CN113360544A

CN113360544A - Short-time catching Knoop force distribution prediction method based on catching time sequence relation

Info

Publication number: CN113360544A
Application number: CN202110627654.7A
Authority: CN
Inventors: 洪锋; 赵中宁; 滕冠龙; 纪晨阳; 田景瑞; 黄海广; 刘超; 冯源; 郭忠文
Original assignee: Ocean University of China
Current assignee: Ocean University of China
Priority date: 2021-06-04
Filing date: 2021-06-04
Publication date: 2021-09-07
Anticipated expiration: 2041-06-04
Also published as: CN113360544B

Abstract

The short-time fishing Nu force distribution prediction method based on the fishing time sequence relation comprises the following steps of 1) preprocessing a track data set of a marine fishery ship; 2) data set feature extraction, and 3) catching time sequence relation definition and quantification; 4) solving a characteristic ship set, namely solving the characteristic ship set for prediction by two greedy principles; 5) constructing a model; 6) and predicting the distribution of the short-time fishing effort force. The method is researched based on the existing ocean trajectory big data, has wide applicability, and does not need other additional data such as weather, ocean current and the like for auxiliary prediction. Meanwhile, the steps designed by the method are easy to understand, the calculation is fast, the cost is low, the designed prediction model based on the convolutional neural network is easy to train, fine quantitative support can be provided for marine fishery resource management and planning in China, and a referential space-time prediction model is provided for marine information research.

Description

Short-time catching Knoop force distribution prediction method based on catching time sequence relation

Technical Field

The invention belongs to the technical field of data mining of track big data, and particularly relates to a short-time fishing effort distribution prediction method based on a fishing time sequence relation.

Background

With the improvement of the fishing capability of modern oceans, the non-sustainable development problems such as year-by-year decline of offshore fishery resources appear. Therefore, fishery resource management departments need to timely acquire and accurately predict the fishing effort, so that effective measures for ecological environment restoration and resource maintenance are formulated, and marine fishery resources and marine ecosystems are effectively recovered.

The ship position monitoring system deployed on the marine fishing ship can regularly acquire various operation parameters of the fishing ship, including the spatial position, the navigational speed, the course and the like of the fishing ship at the sampling moment, and can reconstruct the track of the fishing ship, thereby quantifying the time-space distribution of the fishing operation in the sea area.

When the statistical time period is longer, such as quarterly or year, the distribution of the fishing effort force in a certain sea area can show stability. However, many relevant factors affecting the distribution of fishing efforts, such as weather, ocean currents, and nutrients, have a large tendency to change in a short period of time and are difficult to fully quantify. Due to the fact that the fishing times of most offshore fishery ships are short, the fishing effort force distribution in the short period in the future is known, and the fishery safety and fishery resource management and planning are guaranteed. Therefore, the distribution prediction of the marine fishing effort with short time and high precision is a key problem to be researched urgently in the analysis and application of the current marine track big data.

Since the variation of relevant factors influencing the distribution of the fishing effort amount in a short time period is unstable and difficult to quantify, short-time period prediction cannot be completed depending on external factors, and the stable factors in the distribution of the fishing effort amount need to be mined.

Disclosure of Invention

The invention aims to provide a short-time fishing effort force distribution prediction method based on a fishing time sequence relation, so as to overcome the defects of the prior art.

The short-time fishing Nu force distribution prediction method based on the fishing time sequence relation is characterized by comprising the following steps of:

1) marine fishery ship trajectory data set preprocessing

In the marine fishery ship track data set, the total number of fishing boats is n, and each track data comprises the following fields: ship number id, sampling date d, longitude lon, latitude lat, navigational speed and course; preprocessing comprises abnormal data filtering and interpolation completion;

2) data set feature extraction

The fishing time sequence relation refers to that one fishing boat serving as a characteristic fishing boat and other fishing boats appear in the same unit space in unit time and the fishing behaviors have a sequence;

the quantification and mining of the fishing time sequence relation need to determine the size of unit time and unit space, so the method firstly extracts the relevant characteristics of single fishing of the fishing boat from the track data set:

a) taking the period between two sections of harbor time periods of each fishing vessel as a voyage number, counting the continuous days distribution of the voyage numbers of all the fishing vessels, selecting a percentage N%, correspondingly, completing the voyage numbers of N% within D days, and taking D as the longest voyage number of the method;

b) the fishing boat can sail in a fixed speed interval when fishing, the track record r in the fishing state is screened out by utilizing the speed interval, and the coverage area of single fishing is determined according to the track record r, and the specific method comprises the following steps: counting the size distribution of the geographic area covered by all single catches, selecting the percentages W% and J% of latitude and longitude, and correspondingly, selecting the single catch of W%The latitude range of fishing is all in l_latThe longitude ranges of the inner percent and the J percent of single fishing are all in l_latIn the interior, will (l)_lat,l_lon) As a unit space size;

3) definition and quantification of catching time sequence relation

The invention quantifies the catching time sequence relation from two aspects of space and time:

a) at the spatial level, the sea area to be predicted is represented by (l)_lat,l_lon) Dividing the space grid into X multiplied by Y space grids by taking the space grid as a unit; constructing a catching space-time matrix F through all the track data r in the catching state; the unit (x, y, d) in the fishing space-time matrix F represents the set of fishing boats fishing in the space grid (x, y) within d days, and the calculation method is as the formula (1):

F＝{(x,y,d)}＝{r(id)|(r(lon),r(lat))∈(x,y)∧r(date)＝d} (1)

when the longitude lon, the latitude lat and the date d of the fishing boat with the boat number id in the r satisfy the formula (1), the fishing boat with the boat number id is taken as an element in the set (x, y, d);

b) on the time level, when the characteristic fishing boat appears in a space grid (x, y), if other fishing boats also appear in the space grid after D/2 day and within D day, the other fishing boats and the characteristic fishing boat form a fishing time sequence relation; the fishing time sequence relation meets the conditions as formula (2):

wherein v is_eIs characterized by a fishing boat, v_i∈RS_x,y,d(v_e) Showing other fishing vessels v_iCharacteristic fishing boat v_eFishing activities are carried out on the same spatial grid (x, y) within D days after D/2 days later;

c) after the catching time sequence relation is defined by the formula, the invention further uses a time sequence relation matrix S for X multiplied by Y spatial grids respectively_x,yExpressing the fishing time sequence relation in each space grid (x, y); using X Y adjoint relation matrixes C simultaneously_x,yThe method is optimized by the following steps:

s1: calculating a time-series relationship matrix S_x,y：

S11: initialization S_x,yIs an n multiplied by n zero matrix, wherein n is the number of fishing boats, and each fishing boat corresponds to a serial number;

s12: for each characteristic fishing vessel v in the spatial grid (x, y, d)_eCalculating a set RS_x,y,d(v_e) Each element v in the set_jAre respectively connected with v_eForming a coordinate according to the corresponding serial number of the fishing boat, and recording the coordinate as (e, j), wherein e and j are v respectively_e、v_jThe fishing boat serial number of (1);

s13: will S_x,yThe position corresponding to (e, j) is incremented by 1;

s14: traversing each date d of the whole fishing period, and accumulating the obtained coordinates formed according to the serial number of the fishing boat to S_x,yObtaining the times of forming a fishing time sequence relation among all fishing boats in the space grid (x, y) according to the corresponding positions;

s2: computing an adjoint matrix C_x,y：

S21: initialization C_x,yIs an n multiplied by n zero matrix, wherein n is the number of fishing boats, and each fishing boat corresponds to a serial number;

s22: for each characteristic vessel v in the unit (x, y, d)_eThe set RC is calculated according to formula (3)_x,y,d(v_e)，

Each element v in the set_kAnd v_eForming a coordinate according to the serial number of the fishing boat, and recording the coordinate as (e, k), wherein e and k are v respectively_e、v_kThe fishing boat serial number of (1);

s23: c is to be_x,yThe position corresponding to (e, k) is incremented by 1;

s24: traversing each date d of the whole fishing period, and accumulating the obtained coordinates formed according to the serial number of the fishing boat to C_x,yCorresponding positionObtaining the times of forming the accompanying fishing relation among all fishing boats in the space grid (x, y);

s3: optimizing a timing relationship matrix S_x,y：

S31: judgment S_x,yAll values of the symmetric position elements (i, j) and (j, i) indicate that the fishing vessel v is present if the value at (i, j) is more than twice that at (j, i)_iAt and v_jThe fishing time sequence relation of (a) is dominant and helpful for prediction, and v_jAt and v_iThe fishing timing relationship of (a) is non-dominant and does not contribute to the prediction, so the value at the smaller side (j, i) is set to 0; if the values at (i, j) and (j, i) differ by less than a factor of two, it is an indication that the fishing vessel v is a fishing vessel_iAnd v_jThere is no obvious dominant fishing timing relationship between them, and there is no help to the prediction, so the values at (i, j) and (j, i) are both set to 0;

s32: record each C_x,yIf S is greater than 0_x,yIs also greater than 0, indicating a fishing vessel v_pAnd v_qThe fishing boat has the fishing time sequence relation and the accompanying fishing relation, but the fishing boat with the characteristics cannot predict the fishing behavior of the fishing boat with the accompanying fishing, i.e. the two relations coexist and do not help the prediction of the invention, so S is used_x,yThe middle corresponding position (p, q) is set to 0;

the timing relationship matrix S processed through the above steps S1, S2, S3_x,yThe meaning of any non-zero element (m, n) in (a) is: fishery vessel v_mAnd v_nHas stable catching time sequence relation;

4) feature ship set solution

The invention solves the characteristic ship set for prediction by two greedy principles:

s1: in each space grid (x, y), solving the step 3) to finally obtain a time sequence relation matrix S_x,yRow h with maximum elements of greater than 0, which represents a fishing vessel v_hForming a fishing time sequence relation with the most fishing boats, and connecting the fishing boats v_hCharacteristic fishing vessels as spatial grid (x, y), denoted v_x,yIf the row with the most non-zero elements is provided with a plurality of rows, one row is selected; all in oneWhen the fishing boat corresponding to the column number larger than 0 in the h row is recorded, the set of the fishing boat is recorded as Follow (x, y), and the Follow v in the grid is shown_x,yForming a fishing boat in a fishing time sequence relation;

s2: fishing boat v with selected partial characteristics_x,ySet V as characteristic fishing boat_eThe principle is as follows:

s21: characteristic fishing boat v corresponding to Folow (x, y) with the most elements_x,yShip set with entry features V_eIf v is_x,yIf there are more than one, then taking any one, and simultaneously counting the elements in the Follow (x, y) into the set R;

s22: successively performing the operation of step S21 on meshes other than (x, y), and merging the resultant set with the set R in step S21, thereby expanding R;

when the number of elements in the R reaches a proportion gamma of the number n of all fishery ships, stopping the operation, wherein the gamma is at least 80%;

5) model construction

Set V for solved feature ship_eThe method takes a Convolutional Neural Network (CNN) as a core to construct a prediction model;

characteristic fishing boat set V of prediction model_eThe past 1 to D/2 days (i.e., (0, D/2)]Day) of the course, the distribution of the fishing efforts in each day is input, and the future (D/2, D) is used]The fishing efforts of all fishing boats in the day are distributed as output, and if D is an odd number, D is equal to D + 1;

according to the method, 3 x 3 is used as the size of a CNN convolution kernel, L is used as the depth of the CNN, and L represents the number of spatial grids (x, y) which can be spanned by a fishing boat furthest in D/2 days;

finally, RMSE is used as an error index, and parameters of the CNN are optimized through a back propagation BP algorithm;

6) short-term fishing effort distribution prediction

Training the model constructed in the step 5), and only providing a characteristic fishing boat set V in each prediction_eAt (0, D/2)]The distribution of the fishing effort force every day, the prediction model can output (D/2, D)]And (4) a predicted value of distribution of fishing efforts of all fishing boats in a day.

Advantages of the invention

Because the related factors influencing the distribution of the fishing efforts in a short period are more, the variation is unstable and the quantification is difficult, the short-period prediction is difficult to complete depending on external factors, and the stable factors in the distribution of the fishing efforts are required to be mined. According to the method, by mining the stable fishing time sequence relation among the fishery ships, the statistical characteristics of the data set are extracted after the data set is preprocessed, then the stable fishing time sequence relation among the fishery ships is creatively mined, and the mined characteristic ship set is utilized to realize the prediction of fishing effort force distribution in a short time period. The method can predict the distribution of the fishing effort force with high precision by taking the week as a time unit without external factor variables.

The method is researched based on the existing ocean trajectory big data, has wide applicability, and does not need other additional data such as weather, ocean current and the like for auxiliary prediction. Meanwhile, the steps designed by the method are easy to understand, the calculation is fast, the cost is low, the designed prediction model based on the convolutional neural network is easy to train, fine quantitative support can be provided for marine fishery resource management and planning in China, and a referential space-time prediction model is provided for marine information research.

Drawings

FIG. 1 is a general flow diagram of the present invention.

FIG. 2 schematic diagram of a fishing spatio-temporal matrix F

FIG. 3 timing relationship matrix S_x,ySchematic partial view of

FIG. 4 adjoint relation matrix C_x,ySchematic of a part

FIG. 5 is a timing relationship matrix S after optimization_x,ySchematic partial view of

FIG. 6 prediction error Rate of example 1

Detailed Description

The technical solution of the present invention will be clearly and completely described below with reference to the general flow chart of the present invention, using the large data of the track Monitoring system (VMS) collected by the single-towed fishing Vessel in east China sea from 2015 to 2017 as an embodiment, and it is obvious that the described embodiment is only a part of embodiments of the present invention, but not all embodiments. All other embodiments obtained by a user with ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the protection scope of the present invention.

Example 1:

the ocean trajectory big data used in this embodiment comes from a Beidou satellite-based VMS deployed on a single-tow fishing vessel in the east China sea, and the data set covers two complete fishing periods of 2015 9 to 2016 5 and 2016 9 to 2017 5. The total number of fishing boats in the data set is 1589, and each piece of track data comprises the following fields: ship number id, sampling date d, longitude lon, latitude lat, navigational speed and course; the preprocessing comprises filtering abnormal data and interpolation and completion. The present embodiment mainly includes the following 5 steps, and the step diagram is shown in fig. 1:

1) marine fishery ship trajectory data set preprocessing

The preprocessing comprises filtering abnormal data and interpolation and completion:

a) filtering the track records with the sampling date d, the longitude lon and the latitude lat of 0;

b) filtering track records with longitude lon less than 120 degrees or more than 130 degrees, latitude lat less than 25 degrees or more than 35 degrees, navigational speed over 12 sections and course heading more than 360 degrees or less than 0 degrees;

c) in this embodiment, the sampling rate of the data set is determined to be 3 minutes by counting the sampling time difference between the continuous track records of each single tug, and further, the data set is supplemented in a linear interpolation manner in this embodiment.

2) Data set feature extraction

The fishing time sequence relation refers to that a certain fishery ship and other fishery ships appear in the same unit space in unit time and the fishing behaviors have a sequence; recording a certain fishing boat as a characteristic fishing boat;

the quantification and mining of the fishing time sequence relation need to determine the size of unit time and unit space, so the method firstly extracts the relevant characteristics of single fishing of fishery ships from the track data set:

a) taking the period between two port periods of each fishing boat as a voyage, counting the continuous day number distribution of the voyages of all the fishing boats, and determining that 95% of voyages can be completed within 14 days according to the distribution result, so that 14 days are taken as the longest voyage of the method in the embodiment;

b) when fishing, the fishing boat can sail in a fixed speed range, the track record r in the fishing state is screened out by utilizing the speed range, and the coverage area of single fishing is determined according to the track record r. The specific method comprises the following steps: the size distribution of all the single fishing coverage geographical areas is counted, 91.9% of the single fishing activities cover less than 0.25 degree in latitude and 92% of the single fishing activities cover less than 0.25 degree in longitude according to the distribution result, so the unit space size is set to be (0.25 degree ) in the embodiment.

3) Definition and quantification of catching time sequence relation

The embodiment quantifies the fishing timing relationship from two aspects of space and time:

a) on a spatial level, the sea area studied in this example is the east China sea, and the specific areas are (122.5 ° E,27 ° N) - (128.5 ° E,33 ° N). First, the present embodiment divides the sea area to be predicted into 24 × 24 spatial grids with (0.25 ° ) as a unit space; constructing a catching space-time matrix F through all the track records r in the catching state; the cells (x, y, d) in the fishing spatio-temporal matrix F represent the set of fishing vessels fishing within d days on the spatial grid (x, y), the specific calculation method is shown in formula (1), FIG. 2 is a specific diagram of a certain cell (x, y, d) in F, which shows the date d₁In the meantime, there is a fishing boat v₁、v₂Waiting for catching in the space grid (x, y) at date d₂In the meantime, there is a fishing boat v₃、v₄And fishing in the space grid. Equation (1) represents: when the longitude lon, latitude lat and date d of the record r satisfy formula (1), the ship number id of r is taken as an element in the set (x, y, d);

(x,y,d)＝{r(id)|(r(lon),r(lat))∈(x,y)∧r(date)＝d} (1)

b) on a time level, when a characteristic fishing vessel appears in a spatial grid (x, y), then on day 7 if there are other fishing vesselsAnd then, the spatial grid appears within 14 days, so that the characteristic fishing boat and other fishing boats form a fishing time sequence relation. The fishing time sequence relation meets the conditions, and concretely refers to the formula (2). Wherein v is_eRepresentative characteristic boat, v_i∈RS_x,y,d(v_e) Depict other fishery ships v_iCharacteristic ship v_eFishing activities are carried out on the same spatial grid (x, y) within 14 days after 7 days;

RS_x,y,d(v_e)＝{v_i|v_e∈(x,y,d)∧v_i∈(x,y,t)∧t∈[d+7,d+14]} (2)

c) after the fishing timing relationship is defined by the above formula, the present invention further uses a timing relationship matrix S for each of 24 × 24 spatial grids_x,yExpressing the fishing time sequence relation in each space grid (x, y); using 24 x 24 adjoint relation matrices C simultaneously_x,yThe optimization method comprises the following steps:

s1: calculating a time-series relationship matrix S_x,yThe method comprises the following steps:

s11: initialization S_x,yThe method is a zero matrix of 1589 multiplied by 1589, wherein 1589 is the number of the fishery ships, and each fishery ship corresponds to a serial number;

s12: for each fishing vessel v in the unit (x, y, d)_eSet RS when calculating date d_x,y,d(v_e) Each element v in the set_jAre respectively connected with v_eForming a coordinate according to the corresponding serial number of the fishing boat, and recording the coordinate as (e, j), wherein e and j are v_e、v_jThe fishing boat serial number of (1);

s13: will S_x,yThe position corresponding to (e, j) is incremented by 1;

s14: traversing each date d of the whole fishing period, and accumulating the obtained coordinates formed according to the serial number of the fishing boat to S_x,yObtaining the times of forming a catching time sequence relation among all fishery ships in the space grid (x, y) at the corresponding positions; FIG. 3 is S_x,ySchematic of a part in which S_x,y(13,14) stands for fishing boat v₁₃As a characteristic boat, with fishing vessels v₁₄Form a time sequence of 8 times of fishingRelation, S_x,y(14,13) stands for fishing boat v₁₄As a characteristic boat, with fishing vessels v₁₃Forming a timing relation of 1 time catching, S_x,y(15,16) represents a fishing boat v₁₅As a characteristic boat, with fishing vessels v₁₆Forming a timing relation of 6 times of fishing S_x,y(16,11) stands for fishing boat v₁₆As a characteristic boat, not associated with a fishing vessel v₁₅Forming a catching time sequence relation;

s2: computing an adjoint matrix C_x,yThe method comprises the following steps:

s21: initialization C_x,yThe method is a zero matrix of 1589 multiplied by 1589, wherein 1589 is the number of the fishery ships, and each fishery ship corresponds to a serial number;

s22: for each fishing vessel v in the unit (x, y, d)_eThe set RC of d times of day is calculated according to equation (3)_x,y,d(v_e) Each element v in the set_kAnd v_eForming a coordinate according to the serial number of the fishing boat, and recording the coordinate as (e, k), wherein e and k are v_e、v_kThe fishing boat serial number of (1);

RC_x,y,d(v_e)＝{v_k|v_e∈(x,y,d)∧v_k∈(x,y,T)∧T∈[d,d+7]} (3)

s23: c is to be_x,yThe position corresponding to (e, k) is incremented by 1;

s24: traversing each date d of the whole fishing period, and accumulating the obtained coordinates formed according to the serial number of the fishing boat to C_x,yThe corresponding positions are obtained, the times of forming the relation of fishing between all the fishery ships are obtained, and figure 4 is C_x,ySchematic of part (a), wherein C_x,y(13,14) stands for fishing boat v₁₃With fishing vessels v₁₄ Form 10 times of concomitant fishing relation, C_x,y(15,16) represents a fishing boat v₁₅With fishing vessels v₁₆Forming a 0-time catching time sequence relation;

s3: optimizing a timing relationship matrix S_x,yThe method comprises the following steps:

s31: judgment S_x,yThe values of all the symmetric position elements (i, j) and (j, i) are set to 0 when the difference between the two is more than half, otherwise, they are set to 0, as shown by S in FIG. 4_x,y(13,14) is 8,S_x,y(14,13) is 1, the difference between the two is more than half, and represents v₁₃Should be taken as a characteristic boat, v₁₄Should be used as a following boat, S_x,y(14,13) will be set to 0;

s32: record all C_x,yAll positions (p, q) greater than 0, indicating a fishing vessel v_pAnd v_qThe fishing boat has the fishing time sequence relation and the accompanying fishing relation, the characteristic boat cannot predict the fishing behavior of the fishing boat which accompanies fishing by the characteristic boat, and the coexistence of the two relations does not help the prediction of the invention, so that S is used_x,yThe middle corresponding position (p, q) is set to 0. As shown in FIG. 5, C_x,y(13,14) '10' represents that the fishing boat and the fishing boat are in the accompanying fishing relationship, so S_x,y(13,14) will be set to 0, and S_x,y(15,16) keeping the original value;

the timing relationship matrix S processed through the above steps S1, S2, S3_x,yThe meaning of any non-zero element (p, q) in (a) is: fishery vessel v_pAnd v_qHaving a stable fishing timing relationship, fig. 5 is fig. 3 after being processed through steps S2, S3;

4) characteristic fishery vessel solution

In this embodiment, solving the feature ship set for prediction through two greedy principles specifically includes the following steps:

s1: in each space grid (x, y), solving the step (3) to obtain a time sequence relation matrix S_x,yIn (d), the row i with the most elements greater than 0 is the feature ship v with the spatial grid (x, y)_x,yIf the row with the most non-zero elements is provided with a plurality of rows, one row is selected; at the same time, the column number greater than 0 in the i row, i.e. the corresponding fishing boat, is recorded, the set is marked as Follow (x, y), which indicates that v follows in the grid_x,yForming a fishing boat in a fishing time sequence relation;

s2: boat v with selected partial characteristics_x,ySet V as a characteristic ship_eThe principle is as follows:

s21: feature ship v corresponding to Folow (x, y) with the most elements_x,yShip set with entry features V_eSimultaneously, elements in Folow (x, y) are counted into a set R;

s22: sequentially calculating Follow (x, y) of other grids (x, y) in the manner of step S21, and merging the Follow with the set R obtained in step S21;

until the number of elements in R reaches a certain proportion gamma of the number n of all fishery vessels, gamma being 90%, Table 1 is the variation process of the number of elements in set R, where the first column is the characteristic vessel v_x,yThe second column is the number of elements in R;

TABLE 1 following characteristic ship set V_eNumber of fishing boats

Number of characteristic vessels	Number of fishing boats following characteristic boat
		1	228
2	419
		3	591
4	736
		5	846
6	948
		7	1021
8	1090
		9	1141
10	1189
		11	1236
12	1277
		13	1307
14	1333
		15	1358
16	1381
		17	1403
18	1421
		19	1438(90.5％)

(5) Short-term fishing effort distribution prediction

Set V for solved feature ship_eThe present invention uses convolutional Neural network (convolutional Neural network)Network, CNN) as a core, and constructing a prediction model;

set V of prediction model and characteristic ship_eTaking the distribution of the fishing effort force of each day in the past 1 to 7 days as input, and taking the distribution of the fishing effort force of all fishery ships in the future 1 to 7 days as output;

the method takes 3 x 3 as the size of a CNN convolution kernel and 15 as the depth of the CNN, wherein 15 represents the number of spatial grids (x, y) which can be crossed furthest by the fishery ship in 7 days;

and finally, the RMSE is used as an error index, and the parameters of the CNN are optimized through a back propagation BP algorithm.

(6) Predicted results

In the embodiment, the average error is controlled within 10% in the cycle-by-cycle catching effort distribution prediction of 28 cycles in total, and the result is shown in fig. 6, wherein the horizontal axis is the cycle and the vertical axis is the error rate, and the result achieves short-time and high-precision catching effort distribution prediction.

Claims

1. The short-time fishing Nu force distribution prediction method based on the fishing time sequence relation is characterized by comprising the following steps of:

1) marine fishery ship trajectory data set preprocessing

2) data set feature extraction

firstly, extracting relevant characteristics of single fishing of a fishing boat from a track data set:

b) the fishing boat can sail in a fixed speed interval when fishing, the track record r in the fishing state is screened out by utilizing the speed interval, and the coverage area of single fishing is determined according to the track record r, and the specific method comprises the following steps: counting the size distribution of all the single-catch covered geographic areas, selecting the percentages W% and J% of latitude and longitude, and correspondingly, the latitude range of the single-catch of W% is in l_latThe longitude ranges of the inner percent and the J percent of single fishing are all in l_latIn the interior, will (l)_lat,l_lon) As a unit space size;

3) definition and quantification of catching time sequence relation

Quantifying the fishing time sequence relation from two aspects of space and time:

F＝{(x,y,d)}＝{r(id)|(r(lon),r(lat))∈(x,y)∧r(date)＝d} (1)

c) after the fishing timing relationship is defined by the above formula, a timing relationship matrix S is further used for each of the X Y spatial grids_x,yExpressing the fishing time sequence relation in each space grid (x, y); using X Y adjoint relation matrixes C simultaneously_x,yThe method is optimized by the following steps:

s1: calculating a time-series relationship matrix S_x,y：

s13: will S_x,yThe position corresponding to (e, j) is incremented by 1;

s2: computing an adjoint matrix C_x,y：

Each element v in the set_kAnd v_eAccording to fishing vesselsThe serial numbers form a coordinate, denoted as (e, k), where e and k are v, respectively_e、v_kThe fishing boat serial number of (1);

s23: c is to be_x,yThe position corresponding to (e, k) is incremented by 1;

s24: traversing each date d of the whole fishing period, and accumulating the obtained coordinates formed according to the serial number of the fishing boat to C_x,yObtaining the times of forming the accompanying fishing relation among all fishing boats in the space grid (x, y) according to the corresponding positions;

s3: optimizing a timing relationship matrix S_x,y：

4) feature ship set solution

Solving a characteristic ship set for prediction by two greedy principles:

s1: at each timeIn each space grid (x, y), solving the step 3) to finally obtain a time sequence relation matrix S_x,yRow h with maximum elements of greater than 0, which represents a fishing vessel v_hForming a fishing time sequence relation with the most fishing boats, and connecting the fishing boats v_hCharacteristic fishing vessels as spatial grid (x, y), denoted v_x,yIf the row with the most non-zero elements is provided with a plurality of rows, one row is selected; at the same time, the fishing boat corresponding to the column number greater than 0 in the h row is recorded, and the set is recorded as Follow (x, y) to indicate that v is followed in the grid_x,yForming a fishing boat in a fishing time sequence relation;

5) model construction

taking 3 x 3 as the size of a CNN convolution kernel, taking L as the depth of the CNN, wherein L represents the number of spatial grids (x, y) which can be crossed furthest by the fishing boat in D/2 days;

6) short-term fishing effort distribution prediction