WO2018014658A1 - 一种柔鱼类的中心渔场预测方法 - Google Patents

一种柔鱼类的中心渔场预测方法 Download PDF

Info

Publication number
WO2018014658A1
WO2018014658A1 PCT/CN2017/086000 CN2017086000W WO2018014658A1 WO 2018014658 A1 WO2018014658 A1 WO 2018014658A1 CN 2017086000 W CN2017086000 W CN 2017086000W WO 2018014658 A1 WO2018014658 A1 WO 2018014658A1
Authority
WO
WIPO (PCT)
Prior art keywords
fishery
model
layer
central
neural network
Prior art date
Application number
PCT/CN2017/086000
Other languages
English (en)
French (fr)
Inventor
陈新军
汪金涛
雷林
Original Assignee
上海海洋大学
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Priority claimed from CN201610580962.8A external-priority patent/CN106204314A/zh
Priority claimed from CN201610580969.XA external-priority patent/CN106157162A/zh
Priority claimed from CN201610580774.5A external-priority patent/CN106250980A/zh
Application filed by 上海海洋大学 filed Critical 上海海洋大学
Priority to US16/319,810 priority Critical patent/US11452286B2/en
Publication of WO2018014658A1 publication Critical patent/WO2018014658A1/zh

Links

Images

Classifications

    • A—HUMAN NECESSITIES
    • A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
    • A01K—ANIMAL HUSBANDRY; AVICULTURE; APICULTURE; PISCICULTURE; FISHING; REARING OR BREEDING ANIMALS, NOT OTHERWISE PROVIDED FOR; NEW BREEDS OF ANIMALS
    • A01K99/00—Methods or apparatus for fishing not provided for in groups A01K69/00 - A01K97/00
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N5/00—Computing arrangements using knowledge-based models
    • G06N5/04—Inference or reasoning models
    • G06N5/046—Forward inferencing; Production systems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/02—Agriculture; Fishing; Forestry; Mining
    • A—HUMAN NECESSITIES
    • A01—AGRICULTURE; FORESTRY; ANIMAL HUSBANDRY; HUNTING; TRAPPING; FISHING
    • A01K—ANIMAL HUSBANDRY; AVICULTURE; APICULTURE; PISCICULTURE; FISHING; REARING OR BREEDING ANIMALS, NOT OTHERWISE PROVIDED FOR; NEW BREEDS OF ANIMALS
    • A01K2227/00—Animals characterised by species
    • A01K2227/40—Fish

Definitions

  • the invention relates to a fishery prediction method, in particular to a central fishery prediction method for soft fish.
  • the central fishery forecast is a kind of quick report of fishing conditions. Accurate central fishery forecast can increase catch production and reduce fuel cost for fishing production.
  • the quick report of fishing condition is the location of the central fishery and the trend of fish stocks in the next 24 hours or days. And the possibility of Wangfa is predicted.
  • the forecasting unit of the fishery will promptly and accurately transmit the forecast content to the production vessel through the telecommunications system every day to achieve the purpose of directing on-site production.
  • the technical problem to be solved by the present invention is to provide a method for predicting a central fishery of a soft fish, combining the spatial and temporal scales of the sample of the oceanic economic soft fish and the selection of environmental factors, and establishing a prediction model considering the influence on the central fishery.
  • a method for predicting a central fishery of a soft fish characterized in that it comprises three steps of setting a time and space scale, setting an environmental factor, and establishing a prediction model of a central fishery,
  • the space-time scale setting adopts three levels of spatial scale, and the latitude and longitude are 0.25° ⁇ 0.25°, 0.5° ⁇ 0.5°, 1.0° ⁇ 1.0°, and the time scales of the two levels of week and month;
  • the environmental factor setting adopts SST as the main environmental factor, supplemented by SSH and Chl-a environmental factors;
  • the environmental factors are divided into four cases when establishing the central fishery prediction model: I SST; II SST, SSH; III SST, Chl-a; IV SST, SSH, Chl-a; arranged according to the time and space scale and environmental factor settings
  • the sample plan set of 24 cases is combined;
  • the central fishery prediction model adopts the classical error back propagation BP neural network model, and
  • the BP neural network model is a three-layer structure, namely input layer, hidden layer and output layer, and the input layer is input into the fishery field.
  • the output layer outputs CPUE or the fishery level index transformed by the CPUE; when the BP neural network model is forward-propagating, the sample enters from the input layer, is processed by the hidden layer activation function, and is transmitted to the output layer. If the actual output of the output layer does not meet the error requirement, then it is transferred to the back propagation phase of the error. Backpropagation is to propagate the error back to the input layer layer by layer through the hidden layer, and distribute the error to each. All the nodes in the layer, so as to obtain the error signal of each layer node, this error signal is used as the basis for correction. The forward propagation of this signal and the back propagation of the error are repeated. Carried out continuously adjusting weights, this process continues until the network output error is reduced to an acceptable level or until a preset number of times of learning, the best model is obtained through learning training, using for prediction.
  • the accuracy and stability of the BP neural network model obtained by comparing the sample plan sets of different situations are compared, and the model with the highest accuracy and stability in the environment is obtained, as the central fishery of the environment. Forecast model.
  • the accuracy comparison of the BP neural network model is compared when the output of the model is the CPUE level, and the sample plan set of each different case is compared according to the correct level percentage predicted by the model.
  • the accuracy of the obtained model when the output of the model is the CPUE value, the mean square error of the model is calculated, and the accuracy of the model obtained by the sample plan set of each different case is compared.
  • the stability of the BP neural network model obtained by comparing the sample set of different cases is calculated by calculating the average relative change value, which is defined as:
  • N is the number of comparison data and x(i) is the actual value of the fishery level.
  • x(i) is the actual value of the fishery level.
  • the BP neural network model with the time scale of Zhou, the spatial scale of 1.0° ⁇ 1.0° and the environmental factor of SST sample scheme is used as the prediction model of the central fishery.
  • the time scale is month
  • the spatial scale is 0.5° ⁇ 0.5°
  • the environmental factor is the BP neural network model including the sample plan set of SST and SSH as the prediction model of the central fishery.
  • a BP neural network model with a time scale of week, a spatial scale of 1.0° ⁇ 1.0° and an environmental factor of SST is used as a prediction model for the central fishery.
  • the invention considers the influence of different spatial and temporal scales and environmental factors on the prediction model of the central fishery field, and establishes a set of sample schemes for different situations according to the time and space scale and the setting of environmental factors, and adopts a classic error backpropagation neural network (Error Backpropagation Network, BP), using the error back propagation supervision algorithm, learning and storing a large number of mapping modes for the main operating time, the optimal operating sea range, the appropriate SST range for soft fish habitat, SSH range, Chl-a range forecast Provide technical support for fishing production to increase catch production and reduce fuel costs.
  • the prediction accuracy of the model is 70%-90% or even higher, which is more than 15% higher than the traditional habitat index forecasting method.
  • Figure 1 is a structural diagram of a BP neural network model.
  • Figure 2 is a distribution map of the Pacific Northwest squid in June.
  • Figure 3 is a distribution map of the Pacific Northwest squid in July.
  • Figure 4 is a distribution map of the Pacific Northwest squid in August.
  • Figure 5 is a distribution map of the squid in the Pacific Northwest in September.
  • Figure 6 is a distribution map of the squid in the Pacific Northwest in October.
  • Figure 7 is a distribution map of the squid in the Pacific Northwest in November.
  • Figure 8 is a forecast of the accuracy of the 2011 Pacific Northwest squid fishing ground under the 24 scenarios.
  • Figure 9 is an accuracy forecast of the 2011 Pacific Northwest squid fishing ground under the 24 scenarios.
  • Figure 10 is an ARV diagram of the forecast model of the Pacific Northwest squid fishery from 2003 to 2011 under 24 scenarios.
  • Figure 11 is an ARV diagram of the forecast model of the Pacific Northwest squid fishery from 2003 to 2011 under 24 scenarios.
  • Figure 12 is a plot of sensitivity analysis input variables versus predicted fishery levels on a monthly time scale.
  • Figure 13 is a plot of sensitivity analysis input variables versus predicted fishery levels on a monthly time scale.
  • Figure 14 is a plot of sensitivity analysis input variables versus predicted fishery levels on a monthly time scale.
  • Figure 15 is a plot of sensitivity analysis input variables versus predicted fishery levels on a monthly time scale.
  • Figure 16 is a plot of sensitivity analysis input variables versus predicted fishery levels on a monthly time scale.
  • Figure 17 is a plot of sensitivity analysis input variables versus predicted fishery levels on a monthly time scale.
  • three levels of spatial scale are set, with latitude and longitude of 0.25° ⁇ 0.25°, 0.5° ⁇ 0.5°, 1.0° ⁇ 1.0°, respectively.
  • the time scales for each level are week and month.
  • the resource abundance of fisheries in oceanic economical soft fish is not only affected by spatial and temporal factors, but also by the environmental factors of the habitat.
  • the selected surface temperature (SST) is the main environmental factor, supplemented by sea surface height (SSH) and chlorophyll a (Chl-a). Therefore, the environmental factors are divided into four types when establishing the central fishery prediction model.
  • Situation (Table 1).
  • the sample plan set for establishing the forecast model of the oceanic economic soft fish center fishery has the following 24 cases.
  • the central fishery prediction model uses the classic Error Backpropagation Network (BP).
  • BP Error Backpropagation Network
  • the BP neural network belongs to the multilayer forward neural network. Using the error back propagation supervision algorithm, the BP neural network can learn and store a large number of Mapping mode.
  • the BP neural network model uses a three-layer structure, namely an input layer, an implicit layer, and an output layer (as illustrated in FIG. 1).
  • the input layer is the time and space factor of the fishery and the marine environmental factor.
  • the output layer is the CPUE or the fishery grade index transformed by the CPUE.
  • the method of dividing the different fishery grades refers to the domain knowledge of the fishery experts.
  • the BP algorithm mainly consists of two processes: the forward propagation of the learning process signal and the back propagation of the error.
  • forward propagation the sample enters from the input layer and is processed by the hidden layer activation function and passed to the output layer. If the actual output of the output layer does not meet the error requirement, it is transferred to the back propagation phase of the error.
  • Backpropagation is to propagate the error back to the input layer layer by layer through the hidden layer, and distribute the error to all nodes in each layer to obtain the error signal of each layer node. This error signal is used as the basis for correction.
  • the forward propagation of this kind of signal and the back propagation of the error are repeated from beginning to end, and the weight is constantly adjusted, which is the process of network learning. This process continues until the error in the network output is reduced to an acceptable level or to a predetermined number of learnings.
  • the training method uses the steepest descent method. Assume that the number of input neurons is M, the number of neurons in the hidden layer is I, and the number of neurons in the output layer is J.
  • the mth neuron in the input layer is denoted as X m
  • the i th neuron in the hidden layer is denoted as k i
  • the j th th neuron in the output layer is denoted as Y j .
  • the link weight from Xm to k i is W mi
  • the connection weight from k i to Y j is w ij .
  • the implicit layer transfer function is a Sigmoid function
  • the output layer transfer function is a linear function.
  • u and v represent the input and output of each layer, respectively. Represents the input of the first neuron of the I layer (hidden layer).
  • the actual output of the network can be expressed as:
  • the expected output of the network is:
  • n is the number of iterations.
  • the error signal for the nth iteration is defined as:
  • the training process is the process of reducing the error energy.
  • the adjustment is reversed layer by layer along the network.
  • the gradient of the error pair w ij should be calculated. Adjust in the opposite direction of the direction:
  • the gradient can be obtained by finding the partial derivative. According to the chain rule of differentiation, there is
  • the transfer function is a linear function, so its derivative is 1, ie
  • the error signal propagates forward, and the weight w mi between the input layer and the hidden layer is adjusted, similar to the previous step.
  • f(g) is the sigmoid function, and the previous calculation is visible.
  • the weight adjustment amount ⁇ w the learning rate ⁇ the local gradient ⁇ the upper layer output signal v.
  • the setting of the difference range and the like is performed step by step in a state where the fitting is not performed.
  • the establishment process of BP neural network is completed in matlab (20lOb) software.
  • the sample set is divided into three parts: training sample, verification sample and test sample.
  • the parameters of the network design are: learning rate 0.1, momentum parameter 0.5, the transfer function between the input layer and the hidden layer, the hidden layer and the output layer are respectively S-type tangent function tansig, linear function purelin; the termination of network training
  • the parameters are: the maximum number of training is 1000, and the maximum error is given as 0.001.
  • the model obtains the best model through multiple trainings, and the weight is used for forecasting.
  • the BP forecasting model evaluates from three aspects: forecast accuracy, stability and interpretability:
  • MSE mean square error
  • N is the number of comparison data and x(i) is the actual value of the fishery level.
  • x(i) is the actual value of the fishery level.
  • the variable correlation is used to compare the contribution rate of each input variable to CPUE.
  • the calculation method is the ratio of the sum of the squares of the weight of the input variable to the hidden layer and the sum of the squares of all input layer variables to the hidden layer.
  • Sensitivity analysis is to explore the relationship between input variable changes and output variables. The process is: first calculate the maximum value, minimum value, median value, average value, and mode value of each input variable; then select one of the input variables. Make it gradually change from the minimum value to the maximum value. The other input variables are determined as one of the four special values. The input variable changes in turn, and the change of the output variable is observed.
  • the method is applied to the analysis of squid in the Pacific Northwest, and the production data of the squid squid in the Northwest Pacific Ocean from 2003 to 2011 is processed into a sample with a time resolution of “month” and a spatial resolution of 0.5° ⁇ 0.5°.
  • the Nominal CPUE in the zone and the Nominal CPUE are classified into different fishery grades based on knowledge of fisheries experts (Table 3).
  • the original samples were preprocessed into 24 sample sets according to the sample space-time scale and environmental factor setting method.
  • the data from 2003 to 2010 were used as training and verification samples, and the 2011 data was used as a test.
  • the sample using the matlab neural network tool, according to the modeling method, establishes the prediction model of the Pacific Northwest Squid Center in different time and space scales and environmental factors, and calculates the model accuracy and the ARV value of the whole sample from 2003 to 2011 ( Figure 8 - Figure) 11 is shown).
  • the time scale is week
  • the spatial scale is 1.0° ⁇ 1.0°
  • the environmental factor is SST established BP neural network fishery prediction model.
  • the prediction accuracy is about 85%
  • the ARV value is about 0.2, which has the highest.
  • the comparison between the two is better ( Figure 8 - Figure 11).
  • Table 4 shows the contribution rate of each variable with time, longitude, latitude, SST, SSH, and Chl-a as input variables.
  • Figure 12-2 shows the model prediction (indicated by level) of sensitivity analysis.
  • Sensitivity analysis shows that the weather and time changes of the squid fishing ground in the Northwest Pacific Ocean are complex, and the main (high-yield) operation time is 8, 9, and 10 months.
  • the fishery rank is relatively high and the resources are relatively abundant.
  • the optimal operating sea area is 150°-165. °E, 37°-42°N; suitable SST range for soft fish habitat is 11-18 ° C, SSH range is -10-60 cm, and Chl-a range is 0.1-1.7 mg/m 3 .
  • the method was applied to the analysis of squid squid in the southeastern Pacific, and the squid production data of the squid in the southeastern Pacific was processed into samples with a time resolution of “month” and a spatial resolution of 0.5° ⁇ 0.5°.
  • the Nominal CPUE based on knowledge of fisheries experts (Table 5), divides the Nominal CPUE into different fishery grades and plots the job map based on the fishery grade in the MarineStar software.
  • the original sample was preprocessed into 24 sample sets by FDP software according to the sample space-time scale and environmental factor setting method.
  • Matlab neural network tool BP neural network modeling method was used to establish the Southeast Pacific stem with different spatial and temporal scales and environmental factors.
  • the squid center fishery prediction model calculates the model accuracy and the ARV value of the entire sample.
  • the spatial scale is 0.5° ⁇ 0.5° and the environmental factor is II or III.
  • the prediction accuracy of the established fishery prediction model is around 70%, and the ARV value is around 0.3, which is higher. Accuracy and smaller ARV values.
  • the time scale of schemes 18 and 19 is month, with the spatial scale of 0.5° ⁇ 0.5° and the environmental factor of II or III, the prediction accuracy of the established fishery prediction model is also above 70%, and the ARV value is about 0.2, which has the highest Precision and small ARV values.
  • option 18 is selected, that is, the time scale is monthly, the spatial scale is 0.5° ⁇ 0.5°, and the environmental factor
  • the model established by the sample set of sub-II is used as the final forecasting model.
  • the input variable Nino 3.4 area SSTA has the largest contribution rate to the output CPUE, reaching 28.95%, followed by the variables SST and Latitude, and their contribution rates are 22.1% and 19.68%, respectively.
  • the smallest rate is Month, only 9.87%.
  • the prediction accuracy can be more than 90%, and the ARV value is about 0.2, with the highest precision and the smallest ARV value.
  • the time scale of the scheme 9 is weekly, the prediction accuracy of the fishery prediction model established by the spatial scale of 1.0° ⁇ 1.0° and the environmental factor of SST is over 90%, and the ARV value is about 0.2, with the highest precision and the smallest ARV. value.
  • the time scale of scheme 13 is month, the spatial scale is 0.25.
  • the prediction accuracy of the fishery prediction model established by ° ⁇ 0.25° and the environmental factor SST is also over 90%, and the ARV value is also around 0.2.
  • the scheme 9 is better.
  • the invention considers the influence of different spatial and temporal scales and environmental factors on the prediction model of the central fishery field, and adopts a classic error backpropagation neural network (BP).
  • BP neural network belongs to a multilayer forward neural network, and the error is reversed.
  • BP neural network can learn and store a large number of mapping modes for main (high-yield) operating time, optimal operating sea range, suitable SST range for soft fish habitat, SSH range, Chl-a range Forecasts provide technical support for fishing production to increase catch production and reduce fuel costs.
  • the prediction accuracy of the model established by this prediction method is 70%-90% or even higher, which is more than 15% higher than the traditional habitat index prediction method.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Computing Systems (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Computational Linguistics (AREA)
  • Environmental Sciences (AREA)
  • Animal Husbandry (AREA)
  • Marine Sciences & Fisheries (AREA)
  • Business, Economics & Management (AREA)
  • Molecular Biology (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Computational Mathematics (AREA)
  • Pure & Applied Mathematics (AREA)
  • Mathematical Optimization (AREA)
  • Mathematical Analysis (AREA)
  • Biodiversity & Conservation Biology (AREA)
  • Agronomy & Crop Science (AREA)
  • Mining & Mineral Resources (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • General Business, Economics & Management (AREA)
  • Primary Health Care (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Human Resources & Organizations (AREA)
  • Evolutionary Biology (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Operations Research (AREA)
  • Probability & Statistics with Applications (AREA)

Abstract

本发明涉及渔场预测方法，属于远洋渔业领域。一种柔鱼类的中心渔场预测方法，其特征在于：包括时空尺度设置、环境因子设置、建立中心渔场预测模型三个步骤，时空尺度设置采用三个级别的空间尺度，周和月两个级别的时间尺度；环境因子设置采用SST为主要环境因子，再辅以SSH、Chl-a两种环境因子，分为四种情况；根据时空尺度和环境因子设置情况，排列组合建立24种情况的样本方案集；建立误差反向传播BP神经网络模型，输入层输入样本方案集的数据，输出层输出CPUE或是由CPUE转化而成的渔场等级指标；通过BP神经网络模型的学习，不断修正和调整，得到最佳预报模型。本发明预报精度达到70％～90％，比一般传统方法大幅提高。

Description

一种柔鱼类的中心渔场预测方法 技术领域
本发明涉及渔场预测方法，尤其涉及柔鱼类的中心渔场预测方法。
背景技术
中心渔场预报是渔况速报的一种，准确的中心渔场预报可以为捕捞生产提高渔获产量并降低燃油成本，渔况速报是对未来24h或几天内的中心渔场位置、鱼群动向及旺发的可能性进行预测，由渔讯指挥单位每天定时将预报内容通过电讯***迅速而准确地传播给生产船只，达到指挥现场生产的目的。
目前已有多种方法预报大洋性经济柔鱼类的中心渔场，这些方法的基础是鱼类行动和生物学状况与环境条件之间的关系及其规律，本质都是根据生产统计数据样本获取“经验知识”用于预报，但以往对样本的时空尺度和环境因子的选择均没有深入研究，基本是根据经验设定(如大渔区小渔区等)，没有考虑不同时空尺度和环境因子对中心渔场预报模型的影响；在模型的选择上，也很少考虑海洋环境因子的实时性问题。
发明内容
本发明所要解决的技术问题是提供一种柔鱼类的中心渔场预测方法，结合大洋性经济柔鱼类的样本时空尺度和环境因子的选择，考虑到对其中心渔场的影响建立预测模型。
技术方案
一种柔鱼类的中心渔场预测方法，其特征在于：包括时空尺度设置、环境因子设置、建立中心渔场预测模型三个步骤，
时空尺度设置采用三个级别的空间尺度，经纬度分别为0.25°×0.25°、0.5°×0.5°、1.0°×1.0°，周和月两个级别的时间尺度；
环境因子设置采用SST为主要环境因子，再辅以SSH、Chl-a两种环境因子；
在建立中心渔场预测模型时将环境因子分为四种情况：I SST；Ⅱ SST，SSH；Ⅲ SST，Chl-a；Ⅳ SST，SSH，Chl-a；根据时空尺度和环境因子设置情况，排列组合建立24种情况的样本方案集；中心渔场预测模型采用经典的误差反向传播BP神经网络模型，BP神经网络模型为三层结构，即输入层、隐含层和输出层，输入层输入渔场的时空因子和环境因子，输出层输出CPUE或是由CPUE转化而成的渔场等级指标；BP神经网络模型正向传播时，样本从输入层进入，经隐含层激活函数处理，传向输出层，如输出层的实际输出与期望的输出不符合误差要求，则转入误差的反向传播阶段，反向传播是将误差通过隐含层向输入层逐层反向传播，将误差分摊给各层所有节点，从而获得各层节点的误差信号，此误差信号作为修正的依据，这种信号的正向传播与误差的反向传播周而复始地进行，权值不断调整，此过程一直进行到网络输出的误差减少到可接受的程度或进行到预先设定的学习次数为止，通过学习训练得到最佳模型，供预报使用。
隐含层节点个数由公式Pnum＝2Nnum+l确定，式中：Pnum为隐含层节点个数，Nnum为输入层节点个数。
进一步，所述BP神经网络模型建立后，比较不同情况的样本方案集得到的BP神经网络模型的精度和稳定度，得到该环境下的精度和稳定度最高的模型，作为该环境的中心渔场的预测模型。
进一步，所述BP神经网络模型的精度比较采用当模型的输出为CPUE等级时，根据模型预报出的正确等级百分比，比较各个不同情况的样本方案集 得到的模型的精度；当模型的输出为CPUE数值时，计算模型的均方误差，比较各个不同情况的样本方案集得到的模型的精度。
进一步，通过计算平均相对变动值来比较不同情况的样本方案集得到的BP神经网络模型的稳定度，所述平均相对变动值的定义为：
Figure PCTCN2017086000-appb-000001
其中，N为比较数据的个数，x(i)为渔场等级实际值，
Figure PCTCN2017086000-appb-000002
为渔场等级实际平均值，
Figure PCTCN2017086000-appb-000003
为渔场等级预测值。
经过比较，对于西北太平洋的柔鱼中心渔场的预报，采用时间尺度为周，空间尺度为1.0°×1.0°，环境因子为SST的样本方案集的BP神经网络模型作为中心渔场的预测模型。
对于东南太平洋的茎柔鱼中心渔场的预报，采用时间尺度为月，空间尺度为0.5°×0.5°，环境因子为包括SST和SSH的样本方案集的BP神经网络模型作为中心渔场的预测模型。
对于阿根廷滑柔鱼的中心渔场的预报，采用时间尺度为周，空间尺度为1.0°×1.0°，环境因子为SST的样本方案集的BP神经网络模型作为中心渔场的预测模型。
有益效果
本发明考虑不同时空尺度和环境因子的选择对中心渔场预测模型的影响，根据时空尺度和环境因子设置情况，建立不同情况的样本方案集，采用经典的误差反向传播神经网络(Error Backpropagation Network，BP)，使用误差反向传播的监督算法，学习和存储大量的映射模式，用于主要作业时间，最佳的作业海域范围，柔鱼类栖息适宜的SST范围，SSH范围，Chl-a范围预报，为捕捞生产提高渔获产量并降低燃油成本提供技术支撑。该模型预报精度达到70％～90％，甚至更高，比传统栖息地指数预报方法提高15％以上。
附图说明
图l是BP神经网络模型结构图。
图2是6月西北太平洋柔鱼作业分布图。
图3是7月西北太平洋柔鱼作业分布图。
图4是8月西北太平洋柔鱼作业分布图。
图5是9月西北太平洋柔鱼作业分布图。
图6是10月西北太平洋柔鱼作业分布图。
图7是11月西北太平洋柔鱼作业分布图。
图8是24种方案下2011年西北太平洋柔鱼渔场预报精度图。
图9是24种方案下2011年西北太平洋柔鱼渔场预报精度图。
图10是24种方案下2003～2011年西北太平洋柔鱼渔场预报模型ARV图。
图11是24种方案下2003～2011年西北太平洋柔鱼渔场预报模型ARV图。
图12是月时间尺度下，灵敏度分析输入变量与预测渔场等级的关系图。
图13是月时间尺度下，灵敏度分析输入变量与预测渔场等级的关系图。
图14是月时间尺度下，灵敏度分析输入变量与预测渔场等级的关系图。
图15是月时间尺度下，灵敏度分析输入变量与预测渔场等级的关系图。
图16是月时间尺度下，灵敏度分析输入变量与预测渔场等级的关系图。
图17是月时间尺度下，灵敏度分析输入变量与预测渔场等级的关系图。
具体实施方式
下面结合具体实施例和附图，进一步阐述本发明。
不同海域的海洋环境条件不一样，中心渔场形成的机制也不一样，因此其时间和空间分辨率对中心渔场预测模型的影响也有显著差异。为了解大洋性经济柔鱼类的样本时空尺度和环境因子的选择对其中心渔场预报模型的影响，需要建立最佳时空尺度和环境因子下业务化运行的中心渔场预报模型。
为了能够比较大洋性经济柔鱼类的中心渔场预报模型的最适时空尺度，设置三个级别的空间尺度，经纬度分别为0.25°×0.25°、0.5°×0.5°、1.0°×1.0°，两个级别的时间尺度为周和月。
大洋性经济柔鱼类的渔场的资源丰度不但受时空因子的影响，而且受栖息地的环境因子影响。本方法选定表温(SST)为主要环境因子，再辅以海面高度(SSH)、叶绿素a(Chl-a)两种环境因子，所以在建立中心渔场预报模型时将环境因子分为四种情况(表1)。
表1 环境因子设置
Figure PCTCN2017086000-appb-000004
因此，根据样本的时空尺度(三种空间尺度和两种时间尺度)和四种环境因子设置情况，建立大洋性经济柔鱼类中心渔场预报模型的样本方案集有如下24种情况。
表2 BP预报模型样本集方案
Figure PCTCN2017086000-appb-000005
Figure PCTCN2017086000-appb-000006
中心渔场预报模型采用经典的误差反向传播神经网络(Error Backpropagation Network，BP)，BP神经网络属于多层前向神经网络，使用误差反向传播的监督算法，BP神经网络能够学习和存储大量的映射模式。
BP神经网络模型采用三层结构，即输入层、隐含层和输出层(如附图1所示意)。输入层为渔场的时空因子和海洋环境因子，输出层是CPUE或是由CPUE转化而成的渔场等级指标，不同渔场等级的划分方法参考渔业专家的领域知识。
隐含层节点个数由公式Pnum＝2Nnum+l确定，式中：Pnum为隐含层节点个数，Nnum为输入层节点个数。
BP算法主要包括学习过程信号的正向传播与误差的反向传播两个过程组成。正向传播时，样本从输入层进入，经隐含层激活函数处理，传向输出层，如输出层的实际输出与期望的输出不符合误差要求，则转入误差的反向传播阶段。反向传播是将误差通过隐含层向输入层逐层反向传播，将误差分摊给各层所有节点，从而获得各层节点的误差信号，此误差信号作为修正的依据。这种信号的正向传播与误差的反向传播是周而复始地进行，权值不断调整，也就是网络学习的过程。此过程一直进行到网络输出的误差减少到可接受的程度或进行到预先设定的学习次数为止。
训练方法采用最速下降法。假设输入神经元个数为M，隐含层神经元个数为I，输出层神经元个数为J。输入层第m个神经元记为Xm，隐含层第i个神经元记为ki，输出层第j个神经元记为Yj。从Xm到ki的链接权值为Wmi，从ki到Yj的连接权值为wij。隐含层传递函数为Sigmoid函数，输出层传递函数为线性函数。u和v分别表示每一层的输入和输出，如
Figure PCTCN2017086000-appb-000007
表示I层(隐含层)第一个神经元的输入。网络的实际输出可表示为：
Figure PCTCN2017086000-appb-000008
网络的期望输出为：
Figure PCTCN2017086000-appb-000009
n为迭代次数。第n次迭代的误差信号定义为：
ej(n)＝dj(n)-Yj(n)
将误差能量定义为：
Figure PCTCN2017086000-appb-000010
训练过程即是将误差能量减小的过程。
在权值调整阶段，沿着网络逐层反向进行调整。首先调整隐含层与输出层之间的权值wij，根据最速下降法，应计算误差对wij的梯度
Figure PCTCN2017086000-appb-000011
再沿着该方向的反方向进行调整：
Figure PCTCN2017086000-appb-000012
wij(n+1)＝Δwij(n)+wij(n)
梯度可由求偏导得到，根据微分的链式规则，有
Figure PCTCN2017086000-appb-000013
由于e(n)是ej(n)的二次函数，其微分为一次函数：
Figure PCTCN2017086000-appb-000014
Figure PCTCN2017086000-appb-000015
输出层传递函数的导数：
Figure PCTCN2017086000-appb-000016
Figure PCTCN2017086000-appb-000017
因此，梯度值为
Figure PCTCN2017086000-appb-000018
权值的修正量为
Figure PCTCN2017086000-appb-000019
引入局部梯度的定义：
Figure PCTCN2017086000-appb-000020
所以权值的修正量为：
Figure PCTCN2017086000-appb-000021
在输出层，传递函数为线性函数，因此其导数为1，即
Figure PCTCN2017086000-appb-000022
所以可得
Figure PCTCN2017086000-appb-000023
误差信号向前传播，对输入层与隐含层之间的权值wmi进行调整，与上一步类似应有
Figure PCTCN2017086000-appb-000024
Figure PCTCN2017086000-appb-000025
为输入神经元的输出，即
Figure PCTCN2017086000-appb-000026
Figure PCTCN2017086000-appb-000027
为局部梯度，定义为
Figure PCTCN2017086000-appb-000028
f(g)为sigmoid函数，同时又上一步计算可见，
Figure PCTCN2017086000-appb-000029
故有
Figure PCTCN2017086000-appb-000030
到此，三层BP网络的学习权值调整过程结束，可归结为：
权值调整量Δw＝学习率η·局部梯度δ·上一层输出信号v。至于学习率η、误 差范围等的设定，在不过拟合的状态下进行逐步调优。
BP神经网络的建立过程在matlab(20lOb)软件中完成，使用神经网络工具箱的拟合工具，将样本集分为训练样本、验证样本和测试样本三部分。网络设计的参数为：学习速率0.1，动量参数0.5，输入层与隐含层、隐含层与输出层神经元之间的传递函数分别是S型正切函数tansig、线性函数purelin；网络训练的终止参数为：最大训练次数为1000，最大误差给定为0.001。模型通过多次训练获得最佳模型，取权重供预报使用。
BP预报模型从预报精度、稳定性和可解释性三方面评价：
(l)预报精度评价，当模型的输出为CPUE等级时，根据模型预报出的正确等级百分比，比较各种模型的精度；当模型的输出为CPUE数值时，计算模型的均方误差(MSE)，比较各个模型的精度。
Figure PCTCN2017086000-appb-000031
其中，yk为CPUE的实际值，
Figure PCTCN2017086000-appb-000032
为CPUE的预报值。
(2)稳定性评价，评价不同样本建立的BP模型精度的稳定性，计算平均相对变动值(Average Relative Variance，ARV)，其定义为
Figure PCTCN2017086000-appb-000033
其中，N为比较数据的个数，x(i)为渔场等级实际值，
Figure PCTCN2017086000-appb-000034
为渔场等级实际平均值，
Figure PCTCN2017086000-appb-000035
为渔场等级预测值。平均相对变动值ARV越小，表明预测效果越好，ARV＝0表示达到了理想预测效果，当ARV＝l时，表明模型仅达到了平均值的预测效果。
(3)可解释性评价，首次应用于渔场预报，并作为分析中心渔场预报精度的一个指标。即用变量相关性(Independent variable relevance)和灵敏度分析(Sensitivity Analyses)评价在不同时空尺度和环境因子样本上建立的预报模型的可解释性。
变量相关用来比较各输入变量对CPUE的贡献率，计算方法是输入变量与隐含层连接的权重平方和与所有输入层变量到隐含层连接权重平方和之比。
灵敏度分析是探究输入变量变化与输出变量之间的关系，其过程是：首先计算各个输入变量的最大值、最小值、中值、平均值、众数特殊值；然后选挥其中一个输入变量，使其从最小值到最大值逐渐变化，其他输入变量都确定为四个特殊值中的一个，轮流改变变化的输入变量，观察输出变量的变化情况。
实施例1
例如将本方法应用于西北太平洋柔鱼分析，将2003～2011年西北太平洋柔鱼鱿钓生产数据处理成时间分辨率为“月”，空间分辨率为0.5°×0.5°的样本，计算小渔区内的Nominal CPUE，并根据渔业专家知识(表3)将Nominal CPUE划分为不同的渔场等级。
表3 基于CPUE的西北太平洋柔鱼渔场等级
Figure PCTCN2017086000-appb-000036
利用地理信息***绘制基于渔场等级的作业分布图(图2～图7)。6月、7月渔场分布广泛，经度跨度大，从150°E到180°E均有分布，8月、9月、10月、11月主要集中在西部渔场(165°E以西)，经度跨度相对较小，从6月到11月渔场有逐步向西移的趋势。
用FDP软件，按样本时空尺度和环境因子设置方法将原始样本预处理成24种样本集，2003～2010年数据作为训练和验证样本，2011年数据作为测试 样本，利用matlab神经网络工具，按建模方法，建立不同时空尺度和环境因子的西北太平洋柔鱼中心渔场预报模型，并计算模型精度以及2003—2011年整个样本的ARV值(如图8-图11所示意)。
如采用前述样本方案9，时间尺度为周，空间尺度为1.0°×1.0°、环境因子为SST所建立的BP神经网络渔场预报模型，预报精度在85％左右，ARV值在0.2左右，具有最高的精度和最小的ARV值；样本方案18中，时间尺度为月，空间尺度为0.5°×0.5°、环境因子为SST和SSH所建立的渔场预报模型的预报精度也达80％以上，ARV值在0.3左右，具有较高的精度和较小的ARV值。两者比较方案9则更优(图8-图11)。
为了探讨多种环境因子对渔场的选择作用，因此选择样本方案20建立的模型进行变量相关性分析和灵敏度分析。表4为以时间、经度、纬度、SST、SSH、Chl-a为输入变量，各变量的贡献率；图l2～图17为灵敏度分析的模型预报(以等级表示)变化情况。
表4 预报模型变量相关性分析
Figure PCTCN2017086000-appb-000037
变量相关性分析显示：SST对西北太平洋柔鱼渔场预报模型的贡献率最大，为26.04﹪，其次为时间变量“月”，贡献率最小的为环境因子SSH，仅为5.54﹪。
灵敏度分析显示：西北太平洋柔鱼渔场时空变化复杂，其主要(高产)作业时间为8、9、10三个月，渔场等级较高，资源相对丰富；最佳的作业海 域范围是150°-165°E、37°-42°N；柔鱼栖息适宜的SST范围是11-18℃，SSH范围是-10-60cm，Chl-a范围是0.1-1.7mg/m3。
实施例2
将本方法用于东南太平洋茎柔鱼分析，将东南太平洋茎柔鱼的鱿钓生产数据处理成时间分辨率为“月”，空间分辨率为0.5°×0.5°的样本，计算小渔区内的Nominal CPUE，并根据渔业专家知识(表5)将Nominal CPUE划分为不同的渔场等级，在MarineStar软件中绘制基于渔场等级的作业分布图。
表5 基于CPUE的东南太平洋茎柔鱼渔场等级
Figure PCTCN2017086000-appb-000038
同样，用FDP软件按样本时空尺度和环境因子设置方法将原始样本预处理成24种样本集，利用matlab神经网络工具，按BP神经网络建模方法，建立不同时空尺度和环境因子的东南太平洋茎柔鱼中心渔场预报模型，计算模型精度以及整个样本的ARV值。
方案6、7时间尺度为周时，以空间尺度为0.5°×0.5°、环境因子为Ⅱ或Ⅲ，所建立的渔场预报模型的预报精度在70％附近，ARV值在0.3左右，具有较高的精度和较小的ARV值。方案18、19时间尺度为月时，以空间尺度为0.5°×0.5°、环境因子为Ⅱ或Ⅲ，建立的渔场预报模型的预报精度也达70％以上，ARV值在0.2左右，具有最高的精度和较小的ARV值。但从遥感数据的获得的实时性分析，海面高度比叶绿素a浓度数据更容易获得，方案18则更优。因此选定方案18，即时间尺度为月、空间尺度为0.5°×0.5°、环境因 子为Ⅱ的样本集所建立的模型作为最终的预报模型。
根据计算变量相关计算结果(表6)分析，输入变量Nino 3.4区SSTA对输出CPUE的贡献率最大，达到28.95％，其次是变量SST、和Latitude，其贡献率分别为22.1％和19.68％，贡献率最小的是Month，只有9.87％。
表6 预报模型变量相关性分析
Figure PCTCN2017086000-appb-000039
实施例3
将本方法用于阿根廷滑柔鱼分析，根据渔业专家知识将Nominal CPUE划分为不同的渔场等级，如表7所示意，在MarineStar软件中绘制基于渔场等级的作业分布图。
表7 基于CPUE的阿根廷滑柔鱼渔场等级
Figure PCTCN2017086000-appb-000040
采用本预测方法建立不同时空尺度和环境因子的阿根廷滑柔鱼中心渔场预报模型，能够实现预报精度90％以上，ARV值在0.2左右，具有最高的精度和最小的ARV值。方案9时间尺度为周时，以空间尺度为1.0°×1.0°、环境因子为SST所建立的渔场预报模型的预报精度达90％以上，ARV值在0.2左右，具有最高的精度和最小的ARV值。方案13时间尺度为月时，以空间尺度为0.25 °×0.25°、环境因子为SST所建立的渔场预报模型的预报精度也达90％以上，ARV值也在0.2左右。但从预报精度分析，方案9更优一些。
而变量相关性分析显示：在周时间尺度和月时间尺度下，SST对阿根廷滑柔鱼渔场预报模型的贡献率最大，其次是“纬度”变量(表8)。
表8 预报模型变量相关性分析表
Figure PCTCN2017086000-appb-000041
本发明考虑不同时空尺度和环境因子的选择对中心渔场预测模型的影响，采用经典的误差反向传播神经网络(Error Backpropagation Network，BP)，BP神经网络属于多层前向神经网络，使用误差反向传播的监督算法，BP神经网络能够学习和存储大量的映射模式，用于主要(高产)作业时间，最佳的作业海域范围，柔鱼类栖息适宜的SST范围，SSH范围，Chl-a范围预报，为捕捞生产提高渔获产量并降低燃油成本提供技术支撑。本预测方法建立的模型预报精度达到70％～90％，甚至更高，比传统栖息地指数预报方法提高15％以上。

Claims (8)

  1. 一种柔鱼类的中心渔场预测方法，其特征在于：包括时空尺度设置、环境因子设置、建立中心渔场预测模型三个步骤，
    时空尺度设置采用三个级别的空间尺度，经纬度分别为0.25°×0.25°、0.5°×0.5°、1.0°×1.0°，周和月两个级别的时间尺度；
    环境因子设置采用SST为主要环境因子，再辅以SSH、Chl-a两种环境因子；
    在建立中心渔场预测模型时将环境因子分为四种情况：I SST；Ⅱ SST，SSH；Ⅲ SST，Chl-a；Ⅳ SST，SSH，Chl-a；根据时空尺度和环境因子设置情况，排列组合建立24种情况的样本方案集；中心渔场预测模型采用经典的误差反向传播BP神经网络模型，BP神经网络模型为三层结构，即输入层、隐含层和输出层，输入层输入渔场的时空因子和环境因子，输出层输出CPUE或是由CPUE转化而成的渔场等级指标；BP神经网络模型正向传播时，样本从输入层进入，经隐含层激活函数处理，传向输出层，如输出层的实际输出与期望的输出不符合误差要求，则转入误差的反向传播阶段，反向传播是将误差通过隐含层向输入层逐层反向传播，将误差分摊给各层所有节点，从而获得各层节点的误差信号，此误差信号作为修正的依据，这种信号的正向传播与误差的反向传播周而复始地进行，权值不断调整，此过程一直进行到网络输出的误差减少到可接受的程度或进行到预先设定的学习次数为止，通过学习训练得到最佳模型，供预报使用。
  2. 根据权利要求l所述的柔鱼类的中心渔场预测方法，其特征在于：隐含层节点个数由公式Pnum＝2Nnum+l确定，式中：Pnum为隐含层节点个数，Nnum为输入层节点个数。
  3. 根据权利要求l所述的柔鱼类的中心渔场预测方法，其特征在于：所述BP神经网络模型建立后，比较不同情况的样本方案集得到的BP神经网络模型的精度和稳定度，得到该环境下的精度和稳定度最高的模型，作为该环境 的中心渔场的预测模型。
  4. 根据权利要求3所述的柔鱼类的中心渔场预测方法，其特征在于：所述BP神经网络模型的精度比较采用当模型的输出为CPUE等级时，根据模型预报出的正确等级百分比，比较各个不同情况的样本方案集得到的模型的精度；当模型的输出为CPUE数值时，计算模型的均方误差，比较各个不同情况的样本方案集得到的模型的精度。
  5. 根据权利要求3所述的柔鱼类的中心渔场预测方法，其特征在于：通过计算平均相对变动值来比较不同情况的样本方案集得到的BP神经网络模型的稳定度，所述平均相对变动值的定义为：
    Figure PCTCN2017086000-appb-100001
    其中，N为比较数据的个数，x(i)为渔场等级实际值，
    Figure PCTCN2017086000-appb-100002
    为渔场等级实际平均值，
    Figure PCTCN2017086000-appb-100003
    为渔场等级预测值。
  6. 根据权利要求3所述的柔鱼类的中心渔场预测方法，其特征在于：经过比较，对于西北太平洋的柔鱼中心渔场的预报，采用时间尺度为周，空间尺度为1.0°×1.0°，环境因子为SST的样本方案集的BP神经网络模型作为中心渔场的预测模型。
  7. 根据权利要求3所述的柔鱼类的中心渔场预测方法，其特征在于：对于东南太平洋的茎柔鱼中心渔场的预报，采用时间尺度为月，空间尺度为0.5°×0.5°，环境因子为包括SST和SSH的样本方案集的BP神经网络模型作为中心渔场的预测模型。
  8. 根据权利要求3所述的柔鱼类的中心渔场预测方法，其特征在于：对于阿根廷滑柔鱼的中心渔场的预报，采用时间尺度为周，空间尺度为1.0°×1.0°，环境因子为包括SST的样本方案集的BP神经网络模型作为中心渔场的预测模型。
PCT/CN2017/086000 2016-07-22 2017-05-25 一种柔鱼类的中心渔场预测方法 WO2018014658A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/319,810 US11452286B2 (en) 2016-07-22 2017-05-25 Method of predicting central fishing ground of flying squid family ommastrephidae

Applications Claiming Priority (6)

Application Number Priority Date Filing Date Title
CN201610580969.X 2016-07-22
CN201610580962.8 2016-07-22
CN201610580962.8A CN106204314A (zh) 2016-07-22 2016-07-22 一种东南太平洋茎柔鱼中心渔场预测方法
CN201610580969.XA CN106157162A (zh) 2016-07-22 2016-07-22 一种北太平洋鱿鱼中心渔场预测方法
CN201610580774.5 2016-07-22
CN201610580774.5A CN106250980A (zh) 2016-07-22 2016-07-22 一种阿根廷滑柔鱼中心渔场预测方法

Publications (1)

Publication Number Publication Date
WO2018014658A1 true WO2018014658A1 (zh) 2018-01-25

Family

ID=60991890

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2017/086000 WO2018014658A1 (zh) 2016-07-22 2017-05-25 一种柔鱼类的中心渔场预测方法

Country Status (2)

Country Link
US (1) US11452286B2 (zh)
WO (1) WO2018014658A1 (zh)

Cited By (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109584098A (zh) * 2018-12-20 2019-04-05 上海海洋大学 一种渔场单位捕捞努力量渔获量的预测方法及装置
CN110414742A (zh) * 2019-08-02 2019-11-05 大连海事大学 一种多水位站联动的内河水位智能预测方法
CN110598918A (zh) * 2019-08-28 2019-12-20 宁波海上鲜信息技术有限公司 一种水产品价格预测方法、装置及计算机存储介质
CN110751323A (zh) * 2019-10-14 2020-02-04 浙江海洋大学 一种基于动态建模的渔场渔情预测方法
CN111488974A (zh) * 2020-04-14 2020-08-04 四川北控清洁能源工程有限公司 基于深度学习神经网络的海洋风能降尺度方法
CN113408200A (zh) * 2021-06-17 2021-09-17 中国航空综合技术研究所 航空装备振动环境分析和预测方法

Families Citing this family (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11475310B1 (en) * 2016-11-29 2022-10-18 Perceive Corporation Training network to minimize worst-case error
CA3020884A1 (en) * 2018-01-24 2019-07-24 Wai Kit Ricky Fok Parallel residual neural network architechture and system and method for trainning a residual neural network
CN111583051B (zh) * 2020-05-13 2024-01-23 上海海洋大学 基于生态位模型的太平洋海域大眼金枪鱼栖息地评估方法
CN111784034B (zh) * 2020-06-22 2024-06-11 上海海洋大学 一种影响智利海域美洲赤鱿渔场的关键环境因子的筛选及探测方法
CN113657582B (zh) * 2021-08-20 2024-04-09 西安科技大学 一种基于时空数据的时空图神经网络瓦斯浓度预测方法
CN115014445B (zh) * 2022-08-10 2022-11-15 中国农业大学 一种智慧渔场多维全景感知监控方法、***与装置

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106157162A (zh) * 2016-07-22 2016-11-23 上海海洋大学 一种北太平洋鱿鱼中心渔场预测方法
CN106204314A (zh) * 2016-07-22 2016-12-07 上海海洋大学 一种东南太平洋茎柔鱼中心渔场预测方法
CN106250980A (zh) * 2016-07-22 2016-12-21 上海海洋大学 一种阿根廷滑柔鱼中心渔场预测方法

Family Cites Families (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9298978B1 (en) * 2012-05-03 2016-03-29 Joseph F Hlatky Optical fish recognition
CN102833680B (zh) 2012-09-11 2014-09-24 中国水产科学研究院东海水产研究所 基于位置的海洋渔业信息服务方法
CN103235982A (zh) 2013-04-16 2013-08-07 中国水产科学研究院东海水产研究所 一种基于贝叶斯网络模型的渔场预报方法
US11688154B2 (en) * 2020-05-28 2023-06-27 X Development Llc Analysis and sorting in aquaculture

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106157162A (zh) * 2016-07-22 2016-11-23 上海海洋大学 一种北太平洋鱿鱼中心渔场预测方法
CN106204314A (zh) * 2016-07-22 2016-12-07 上海海洋大学 一种东南太平洋茎柔鱼中心渔场预测方法
CN106250980A (zh) * 2016-07-22 2016-12-21 上海海洋大学 一种阿根廷滑柔鱼中心渔场预测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
WANG, JINTAO ET AL.: "Analysis of the Optimum Temporal and Spatial Scale as well as Environmental Factors of the Fishery Forecasting Models for Illex argentinus", JOURNAL OF FISHERY SCIENCES IN CHINA, vol. 22, no. 5, 30 September 2015 (2015-09-30), pages 1007 - 1014, ISSN: 1005-8737 *
WANG, JINTAO ET AL.: "Impacts of Oceanic Environmental Factors of Different Temporal and Spatial Scales of Fishery Forecasting Models for Illex argentinus in the Southwest Atlantic", PROCEEDINGS OF THE 2014 ACADEMIC ANNUAL CONFERENCE ON CHINA GEOGRAPHICAL INFORMATION SCIENCE, 12 October 2014 (2014-10-12), pages 203 - 210 *

Cited By (11)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN109584098A (zh) * 2018-12-20 2019-04-05 上海海洋大学 一种渔场单位捕捞努力量渔获量的预测方法及装置
CN109584098B (zh) * 2018-12-20 2023-07-25 上海海洋大学 一种渔场单位捕捞努力量渔获量的预测方法及装置
CN110414742A (zh) * 2019-08-02 2019-11-05 大连海事大学 一种多水位站联动的内河水位智能预测方法
CN110598918A (zh) * 2019-08-28 2019-12-20 宁波海上鲜信息技术有限公司 一种水产品价格预测方法、装置及计算机存储介质
CN110598918B (zh) * 2019-08-28 2021-06-11 宁波海上鲜信息技术有限公司 一种水产品价格预测方法、装置及计算机存储介质
CN110751323A (zh) * 2019-10-14 2020-02-04 浙江海洋大学 一种基于动态建模的渔场渔情预测方法
CN110751323B (zh) * 2019-10-14 2023-06-23 浙江海洋大学 一种基于动态建模的渔场渔情预测方法
CN111488974A (zh) * 2020-04-14 2020-08-04 四川北控清洁能源工程有限公司 基于深度学习神经网络的海洋风能降尺度方法
CN111488974B (zh) * 2020-04-14 2023-05-02 四川北控清洁能源工程有限公司 基于深度学习神经网络的海洋风能降尺度方法
CN113408200A (zh) * 2021-06-17 2021-09-17 中国航空综合技术研究所 航空装备振动环境分析和预测方法
CN113408200B (zh) * 2021-06-17 2022-11-01 中国航空综合技术研究所 航空装备振动环境分析和预测方法

Also Published As

Publication number Publication date
US11452286B2 (en) 2022-09-27
US20190230913A1 (en) 2019-08-01

Similar Documents

Publication Publication Date Title
WO2018014658A1 (zh) 一种柔鱼类的中心渔场预测方法
CN109858647B (zh) 一种耦合gis和gbdt算法的区域洪涝灾害风险评价与预估方法
WO2021098472A1 (zh) 一种深度时空特征联合学习的农作物产量估测方法
CN110084367B (zh) 一种基于lstm深度学习模型的土壤墒情预测方法
Matsumura et al. Maize yield forecasting by linear regression and artificial neural networks in Jilin, China
CN110751318B (zh) 一种基于ipso-lstm的超短期电力负荷预测方法
CN109871995A (zh) Spark框架下分布式深度学习的量子优化调参方法
CN108416460B (zh) 基于多因素时序—随机深度置信网络模型的蓝藻水华预测方法
CN106295833B (zh) 一种太平洋褶柔鱼资源补充量预测方法及其应用
CN106485093B (zh) 基于改进马尔可夫链的太阳辐照度时间序列合成方法
CN110472840B (zh) 一种基于神经网络技术的农业水利调度方法及***
CN106157162A (zh) 一种北太平洋鱿鱼中心渔场预测方法
CN108921359A (zh) 一种分布式瓦斯浓度预测方法及装置
CN109346177A (zh) 一种基于改进ga-bp算法的疾病症状数量预测方法
CN110069032A (zh) 一种基于小波神经网络的茄子温室环境智能检测***
CN113393057A (zh) 一种基于深度融合机器学习模型的小麦产量集成预测方法
CN109034500A (zh) 一种多时序协同的中期电力负荷预测方法
CN114862035B (zh) 一种基于迁移学习的组合式海湾水温预测方法
CN114861980A (zh) 一种基于bp-lstm模型的碳预测方法
CN115640956A (zh) 一种未来水资源供需平衡分析方法
Nhita et al. Planting calendar forecasting system using evolving neural network
Bashar et al. Investigation of factors affecting rural drinking water consumption using intelligent hybrid models
CN106204314A (zh) 一种东南太平洋茎柔鱼中心渔场预测方法
CN114971005B (zh) 基于lstm与差分回归模型动态加权的海湾水温组合预测方法
CN106250980A (zh) 一种阿根廷滑柔鱼中心渔场预测方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 17830295

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 17830295

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205 DATED 28/10/2019)

122 Ep: pct application non-entry in european phase

Ref document number: 17830295

Country of ref document: EP

Kind code of ref document: A1