WO2022193681A1 - 一种基于时间卷积网络的防洪调度方案优选方法 - Google Patents

一种基于时间卷积网络的防洪调度方案优选方法 Download PDF

Info

Publication number
WO2022193681A1
WO2022193681A1 PCT/CN2021/128854 CN2021128854W WO2022193681A1 WO 2022193681 A1 WO2022193681 A1 WO 2022193681A1 CN 2021128854 W CN2021128854 W CN 2021128854W WO 2022193681 A1 WO2022193681 A1 WO 2022193681A1
Authority
WO
WIPO (PCT)
Prior art keywords
flood control
control scheduling
evaluation
value
convolutional network
Prior art date
Application number
PCT/CN2021/128854
Other languages
English (en)
French (fr)
Inventor
胡鹤轩
胡强
张晔
胡震云
Original Assignee
河海大学
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
Application filed by 河海大学 filed Critical 河海大学
Priority to US17/909,003 priority Critical patent/US20230196076A1/en
Publication of WO2022193681A1 publication Critical patent/WO2022193681A1/zh

Links

Images

Classifications

    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • 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
    • G06Q10/00—Administration; Management
    • G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
    • G06Q10/063—Operations research, analysis or management
    • G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
    • G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/043—Architecture, e.g. interconnection topology based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • 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
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • 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
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02A—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE
    • Y02A10/00—TECHNOLOGIES FOR ADAPTATION TO CLIMATE CHANGE at coastal zones; at river basins
    • Y02A10/40—Controlling or monitoring, e.g. of flood or hurricane; Forecasting, e.g. risk assessment or mapping

Definitions

  • the invention belongs to the technical field of flood control scheduling of reservoirs, and relates to a method for optimizing a flood control scheduling scheme based on a time convolution network.
  • Reservoir group flood control scheduling has a strong practicality. It is affected by many factors such as scheduling objectives, water inflow conditions, and schedulers' knowledge and experience, and is closely related to social, economic, natural, ecological and other factors. Therefore, the evaluation of flood control scheduling schemes for reservoir groups is A multi-objective, multi-attribute, multi-level evaluation index model, and real-time flood control scheduling is an irreversible real-time dynamic correction process. For such a multi-index model, its evaluation has many levels of influencing factors, complex index system composition, and many Qualitative indicators are difficult to quantify. Therefore, it has important theoretical significance and practical value to comprehensively and comprehensively consider a variety of influencing factors, build a reasonable evaluation index system, and select the optimal flood control scheduling scheme for reservoir groups.
  • multi-scheme comparison and selection methods are widely used, such as expert system evaluation method, grey relational decision-making method, projection pursuit method, fuzzy comprehensive evaluation method, analytic hierarchy process (ahp) and pros and cons Solution (Topsis) and so on.
  • the purpose of the invention is to overcome the defects of the prior art, and provide a flood control scheduling scheme optimization method based on time convolution network, which can carry out comprehensive, comprehensive and comprehensive flood control scheduling scheme for large-scale reservoir groups in the basin under the influence of multiple uncertainties. Intelligent optimization of multi-angle solutions.
  • the present invention adopts the following technical solutions.
  • a method for optimizing a flood control scheduling scheme based on a time convolutional network of the present invention includes the following steps:
  • Step 1 Establish the evaluation index system of the flood control scheduling scheme of the reservoir group
  • Step 2 Construct the time series evaluation index matrix of the flood control scheduling scheme.
  • the matrix is used as the input of the time convolution network.
  • the comprehensive score of the training samples of the flood control scheduling scheme is calculated: establish a fuzzy decision matrix, determine the quantitative and The relative membership degree of the qualitative evaluation index is obtained, and the relative membership degree matrix is obtained, so as to construct the time series evaluation index matrix of the flood control scheduling scheme; the matrix is used as the input of the time convolution network, and the calculation formula of the entropy weight is carried out according to different types of evaluation indicators.
  • the comprehensive evaluation value of the flood control scheduling scheme is calculated, and the final comprehensive evaluation value used to judge the pros and cons of the scheme is used as the output; wherein, the comprehensive evaluation value is obtained by using the fuzzy comprehensive evaluation method.
  • a supervised interpolation-based multi-sample data enhancement method SMOTE is used to augment the training samples of the temporal convolutional network to generate new samples for the small sample class;
  • Step 3 Determine the structure of the temporal convolutional network, including an input layer, a causal dilated convolution, an activation function, a residual connection, a fully connected layer and an output layer;
  • Step 4 Use the loss function combining the mean square error and the Nash efficiency coefficient to train the temporal convolutional network
  • Step 5 Input the time series evaluation index matrix of the flood control scheduling scheme into the time convolution network to obtain the comprehensive evaluation value of the scheme, and the optimal comprehensive evaluation value is the optimal scheme for flood control scheduling of the reservoir group.
  • step 2 specifically includes the following steps:
  • Step 2.1 set the evaluation index weights of reservoirs, flood storage and detention areas, and hydrological stations
  • l represents the evaluation index
  • q represents the evaluation target
  • i 1,2,...,q
  • j 1,2,...,l
  • x ij is the eigenvalue of the target i index j
  • Step 2.1.2 calculate the entropy weight ⁇ hj of the evaluation index
  • ⁇ hj H s ⁇ hsj +(1-H s ) ⁇ hkj
  • ⁇ hsj and ⁇ hkj are the entropy value separation magnitude weight coefficients
  • H s is the same part of each entropy value in the entropy value vector from the first decimal point
  • r ij is the relative superiority degree of the target i index j value
  • H j is the entropy value corresponding to r ij
  • e ij is the relative importance of r ij
  • Step 2.2 set the weights of risk and benefit evaluation indicators
  • Step 2.2.1 build a risk-benefit negotiation decision model for flood control scheduling
  • the systematic evaluation indicators constitute a risk set DM 1 and a benefit set DM 2 , and define u 1 (x) and u 2 (x) is the utility earning function of risk and benefit; the utility earning function is as follows:
  • ⁇ i is the weight of the risk or benefit evaluation index
  • r ji is the relative superiority value of the evaluation index j target i
  • the multi-attribute decision-making optimization problem is transformed into a nonlinear programming problem; for the utility gain function, it can form a two-dimensional surface in space; for the risk and benefit constraints, it can form a space in the space. plane; according to the utility earning function and the risk-benefit constraints, the goal is to obtain the maximum value of the utility earning function among the plane and surface nodes, which can be expressed as:
  • Step 2.2.2 calculate the risk benefit weight
  • the risk indicator weights are:
  • the weight of the benefit index is:
  • H k is the entropy value of the evaluation index j
  • Step 2.3 carry out fuzzy comprehensive evaluation calculation
  • Y is the time series evaluation index matrix of the flood control scheduling scheme
  • a and b are the time and the serial number of the evaluation index respectively
  • Step 2.4 define the feature space, correspond each sample to a certain point in the feature space, and determine a sampling ratio N according to the sample imbalance ratio; for each small sample class sample (x, y), find it according to the Euclidean distance.
  • K nearest neighbor samples randomly select a sample point from them, assuming that the selected nearest neighbor point is (x n , y n ), randomly select a point on the line segment between the sample point and the nearest neighbor sample point in the feature space as a new sample point, The following formulas are satisfied:
  • step 3 includes the following steps:
  • step 4 includes the following steps:
  • yi is the output value of the i sample
  • yi ' is the target value of the i sample
  • ⁇ is the Nash correction parameter
  • T is the time
  • the values of weights and parameters are updated by gradient descent to minimize the output error.
  • the training ends.
  • the present invention has the following advantages and beneficial effects:
  • the present invention introduces the time convolution network to optimize the flood control scheduling scheme, establishes the evaluation index system of the flood control scheduling scheme of the reservoir group, and constructs the time series evaluation index matrix combining the comprehensive evaluation index of the flood control scheduling scheme with the time series, and the The matrix is input into the time convolution network to obtain the comprehensive evaluation value of the scheme, and the optimal comprehensive evaluation value is the optimal scheme for flood control scheduling of the reservoir group.
  • the evaluation index system of the constructed flood control scheduling scheme is very large, which increases the complexity of the scheme optimization modeling. Therefore, for the optimization problem of the flood control scheduling scheme for large-scale reservoir groups, the present invention uses the time convolution network to carry out the flood control scheduling scheme.
  • the optimization can greatly reduce the number of parameters of complex models, better explore the relationship between evaluation indicators, and fully consider the optimization process of flood control scheduling schemes under time changes. Based on the idea of transfer learning, fine-tuning technology is widely used, which improves the The optimal evaluation performance of the model.
  • the present invention deeply excavates the characteristics of each evaluation index and the internal relationship between the evaluation indexes, makes different improvements to the calculation formula of the entropy weight of each evaluation index, and is based on fuzzy set theory. and the improved entropy weight method to calculate the comprehensive score of the flood control scheduling scheme, so as to obtain the training samples of the temporal convolutional network.
  • the calculation process of the invention is simple, clear in logic and easy to understand, and overcomes the disadvantages of traditional optimization methods, such as most of the traditional optimization methods only take a single database as the research object, are highly subjective, are affected by the correlation of evaluation indicators, and require experts to give the weights of different indicators.
  • the weights change with the flood control situation, fully considering the time changes in the flood control scheduling program optimization process, and its benefit risk decision-making index may be expected to be caused by the risk (water level or flow exceeding the safety threshold) of the reservoir or flood control control point.
  • the magnitude of the loss is also adequately measured so that the risk of socioeconomic loss can be minimized.
  • the present invention uses a supervised interpolation-based multi-sample data enhancement method (SMOTE) to expand the training samples of the temporal convolutional network, and generate new samples for the small sample class, effectively ensuring the temporal convolutional network.
  • SMOTE supervised interpolation-based multi-sample data enhancement method
  • Training accuracy; the loss function combining mean square error and Nash efficiency coefficient is used to train the time convolution network, which greatly improves the scientificity and rationality of the flood control scheduling scheme selection, and is convenient to be coupled with the decision support system for the flood control scheduling of the reservoir. It provides decision-making support and provides a powerful tool for the comprehensive evaluation and optimization of flood control scheduling schemes for reservoir groups.
  • FIG. 1 is a flow chart of a method according to an embodiment of the present invention.
  • FIG. 2 is a preferred model diagram of a flood control scheduling scheme according to an embodiment of the present invention.
  • FIG. 3 is a structural diagram of a temporal convolutional network according to an embodiment of the present invention.
  • FIG. 4 is a diagram of a temporal convolutional network training process according to an embodiment of the present invention.
  • the evaluation index of the flood control scheduling scheme of the reservoir is a series of measurement sets that measure the pros and cons of the scheduling scheme from different target levels and different evaluation angles, such as reservoirs, flood storage and detention areas, and hydrological stations.
  • the plan evaluation is essentially to use a certain mathematical model to integrate each index value of each plan into a comprehensive evaluation value, and to sort and optimize the plans according to the size of the comprehensive evaluation value.
  • the present invention uses the time convolution network to optimize the scheme, which can greatly reduce the number of parameters of complex models and better explore the relationship between evaluation indicators.
  • a method for optimizing a flood control scheduling scheme based on a temporal convolutional network includes the following steps:
  • Step 1 establish the evaluation index system of the flood control scheduling scheme of the reservoir group.
  • Step 1 specifically includes the following sub-steps:
  • Step 1.2 the specific evaluation index system for flood control dispatching of reservoir groups is as follows:
  • Step 2 constructing the time series evaluation index matrix of the flood control scheduling scheme, the matrix is used as the input of the time convolution network, and the comprehensive score of the training samples of the flood control scheduling scheme is calculated based on the fuzzy set theory and the improved entropy weight method;
  • Temporal convolutional network is the problem of using convolutional network to deal with time series data.
  • its main task is to make the trained temporal convolutional network have the ability to make self-decision under time change.
  • network training samples are generated, and the calculation formula of entropy weight is improved for different evaluation indicators.
  • the input of the temporal convolutional network takes the final comprehensive evaluation value used to judge the pros and cons of the scheme as the output.
  • the comprehensive evaluation value is obtained by the fuzzy comprehensive evaluation method; considering that the number of schemes generated in the actual flood control dispatch is often limited, a supervised interpolation-based multi-sample data enhancement method (SMOTE) is used to evaluate the time
  • SMOTE supervised interpolation-based multi-sample data enhancement method
  • Step 2 specifically includes the following sub-steps:
  • the purpose is to determine the "excellent" membership degree of each scheme to the fuzzy concept, and the scheme with the largest membership degree is the optimal scheme. Then determine the relative degree of membership of the quantitative and qualitative evaluation indicators.
  • the absolute quantity of the evaluation index should be converted into a relative quantity, which is the relative degree of membership, usually each The relative membership of the indicator is a decimal in the interval [0,1].
  • Step 2.1.1 the standardization of quantitative targets.
  • the evaluation indicators involve two types of indicators: benefit type (the bigger the better type) and cost type (the smaller the better type). Different linear scale transformation methods are used for different types of evaluation indicators.
  • the data indicators are normalized and standardized;
  • o ij is the j-th evaluation index value of the ith scheme
  • o jmax and o jmin are the maximum and minimum values of the j-th evaluation index values of each scheme.
  • Step 2.1.2 using the bipolar ratio method to convert qualitative indicators into quantitative indicators.
  • Step 2.2.1 setting the evaluation index weights of reservoirs, flood storage and detention areas, and hydrological stations
  • ⁇ hj H s ⁇ hsj +(1-H s ) ⁇ hkj ;
  • H s is the same part of each entropy value from the first decimal place in the entropy value vector
  • ⁇ hsj , ⁇ hkj are the entropy value separation magnitude weight coefficients
  • ri ij is the relative value of the target i index j.
  • Attribute value, e ij is the relative importance of r ij ;
  • Step 2.2.2 setting the weight of benefit and risk evaluation indicators
  • a risk-benefit negotiation decision-making model for flood control dispatch is constructed; it is assumed that there are w+p comprehensive benefit risk evaluation indicators, including w risk indicators and p benefit indicators.
  • the systematic evaluation index constitutes the risk set DM 1 and the benefit set DM 2 . Define u 1 (x) and u 2 (x) as utility earning functions of risk and benefit;
  • ⁇ i is the weight of the risk or benefit evaluation index
  • r ji is the relative superiority value of the target i of the index j.
  • the multi-attribute decision-making optimization problem is transformed into a nonlinear programming problem.
  • the utility earning function it can form a two-dimensional surface in space, and for the risk and benefit constraints, it can form a plane in space.
  • the goal is to obtain the maximum value of the utility earning function among the plane and surface nodes, which can be expressed as:
  • H k is the entropy value of the evaluation index j.
  • Step 2.3 carry out fuzzy comprehensive evaluation calculation
  • Y is the time series evaluation index matrix of the flood control scheduling scheme
  • a and b are the time and the serial number of the evaluation index, respectively.
  • Step 2.4 using a supervised interpolation-based multi-sample data enhancement method (SMOTE) to augment the training samples of the temporal convolutional network to generate new samples for the small sample class;
  • SMOTE supervised interpolation-based multi-sample data enhancement method
  • Step 2.4.1 define the feature space, correspond each sample to a certain point in the feature space, and determine a sampling ratio N according to the sample imbalance ratio;
  • Step 2.4.2 for each small sample class sample (x, y), find the K nearest neighbor samples according to the Euclidean distance, and randomly select a sample point from them, assuming that the selected nearest neighbor point is (x n , y n ) , randomly select a point on the line segment between the sample point and the nearest neighbor sample point in the feature space as a new sample point, which satisfies the formula:
  • Step 2.4.3 repeat the above steps until the number of large and small samples is balanced.
  • Step 3 Determine the structure of the temporal convolutional network, which consists of an input layer, a causal dilated convolution, an activation function (parameterized ReLU), a residual connection, a fully connected layer and an output layer;
  • step 3 specifically includes the following steps:
  • Step 3.1 causal convolution
  • the temporal convolutional network uses causal convolution to process the input data to calculate and extract the feature information of the underlying data.
  • Step 3.2 dilated convolution
  • the dilation factor d represents the distance between the two elements of the convolution kernel. First, expand the size of the convolution kernel, and then set the elements of the expanded part of the convolution kernel to 0. Set the size of the convolution kernel to 3, and the expansion factor to be 1, 2, and 4 in turn.
  • Step 3.3 use the parameterized ReLU as the activation function
  • the activation function can perform nonlinear operations on the extracted features to increase the fitting ability of the network.
  • the gradient of the ReLU function is 1 when x ⁇ 0 and 0 when x ⁇ 0, and it is difficult to train when encountering convolution kernels less than 0.
  • a parameterized ReLU function is used as the activation function:
  • the parameter a is a learnable variable in the network model, and the optimal value can be automatically obtained in the overall training process of the network model.
  • the residual block contains two nonlinear transformation branches (F 1 , F 2 ).
  • the output of the residual block can be regarded as a linear operation of the residual output F 1 and the baseline output F 2 :
  • Step 3.5 fully connected layer
  • the output of the last residual block is connected to a fully connected layer with a sigmoid activation function.
  • the calculation formula from the fully connected layer to the output is:
  • w out , b out represent the weight matrix and bias, respectively
  • h k is the hidden output tensor of the last residual fast at time step k
  • ⁇ represents the sigmoid function
  • Step 4 using the loss function combining the mean square error and the Nash efficiency coefficient to train the time convolution network
  • step 4 specifically includes the following steps:
  • step 4.1 the training data is selected by random sampling, and the network training samples are used for network training and network testing, respectively, with a ratio of 80% and 20%.
  • the learning and training of temporal convolutional networks includes forward propagation and back propagation stages.
  • Step 4.2 in the forward propagation phase, initialize all filters and parameters/weights with random numbers; input the time series evaluation index matrix combining the comprehensive evaluation index of the flood control scheduling scheme and the time series, and then go through the causal expansion convolution, activation function, The output value is obtained by forward propagation of residual connection and fully connected layer;
  • Step 4.3 in the back-propagation stage, a loss function MSE' that combines mean square error and Nash efficiency coefficient is constructed, and MSE' is used to train the time convolution network, and the error between the output value of the network and the target value is obtained. ;
  • the loss function can be expressed as MSE'(y,y');
  • yi is the output value of the i sample
  • yi ' is the target value of the i sample
  • ⁇ is the Nash correction parameter
  • T is the time
  • Step 5 Input the time series evaluation index matrix of the flood control scheduling scheme into the time convolution network to obtain the comprehensive evaluation value of the scheme, and the optimal comprehensive evaluation value is the optimal scheme for flood control scheduling of the reservoir group.
  • a method for optimizing a flood control scheduling scheme based on a time convolutional network of the present invention can greatly reduce the number of complex model parameters, better explore the relationship between evaluation indicators, and fully consider the flood control scheduling scheme optimization under time changes.
  • the process based on the transfer learning idea, widely uses fine-tuning techniques to improve the optimal evaluation performance of the model.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Software Systems (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Biomedical Technology (AREA)
  • Evolutionary Computation (AREA)
  • General Health & Medical Sciences (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Artificial Intelligence (AREA)
  • Mathematical Physics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Health & Medical Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Strategic Management (AREA)
  • Development Economics (AREA)
  • Educational Administration (AREA)
  • Economics (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Tourism & Hospitality (AREA)
  • Computational Mathematics (AREA)
  • Operations Research (AREA)
  • Quality & Reliability (AREA)
  • Game Theory and Decision Science (AREA)
  • General Business, Economics & Management (AREA)
  • Automation & Control Theory (AREA)
  • Marketing (AREA)
  • Fuzzy Systems (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Feedback Control In General (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本发明公开了一种基于时间卷积网络的防洪调度方案优选方法，包括：建立水库群防洪调度方案的评价指标体系；构建综合评价指标与时间序列的时序评价指标矩阵，该矩阵作为时间卷积网络的输入，基于模糊集理论和改进熵权法计算防洪调度方案训练样本的综合评分；确定时间卷积网络的结构；采用均方误差与纳什效率系数相结合的损失函数训练时间卷积网络；将防洪调度方案的时序评价指标矩阵输入到时间卷积网络中得到方案的综合评判值，综合评判值最优的即为水库群防洪调度的最优方案。本发明能够充分考虑时间变化下的防洪调度方案优选过程，极大的减少复杂模型参数的数量，为水库群防洪调度方案决策及优选提供了有力工具。

Description

一种基于时间卷积网络的防洪调度方案优选方法 技术领域
本发明属于水库防洪调度技术领域，涉及一种基于时间卷积网络的防洪调度方案优选方法。
背景技术
水库群防洪调度具有很强的实践性，受调度目标、来水情况、调度者知识经验等众多因素影响，与社会、经济、自然、生态等因素密切相关，因而水库群防洪调度方案的评价是一个多目标、多属性、多层次的评价指标模型，实时防洪调度又是一个不可逆的实时动态校正过程，对于这样的多指标模型来说，其评价具有影响因素层次多、指标体系构成复杂、很多定性指标难以得到量化等特点，因此，全面、综合地考虑多种影响因素，并构建合理的评价指标体系，选择水库群防洪调度最优方案具有重要的理论意义和实用价值。
在防洪调度方案优选问题上，应用较多的是多方案比选方法，如专家系统评价法、灰色关联度决策法、投影寻踪法、模糊综合评价法、层次分析法(ahp)和优劣解(Topsis)等。
上述方法主观性强、受指标相关性影响、需要专家给出不同指标的权值大小且权值不会随着防洪形势发生改变、未能充分考虑方案优选过程中的时间变化、方法大多仅以单库为研究对象，其效益风险决策指标缺乏对水库或防洪控制点发生风险(水位或流量超过安全阈值)后可能造成的预期损失大小进行衡量。为此，现有方法尚不能准确地对众多可行的调度方案进行决策优选。如何在多重不确定性影响下，对流域大规模水库群联合防洪调度方案进行全方位、多角度的方案智能优选是一项亟待解决的技术难题。
发明内容
本发明目的在于克服现有技术的缺陷，提供了一种基于时间卷积网络的防洪调度方案优选方法，可以在多重不确定性影响下，对流域大规模水库群联合防洪调度方案进行全方位、多角度的方案智能优选。
为了解决上述技术问题，本发明采用以下技术方案。
本发明的一种基于时间卷积网络的防洪调度方案优选方法，包括如下步骤：
步骤1、建立水库群防洪调度方案的评价指标体系；
步骤2、构建防洪调度方案的时序评价指标矩阵，该矩阵作为时间卷积网络的输入，基于模糊集理论和改进熵权法计算防洪调度方案训练样本的综合评分：建立模糊决策矩阵，确定定量与定性评价指标的相对隶属度，得到相对隶属度矩阵，从而构建防洪调度方案的时序评价指标矩阵；将该矩阵作为时间卷积网络的输入，针对不同类型的评价指标，对熵权的计算公式进行改进；基于模糊集理论和改进后的熵权法计算防洪调度方案的综合评判值，将最 终用于评判方案优劣的综合评判值作为输出；其中，所述综合评判值采用模糊综合评判法求得；采用一种有监督的基于插值的多样本数据增强方法SMOTE对时间卷积网络的训练样本进行扩增，为小样本类生成新的样本；
步骤3、确定时间卷积网络的结构，包括输入层、因果膨胀卷积、激活函数、残差连接、全连接层和输出层；
步骤4、采用均方误差与纳什效率系数相结合的损失函数训练时间卷积网络；
步骤5、将防洪调度方案的时序评价指标矩阵输入到时间卷积网络中得到方案的综合评判值，综合评判值最优的即为水库群防洪调度的最优方案。
进一步地，所述步骤2的过程具体包括以下步骤：
步骤2.1，设定水库、蓄滞洪区、水文站评价指标权重；
步骤2.1.1，将评价指标矩阵X＝(x ij) l×q进行归一化处理，得到相对优属度矩阵R＝(r ij) l×q,r ij∈[0,1]；
其中，l表示评价指标，q表示评价目标，i＝1,2,…,q；j＝1,2,…,l；x ij是目标i指标j的特征值；
步骤2.1.2，计算评价指标的熵权ω hj；
ω hj＝H sω hsj+(1-H s)ω hkj
其中，ω hsj，ω hkj均为熵值分离量级权重系数，H s为熵值向量中各熵值自小数点后第一位起的相同部分，r ij为目标i指标j的相对优属度值，H j为r ij所对应的熵值，e ij为r ij的相对重要程度，
Figure PCTCN2021128854-appb-000001
步骤2.2，设定风险及效益评价指标权重；
步骤2.2.1，构建防洪调度风险-效益协商决策模型；
假设综合效益风险评价指标共有w+p个，其中风险评价指标有w个，效益评价指标有p个，系统评价指标构成了风险集DM 1和效益集DM 2，定义u 1(x)和u 2(x)为风险及效益的效用赢得函数；效用赢得函数如下：
Figure PCTCN2021128854-appb-000002
其中，ω i为风险或效益评价指标的权重，r ji为评价指标j目标i的相对优属度值；
由此，多属性决策优选问题转化为非线性规划问题；对于效用赢得函数而言，其可以在空间中形成一个二维的曲面；对于风险、效益约束条件而言，其能够在空间中形成一个平面；根据效用赢得函数和风险效益约束条件可知，目标就是获取到平面和曲面节点的中使得效用赢得函数最大的值，可表示为：
max{F(x)＝[u 1(x),u 2(x)]}
步骤2.2.2，计算风险效益权重；
风险指标权重为：
Figure PCTCN2021128854-appb-000003
效益指标权重为：
Figure PCTCN2021128854-appb-000004
其中i＝1,2,…,l，
Figure PCTCN2021128854-appb-000005
H k为评价指标j的熵值；
步骤2.3，进行模糊综合评价计算；
模糊综合评判模型为
Figure PCTCN2021128854-appb-000006
按照最大隶属度原则对方案进行排序，选出防洪调度方案中的最优方案，即B opt＝max{b j}；将时间卷积网络训练样本表示为：{Y,b j|Y＝(x ab) t×12}；
其中，Y为防洪调度方案的时序评价指标矩阵，a，b分别为时间和评价指标的序号；
步骤2.4，定义特征空间，将每个样本对应到特征空间中的某一点，根据样本不平衡比例确定好一个采样倍率N；对每一个小样本类样本(x,y)，按欧氏距离找出K个最近邻样本，从中随机选取一个样本点，假设选择的近邻点为(x n,y n)，在特征空间中样本点与最近邻样本点的线段上随机选取一点作为新样本点，满足以下公式：
(x new,y new)＝(x,y)+rand(0-1)*((x n-x),(y n-y))
重复以上步骤，直到大、小样本数量平衡。
进一步地，所述步骤3包括以下步骤：
设置因果卷积的卷积核大小为3；膨胀卷积的卷积核大小也为3且膨胀因子依次为1，2，4；采用参数化的ReLU＝max{ax,x}，0<a<1作为激活函数；在残差线上分别有两个膨胀因果卷积层和激活函数，设置堆叠了6个残差块，从左至右残差块的膨胀因子从20到25；最后一个残差块的输出连接了一层带有sigmoid激活函数的全连接层。
进一步地，所述步骤4包括以下步骤：
构造一种均方误差与纳什效率系数相结合的损失函数MSE′，采用MSE′来训练时间卷积网络，将其表示为：
Figure PCTCN2021128854-appb-000007
其中，y i为i样本的输出值，y i′为i样本的目标值，
Figure PCTCN2021128854-appb-000008
为i样本输出值的平均值，α为纳什修正参数，T为时间；
根据求得的误差，用梯度下降法更新权重和参数的值，以使输出误差最小化，当训练迭代次数满足要求且误差小于等于我们的期望值时，结束训练。
与现有技术相比，本发明具有以下优点和有益效果：
1.本发明引入时间卷积网络进行防洪调度方案的优选，建立了水库群防洪调度方案的评价指标体系，构建了防洪调度方案的综合评价指标与时间序列相结合的时序评价指标矩阵，并将该矩阵输入到时间卷积网络中得到方案的综合评判值，综合评判值最优的即为水库群防 洪调度的最优方案。通常，所构建的防洪调度方案的评价指标体系十分庞大，增加了方案优选建模的复杂度，因此，对于大规模水库群防洪调度方案的优选问题，本发明利用时间卷积网络进行防洪调度方案的优选，能够极大减少复杂模型的参数数量，更好的挖掘评价指标之间的关系，并且充分考虑了时间变化下的防洪调度方案优选过程，基于迁移学习思想，广泛使用微调技术，提高了模型的优选评估性能。
2.本发明针对防洪调度方案中不同类型的评价指标，深度挖掘各评价指标的特征及评价指标之间的内在关联，对各评价指标熵权的计算公式进行不同的改进，并且基于模糊集理论和改进后的熵权法计算防洪调度方案的综合评分，从而得到时间卷积网络的训练样本。本发明的计算过程简单、逻辑清晰、易于理解，克服了传统优选方法大多仅以单库为研究对象、主观性强、受评价指标相关性影响、需要专家给出不同指标的权值大小等弊端，且权值随着防洪形势发生改变，充分考虑了防洪调度方案优选过程中的时间变化，其效益风险决策指标对水库或防洪控制点发生风险(水位或流量超过安全阈值)后可能造成的预期损失大小也进行了充分的衡量，从而可以将社会经济损失的风险降到最低。
3.本发明采用一种有监督的基于插值的多样本数据增强方法(SMOTE)对时间卷积网络的训练样本进行扩增，为小样本类生成新的样本，有效保证了时间卷积网络的训练精度；采用均方误差与纳什效率系数相结合的损失函数训练时间卷积网络，大大提高了防洪调度方案优选的科学性和合理性，便于和水库防洪调度决策支持系统相藕合，为决策者提供决策支持，为水库群防洪调度方案综合评价优选提供了有力工具。
附图说明
图1是本发明的一种实施例的方法流程图。
图2是本发明的一种实施例的防洪调度方案优选模型图。
图3是本发明的一种实施例的时间卷积网络结构图。
图4是本发明的一种实施例的时间卷积网络训练过程图。
具体实施方式
水库防洪调度方案的评价指标是一系列从不同的目标层次和不同的评价角度例如水库、蓄滞洪区、水文站等来衡量调度方案优劣的测度集合。而方案评价实质上是采用一定的数学模型将每个方案的各个指标值整合为一个综合评价值，并根据该综合评价值的大小进行方案排序和优选。
研究发现现有水库防洪调度方案评价研究主要侧重于传统评价方法及其改进，在指标的筛选上也存在较大的主观任意性、未考虑指标相关性对方案综合评价结果的影响、需要专家给出不同指标的权值大小且权值不会随着防洪形势改变以及现有研究多针对单一水库，未考虑水库群防洪调度的情景。
为了更精准的对防洪调度方案进行优选，方案综合评价问题显得更为复杂，因为需要为水库群系统中的每个一级评价指标选取二级评价指标，以此类推，最终所构建的指标体系通常十分庞大，这增加了方案优选建模的复杂度。因此，对于大规模水库群防洪调度方案的优选问题，本发明利用时间卷积网络进行方案优选，能够极大减少复杂模型的参数数量，更好的挖掘评价指标之间的关系。
以下结合附图对本发明做进一步详细说明。
如图1和图2所示，本发明的一种实施例的基于时间卷积网络的防洪调度方案优选方法，包括以下步骤：
步骤1，建立水库群防洪调度方案的评价指标体系。
在建立防洪调度评价指标体系时需遵循以下几个原则：(1)客观性，(2)独立性，(3)系统性，(4)可操作性，(5)层次性。传统方法的评价指标数量较少，且未考虑评价指标之间的相关性。根据上述五个原则和传统方法的弊端，综合考虑防洪调度方案的风险及效益，选取水库、蓄滞洪区、水文站作为评价目标来建立防洪调度方案的评价指标体系。
步骤1具体包括以下子步骤：
步骤1.1，评价指标体系为E i＝{e i1,e i2,e i3,…,e ij}，ei j(i＝1,2,…,n；j＝1,2,…,m)为第i个方案的第j个评价指标值。
步骤1.2，水库群防洪调度评价指标体系具体为：
Figure PCTCN2021128854-appb-000009
步骤2，构建防洪调度方案的时序评价指标矩阵，该矩阵作为时间卷积网络的输入，基于模糊集理论和改进熵权法计算防洪调度方案训练样本的综合评分；
时间卷积网络是使用卷积网络来处理时间序列数据的问题，对于水库防洪调度方案优选而言，其主要任务是使得训练后的时间卷积网络具有时间变化下自我决策的能力。基于模糊集理论生成网络训练样本，针对不同评价指标，对熵权的计算公式进行改进，利用改进后的熵权法设定评价指标的权重，构建防洪调度方案的时序评价指标矩阵，该矩阵作为时间卷积网络的输入，将最终用于评判方案优劣的综合评判值作为输出。其中，所述综合评判值采用模糊综合评判法求得；考虑到在实际防洪调度中生成的方案数目往往是有限的，采用一种有监督的基于插值的多样本数据增强方法(SMOTE)对时间卷积网络的训练样本进行扩增，以满足神经网络的训练精度要求。
步骤2具体包括以下子步骤：
步骤2.1，建立模糊决策矩阵O＝(o ij) m×n，O表示m个目标对n个决策方案的目标特征值矩阵。其中，i＝1,2,…,n；j＝1,2,…,m；o ij是方案i目标j的特征值。对于模糊综合评判法而言，目的在于确定每个方案对于模糊概念“优”的隶属度，其中隶属度最大的方案，即为最优方案。然后确定定量与定性评价指标的相对隶属度，为了消除量纲和量纲单位不同所带来的不可公度性，应将评价指标的绝对量转化为相对量，这就是相对隶属度，通常各个指标的相对隶属度为[0,1]区间的小数。
步骤2.1.1，定量目标的规范化，评价指标中涉及效益型(越大越好型)和成本型(越小越好型)两类指标，对不同类型的评价指标采用不同线性刻度变换法，对数据指标进行归一化规范处理；
效益性指标：
Figure PCTCN2021128854-appb-000010
成本型指标：
Figure PCTCN2021128854-appb-000011
式中，o ij为第i个方案第j个评价指标值，o jmax、o jmin为各方案第j个评价指标值中的最大最小值。
步骤2.1.2，采用两极比例方法将定性指标转化为定量指标。对评价指标进行归一化规范处理后，由此确定n个方案对m个评价指标的相对隶属度矩阵R＝(r ij) m×n，其中，r ij(i＝1,2,…,m；j＝1,2,…,n)为第i个方案第j个评价指标的相对优属度。
步骤2.2，针对不同的评价指标，分别采用改进后的熵权法确定权重集，评价指标权重向量为W＝(ω(1),ω(2),…,ω(n))，
Figure PCTCN2021128854-appb-000012
步骤2.2.1，水库、蓄滞洪区、水文站评价指标权重的设定；
构建评价指标矩阵X＝(x ij) l×q，其中，l表示评价指标，q表示评价目标，i＝1,2,…,q；j＝1,2,…,l；x ij是目标i指标j的特征值。
归一化处理后得到相对优属度矩阵R＝(r ij) l×q，r ij∈[0,1]；
计算评价指标的熵值；
Figure PCTCN2021128854-appb-000013
其中，
Figure PCTCN2021128854-appb-000014
计算评价指标的熵权；
ω hj＝H sω hsj+(1-H s)ω hkj；
其中，
Figure PCTCN2021128854-appb-000015
需要注意的是，H s为熵值向量中各熵值自小数点后第一位起的相同部分，ω hsj，ω hkj均为熵值分离量级权重系数，r ij为目标i指标j的相对优属度值，e ij为r ij的相对重要程度；
步骤2.2.2，效益及风险评价指标权重的设定；
构建防洪调度风险-效益协商决策模型；假设综合效益风险评价指标共有w+p个，其中风险指标有w个，效益指标有p个。系统评价指标构成了风险集DM 1和效益集DM 2。定义u 1(x)和u 2(x)为风险及效益的效用赢得函数；
Figure PCTCN2021128854-appb-000016
其中，ω i为风险或效益评价指标的权重，r ji为指标j目标i的相对优属度值。
因此，多属性决策优选问题转化为非线性规划问题，对于效用赢得函数而言，其可以在空间中形成一个二维的曲面，对于风险、效益约束条件而言，其能够在空间中形成一个平面，根据效用赢得函数和风险效益约束条件可知，目标就是获取到平面和曲面节点的中使得效用赢得函数最大的值，可表示为：
max{F(x)＝[u 1(x),u 2(x)]}；
计算风险权重
Figure PCTCN2021128854-appb-000017
和效益权重
Figure PCTCN2021128854-appb-000018
其中，i＝1，2，…,l，
Figure PCTCN2021128854-appb-000019
H k为评价指标j的熵值。
步骤2.3，进行模糊综合评价计算；
模糊综合评判模型为
Figure PCTCN2021128854-appb-000020
按照最大隶属度原则对方案进行排序，选出防洪调度方案中的最优方案，即B opt＝max{b j}；
将时间卷积网络训练样本表示为：{Y,b j|Y＝(x ab) t×12}；
其中，Y为防洪调度方案的时序评价指标矩阵，a,b分别为时间和评价指标的序号。
步骤2.4，采用一种有监督的基于插值的多样本数据增强方法(SMOTE)对时间卷积网络的训练样本进行扩增，为小样本类生成新的样本；
步骤2.4.1，定义特征空间，将每个样本对应到特征空间中的某一点，根据样本不平衡比例确定好一个采样倍率N；
步骤2.4.2，对每一个小样本类样本(x,y)，按欧氏距离找出K个最近邻样本，从中随机 选取一个样本点，假设选择的近邻点为(x n,y n)，在特征空间中样本点与最近邻样本点的线段上随机选取一点作为新样本点，满足公式：
(x new,y new)＝(x,y)+rand(0-1)*((x n-x),(y n-y))；
步骤2.4.3，重复以上步骤，直到大、小样本数量平衡。
步骤3，确定时间卷积网络的结构，由输入层、因果膨胀卷积、激活函数(参数化ReLU)、残差连接、全连接层和输出层组成；
结合附图3，步骤3具体包括以下步骤：
步骤3.1，因果卷积；
时间卷积网络使用因果卷积对输入数据进行处理，用以计算并提取底层数据的特征信息。在任意时刻t，进行卷积核大小为3的单次卷积操作，滤波器F＝(f 1,f 2,…,f k)，序列X＝(x 1,x 2,…,x t)，在x t处的因果卷积过程可表示为：
Figure PCTCN2021128854-appb-000021
步骤3.2，膨胀卷积；
设置不同大小的膨胀卷积，膨胀因子d表示卷积核两个元素之间的距离，先扩充卷积核的大小，然后将卷积核扩展部分的元素设置为0。设置卷积核的大小为3，膨胀因子依次为1，2，4。滤波器F＝(f 1,f 2,…,f k)，序列X＝(x 1,x 2,…,x t)，膨胀因子为d，在x t处的膨胀卷积过程可表示为：
Figure PCTCN2021128854-appb-000022
步骤3.3，使用参数化的ReLU作为激活函数；
激活函数可以对提取的特征进行非线性操作，增加网络的拟合能力。ReLU函数的梯度在x≥0时为1，在x<0时为0，在遇到小于0的卷积核时训练困难。为了解决上述缺陷，使用参数化的ReLU函数作为激活函数：
ReLU(x)＝max{ax,x}，0<a<1
其中，参数a作为一个网络模型中可学习的变量，在网络模型的整体训练过程中可以自动求得最优值。
步骤3.4，残差连接；
在残差线上分别有两个膨胀因果卷积层和激活函数，设置堆叠了6个残差块，从左至右残差块的膨胀因子从20到25，每个残差块的基线都通过跳连连接到最后一个残差块的输出上并进行张量相加，有利于网络在任意残差模块后进行恒等学习，最大程度地缓和了网络退化问题。残差块中包含两个非线形变换分支(F 1,F 2)，残差块输出可以看作残差的输出F 1与基线输出F 2的线性运算：
o＝F 1(x)+F 2(x)；
步骤3.5，全连接层；
最后一个残差块的输出连接了一层带有sigmoid激活函数的全连接层。全连接层到输出的计算公式为：
Figure PCTCN2021128854-appb-000023
其中，w out，b out分别表示权重矩阵和偏置，h k是时间步长k的最后一个残差快的隐藏输出张量，σ表示sigmoid函数。
步骤4，采用均方误差与纳什效率系数相结合的损失函数训练时间卷积网络；
结合附图4，步骤4具体包括以下步骤：
步骤4.1，训练数据按照随机抽样方式抽取，将所述网络训练样本分别用于网络训练和网络检验，比例为80％，20％。时间卷积网络的学习训练包括前向传播和反向传播阶段。
步骤4.2，前向传播阶段，用随机数初始化所有的滤波器和参数/权重；输入防洪调度方案的综合评价指标与时间序列相结合的时序评价指标矩阵，依次经过因果膨胀卷积、激活函数、残差连接、全连接层的前向传播得到输出值；
步骤4.3，反向传播阶段，构造了一种均方误差与纳什效率系数相结合的损失函数MSE′，采用MSE′来训练时间卷积网络，求出网络的输出值与目标值之间的误差；
损失函数可表示为MSE′(y,y′)；
Figure PCTCN2021128854-appb-000024
其中，y i为i样本的输出值，y i′为i样本的目标值，
Figure PCTCN2021128854-appb-000025
为i样本输出值的平均值，α为纳什修正参数，T为时间；
根据求得误差，用梯度下降法更新所有的滤波器/权重和参数的值，以使输出误差最小化，当训练迭代次数满足要求且误差小于等于我们的期望值时，结束训练，时间卷积网络达到收敛。
步骤5，将防洪调度方案的时序评价指标矩阵输入到时间卷积网络中，得到方案的综合评判值，综合评判值最优的即为水库群防洪调度的最优方案。
总之，本发明一种基于时间卷积网络的防洪调度方案优选方法，能够极大地减少复杂模型参数的数量，更好的挖掘评价指标之间的关系，充分考虑了时间变化下的防洪调度方案优选过程，基于迁移学习思想，广泛使用微调技术，提高了模型的优选评估性能。

Claims (4)

  1. 一种基于时间卷积网络的防洪调度方案优选方法，其特征在于，包括如下步骤：
    步骤1、建立水库群防洪调度方案的评价指标体系；
    步骤2、构建防洪调度方案的时序评价指标矩阵，该矩阵作为时间卷积网络的输入，基于模糊集理论和改进熵权法计算防洪调度方案训练样本的综合评分：建立模糊决策矩阵，确定定量与定性评价指标的相对隶属度，得到相对隶属度矩阵，从而构建防洪调度方案的时序评价指标矩阵；将该矩阵作为时间卷积网络的输入，针对不同类型的评价指标，对熵权的计算公式进行改进；基于模糊集理论和改进后的熵权法计算防洪调度方案的综合评判值，将最终用于评判方案优劣的综合评判值作为输出；其中，所述综合评判值采用模糊综合评判法求得；采用一种有监督的基于插值的多样本数据增强方法SMOTE对时间卷积网络的训练样本进行扩增，为小样本类生成新的样本；
    步骤3、确定时间卷积网络的结构，包括输入层、因果膨胀卷积、激活函数、残差连接、全连接层和输出层；
    步骤4、采用均方误差与纳什效率系数相结合的损失函数训练时间卷积网络；
    步骤5、将防洪调度方案的时序评价指标矩阵输入到时间卷积网络中得到方案的综合评判值，综合评判值最优的即为水库群防洪调度的最优方案。
  2. 根据权利要求1所述的一种基于时间卷积网络的防洪调度方案优选方法，其特征在于，所述步骤2的过程具体包括以下步骤：
    步骤2.1，设定水库、蓄滞洪区、水文站评价指标权重；
    步骤2.1.1，将评价指标矩阵X＝(x ij) l×q进行归一化处理，得到相对优属度矩阵R＝(r ij) l×q,r ij∈[0,1]；
    其中，l表示评价指标，q表示评价目标，i＝1,2,…,q；j＝1,2,…,l；x ij是目标i指标j的特征值；
    步骤2.1.2，计算评价指标的熵权ω hj；
    ω hj＝H sω hsj+(1-H s)ω hkj
    其中，ω hsj，ω hkj均为熵值分离量级权重系数，H s为熵值向量中各熵值自小数点后第一位起的相同部分，r ij为目标i指标j的相对优属度值，H j为r ij所对应的熵值，e ij为r ij的相对重要程度，
    Figure PCTCN2021128854-appb-100001
    步骤2.2，设定风险及效益评价指标权重；
    步骤2.2.1，构建防洪调度风险-效益协商决策模型；
    假设综合效益风险评价指标共有w+p个，其中风险评价指标有w个，效益评价指标有p个，系统评价指标构成了风险集DM 1和效益集DM 2，定义u 1(x)和u 2(x)为风险及效益的效用赢得函数；效用赢得函数如下：
    Figure PCTCN2021128854-appb-100002
    其中，ω i为风险或效益评价指标的权重，r ji为评价指标j目标i的相对优属度值；
    由此，多属性决策优选问题转化为非线性规划问题；对于效用赢得函数而言，其可以在空间中形成一个二维的曲面；对于风险、效益约束条件而言，其能够在空间中形成一个平面；根据效用赢得函数和风险效益约束条件可知，目标就是获取到平面和曲面节点的中使得效用赢得函数最大的值，可表示为：
    max{F(x)＝[u 1(x),u 2(x)]}
    步骤2.2.2，计算风险效益权重；
    风险指标权重为：
    Figure PCTCN2021128854-appb-100003
    效益指标权重为：
    Figure PCTCN2021128854-appb-100004
    其中i＝1,2,…,l，
    Figure PCTCN2021128854-appb-100005
    H k为评价指标j的熵值；
    步骤2.3，进行模糊综合评价计算；
    模糊综合评判模型为
    Figure PCTCN2021128854-appb-100006
    按照最大隶属度原则对方案进行排序，选出防洪调度方案中的最优方案，即B opt＝max{b j}；将时间卷积网络训练样本表示为：{Y,b j|Y＝(x ab) t×12}；
    其中，Y为防洪调度方案的时序评价指标矩阵，a，b分别为时间和评价指标的序号；
    步骤2.4，定义特征空间，将每个样本对应到特征空间中的某一点，根据样本不平衡比例确定好一个采样倍率N；对每一个小样本类样本(x,y)，按欧氏距离找出K个最近邻样本，从中随机选取一个样本点，假设选择的近邻点为(x n,y n)，在特征空间中样本点与最近邻样本点的线段上随机选取一点作为新样本点，满足以下公式：
    (x new,y new)＝(x,y)+rand(0-1)*((x n-x),(y n-y))
    重复以上步骤，直到大、小样本数量平衡。
  3. 根据权利要求1所述的一种基于时间卷积网络的防洪调度方案优选方法，其特征在于，所述步骤3包括以下步骤：
    设置因果卷积的卷积核大小为3；膨胀卷积的卷积核大小也为3且膨胀因子依次为1，2，4；采用参数化的ReLU＝max{ax,x}，0<a<1作为激活函数；在残差线上分别有两个膨胀因果卷积层和激活函数，设置堆叠了6个残差块，从左至右残差块的膨胀因子从20到25；最后一个残差块的输出连接了一层带有sigmoid激活函数的全连接层。
  4. 根据权利要求1所述的一种基于时间卷积网络的防洪调度方案优选方法，其特征在于，所述步骤4进一步包括以下步骤：
    构造一种均方误差与纳什效率系数相结合的损失函数MSE′，采用MSE′来训练时间卷积网 络，将其表示为：
    Figure PCTCN2021128854-appb-100007
    其中，y i为i样本的输出值，y i′为i样本的目标值，
    Figure PCTCN2021128854-appb-100008
    为i样本输出值的平均值，α为纳什修正参数，T为时间；
    根据求得的误差，用梯度下降法更新权重和参数的值，以使输出误差最小化，当训练迭代次数满足要求且误差小于等于我们的期望值时，结束训练。
PCT/CN2021/128854 2021-03-15 2021-11-05 一种基于时间卷积网络的防洪调度方案优选方法 WO2022193681A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US17/909,003 US20230196076A1 (en) 2021-03-15 2021-11-05 Method for optimally selecting flood-control operation scheme based on temporal convolutional network

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110289848.0A CN112966954B (zh) 2021-03-15 2021-03-15 一种基于时间卷积网络的防洪调度方案优选方法
CN202110289848.0 2021-03-15

Publications (1)

Publication Number Publication Date
WO2022193681A1 true WO2022193681A1 (zh) 2022-09-22

Family

ID=76279122

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/128854 WO2022193681A1 (zh) 2021-03-15 2021-11-05 一种基于时间卷积网络的防洪调度方案优选方法

Country Status (3)

Country Link
US (1) US20230196076A1 (zh)
CN (1) CN112966954B (zh)
WO (1) WO2022193681A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116187745A (zh) * 2022-12-12 2023-05-30 安徽继远软件有限公司 一种基于态势感知的电网风险评估方法
CN116630826B (zh) * 2023-07-25 2023-10-03 中国石油大学(华东) 基于卫星遥感观测的甲烷排放重点区域识别定位方法
CN117933712A (zh) * 2024-01-25 2024-04-26 中国水利水电科学研究院 一种基于模糊云的水库调度方案风险评估方法

Families Citing this family (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112966954B (zh) * 2021-03-15 2022-10-18 河海大学 一种基于时间卷积网络的防洪调度方案优选方法
CN116862231A (zh) * 2023-07-07 2023-10-10 上海福智宠物食品有限公司 一种宠食生产全链路的风险评估系统及方法
CN116882696B (zh) * 2023-07-18 2024-03-08 黄河水利委员会黄河水利科学研究院 一种耦合模糊识别及概率推理的水库群防洪安全双层判别方法
CN117010575B (zh) * 2023-09-27 2023-12-19 浙江远算科技有限公司 基于数字孪生的水库群多目标优化调度仿真方法以及系统
CN117057335B (zh) * 2023-10-12 2024-01-05 中国水利水电科学研究院 一种流域防洪四预方案智能化生成方法
CN117093873B (zh) * 2023-10-19 2024-01-30 国网浙江省电力有限公司丽水供电公司 基于天然来水量预测的水电站库容能力评估方法和系统
CN117479510B (zh) * 2023-12-28 2024-03-08 海斯福(深圳)科技有限公司 基于氟化液性能退化评价的冷却系统智能控制方法及系统
CN118014325B (zh) * 2024-04-09 2024-05-31 水利部水利水电规划设计总院 水网系统中调蓄工程调度规则提取方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070079652A1 (en) * 2005-10-07 2007-04-12 Craig David P Methods and systems for determining reservoir properties of subterranean formations
CN110033164A (zh) * 2019-03-04 2019-07-19 华中科技大学 一种水库群联合防洪调度的风险评估与决策方法
CN111191559A (zh) * 2019-12-25 2020-05-22 国网浙江省电力有限公司泰顺县供电公司 基于时间卷积神经网络的架空线预警系统障碍物识别方法
CN112966954A (zh) * 2021-03-15 2021-06-15 河海大学 一种基于时间卷积网络的防洪调度方案优选方法

Family Cites Families (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6330553B1 (en) * 1997-04-09 2001-12-11 Yamaha Hatsudoki Kabushiki Kaisha Autonomic system for updating fuzzy neural network and control system using the fuzzy neural network
CN105046377A (zh) * 2015-09-06 2015-11-11 河海大学 基于bp神经网络筛选水库防洪调度方案优选指标的方法
CN108875161B (zh) * 2018-05-31 2022-11-29 长江勘测规划设计研究有限责任公司 基于卷积神经网络深度学习的流量等级预测方法
US20200160565A1 (en) * 2018-11-19 2020-05-21 Zhan Ma Methods And Apparatuses For Learned Image Compression
EP3848836A1 (en) * 2020-01-07 2021-07-14 Robert Bosch GmbH Processing a model trained based on a loss function
US11790565B2 (en) * 2021-03-04 2023-10-17 Snap Inc. Compressing image-to-image models with average smoothing

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20070079652A1 (en) * 2005-10-07 2007-04-12 Craig David P Methods and systems for determining reservoir properties of subterranean formations
CN110033164A (zh) * 2019-03-04 2019-07-19 华中科技大学 一种水库群联合防洪调度的风险评估与决策方法
CN111191559A (zh) * 2019-12-25 2020-05-22 国网浙江省电力有限公司泰顺县供电公司 基于时间卷积神经网络的架空线预警系统障碍物识别方法
CN112966954A (zh) * 2021-03-15 2021-06-15 河海大学 一种基于时间卷积网络的防洪调度方案优选方法

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116187745A (zh) * 2022-12-12 2023-05-30 安徽继远软件有限公司 一种基于态势感知的电网风险评估方法
CN116187745B (zh) * 2022-12-12 2023-11-14 安徽继远软件有限公司 一种基于态势感知的电网风险评估方法
CN116630826B (zh) * 2023-07-25 2023-10-03 中国石油大学(华东) 基于卫星遥感观测的甲烷排放重点区域识别定位方法
CN117933712A (zh) * 2024-01-25 2024-04-26 中国水利水电科学研究院 一种基于模糊云的水库调度方案风险评估方法

Also Published As

Publication number Publication date
US20230196076A1 (en) 2023-06-22
CN112966954A (zh) 2021-06-15
CN112966954B (zh) 2022-10-18

Similar Documents

Publication Publication Date Title
WO2022193681A1 (zh) 一种基于时间卷积网络的防洪调度方案优选方法
CN104463381B (zh) 基于kpca与wlssvm的建筑能耗预测方法
CN112232561B (zh) 基于约束并行lstm分位数回归的电力负荷概率预测方法
WO2021139753A1 (zh) 一种数据库的处理方法和装置
CN113190688A (zh) 基于逻辑推理和图卷积的复杂网络链接预测方法及系统
CN114777192A (zh) 基于数据关联和深度学习的二级网供热自主优化调控方法
CN112036651A (zh) 基于量子免疫优化bp神经网络算法的电价预测方法
CN111832839A (zh) 基于充分增量学习的能耗预测方法
Sun et al. Research on prediction of housing prices based on GA-PSO-BP neural network model: evidence from Chongqing, China
CN113128666A (zh) 基于Mo-S-LSTMs模型的时间序列多步预测方法
CN116632834A (zh) 一种基于SSA-BiGRU-Attention的短期电力负荷预测方法
CN116258234A (zh) 一种基于bp神经网络模型的能源企业碳排测算与预测方法
Wang et al. Digital Management Mode of Enterprise Human Resources under the Background of Digital Transformation
Zhang et al. Prediction of stock closing prices based on attention mechanism
CN113836174A (zh) 基于强化学习dqn算法的异步sql连接查询优化方法
Ying et al. A novel optimization algorithm for BP neural network based on RS-MEA
Xue The prediction on residential real estate price based on BPNN
Yizhen et al. The forecasting of shanghai index trend based on genetic algorithm and back propagation artificial neural network algorithm
Liu The analysis of differential high-quality development of economy by deep neural network and internet of things
Zhiyuan et al. Research on the evaluation of enterprise competitiveness based on the wavelet neural network forecasting system
Wang et al. [Retracted] Construction Principles of Physical Fitness Training Objective System Based on Machine Learning and Data Mining
CN104820691A (zh) 一种交通流预测中数据库设计方法及其查询向量得出方法
CN114936296B (zh) 超大规模知识图谱存储的索引方法、系统及计算机设备
Li et al. A Study on House Price Prediction Based on Stacking-Sorted-Weighted-Ensemble Model
CN117648646B (zh) 基于特征选择和堆叠异构集成学习的钻采成本预测方法

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: 21931256

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: 21931256

Country of ref document: EP

Kind code of ref document: A1