CN112884590A - Power grid enterprise financing decision method based on machine learning algorithm - Google Patents

Power grid enterprise financing decision method based on machine learning algorithm Download PDF

Info

Publication number
CN112884590A
CN112884590A CN202110102452.0A CN202110102452A CN112884590A CN 112884590 A CN112884590 A CN 112884590A CN 202110102452 A CN202110102452 A CN 202110102452A CN 112884590 A CN112884590 A CN 112884590A
Authority
CN
China
Prior art keywords
financing
index
decision
scheme
indexes
Prior art date
Legal status (The legal status 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 status listed.)
Pending
Application number
CN202110102452.0A
Other languages
Chinese (zh)
Inventor
綦方中
周家恺
曹聪
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Zhejiang University of Technology ZJUT
Original Assignee
Zhejiang University of Technology ZJUT
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 Zhejiang University of Technology ZJUT filed Critical Zhejiang University of Technology ZJUT
Priority to CN202110102452.0A priority Critical patent/CN112884590A/en
Publication of CN112884590A publication Critical patent/CN112884590A/en
Pending legal-status Critical Current

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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or 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
    • 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
    • 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—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas 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

Abstract

A power grid enterprise financing decision method based on a machine learning algorithm obtains data, calculates monthly financing gaps, and calculates monthly financing risk indexesCRIThe method comprises the steps of predicting financing risks based on an XGboost model, obtaining financing schemes and financing decision indexes based on different scenes by combining internal and external environments, financing conditions and financing efficiency faced by a power grid enterprise, weighting the decision indexes through an AHP-coefficient of variation method, selecting an optimal monthly financing scheme based on an improved TOPSIS algorithm, and outputting an optimal financing scheme. The method can effectively form the financing decision of the power grid enterprise and better assist a manager in selecting the optimal financing scheme.

Description

Power grid enterprise financing decision method based on machine learning algorithm
Technical Field
The invention belongs to a power grid enterprise financing decision method, and particularly relates to a power grid enterprise financing decision method based on a machine learning algorithm.
Background
The provincial power grid enterprise belonging to the asset intensive industry has the characteristics of large investment capital requirement, large collection and payment fund flow scale, large social influence on payment safety and the like. As investment scale and investment requirements grow, more funds are required to support normal business activities; the investment demand of provincial power grid enterprises is continuously expanded, and the financial cost caused by paying interest for borrowing is continuously increased every year; but the income brought by the power grid enterprise through the self operation activity can not completely meet the investment demand and the financial expense increase caused by the expansion of the financing scale at present. Meanwhile, national economic environment changes such as controlling the investment scale of fixed assets, setting energy-saving and consumption-reducing targets and the like and government policy regulation and control requirements enlarge the investment uncertainty of power grid enterprises, and the financing risk is also increased continuously. Therefore, how to better evaluate and predict the financing risk and make an optimal financing decision scheme according to the financing environment and conditions is one of the important problems to be solved urgently faced by provincial power grid enterprises.
The financing gap can well reflect the financing requirement of the power grid enterprise at a certain stage to a certain extent, but the currently feasible financing gap calculation method mainly aims at small and medium-sized enterprises, and corresponding estimated values are provided mostly by explaining the reason for forming the financing gap. Aiming at large enterprises, particularly large enterprises in China such as provincial power grid enterprises, the existing financing gap calculation method is often too simple; and moreover, according to the characteristics of the power grid enterprise, the profit, the asset liability and the cash flow condition of the power grid enterprise are predicted through the analysis of the financial system data of the enterprise, and then the financing gap is calculated relatively accurately. For a long time, China provincial power grid enterprises always maintain a traditional financing mode with bank lending and bond issuing as the main and internal fund market financing as the auxiliary. The application research results in the field of power grid enterprise financing management currently focus on the innovation of financing modes or the combination of multiple financing modes. In view of the particularity and importance of provincial power grid enterprises in China, the innovative financing mode is difficult to be promoted, and the realization process and the actual effect have great unknown; and based on the combination of multiple financing modes, the change of financing risk is not considered, and great uncertainty is provided.
Disclosure of Invention
In order to solve the problems in the prior art, a power grid enterprise financing decision method based on a machine learning algorithm is provided, and the method can effectively form a power grid enterprise financing decision and better assist a manager in selecting an optimal financing scheme.
In order to achieve the purpose, the invention adopts the following technical scheme:
the power grid enterprise financing decision method based on the machine learning algorithm comprises the following steps:
step 1, acquiring data, and calculating a monthly financing gap:
the acquired data includes: net cash flow rate C generated by business activities1Net cash flow C from investment activities2Net cash flow C from financing activities3Initial balance of cash and cash equivalents C0The minimum cash holding amount S;
when (C)1+C2+C3+C0)>S, the financing gap G is equal to 0, otherwise G is equal to S- (C)1+C2+C3+C0)；
Step 2, based on a financing risk index system, calculating a monthly financing risk index CRI:
step 2.1, inputting index data: inputting 15 index data of a flowing ratio, an asset liability ratio, a net cash flow for operation, a total liability, a flowing asset turnover ratio, a total asset advanced recovery rate, a total asset turnover rate, a business profit rate, an asset reward rate, a net profit growth rate, a business profit growth rate, a net asset growth rate, a total asset growth rate, a financing gap size and SHIBOR according to a set time range;
step 2.2, establishing an original evaluation index matrix:
for m months financing risk to be evaluated and n evaluation indexes, wherein n is 15, an original evaluation index matrix A is set as:
A＝(aij)m×n
wherein, aijTaking a value for the jth index of the ith financing risk to be evaluated;
step 2.3, standardization of evaluation indexes: standardizing the 5 indexes in the step 2.1 to enable aijAfter standardization is bijFor the benefit index, there are:
Figure BDA0002915811490000021
for cost-type indicators, there are:
Figure BDA0002915811490000022
step 2.4, calculating the information entropy value of the evaluation index:
Figure BDA0002915811490000023
wherein E isjIs the information entropy, p, of the index jijExpressed as:
Figure BDA0002915811490000024
to avoid the meaningless case of ln0 that may occur after normalization by entropy, for simplicity of calculation, if p ij0, then lnp is definedij＝0；
Step 2.5, calculating the weight of each evaluation index, namely the weight w of the index jjThe calculation formula of (2) is as follows:
Figure BDA0002915811490000025
step 2.6, calculating a financing risk index CRI:
Figure BDA0002915811490000026
wherein, cjIs a normalized value of the value;
step 3, predicting financing risk based on the XGboost model:
step 3.1, constructing an XGboost model:
when t decision trees exist in the model, the predicted value y of the ith sample after the decision trees are finished for t timesi (t)Can be expressed as:
Figure BDA0002915811490000031
where s is the number of samples, fkIs the kth regression tree between 1 and t regression trees, ftF is the set space of all classification and regression trees; the loss function of the model is expressed as:
Figure BDA0002915811490000032
Figure BDA0002915811490000033
wherein l is the deviation between the true value and the predicted value, w represents the leaf node weight, gamma and lambda are regularization coefficients, and T is the number of decision tree leaf nodes;
the loss function is expanded by taylor series, which can be given by:
Figure BDA0002915811490000034
wherein C is a constant term, giAnd hiIs a first partial derivative
Figure BDA0002915811490000038
Second partial derivative
Figure BDA0002915811490000039
The method specifically comprises the following steps:
Figure BDA0002915811490000035
define G separatelyjAnd HjComprises the following steps:
Figure BDA0002915811490000036
the optimal value of the objective function can be finally obtained as follows:
Figure BDA0002915811490000037
step 3.2, optimizing the following parameter variables in the XGboost model by using a GridSearch method:
the method comprises the steps of determining the number n _ estimators of regression trees, the minimum leaf node sample weight and min _ child _ weight, the maximum tree depth max _ depth of a decision tree, specifying the descending value gamma of a minimum loss function required by node splitting, controlling the random sampling proportion subsample of each tree, controlling the feature ratio subsample of each tree during random sampling, determining whether a target function can converge to a local minimum value and when the target function converges to the minimum value, and 8 parameter variables such as regularization alpha of L1 of the weight;
firstly, selecting n _ estimators with the highest harmony value by adopting a K-fold cross validation method, then optimizing other parameters by adopting a GridSearch mode, and selecting the best value;
and 3.3, predicting the power grid financing risk index value based on XGboost: firstly, dividing risk index values into two sets of training and testing, carrying out standardization processing on features, and predicting the risk index values based on the constructed XGboost model;
step 4, based on the result of financing risk calculation/prediction, combining the internal and external environments, financing conditions and financing efficiency faced by the power grid enterprise, obtaining financing schemes and financing decision indexes based on different scenes, comprising the following steps:
step 4.1, setting financing scenes, namely setting three scenes, namely a loose scene, a stable scene, a tight scene and the like according to different external scene types, wherein the scene factors corresponding to various scenes comprise interest rate borrowing, electricity selling amount increasing speed, electricity selling price and electricity purchasing price, and the regulation and control range of the scene factors can be set according to different scene types;
step 4.2, setting corresponding financing schemes according to financing situations, and aiming at three different external situations, decision makers can select three financing schemes such as aggressive type, moderate type or conservative type;
step 4.3, establishing a financing decision scheme evaluation index system:
aiming at three primary indexes of financing risk, financing conditions and financing efficiency and six secondary indexes of corresponding financing risk index CRI, interest cost, asset liability structure, financing structure, net asset profitability and total asset profitability, establishing a financing scheme evaluation index system;
and 5, weighting the decision index by an AHP-variation coefficient method, and selecting an optimal monthly financing scheme based on an improved TOPSIS algorithm:
step 5.1, weighting the decision index by an AHP-coefficient of variation method, comprising the following steps:
step 5.1.1, constructing a judgment matrix Q:
Figure BDA0002915811490000041
wherein r is the number of indexes to be compared and q isijThe numerical expression of the relative importance of the ith index to the jth index in 1 to r indexes to be compared is obtained;
step 5.1.2, carrying out consistency check on the judgment matrix, and calculating the weight of each decision index to obtain a weight vector w':
w'＝(w'1,w'2,…,w'r)
step 5.1.3, weighting according to the variation degree of the observed values of different indexes on the decision object, calculating a standard deviation representing the variation degree of each index, and performing normalization processing to obtain a weight vector w' of each index:
w"＝(w"1,w"2,…,w"r)
step 5.1.4, to the decision indexPerforming combined weighting, and obtaining the combined weighting w of the ith decision index according to the weights obtained by the AHP and the variation coefficient methodi *Comprises the following steps:
Figure BDA0002915811490000051
step 5.2, the optimal monthly financing scheme selected based on the improved TOPSIS algorithm comprises the following steps:
step 5.2.1, constructing a decision matrix
For v to-be-evaluated scheme sets with u decision indexes in each decision scheme set, the original decision matrix can be expressed as:
Figure BDA0002915811490000052
step 5.2.2, index standardization
Similarly, as the larger the decision index is, the better the benefit index is and the smaller the cost index is, the j decision index x of the ith scheme set in the v to-be-evaluated method setsijThe standardization process is required, and if the index is a benefit type index, the standardization process can be as follows:
Figure BDA0002915811490000053
if the index is a cost-type index, the index is normalized as follows:
Figure BDA0002915811490000054
a normalized decision matrix Z can be obtained:
Figure BDA0002915811490000055
step 5.2.3, determination of optimal/worst decision scheme:
for the optimal decision scheme, there are:
Figure BDA0002915811490000056
for the worst decision scheme, there are:
Figure BDA0002915811490000057
and 5.2.4, calculating the proximity degree of each decision scheme to be evaluated and the optimal and the worst schemes, and adopting an Euclidean distance measurement mode:
Figure BDA0002915811490000061
the Hamming distance measurement mode is adopted, and the method comprises the following steps:
Figure BDA0002915811490000062
step 5.2.5, the Euclidean distance and the Hamming distance are subjected to standardization treatment:
Figure BDA0002915811490000063
Figure BDA0002915811490000064
step 5.2.6, integrating the four distances to obtain the centralized distance between each decision scheme and the optimal/worst scheme:
Figure BDA0002915811490000065
wherein α + β ═ 1, i ═ 1,2,3, …, v;
step 5.2.7, calculating the comprehensive evaluation result M of each decision scheme to be evaluatedi：
Figure BDA0002915811490000066
Step 6, outputting an optimal financing scheme:
for the comprehensive evaluation result MiSequencing to obtain the optimal financing scheme M*：
M*＝max(Mi),i＝1,2,3,...,v。
In step 2.6, a linear weighting method is used to calculate the financing risk index CRI. The CRI reflects the condition of the financing risk of the power grid enterprise, and the value of the CRI is from 0 to 1 to indicate the degree of the financing risk of the enterprise from small to large.
The specific indexes in the financing risk evaluation index system come from seven aspects including repayment capacity, cash flow, capital operation capacity, profit capacity, development capacity, financing scale and economic environment; specific financing risk evaluation indexes include 15 items of liquidity ratio, equity rate, net cash flow in operation, total equity, liquidity turnover rate, total equity advance recovery rate, total equity turnover rate, business profit rate, equity reward rate, net profit growth rate, business profit growth rate, net equity growth rate, total equity growth rate, financing gap size, and SHIBOR.
In step 4.2, three different external scenario decision makers can select three financing schemes, such as aggressive type, moderate type or conservative type, and the difference of the financing schemes is mainly reflected in the financing amount and the long-short term financing proportion.
The advantages of a machine learning algorithm, particularly an XGboost model and a BP neural network model, in the field of small sample data prediction are considered, and meanwhile, the higher accuracy of the XGboost model in the aspect of financing risk index prediction is considered, so that on the basis of constructing a provincial power grid enterprise financing risk evaluation index system, the XGboost-based machine learning algorithm is introduced to predict monthly financing risk indexes, and the influence of risk factors is fully considered in the financing decision process of an enterprise; factors such as financing risk, financing schemes of different financing situations, interest rate and the like are comprehensively considered, an optimal monthly financing scheme can be selected by utilizing an AHP-variation coefficient method and an improved TOPSIS algorithm, the influence of risk factors is considered when financing decision is made and a reasonable financing plan is made, the influence of the financing situations and the financing preference is also considered, and the provincial power grid enterprise is effectively helped to make fast and objective financing decision.
Drawings
FIG. 1 is a flow diagram of the present invention.
Detailed Description
The specific implementation mode of the power grid enterprise financing decision method based on the machine learning algorithm comprises the following steps:
step 1, acquiring related data, and calculating a monthly financing gap:
acquiring related data: net cash flow rate C generated by business activities1Net cash flow C from investment activities2Net cash flow C from financing activities3Initial balance of cash and cash equivalents C0The minimum cash holding amount S;
when (C)1+C2+C3+C0)>S, the financing gap G is equal to 0, otherwise G is equal to S- (C)1+C2+C3+C0)；
Acquiring related data of 48 months in 4 years, respectively calculating data of financing gaps G with monthly IDs of 1 to 44, wherein the data of the financing gaps G is shown in the table 1:
table 1: monthly financing gap data
ID G ID G ID G ID G
1 6173 12 5243 23 8063 34 7662
2 18161 13 14632 24 16819 35 22679
3 24712 14 19066 25 23319 36 29170
4 40144 15 32103 26 35153 37 38503
5 48881 16 45362 27 44844 38 50916
6 59566 17 58417 28 56722 39 65023
7 69426 18 67935 29 68864 40 81364
8 81611 19 81161 30 83725 41 90370
9 94364 20 95769 31 101529 42 93801
10 103532 21 105218 32 110302 43 120211
11 115765 22 117346 33 126416 44 156204
Step 2, based on a financing risk index system, calculating a monthly financing risk index CRI:
step 2.1, inputting index data: inputting 15 index data of a flowing ratio, an asset liability ratio, a net cash flow, a total liability, a flowing asset turnover ratio, a total asset advance recovery ratio, a total asset turnover ratio, an operating profit ratio, an asset reward ratio, a net profit growth ratio, an operating profit growth ratio, a net asset growth ratio, a total asset growth ratio, a financing gap size and SHIBOR according to a set time range, wherein the SHIBOR data corresponding to each month is as follows:
table 2: SHIBOR data corresponding to each month
ID SHIBOR ID SHIBOR ID SHIBOR ID SHIBOR
1 4.785 12 3.236 23 3.957 34 4.744
2 4.780 13 3.088 24 4.115 35 4.731
3 4.726 14 3.044 25 4.209 36 4.492
4 4.113 15 3.047 26 4.288 37 4.365
5 3.402 16 3.050 27 4.417 38 4.389
6 3.390 17 3.046 28 4.403 39 3.965
7 3.396 18 3.028 29 4.401 40 3.432
8 3.410 19 3.027 30 4.403 41 3.513
9 3.404 20 3.029 31 4.404 42 3.522
10 3.351 21 3.068 32 4.522 43 3.544
11 3.350 22 3.284 33 4.698 44 3.522
Step 2.2, establishing an original evaluation index matrix:
for m months of financing risk to be evaluated and n evaluation indexes (where n is 15), an original evaluation index matrix a is set as:
A＝(aij)m×n
wherein, aijTaking a value for the jth index of the ith financing risk to be evaluated;
and 2.3, standardizing the evaluation indexes. Let aijAfter standardization is bijFor the benefit index, there are:
Figure BDA0002915811490000081
for cost-type indicators, there are:
Figure BDA0002915811490000082
step 2.4, calculating the information entropy value of the evaluation index:
Figure BDA0002915811490000083
wherein E isjIs the information entropy, p, of the index jijExpressed as:
Figure BDA0002915811490000084
to avoid the meaningless case of ln0 that may occur after normalization by entropy, for simplicity of calculation, if p ij0, then lnp is definedij0; and 2.5, calculating the weight of each evaluation index. Weight w of index jjThe calculation formula of (2) is as follows:
Figure BDA0002915811490000091
the weights of the 15 evaluation indexes in the financing risk evaluation index system and the calculation result of the entropy value in the step 2.4 are shown in table 3:
table 3: financing risk evaluation index weight distribution condition
Index (I) Type of index Entropy value Weight of
Flow rate/%) Benefit type 0.9673 0.0550
Percentage of assets liability/%) Cost type 0.9360 0.1077
Operating net cash flow/million yuan Benefit type 0.9728 0.0457
Total liability per million yuan Cost type 0.9792 0.0350
Turnover/percent of flowing assets Benefit type 0.9558 0.0744
Total asset cash recovery/%) Benefit type 0.9690 0.0521
Total asset turnover/%) Benefit type 0.9621 0.0638
Business profit margin/%) Benefit type 0.9783 0.0364
Asset remuneration/%) Benefit type 0.9711 0.0487
Net profit growth rate/%) Benefit type 0.9938 0.0105
Business profit growth rate/%) Benefit type 0.9932 0.0114
Net asset growth rate/%) Benefit type 0.9843 0.0262
Total asset growth rate/%) Benefit type 0.8361 0.2759
Financing gap/million yuan Cost type 0.9741 0.0437
SHIBOR/％ Cost type 0.9326 0.1133
Step 2.6, calculating a financing risk index CRI:
Figure BDA0002915811490000092
wherein, cjIs a normalized value of the value;
the results of the calculation of the financing risk index for each month are shown in Table 4:
table 4: financing risk index corresponding to monthly
ID CRI ID CRI ID CRI ID CRI
1 0.1991 12 0.3060 23 0.3521 34 0.3745
2 0.2618 13 0.3957 24 0.4232 35 0.7472
3 0.3108 14 0.4025 25 0.3923 36 0.4375
4 0.3552 15 0.3752 26 0.4473 37 0.4230
5 0.3897 16 0.3881 27 0.3981 38 0.3822
6 0.3777 17 0.4446 28 0.4312 39 0.4159
7 0.4206 18 0.4951 29 0.4369 40 0.4484
8 0.3296 19 0.3924 30 0.3574 41 0.4194
9 0.3299 20 0.4199 31 0.3659 42 0.3025
10 0.3440 21 0.4635 32 0.3517 43 0.3585
11 0.2932 22 0.3677 33 0.2895 44 0.3643
Step 3, predicting financing risk based on the XGboost model:
step 3.1, constructing an XGboost model:
when t decision trees exist in the model, the predicted value y of the ith sample after the decision trees are finished for t timesi (t)Can be expressed as:
Figure BDA0002915811490000101
wherein s is a sampleNumber fkIs the kth regression tree between 1 and t regression trees, ftF is the set space of all classification and regression trees; the loss function of the model is expressed as:
Figure BDA0002915811490000102
Figure BDA0002915811490000103
wherein l is the deviation between the true value and the predicted value, w represents the leaf node weight, gamma and lambda are regularization coefficients, and T is the number of decision tree leaf nodes;
the loss function is expanded by taylor series, which can be given by:
Figure BDA0002915811490000104
wherein C is a constant term, giAnd hiIs a first partial derivative
Figure BDA0002915811490000105
Second partial derivative
Figure BDA0002915811490000106
The method specifically comprises the following steps:
Figure BDA0002915811490000107
define G separatelyjAnd HjComprises the following steps:
Figure BDA0002915811490000108
the optimal value of the objective function can be finally obtained as follows:
Figure BDA0002915811490000109
step 3.2, optimizing some parameter variables in the XGboost model by using a GridSearch method:
the method mainly comprises 8 parameter variables such as regression tree number n _ estimators, minimum leaf node sample weight and min _ child _ weight, maximum tree depth max _ depth of a decision tree, a drop value gamma of a minimum loss function required by node splitting, random sampling proportion subsample of each tree, feature ratio colsample _ byte when each tree is randomly sampled, whether a target function can converge to a local minimum value and when the target function converges to the minimum value learning _ rate and regularization alpha of L1 of weight;
firstly, selecting n _ estimators with the highest harmony value by adopting a K-fold cross validation method, then optimizing other parameters by adopting a GridSearch mode, and selecting the best value;
and 3.3, predicting the power grid financing risk index value based on XGboost: firstly, dividing risk index values into two sets of training and testing, carrying out standardization processing on features, and predicting the risk index values based on the constructed XGboost model;
the results of five sample predictions with IDs of 30, 37, 27, 4, and 10 are shown in table 4:
table 4: predicted value situation of XGboost model
ID 4 10 27 30 37
Prediction value 0.3875 0.3125 0.4322 0.3678 0.3839
Step 4, based on the result of financing risk calculation/prediction, combining the internal and external environments, financing conditions and financing efficiency faced by the power grid enterprise, obtaining financing schemes and financing decision indexes based on different scenes, comprising the following steps:
step 4.1, setting financing scenes, namely setting three scenes, namely a loose scene, a stable scene, a tight scene and the like according to different external scene types, wherein the scene factors corresponding to various scenes comprise interest rate borrowing, electricity selling amount increasing speed, electricity selling price and electricity purchasing price, and the regulation and control range of the scene factors can be set according to different scene types;
step 4.2, setting corresponding financing schemes according to financing situations, and aiming at three different external situations, decision makers can select three financing schemes such as aggressive type, moderate type or conservative type;
step 4.3, establishing a financing decision scheme evaluation index system:
aiming at three primary indexes of financing risk, financing conditions and financing efficiency and six secondary indexes of corresponding financing risk index CRI, interest cost, asset liability structure, financing structure, net asset profitability and total asset profitability, establishing a financing scheme evaluation index system;
and 5, weighting the decision index by an AHP-variation coefficient method, and selecting an optimal monthly financing scheme based on an improved TOPSIS algorithm:
step 5.1, weighting the decision index by an AHP-coefficient of variation method, comprising the following steps:
step 5.1.1, constructing a judgment matrix Q:
Figure BDA0002915811490000111
wherein r is the number of indexes to be compared and q isijThe numerical expression of the relative importance of the ith index to the jth index in 1 to r indexes to be compared is obtained;
step 5.1.2, carrying out consistency check on the judgment matrix, and calculating the weight of each decision index to obtain a weight vector w':
w'＝(w'1,w'2,…,w'r)
step 5.1.3, weighting according to the variation degree of the observed values of different indexes on the decision object, calculating a standard deviation representing the variation degree of each index, and performing normalization processing to obtain a weight vector w' of each index:
w"＝(w"1,w"2,…,w"r)
step 5.1.4, the decision indexes are combined and weighted, and the combined weighting w of the ith decision index can be obtained according to the weights obtained by the AHP and the variation coefficient methodi *Comprises the following steps:
Figure BDA0002915811490000121
table 5 shows the calculation results of the combined weighting of each index in the financing scheme evaluation index system:
table 5: weight distribution of financing scheme evaluation index
Figure BDA0002915811490000122
Step 5.2, the optimal monthly financing scheme selected based on the improved TOPSIS algorithm comprises the following steps:
step 5.2.1, constructing a decision matrix
For v to-be-evaluated scheme sets with u decision indexes in each decision scheme set, the original decision matrix can be expressed as:
Figure BDA0002915811490000123
step 5.2.2, index standardization
For the jth decision index x of the ith scheme set in the v to-be-evaluated method setsijAnd (3) carrying out standardization treatment, if the standard is a benefit type index, standardizing the standard into the following steps:
Figure BDA0002915811490000124
if the index is a cost-type index, the index is normalized as follows:
Figure BDA0002915811490000125
a normalized decision matrix Z can be obtained:
Figure BDA0002915811490000126
step 5.2.3, determination of optimal/worst decision scheme:
for the optimal decision scheme, there are:
Figure BDA0002915811490000131
for the worst decision scheme, there are:
Figure BDA0002915811490000132
and 5.2.4, calculating the proximity degree of each decision scheme to be evaluated and the optimal and the worst schemes, and adopting an Euclidean distance measurement mode:
Figure BDA0002915811490000133
the Hamming distance measurement mode is adopted, and the method comprises the following steps:
Figure BDA0002915811490000134
step 5.2.5, the Euclidean distance and the Hamming distance are subjected to standardization treatment:
Figure BDA0002915811490000135
Figure BDA0002915811490000136
step 5.2.6, integrating the four distances to obtain the centralized distance between each decision scheme and the optimal/worst scheme:
Figure BDA0002915811490000137
wherein α + β ═ 1, i ═ 1,2,3, …, v;
step 5.2.7, calculating the comprehensive evaluation result M of each decision scheme to be evaluatedi：
Figure BDA0002915811490000138
For the given nine financing scheme combinations, the evaluation results can be calculated separately, as shown in table 6:
table 6: comprehensive evaluation result of financing scheme
Figure BDA0002915811490000141
Step 6, outputting an optimal financing scheme:
for the comprehensive evaluation result MiSequencing to obtain the optimal financing scheme M*：
M*＝max(Mi),i＝1,2,3,...,v
In the above calculation example, M is obtained after sorting7Namely, the scheme seven is the optimal financing scheme.

Claims (1)

1. A power grid enterprise financing decision method based on a machine learning algorithm is characterized by comprising the following steps:
step 1, acquiring data, and calculating a monthly financing gap:
the acquired data includes: net cash flow rate C generated by business activities1Net cash flow C from investment activities2Net cash flow C from financing activities3Initial balance of cash and cash equivalents C0The minimum cash holding amount S;
when (C)1+C2+C3+C0)>S, the financing gap G is equal to 0, otherwise G is equal to S- (C)1+C2+C3+C0)；
Step 2, based on a financing risk index system, calculating a monthly financing risk index CRI:
step 2.1, inputting index data: inputting 15 index data of a flowing ratio, an asset liability ratio, a net cash flow for operation, a total liability, a flowing asset turnover ratio, a total asset advanced recovery rate, a total asset turnover rate, a business profit rate, an asset reward rate, a net profit growth rate, a business profit growth rate, a net asset growth rate, a total asset growth rate, a financing gap size and SHIBOR according to a set time range;
step 2.2, establishing an original evaluation index matrix:
for m months financing risk to be evaluated and n evaluation indexes, wherein n is 15, an original evaluation index matrix A is set as:
A＝(aij)m×n
wherein, aijTaking a value for the jth index of the ith financing risk to be evaluated;
step 2.3, standardization of evaluation indexes: standardizing the 5 indexes in the step 2.1 to enable aijAfter standardization is bijFor the benefit index, there are:
Figure FDA0002915811480000011
for cost-type indicators, there are:
Figure FDA0002915811480000012
step 2.4, calculating the information entropy value of the evaluation index:
Figure FDA0002915811480000013
wherein E isjIs the information entropy, p, of the index jijExpressed as:
Figure FDA0002915811480000014
to avoid the meaningless case of ln0 that may occur after normalization by entropy, for simplicity of calculation, if pij0, then lnp is definedij＝0；
Step 2.5, calculating the weight of each evaluation index, namely the weight w of the index jjThe calculation formula of (2) is as follows:
Figure FDA0002915811480000021
step 2.6, calculating a financing risk index CRI:
Figure FDA0002915811480000022
wherein, cjIs a normalized value of the value;
step 3, predicting financing risk based on the XGboost model:
step 3.1, constructing an XGboost model:
when t decision trees exist in the model, the predicted value y of the ith sample after the decision trees are finished for t timesi (t)Can be expressed as:
Figure FDA0002915811480000023
where s is the number of samples, fkIs the kth regression tree between 1 and t regression trees, ftF is the set space of all classification and regression trees; the loss function of the model is expressed as:
Figure FDA0002915811480000024
Figure FDA0002915811480000025
wherein l is the deviation between the true value and the predicted value, w represents the leaf node weight, gamma and lambda are regularization coefficients, and T is the number of decision tree leaf nodes;
the loss function is expanded by taylor series, which can be given by:
Figure FDA0002915811480000026
wherein C is a constant term, giAnd hiIs a first partial derivative
Figure FDA0002915811480000027
Second partial derivative
Figure FDA0002915811480000028
The method specifically comprises the following steps:
Figure FDA0002915811480000029
define G separatelyjAnd HjComprises the following steps:
Figure FDA0002915811480000031
the optimal value of the objective function can be finally obtained as follows:
Figure FDA0002915811480000032
step 3.2, optimizing the following parameter variables in the XGboost model by using a GridSearch method:
the method comprises the steps of determining the number n _ estimators of regression trees, the minimum leaf node sample weight and min _ child _ weight, the maximum tree depth max _ depth of a decision tree, specifying the descending value gamma of a minimum loss function required by node splitting, controlling the random sampling proportion subsample of each tree, controlling the feature ratio subsample of each tree during random sampling, determining whether a target function can converge to a local minimum value and when the target function converges to the minimum value, and 8 parameter variables such as regularization alpha of L1 of the weight;
firstly, selecting n _ estimators with the highest harmony value by adopting a K-fold cross validation method, then optimizing other parameters by adopting a GridSearch mode, and selecting the best value;
and 3.3, predicting the power grid financing risk index value based on XGboost: firstly, dividing risk index values into two sets of training and testing, carrying out standardization processing on features, and predicting the risk index values based on the constructed XGboost model;
step 4, based on the result of financing risk calculation/prediction, combining the internal and external environments, financing conditions and financing efficiency faced by the power grid enterprise, obtaining financing schemes and financing decision indexes based on different scenes, comprising the following steps:
step 4.1, setting financing scenes, namely setting three scenes, namely a loose scene, a stable scene, a tight scene and the like according to different external scene types, wherein the scene factors corresponding to various scenes comprise interest rate borrowing, electricity selling amount increasing speed, electricity selling price and electricity purchasing price, and the regulation and control range of the scene factors can be set according to different scene types;
step 4.2, setting corresponding financing schemes according to financing situations, and aiming at three different external situations, decision makers can select three financing schemes such as aggressive type, moderate type or conservative type;
step 4.3, establishing a financing decision scheme evaluation index system:
aiming at three primary indexes of financing risk, financing conditions and financing efficiency and six secondary indexes of corresponding financing risk index CRI, interest cost, asset liability structure, financing structure, net asset profitability and total asset profitability, establishing a financing scheme evaluation index system;
and 5, weighting the decision index by an AHP-variation coefficient method, and selecting an optimal monthly financing scheme based on an improved TOPSIS algorithm:
step 5.1, weighting the decision index by an AHP-coefficient of variation method, comprising the following steps:
step 5.1.1, constructing a judgment matrix Q:
Figure FDA0002915811480000041
wherein r is the number of indexes to be compared and q isijThe numerical expression of the relative importance of the ith index to the jth index in 1 to r indexes to be compared is obtained;
step 5.1.2, carrying out consistency check on the judgment matrix, and calculating the weight of each decision index to obtain a weight vector w':
w'＝(w'1,w'2,…,w'r)
step 5.1.3, weighting according to the variation degree of the observed values of different indexes on the decision object, calculating a standard deviation representing the variation degree of each index, and performing normalization processing to obtain a weight vector w' of each index:
w"＝(w"1,w"2,…,w"r)
step 5.1.4, the decision indexes are combined and weighted, and the combined weighting w of the ith decision index can be obtained according to the weights obtained by the AHP and the variation coefficient methodi *Comprises the following steps:
Figure FDA0002915811480000042
step 5.2, the optimal monthly financing scheme selected based on the improved TOPSIS algorithm comprises the following steps:
step 5.2.1, constructing a decision matrix
For v to-be-evaluated scheme sets with u decision indexes in each decision scheme set, the original decision matrix can be expressed as:
Figure FDA0002915811480000043
step 5.2.2, index standardization
Similarly, as the larger the decision index is, the better the benefit index is and the smaller the cost index is, the j decision index x of the ith scheme set in the v to-be-evaluated method setsijThe standardization process is required, and if the index is a benefit type index, the standardization process can be as follows:
Figure FDA0002915811480000051
if the index is a cost-type index, the index is normalized as follows:
Figure FDA0002915811480000052
a normalized decision matrix Z can be obtained:
Figure FDA0002915811480000053
step 5.2.3, determination of optimal/worst decision scheme:
for the optimal decision scheme, there are:
Figure FDA0002915811480000054
for the worst decision scheme, there are:
Figure FDA0002915811480000055
and 5.2.4, calculating the proximity degree of each decision scheme to be evaluated and the optimal and the worst schemes, and adopting an Euclidean distance measurement mode:
Figure FDA0002915811480000056
the Hamming distance measurement mode is adopted, and the method comprises the following steps:
Figure FDA0002915811480000057
step 5.2.5, the Euclidean distance and the Hamming distance are subjected to standardization treatment:
Figure FDA0002915811480000061
Figure FDA0002915811480000062
step 5.2.6, integrating the four distances to obtain the centralized distance between each decision scheme and the optimal/worst scheme:
Figure FDA0002915811480000063
wherein α + β ═ 1, i ═ 1,2,3, …, v;
step 5.2.7, calculating the comprehensive evaluation result M of each decision scheme to be evaluatedi：
Figure FDA0002915811480000064
Step 6, outputting an optimal financing scheme:
for the comprehensive evaluation result MiSequencing to obtain the optimal financing scheme M*：
M*＝max(Mi),i＝1,2,3,...,v。
CN202110102452.0A 2021-01-26 2021-01-26 Power grid enterprise financing decision method based on machine learning algorithm Pending CN112884590A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110102452.0A CN112884590A (en) 2021-01-26 2021-01-26 Power grid enterprise financing decision method based on machine learning algorithm

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110102452.0A CN112884590A (en) 2021-01-26 2021-01-26 Power grid enterprise financing decision method based on machine learning algorithm

Publications (1)

Publication Number Publication Date
CN112884590A true CN112884590A (en) 2021-06-01

Family

ID=76053201

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110102452.0A Pending CN112884590A (en) 2021-01-26 2021-01-26 Power grid enterprise financing decision method based on machine learning algorithm

Country Status (1)

Country Link
CN (1) CN112884590A (en)

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113554366A (en) * 2021-09-23 2021-10-26 山东大学 Classification supervision method for disinfection product production enterprises and related equipment
CN113657812A (en) * 2021-09-02 2021-11-16 谭维敏 Method and system for intelligent decision-making of store operation based on big data and algorithm
CN113762802A (en) * 2021-09-17 2021-12-07 昆明理工大学 Artificial intelligence PPP project financing evaluation system
CN113886372A (en) * 2021-09-08 2022-01-04 天元大数据信用管理有限公司 User portrait construction method based on improved analytic hierarchy process

Cited By (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN113657812A (en) * 2021-09-02 2021-11-16 谭维敏 Method and system for intelligent decision-making of store operation based on big data and algorithm
CN113886372A (en) * 2021-09-08 2022-01-04 天元大数据信用管理有限公司 User portrait construction method based on improved analytic hierarchy process
CN113762802A (en) * 2021-09-17 2021-12-07 昆明理工大学 Artificial intelligence PPP project financing evaluation system
CN113554366A (en) * 2021-09-23 2021-10-26 山东大学 Classification supervision method for disinfection product production enterprises and related equipment

Similar Documents

Publication Publication Date Title
CN112884590A (en) Power grid enterprise financing decision method based on machine learning algorithm
Rico-Juan et al. Machine learning with explainability or spatial hedonics tools? An analysis of the asking prices in the housing market in Alicante, Spain
Brauers et al. MULTIMOORA for the EU Member States updated with fuzzy number theory
CN110956273A (en) Credit scoring method and system integrating multiple machine learning models
Tanuwijaya et al. LQ45 stock index prediction using k-nearest neighbors regression
CN107833137A (en) Quantization trading strategies generation method and device, equipment and storage medium based on multiple-objection optimization
CN107609771A (en) A kind of supplier's value assessment method
Boguslauskas et al. The selection of financial ratios as independent variables for credit risk assessment
CN116468536A (en) Automatic risk control rule generation method
CN111126499A (en) Secondary clustering-based power consumption behavior pattern classification method
Lukić Analysis of the performance of the Serbian economy based on the MEREC-WASPAS method
Li Evaluation of the financial and economic development of the European Union member states on basis of multiple indicators changed to multiple objectives
CN110807543A (en) Investment portfolio optimization method and device based on group decision intelligent search
Önder et al. Comparing macroeconomic performance of the union for the mediterranean countries using grey relational analysis and multi-dimensional scaling
Zhang et al. Multi-factor stock selection model based on kernel support vector machine
Nicita Who benefits from export-led growth? Evidence from Madagascar's textile and apparel industry
Rong Dynamic Cause Analysis of Quantitative Investment Using Grey Correlation Analysis
CN112365120A (en) Intelligent business strategy generation method based on three decisions
Lin et al. Smart risk management with financial big data
Wei et al. Stock Investment Strategy Driven by EIS Events Based on Decision Tree Model
Mishra et al. Financial trends prediction using the back propagation neural network and YQL
CN113095941A (en) Financial data analysis method and server
Zhulavskyi et al. Investment Potential of the Enterprise: Theory and Practice of Assessment
Zhang et al. A Research on the Target Customer Mining of Investment Advisory Products—A Case Study of a Securities Company
Niranjanamurthy et al. Research on Optimization Algorithm of Real Estate Enterprise Value Evaluation Based on BP Neural Network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination