three categories: no influence, weak influence and strong influence; Step B2, select three data α 21 , α 22 , α 23 , where α 21 < α 22 < α 23 , α 21 represents the wind membership of no influence, α 22 represents the wind membership of weak influence, and α 23 represents the wind membership of strong influence, and according to fuzzy logic, judge the wind membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to no influence, then the data α 21 representing the wind is no influence is taken as the wind data of the corresponding interval, if a certain interval belongs to weak influence, then the data α 22 representing the wind is weak influence is taken as the wind data of the corresponding interval, if a certain interval belongs to strong influence, then the data α 23 representing the wind is strong influence is taken as the wind data of the corresponding interval; In the step two, the rain data of the climate data includes the following steps: Step C1, using fuzzy logic, the rain is divided into three categories: no influence, weak influence and strong influence; Step C2, select three data α 31 , α 32 , α 33 , where α 31 < α 32 < α 33 , α 31 represent the rain membership of no impact, α 32 represent the rain membership of weak impact, α 33 represent the rain membership of strong impact, and according to the fuzzy logic to judge the rain membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to no impact, then the data α 31 representing the rain is no impact is taken as the rain data of the corresponding interval, if a certain interval belongs to weak impact, then the data α 32 representing the rain is weak impact is taken as the rain data of the corresponding interval, if a certain interval belongs to strong impact, then the data α 33 representing the rain is strong impact is taken as the rain data of the corresponding interval; In the step two, the thunder data of the climate data includes the following steps: Step D1, using fuzzy logic, the thunder is divided into three categories: no influence, weak influence and strong influence; Step D2, select three data α 41 , α 42 , α 43 , where α 41 < α 42 < α 43 , α 41 represent the rain belongs to no influence, α 42 represent the rain belongs to weak influence, α 43 represent the rain belongs to strong influence, and according to the fuzzy logic to judge the rain belonging of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval belongs to no influence, then the data α 41 representing the rain is no influence is taken as the rain data of the corresponding interval, if a certain interval belongs to weak influence, then the data α 42 representing the rain is weak influence is taken as the rain data of the corresponding interval, if a certain interval belongs to strong influence, then the data α 43 representing the rain is strong influence is taken as the rain data of the corresponding interval.

3. The method of acquiring a qualified port load forecasting neural network according to claim 2, characterized in that; In the step two, the port type data includes the following steps: Step E1, divide the day into holiday, double holiday and workday, select three data α 51 , α 52 , α 53 , wherein α 51 < α 52 < α 53 , let α 51 represent the holiday to which the day belongs, α 52 represent the double holiday to which the day belongs, and α 53 represent the workday to which the day belongs; Step E2, judging the day membership of each interval of the interval sequence [t0, t1], [t1, t2],..., [t m-1 , t m ], if a certain interval is subordinate to a holiday, then data α 51 representing that the day is a holiday is taken as the day data of the corresponding interval, if a certain interval is subordinate to a double holiday, then data α 52 representing that the day is a double holiday is taken as the day data of the corresponding interval, if a certain interval is subordinate to a weekday, then data α 53 representing that the day is a weekday is taken as the day data of the corresponding interval.

4. The method of acquiring a qualified port load forecasting neural network according to claim 3, characterized in that: In the step two, the season type data includes the following steps: Step F1, divide the season into four seasons of spring, summer, autumn and winter, select four data β1, β2, β3, β4, wherein β1<β 2< β 3< β4, β1 represents that the season is spring, β2 represents that the season is summer, β3 represents that the season is autumn, and β4 represents that the season is winter; Step F2, select the data β1, β2, β3, β4, and the data β1, β2, β3, β4 are respectively represented as follows: Step F2, judging the seasonal membership of each interval in the interval sequence [t0, t1], [t1, t2],..., [t m-1 , m ] If a certain interval belongs to spring, then the data β1 representing the season is spring as the seasonal data of the corresponding interval, if a certain interval belongs to summer, then the data β2 representing the season is summer as the seasonal data of the corresponding interval, if a certain interval belongs to autumn, then the data β3 representing the season is autumn as the seasonal data of the corresponding interval, if a certain interval belongs to winter, then the data β4 representing the season is winter as the seasonal data of the corresponding interval.

5. The method of acquiring a qualified port load forecasting neural network according to claim 4, characterized in that: The job amount data acquired in the step two includes the total amount of jobs in each of the intervals [t0, t1], [t1, t2],..., [t m-1 , m ] acquired in the step one.

6. The method of acquiring a qualified port load forecasting neural network according to claim 5, characterized in that: In the step two, the port ship data includes type data, shore power connection state data, port ship auxiliary engine maximum sustainable power and port ship time data; In the step two, the port ship data includes the following steps: Step G1, the type of the port ship is divided into bulk carrier, container ship, oil tanker and other, data a represents the type of the port ship is bulk carrier, data b represents the type of the port ship is container ship, data c represents the type of the port ship is oil tanker, and data d represents the type of the port ship is other, at the same time, if there is no port ship shore power connection in a certain interval, 0 is taken as the shore power connection state data of the corresponding interval, if there is port ship shore power connection in a certain interval, 1 is taken as the shore power connection state data of the corresponding interval; Step G2, obtaining the ported ship shore power connection state data and the type data of the ported ship, the maximum sustainable power of the ported ship auxiliary machinery, and the ported time data in each interval of the interval sequence [t0, t1], [t1, t2],..., [t m-1 , m ] 7. The method of acquiring a qualified port load forecasting neural network according to claim 6, characterized in that: The ratio of the data amount of the training set C to the data amount of the test set is 7:

3.

8. The method of acquiring a qualified port load forecasting neural network according to claim 7, characterized in that: The convolutional neural network includes an input layer, a convolutional layer C1, a pooling layer S1, a convolutional layer C2, a pooling layer S2, a full connection layer and an output layer, and the input layer, the convolutional layer C1, the pooling layer S1, the convolutional layer C2, the pooling layer S2, the full connection layer and the output layer are connected in turn according to the running order.

9. A port load forecasting method characterized by: The qualified port load prediction neural network obtained by the method for obtaining a qualified port load prediction neural network according to any one of claims 1-8 comprises the following steps: Step 10.1, select a time period as a prediction time period; Step 10.2, obtaining the climate data, day type data, season type data, operation amount data, port-visited ship data, power of each hydrogen energy equipment used, and power of each diesel equipment used of the predicted time period, and taking these data as prediction input data; Step 10.3, inputting the prediction input data into the qualified port load prediction neural network, and obtaining output power data, hydrogen consumption data, and diesel consumption data.

Images