WO2023142424A1 - 基于gru-lstm神经网络的电力金融业务风控方法及*** - Google Patents

基于gru-lstm神经网络的电力金融业务风控方法及*** Download PDF

Info

Publication number
WO2023142424A1
WO2023142424A1 PCT/CN2022/110593 CN2022110593W WO2023142424A1 WO 2023142424 A1 WO2023142424 A1 WO 2023142424A1 CN 2022110593 W CN2022110593 W CN 2022110593W WO 2023142424 A1 WO2023142424 A1 WO 2023142424A1
Authority
WO
WIPO (PCT)
Prior art keywords
risk
data
gru
neural network
financial
Prior art date
Application number
PCT/CN2022/110593
Other languages
English (en)
French (fr)
Inventor
郑祥华
毕晓蓉
张�杰
李�杰
王惠
王俊康
徐中中
Original Assignee
国网江苏省电力有限公司南京供电分公司
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 国网江苏省电力有限公司南京供电分公司 filed Critical 国网江苏省电力有限公司南京供电分公司
Publication of WO2023142424A1 publication Critical patent/WO2023142424A1/zh

Links

Images

Classifications

    • 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
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • 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
    • G06N3/044—Recurrent networks, e.g. Hopfield 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
    • G06N3/08—Learning methods
    • 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
    • G06Q10/0635—Risk analysis of enterprise or organisation activities
    • 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
    • G06Q50/06—Energy or water supply
    • 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
    • Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
    • Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
    • Y04S10/00—Systems supporting electrical power generation, transmission or distribution
    • Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications

Definitions

  • the invention belongs to the technical field of data analysis, relates to risk analysis of financial business, and is a method and system for risk control of electric power financial business based on GRU-LSTM neural network.
  • Deep learning technology is the process of digging out potential and valuable information from massive raw data. It is a data processing technology. It can be known from its definition: raw data must be real, objective and noisy, and finally mining The information and knowledge that come out are interesting, acceptable and easy to understand for users, and can also be used in real life.
  • an effective analysis tool or algorithm is selected, and after being decomposed into a multivariate data set, an in-depth mining analysis is performed to provide users with potentially valuable information and knowledge. How to apply deep learning technology to power financial business and improve the ability to analyze business data is a problem to be solved.
  • the purpose of the present invention is to aim at the risk control problem of financial business, analyze the data through deep learning technology, combine the advantages of short prediction time of GRU and high prediction accuracy of LSTM, combine the two, thus propose a neural network based on GRU-LSTM Effectively control the risk control system of the electric power financial business.
  • the technical solution of the present invention is: based on the GRU-LSTM neural network risk control method for electric power financial business, data analysis is performed on the historical data of electric power financial business through deep learning, the risk factors of electric power financial business are obtained, and the risk early warning index system is established according to the risk factors , build a risk early warning model;
  • deep learning is GRU-LSTM network combination prediction: first analyze the financial business data through the GRU network, and then input the LSTM neural network, use the LSTM neural network to simulate risk control and predict risk factors, according to Risk early warning indicators determine the corresponding risk level; use historical data to train the GRU-LSTM network to obtain a risk early warning model, use the risk early warning model to predict the risk level of real-time financial business data, and realize financial business risk control.
  • the present invention comprises the following steps:
  • Step 1 data collection: by monitoring the financial business data, recording the financial data, and collecting the required target financial business historical data from the relevant financial project business according to the financial business risk to be analyzed by deep learning, and training the GRU network for data analysis;
  • Step 2 comprehensive analysis, risk assessment: According to the data analyzed by GRU, the LSTM neural network is used to simulate risk control, and the LSTM neural network is trained to predict and judge the risk factors and risk levels corresponding to financial data through historical data;
  • Step 3 real-time detection, risk prevention and control: real-time monitoring of financial business data, input into the trained GRU-LSTM network, output predicted risk factors and risk levels, and realize financial business risk control.
  • the electric power financial business is collected, and information is randomly extracted from the financial database, business database and human database.
  • the present invention also proposes a wind control system for electric power financial business based on GRU-LSTM neural network, which has a computer-readable storage medium, and a computer program is configured in the computer-readable storage medium, and the above-mentioned risk early warning model is realized when the computer program is executed .
  • the invention can assist the power company to monitor some financial business risk problems when entering the financial market, effectively carry out risk control, and make reasonable business allocation decisions.
  • the method of the present invention uses deep learning technology to learn and analyze complex financial problem data, and analyzes related business information to analyze the relationship between business and problem data, which is conducive to avoiding risks and helping to improve decision-making management capabilities.
  • the GRU network structure is simple so that the parameters are easier to converge, greatly reducing the training time, but the accuracy of GRU is not as good as LSTM, and then through LSTM, the prediction accuracy is greatly improved.
  • the proposed GRU -LSTM combination forecasting can be applied to the risk analysis of power financial business, explore data processing and solutions, and provide a practical method for the prevention of financial risks.
  • Fig. 1 is a schematic structural diagram of the wind control system for electric power financial business of the present invention.
  • Fig. 2 is a schematic diagram of the deep learning process of the present invention.
  • Fig. 3 is a schematic flow chart of the LSTM neural network prediction model adopted in the present invention.
  • the present invention aims at the financial business risks faced by current electric power companies when they go deep into the financial market, combines deep learning technology to carry out risk control, takes the application of deep learning technology in financial business as the core, and conducts in-depth research on complex financial models, thereby helping The extension of power finance business.
  • Use deep learning technology to analyze relevant business information, analyze the stakes, improve internal decision-making management, and provide a scientific and powerful foundation for the efficient development of power financial business by mining valuable data.
  • the theoretical basis of the risk control system of the present invention includes absolute income theory and life cycle theory, etc., which are mainly used to help judge the risk of the electric power financial business, assist the company in making decisions, and when the revenue reaches the maximum value, it is the optimal business performance. Selection enables reasonable allocation of business and provides directions for finding potential benefits.
  • the present invention proposes a power financial business risk control system based on the GRU-LSTM neural network.
  • GRU-LSTM neural network As shown in Figure 1, it is first necessary to randomly extract information from the financial database, business database, and human database, and obtain relevant financial and financial information through the business database.
  • Manpower information, etc., data processing using the advantages of short prediction time of GRU and high precision of LSTM, the combination of the two is the most appropriate for processing.
  • the simple network structure of GRU makes the parameters easier to converge, greatly reducing the training time.
  • the accuracy of GRU is not as good as that of LSTM.
  • After LSTM the prediction accuracy is greatly improved.
  • the processed data can be stored and continue to provide the basis for future training. Diagnose the problem after processing the data, and judge the output results.
  • a first-level alarm will be issued and specific abnormal data will be reported; if 3 or 4 parameters are abnormal, then Issue a second-level alarm and report specific abnormal data; if 5 or more parameters are abnormal, issue a third-level alarm and report specific abnormal data.
  • the computer performs two or more simulations to ensure that there will be no second data abnormality, and to a certain extent achieve the effect of risk control.
  • the process of deep learning of power financial business data by the power financial business risk control system based on GRU-LSTM neural network is shown in Figure 2.
  • Data collection is collected from financial business, and risk feature description is obtained through data mining.
  • Financial business defines risk, combines risk definition with risk characteristic description, and comprehensively determines actual risk factors.
  • the present invention realizes following steps:
  • Step 1 data collection: through the data monitoring of financial business, record the financial data, and according to the theory of deep learning, collect the required target financial data from the relevant financial project business, so as to understand it more accurately, Master and use the financial operation rules to prevent and control risks and crises, making the collected data more accurate and effective.
  • the present invention first needs to randomly extract information from the financial database, business database, and manpower database, and obtain relevant financial and manpower information through the business database, and perform data processing on it.
  • the GRU network is first trained for data analysis.
  • Step 2 comprehensive analysis, risk assessment: According to the data analyzed by GRU, the LSTM neural network is used to simulate risk control, and the LSTM neural network is trained to predict and judge the risk factors and risk levels corresponding to financial data through historical data;
  • Step 3 real-time detection, risk prevention and control: real-time monitoring of financial business data, input into the trained GRU-LSTM network, output predicted risk factors and risk levels, and realize financial business risk control.
  • the data that has been analyzed and processed by the deep learning of the GRU network is put into the LSTM neural network for simulation analysis, the risk level is divided, and the final risk level is judged according to the simulation analysis results, providing managers with scientific and effective judgment basis, so as to advance Risk avoidance, some preventive suggestions for the assessment and control of power financial business risks, and has certain reliability.
  • FIG. 3 it is a schematic flow chart of the LSTM predictive analysis of the present invention.
  • an index that reflects financial risks such as time, number of people, progress, finance, etc.
  • the number of nodes is 12, and the data needs to be normalized before inputting parameters to become a standard value between 0 and 1.
  • the normalization equation is as follows:
  • x max and x min are the maximum and minimum values of the input sample parameters respectively
  • x t and x′ t are the real initial input sample parameters and normalized processing results respectively.
  • n the number of indicators, that is, the input parameters
  • n the number of indicators, that is, the input parameters
  • the number of neurons contained in the hidden layer is assumed to be a j
  • the weight from the input layer to the hidden layer is w i
  • the number of neurons contained in the hidden layer is j
  • the transfer function in the risk prediction model is as follows:
  • the input unit ⁇ i of the hidden layer can be described as:
  • the data After calculation and screening, the data reaches the output layer, and the input item s t and output item ⁇ t of the neurons in the output layer are respectively:
  • v jt is the weight from the hidden layer to the output layer
  • ⁇ t is the threshold of the output layer.
  • 12 parameters are input, including time, number of people, workload, progress, etc., and judged by the results output by the model. If one or two parameters are abnormal, a first-level alarm will be issued, and the specific Abnormal data; if 3 or 4 parameters are abnormal, a second-level alarm will be issued, and specific abnormal data will be reported; if 5 or more parameters are abnormal, a third-level alarm will be issued, specific abnormal data will be reported, and the administrator Perform a second simulation after modifying the business data, so as to ensure that the business data will not have a second abnormality.
  • the present invention proposes a power financial business risk control system based on GRU-LSTM neural network, which has a computer-readable storage medium, and a computer program is configured in the computer-readable storage medium.
  • the computer program is executed, the above-mentioned risk warning model.

Landscapes

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

Abstract

基于GRU-LSTM神经网络的电力金融业务风控方法及***，通过深度学习对电力金融业务历史数据进行数据分析，得到电力金融业务的风险因素，根据风险因素建立风险预警指标体系，构建风险预警模型，其中深度学习为GRU-LSTM组合预测：先将金融业务数据通过GRU网络进行分析，再放入LSTM网络进行模拟分析。本发明方法利用深度学习技术对复杂的金融问题数据进行学习分析，并对相关的业务信息进行分析，剖析业务与问题数据之间的关联，有利于规避风险，帮助提升决策管理能力。

Description

基于GRU-LSTM神经网络的电力金融业务风控方法及*** 技术领域
本发明属于数据分析技术领域，涉及金融业务的风险分析，为一种基于GRU-LSTM神经网络的电力金融业务风控方法及***。
背景技术
随着经济水平不断提高，金融业务量急速增长，金融市场规模会越来越大，潜力巨大，但同时也会带来金融风险，如何进行风险控制是当前备受瞩目的问题。
电力金融业务的发展与数据管理息息相关，随着不断市场化、信息化与数字化建设，大量数据涌入金融业务，需要从中挖掘出有价值的信息，对其复杂的形式进行深入探讨，从而帮助用户分析其中利弊，推进电力金融业务全面升级。
金融业务的数据与业务资源类型复杂且海量，虽然可以通过一些规则进行整理归纳，但仍需管理者通过主观分析判断，所做出的决策往往具有一定的局限性，从而无法准确有效地处理金融业务风险。近几年随着计算机人工智能技术的发展，利用机器学习辅助处理金融业务风险越来越常见。深度学习技术是从海量的原始数据中挖掘出潜在、有价值的信息的过程，它是一种数据处理技术，通过其定义可以得知：原始数据必须是真实、客观且有噪声的，最后挖掘出来的信息和知识是用户感兴趣、可接受、易理解的，也能够在现实生活中运用的。收集过来的数据进行处理后选择有效的分析工具或者算法，分解为多元数据集之后，进行深入挖掘分析，从而为用户提供潜在的有价值的信息与知识。如何将深度学习技术应用到电力金融业务中，提高对业务数据的分析能力，是一个函待解决的问题。
发明内容
本发明的目的是针对金融业务的风控问题，通过深度学习技术对数据进行分析，结合GRU预测时间短与LSTM预测精度高的优点，两者相结合，从而提出一种基于GRU-LSTM神经网络的电力金融业务风控***进行有效控制。
本发明的技术方案为：基于GRU-LSTM神经网络的电力金融业务风控方法，通过深度学习对电力金融业务历史数据进行数据分析，得到电力金融业务的风险因素，根据风险因素建立风险预警指标体系，构建得到风险预警模型；其中深度学习为GRU-LSTM网络组合预测：先将金融业务数据通过GRU网络进行分析，再输入LSTM神经网络，利用LSTM神经网络对风 险控制进行模拟，预测风险因素，根据风险预警指标判断其对应的风险等级；利用历史数据训练GRU-LSTM网络得到风险预警模型，利用风险预警模型对实时金融业务数据进行风险等级预测，实现金融业务风控。
作为优选方式，本发明包括以下步骤：
步骤一，数据收集：通过对金融业务的数据进行监测，记录其中的金融数据，根据深度学习所要分析的金融业务风险，从相关的金融项目业务中搜集出所需要的目标金融业务历史数据，训练GRU网络进行数据分析；
步骤二，综合分析，风险评定：根据GRU分析的数据，利用LSTM神经网络对风险控制进行模拟，通过历史数据训练LSTM神经网络预测判断金融数据所对应的风险因素及风险等级；
步骤三，实时检测，风险防控：实时监测金融业务数据，输入训练好的GRU-LSTM网络，输出预测的风险因素及风险等级，实现金融业务风控。
进一步的，采集电力金融业务，对财务数据库、业务数据库和人力数据库进行随机抽取信息。
本发明还提出基于GRU-LSTM神经网络的电力金融业务风控***，其具有计算机可读存储介质，计算机可读存储介质中配置有计算机程序，所述计算机程序被执行时实现上述的风险预警模型。
本发明能够协助电力公司在进入金融市场时监控一些金融业务风险问题，有效地进行风险控制，对业务进行合理分配决策。本发明方法利用深度学习技术对复杂的金融问题数据进行学习分析，并对相关的业务信息进行分析，剖析业务与问题数据之间的关联，有利于规避风险，帮助提升决策管理能力。
本发明具有以下优点：
1、由于金融数据的类型多样、数量大、关系复杂、动态波动等特点，并且多是大量样本、高维度的数据，利用GRU预测时间短与LSTM精度高的优点，两者相结合对其进行处理最为妥当，GRU网络结构简单从而使得参数更容易收敛，大大减少了训练时间，但是GRU精度不如LSTM，再经过LSTM，使得预测精度大大提高，考虑到预测精度与预测速度相结合，提出的GRU-LSTM组合预测，可以将其应用于电力金融业务的风险分析，探索数据处理和解决方法，为金融风险的预防提供了切实可行的方法。
2、风险判断可靠。通过深度学习分析每个数据，并且从大量的数据中寻找潜在的规律，因此具有一定的可靠性，将搜集的数据进行处理后，对电力金融业务风险评估提供了可靠的基 础，并且随着数据量的不断增加，其中挖掘的有价值的信息也会逐渐增加，最终结果也会越来越准确。
3、风险控制及时。通过深度学习处理数据的反馈，管理者可以快速准确的定位电力金融业务风险的要素，并且根据其要素分析其中的问题所在，并可通过风险预警模型对调整的业务进行模拟预测，验证决策的有效性及可靠性，从而在最短的时间内帮助管理者做出有效的决策，大大降低了可能面临金融业务的风险。
附图说明
图1为本发明电力金融业务风控***的结构示意图。
图2为本发明的深度学习流程示意图。
图3为本发明采用的LSTM神经网络预测模型的流程示意图。
具体实施方式
本发明针对当前电力公司深入金融市场所面临的金融业务风险，结合深度学习技术对其进行风险控制，以深度学习技术在金融业务的应用为核心，对形式复杂的金融模式进行深入研究，从而帮助电力金融业务的延拓。利用深度学习技术分析相关业务信息，剖析其中利害关系，提升内部决策管理，并通过挖掘其中有价值的数据，为电力金融业务高效发展提供科学有力的基础。
本发明风控***的理论基础包括绝对收入理论与生命周期理论等，主要用于帮助评判该电力金融业务的风险，辅助公司进行决策，在收益达到最大值时，则是业务发挥效用的最优选择，使得业务进行合理配置，为寻找潜在的效益提供方向。
本发明提出一种基于GRU-LSTM神经网络的电力金融业务风控***，通如图1所示，首先需要对财务数据库、业务数据库、人力数据库进行随机抽取信息，通过业务数据库获得相关的财务与人力信息等，对其进行数据处理，利用GRU预测时间短与LSTM精度高的优点，两者相结合对其进行处理最为妥当，GRU网络结构简单从而使得参数更容易收敛，大大减少了训练时间，但是GRU精度不如LSTM，再经过LSTM，使得预测精度大大提高，经过处理后的数据可以存放起来，继续为以后的训练提供基础。将数据通过处理后进行问题诊断，对所输出的结果进行判断，若有1个或2个参数异常，则发出一级警报，并报告具体异常数据；若有3个或4个参数异常，则发出二级警报，并报告具体异常数据；若有5个及5个以上参数异常，则发出三级警报，报告具体异常数据。管理者修改电力金融业务数据后计算机进行二次或者多次模拟，从而确保不会出现第二次数据异常，从一定程度上达到风险控制的效果。
所述基于GRU-LSTM神经网络的电力金融业务风控***对电力金融业务数据进行深度学习的流程如图2所示，从金融业务中采集数据集合，通过数据挖掘得到风险特征描述，同时结合具体金融业务对风险进行定义，将风险定义和风险特征描述结合，综合确定实际风险因素。本发明实现以下步骤：
步骤一，数据收集：通过对金融业务的数据监测，记录其中的金融数据，并且根据深度学习的理论，从相关的金融项目业务中搜集出所需要的目标金融数据，从而对其更加准确的认识、掌握，并且利用其中的金融运作规律，去防控风险危机，使得搜集的数据更加准确有效。本发明首先需要对财务数据库、业务数据库、人力数据库进行随机抽取信息，通过业务数据库获得相关的财务与人力信息等，对其进行数据处理，利用GRU预测时间短与LSTM精度高的优点，两者相结合对其进行处理，首先训练GRU网络进行数据分析。
步骤二，综合分析，风险评定：根据GRU分析的数据，利用LSTM神经网络对风险控制进行模拟，通过历史数据训练LSTM神经网络预测判断金融数据所对应的风险因素及风险等级；
步骤三，实时检测，风险防控：实时监测金融业务数据，输入训练好的GRU-LSTM网络，输出预测的风险因素及风险等级，实现金融业务风控。
经过GRU网络深度学习分析处理过的数据，放入LSTM神经网络中进行模拟分析，将风险等级进行划分，根据模拟分析结果判断最终的风险等级，为管理者提供科学的有效的判断依据，从而提前对其进行风险规避，为电力金融业务风险的评估和管控一些预防性的建议，且具有一定的可靠性。
如图3所示，为本发明LSTM预测分析的流程示意图，通过运用LSTM神经网络，向网络中输入体现金融风险的指数，例如时间、人数、进度、财务等，假设这里包含12个参数，输入节点数为12，在输入参数前需要对数据进行归一化处理，变为0到1之间的标准值，归一化方程如下：
x′ t＝(x t-x min)/(x max-x min)                                             (1)
其中x max与x min分别是输入样本参数中的最大值与最小值，x t与x′ t则分别是真实的初始输入样本参数以及归一化处理结果。
对于LSTM神经网络，假定网络的预期输出结果为X k，真实的输出结果是
Figure PCTCN2022110593-appb-000001
那么预期的误差e等于
Figure PCTCN2022110593-appb-000002
误平方和SSE可以用下述等式表示：
Figure PCTCN2022110593-appb-000003
LSTM训练时，直至网络在限定的误差区间上收敛表示训练完成，反之则代表网络训练失败，同时需要调整网络的初始权重乃至整个网络的拓扑结构，直到获得理想的输出结果。
对于本发明风险防控方案，在GRU-LSTM神经网络中，设输入LSTM神经网络的风险指标向量集合为X k(k＝1,2...n)，n为指标数量，即输入的参数个数，设隐含层中包含的各个神经元输入为a j，输入层到隐含层的权值为w i，隐含层中包含的神经元的数量为j，则公式如下：
Figure PCTCN2022110593-appb-000004
风险预测模型中的传递函数如下：
Figure PCTCN2022110593-appb-000005
此时隐含层的输入单元λ i可以描述为：
Figure PCTCN2022110593-appb-000006
经过运算筛选处理后，数据达到输出层，输出层神经元的输入项s t与输出项η t分别为：
Figure PCTCN2022110593-appb-000007
其中v jt是隐含层到输出层的权值，ξ t是输出层的阈值。通过LSTM神经网络，由输入的n指标参数得到t个预测输出，对输出的预测结果进行判断，根据结果的异常数量，即存在的风险因素数量，根据风险预警指标判断其对应的风险等级。
在本实施例中，输入12个参数，包含时间、人数、工作量、进度等，通过模型所输出的结果进行判断，若有1个或2个参数异常，则发出一级警报，并报告具体异常数据；若有3个或4个参数异常，则发出二级警报，并报告具体异常数据；若有5个及5个以上参数异常，则发出三级警报，报告具体异常数据，且管理者修改业务数据后进行二次模拟，从而保证业务数据不会出现第二次异常。
作为上述方法的载体，本发明提出基于GRU-LSTM神经网络的电力金融业务风控***，具有计算机可读存储介质，计算机可读存储介质中配置有计算机程序，所述计算机程序被执行时实现上述的风险预警模型。

Claims (7)

  1. 基于GRU-LSTM神经网络的电力金融业务风控方法，其特征是通过深度学习对电力金融业务历史数据进行数据分析，得到电力金融业务的风险因素，根据风险因素建立风险预警指标体系，构建得到风险预警模型；其中深度学习为GRU-LSTM网络组合预测：先将金融业务数据通过GRU网络进行分析，再输入LSTM神经网络，利用LSTM神经网络对风险控制进行模拟，预测风险因素，根据风险预警指标判断其对应的风险等级；利用历史数据训练GRU-LSTM网络得到风险预警模型，利用风险预警模型对实时金融业务数据进行风险等级预测，实现金融业务风控。
  2. 根据权利要求1所述的基于GRU-LSTM神经网络的电力金融业务风控方法，其特征是包括以下步骤：
    步骤一，数据收集：通过对金融业务的数据进行监测，记录其中的金融数据，根据深度学习所要分析的金融业务风险，从相关的金融项目业务中搜集出所需要的目标金融业务历史数据，训练GRU网络进行数据分析；
    步骤二，综合分析，风险评定：根据GRU分析的数据，利用LSTM神经网络对风险控制进行模拟，通过历史数据训练LSTM神经网络预测判断金融数据所对应的风险因素及风险等级；
    步骤三，实时检测，风险防控：实时监测金融业务数据，输入训练好的GRU-LSTM网络，输出预测的风险因素及风险等级，实现金融业务风控。
  3. 根据权利要求1或2所述的基于GRU-LSTM神经网络的电力金融业务风控方法，其特征是采集电力金融业务历史数据，包括对财务数据库、业务数据库和人力数据库进行随机信息抽取。
  4. 根据权利要求1或2所述的基于GRU-LSTM神经网络的电力金融业务风控方法，其特征是LSTM神经网络的具体实现为：向LSTM神经网络输入金融数据，所述金融数据为体现金融风险的指标参数，设包含n个参数，则输入节点数为n，在输入参数前需要对数据进行归一化处理，变为0到1之间的标准值，归一化方程如下：
    x′ t＝(x t-x min)/(x max-x min)      (1)
    其中x max与x min分别是输入样本参数中的最大值与最小值，x t与x′ t则分别是真实的初始输入样本参数以及归一化处理结果；
    通过历史数据训练LSTM神经网络，直至在限定的误差区间上收敛，反之则代表网络训练失败，同时调整神经网络的初始权重及整个网络的拓扑结构，直到获得理想的输出结果；
    设输入的风险指标向量集合为X k，k＝1,2...n，隐含层中包含的神经元的数量为j，隐含层中包含的各个神经元输入为a j，输入层到隐含层的权值为w ij，则：
    Figure PCTCN2022110593-appb-100001
    b j是隐含层的阈值，风险预测模型中的传递函数如下：
    Figure PCTCN2022110593-appb-100002
    此时隐含层的输出λ j为：
    Figure PCTCN2022110593-appb-100003
    经过运算筛选处理后，数据达到输出层，输出层神经元的输入项s t与输出项η t分别为：
    Figure PCTCN2022110593-appb-100004
    其中v jt是隐含层到输出层的权值，ξ t是输出层的阈值；
    通过LSTM神经网络，由输入的n指标参数得到t个预测输出，对输出的预测结果进行判断，根据结果的异常数量，即存在的风险因素数量，根据风险预警指标判断其对应的风险等级。
  5. 根据权利要求4所述的基于GRU-LSTM神经网络的电力金融业务风控方法，其特征是若有1个或2个参数异常，则发出一级警报，并报告具体异常数据；若有3个或4个参数异常，则发出二级警报，并报告具体异常数据；若有5个及5个以上参数异常，则发出三级警报，报告具体异常数据。
  6. 根据权利要求1或2所述的基于GRU-LSTM神经网络的电力金融业务风控方法，其特征是，根据风险等级预测结果，调整金融业务，对实时监测的数据进行再次模拟，直至风险预警模型不再输出异常风险因素。
  7. 基于GRU-LSTM神经网络的电力金融业务风控***，其特征是具有计算机可读存储介质，计算机可读存储介质中配置有计算机程序，所述计算机程序被执行时实现权利要求1-6任一项所述的风险预警模型。
PCT/CN2022/110593 2022-01-25 2022-08-05 基于gru-lstm神经网络的电力金融业务风控方法及*** WO2023142424A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210087496.5 2022-01-25
CN202210087496.5A CN114580706A (zh) 2022-01-25 2022-01-25 基于gru-lstm神经网络的电力金融业务风控方法及***

Publications (1)

Publication Number Publication Date
WO2023142424A1 true WO2023142424A1 (zh) 2023-08-03

Family

ID=81772927

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2022/110593 WO2023142424A1 (zh) 2022-01-25 2022-08-05 基于gru-lstm神经网络的电力金融业务风控方法及***

Country Status (2)

Country Link
CN (1) CN114580706A (zh)
WO (1) WO2023142424A1 (zh)

Cited By (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117118745A (zh) * 2023-10-20 2023-11-24 山东慧贝行信息技术有限公司 一种基于深度学习的网络安全动态预警***
CN117172721A (zh) * 2023-10-31 2023-12-05 深圳薪汇科技有限公司 用于融资业务的数据流转监管预警方法及***
CN117235063A (zh) * 2023-11-10 2023-12-15 广州汇通国信科技有限公司 一种基于人工智能技术的数据质量管理方法
CN117314424A (zh) * 2023-09-18 2023-12-29 纬创软件(武汉)有限公司 面向金融大数据的区块链交易***及方法
CN117374977A (zh) * 2023-12-07 2024-01-09 能拓能源股份有限公司 一种储能***负荷预测及风险分析方法
CN117595231A (zh) * 2023-10-20 2024-02-23 国网安徽省电力有限公司六安供电公司 一种智能电网配网管理***及其方法
CN117829787A (zh) * 2024-03-05 2024-04-05 中亿丰数字科技集团有限公司 一种基于bim和ai大模型的工程安全智能管理方法及***

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114580706A (zh) * 2022-01-25 2022-06-03 国网江苏省电力有限公司南京供电分公司 基于gru-lstm神经网络的电力金融业务风控方法及***

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180336452A1 (en) * 2017-05-22 2018-11-22 Sap Se Predicting wildfires on the basis of biophysical indicators and spatiotemporal properties using a long short term memory network
CN111429006A (zh) * 2020-03-24 2020-07-17 北京明略软件***有限公司 金融风险指标预测模型构建、风险态势预测方法和装置
CN112381591A (zh) * 2020-12-04 2021-02-19 四川长虹电器股份有限公司 基于lstm深度学习模型的销售预测优化方法
CN114580706A (zh) * 2022-01-25 2022-06-03 国网江苏省电力有限公司南京供电分公司 基于gru-lstm神经网络的电力金融业务风控方法及***

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20180336452A1 (en) * 2017-05-22 2018-11-22 Sap Se Predicting wildfires on the basis of biophysical indicators and spatiotemporal properties using a long short term memory network
CN111429006A (zh) * 2020-03-24 2020-07-17 北京明略软件***有限公司 金融风险指标预测模型构建、风险态势预测方法和装置
CN112381591A (zh) * 2020-12-04 2021-02-19 四川长虹电器股份有限公司 基于lstm深度学习模型的销售预测优化方法
CN114580706A (zh) * 2022-01-25 2022-06-03 国网江苏省电力有限公司南京供电分公司 基于gru-lstm神经网络的电力金融业务风控方法及***

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
"Master's Thesis", 1 June 2021, HARBIN INSTITUTE OF TECHNOLOGY, Heilongjiang, CN, article JIANG YIBING: "PTA Futures Price Prediction Research based on a GRU-LSTM Hybrid Model", pages: 1 - 61, XP009548028, DOI: 10.27061/d.cnki.ghgdu.2021.001424 *
LI, JIYONG; ZHAO, DE-BIAO; ZHANG, JING: "Research on risk forewarning of the supply chain based on BP neural network", JOURNAL OF HEBEI UNIVERSITY OF ENGINEERING (NATURAL SCIENCE EDITION), vol. 28, no. 3, 30 September 2011 (2011-09-30), CN, pages 83 - 87, XP009548031, ISSN: 1673-9469 *

Cited By (13)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117314424B (zh) * 2023-09-18 2024-03-29 纬创软件(武汉)有限公司 面向金融大数据的区块链交易***及方法
CN117314424A (zh) * 2023-09-18 2023-12-29 纬创软件(武汉)有限公司 面向金融大数据的区块链交易***及方法
CN117118745B (zh) * 2023-10-20 2024-01-05 山东慧贝行信息技术有限公司 一种基于深度学习的网络安全动态预警***
CN117595231A (zh) * 2023-10-20 2024-02-23 国网安徽省电力有限公司六安供电公司 一种智能电网配网管理***及其方法
CN117118745A (zh) * 2023-10-20 2023-11-24 山东慧贝行信息技术有限公司 一种基于深度学习的网络安全动态预警***
CN117595231B (zh) * 2023-10-20 2024-05-31 国网安徽省电力有限公司六安供电公司 一种智能电网配网管理***及其方法
CN117172721A (zh) * 2023-10-31 2023-12-05 深圳薪汇科技有限公司 用于融资业务的数据流转监管预警方法及***
CN117172721B (zh) * 2023-10-31 2024-02-20 深圳薪汇科技有限公司 用于融资业务的数据流转监管预警方法及***
CN117235063A (zh) * 2023-11-10 2023-12-15 广州汇通国信科技有限公司 一种基于人工智能技术的数据质量管理方法
CN117235063B (zh) * 2023-11-10 2024-03-29 广州汇通国信科技有限公司 一种基于人工智能技术的数据质量管理方法
CN117374977A (zh) * 2023-12-07 2024-01-09 能拓能源股份有限公司 一种储能***负荷预测及风险分析方法
CN117374977B (zh) * 2023-12-07 2024-02-20 能拓能源股份有限公司 一种储能***负荷预测及风险分析方法
CN117829787A (zh) * 2024-03-05 2024-04-05 中亿丰数字科技集团有限公司 一种基于bim和ai大模型的工程安全智能管理方法及***

Also Published As

Publication number Publication date
CN114580706A (zh) 2022-06-03

Similar Documents

Publication Publication Date Title
WO2023142424A1 (zh) 基于gru-lstm神经网络的电力金融业务风控方法及***
CN110929918B (zh) 一种基于CNN和LightGBM的10kV馈线故障预测方法
CN112859822B (zh) 基于人工智能的设备健康分析及故障诊断的方法及***
CN110636066B (zh) 基于无监督生成推理的网络安全威胁态势评估方法
CN112529341B (zh) 一种基于朴素贝叶斯算法的钻井漏失机率预测方法
CN113807570B (zh) 基于XGBoost的水库大坝风险等级评估方法及***
CN111178675A (zh) 基于LR-Bagging算法的电费回收风险预测方法、***、存储介质及计算机设备
CN113378990B (zh) 基于深度学习的流量数据异常检测方法
CN115660293B (zh) 基于数字孪生的复杂机电产品全生命周期综合评价方法
CN112906764A (zh) 基于改进bp神经网络的通信安全设备智能诊断方法及***
Hu et al. A hybrid model for predicting the degradation trend of hydropower units based on deep learning
CN117032165A (zh) 一种工业设备故障诊断方法
Li et al. A lstm-based method for comprehension and evaluation of network security situation
CN117575328A (zh) 一种基于图注意力神经网络的产业链风险评估方法及装置
CN115545342A (zh) 一种企业电费回收的风险预测方法与***
CN117455681A (zh) 业务风险预测方法和装置
CN115392618A (zh) 一种基于ahp-ew和ae-rnn融合的食品安全风险预警模型及其建立方法
CN114065624A (zh) 一种炼化装置管网完整性管理水平智能评估方法
Zhao et al. Application Research of Artificial Neural Network in Environmental Quality Monitoring
Xu et al. Research on health assessment of electric power information system based on deep belief networks and cluster analysis
Li et al. On Testing and Evaluation of Artificial Intelligence Models
CN118246744A (zh) 一种特长隧道施工现场风险评估方法及***
CN113627735B (zh) 工程建设项目安全风险的预警方法及***
Wang et al. Product Key Reliability Characteristics Identification Method Based on XGBoost in Manufacturing Process
An Research on early warning of operation and maintenance cost of engineering construction projects

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: 22923228

Country of ref document: EP

Kind code of ref document: A1