CN115310999A - Enterprise power utilization behavior analysis method and system based on multilayer perceptron and sequencing network - Google Patents
Enterprise power utilization behavior analysis method and system based on multilayer perceptron and sequencing network Download PDFInfo
- Publication number
- CN115310999A CN115310999A CN202210737510.1A CN202210737510A CN115310999A CN 115310999 A CN115310999 A CN 115310999A CN 202210737510 A CN202210737510 A CN 202210737510A CN 115310999 A CN115310999 A CN 115310999A
- Authority
- CN
- China
- Prior art keywords
- enterprise
- power utilization
- model
- enterprises
- behavior analysis
- 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.)
- Granted
Links
- 238000004458 analytical method Methods 0.000 title claims abstract description 80
- 238000012163 sequencing technique Methods 0.000 title claims abstract description 52
- 230000006399 behavior Effects 0.000 claims abstract description 204
- 239000013598 vector Substances 0.000 claims abstract description 80
- 238000012549 training Methods 0.000 claims abstract description 74
- 230000005611 electricity Effects 0.000 claims abstract description 73
- 238000000034 method Methods 0.000 claims abstract description 26
- 238000012545 processing Methods 0.000 claims abstract description 11
- 238000011156 evaluation Methods 0.000 claims description 99
- 230000006870 function Effects 0.000 claims description 33
- 238000005457 optimization Methods 0.000 claims description 14
- 238000012544 monitoring process Methods 0.000 claims description 11
- 238000010276 construction Methods 0.000 claims description 9
- 230000003993 interaction Effects 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000007405 data analysis Methods 0.000 claims description 4
- 230000001174 ascending effect Effects 0.000 claims description 3
- 210000002569 neuron Anatomy 0.000 description 8
- 238000004364 calculation method Methods 0.000 description 6
- 230000007547 defect Effects 0.000 description 3
- 238000012417 linear regression Methods 0.000 description 3
- 238000007477 logistic regression Methods 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 230000001186 cumulative effect Effects 0.000 description 2
- 238000013210 evaluation model Methods 0.000 description 2
- 239000000284 extract Substances 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000002372 labelling Methods 0.000 description 2
- 239000011159 matrix material Substances 0.000 description 2
- 238000012986 modification Methods 0.000 description 2
- 230000004048 modification Effects 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 238000011002 quantification Methods 0.000 description 2
- 238000011038 discontinuous diafiltration by volume reduction Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002452 interceptive effect Effects 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 230000009466 transformation Effects 0.000 description 1
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
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
-
- 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
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
- Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
- Y02P90/80—Management or planning
- Y02P90/82—Energy audits or management systems therefor
Landscapes
- Business, Economics & Management (AREA)
- Engineering & Computer Science (AREA)
- Strategic Management (AREA)
- Finance (AREA)
- Economics (AREA)
- Development Economics (AREA)
- Accounting & Taxation (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Health & Medical Sciences (AREA)
- Entrepreneurship & Innovation (AREA)
- Marketing (AREA)
- Physics & Mathematics (AREA)
- General Business, Economics & Management (AREA)
- Tourism & Hospitality (AREA)
- General Health & Medical Sciences (AREA)
- Water Supply & Treatment (AREA)
- Data Mining & Analysis (AREA)
- Public Health (AREA)
- Human Resources & Organizations (AREA)
- Primary Health Care (AREA)
- Game Theory and Decision Science (AREA)
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses an enterprise electricity utilization behavior analysis method and system based on a multilayer perceptron and a sequencing network, wherein the method comprises the following steps: constructing a power utilization characteristic index set to obtain a model index characteristic vector, and grading power utilization behaviors of all indexes; constructing a training data set which comprises a feature vector set formed by model index feature vectors of a plurality of enterprises and supervision information; respectively carrying out multilayer perceptron processing on model index feature vectors of a plurality of enterprises in the training data set to obtain corresponding model index feature coding sets, and training an enterprise power utilization behavior analysis model based on a sequencing network; and analyzing the power utilization behaviors of the enterprises according to the enterprise power utilization behavior analysis model and the benchmark enterprises screened from the training data set. The invention can better utilize multi-dimensional power big data, improve the efficiency, accuracy and expandability of the model for analyzing the power utilization behavior of enterprises, and greatly reduce the cost.
Description
Technical Field
The invention belongs to the technical field of electric power big data processing, and relates to an enterprise electricity utilization behavior analysis method and system based on a multilayer perceptron and a sequencing network.
Background
In the prior art, the traditional expert experience model is generally used for constructing the enterprise credit score, and the traditional expert experience model generally selects a few important data indexes and utilizes expert experience to make judgment so as to obtain the evaluation of the enterprise power utilization behavior. However, the expert experience model also has the disadvantages that: firstly, when a large amount of power data needs to be analyzed and processed, the labor cost of the expert experience model is overlarge; secondly, the expert experience model can only evaluate the power behavior of the enterprise in the current situation, and the model is difficult to expand; thirdly, when the dimensionality of the large power data is large, the expert experience model has the problems that the subjectivity is strong, and the obtained analysis result of the power utilization behavior of the enterprise is often not accurate enough.
Based on this, when an evaluation model is constructed to evaluate the enterprise electricity utilization behavior, the following problems are often faced: the manual labeling of a large amount of power data requires quantitative analysis labeling, the cost is high, the labeled data has strong subjectivity, the accuracy cannot be guaranteed, and the evaluation model of the power utilization behavior of an enterprise is difficult to obtain by using methods such as logistic regression, numerical value regression and linear regression.
Disclosure of Invention
In order to solve the defects in the prior art, the invention aims to provide the enterprise power utilization behavior analysis method and system based on the multilayer perceptron and the sequencing network, so that multi-dimensional power big data can be better utilized, the efficiency, the accuracy and the expandability of the model for enterprise power utilization behavior analysis are improved, and the cost is greatly reduced.
In order to achieve the above purpose, the invention adopts the following technical scheme:
the enterprise electricity utilization behavior analysis method based on the multilayer perceptron and the sequencing network comprises the following steps:
step 1, constructing a power utilization characteristic index set to obtain a model index characteristic vector, and grading power utilization behaviors of all indexes;
step 2, constructing a training data set, wherein the training data set comprises a feature vector set formed by model index feature vectors of a plurality of enterprises and monitoring information which is obtained based on the power utilization behavior scores of all indexes and represents the primary evaluation results of the power utilization behaviors among different enterprises;
step 3, respectively carrying out multilayer perceptron processing on model index feature vectors of a plurality of enterprises in the training data set to obtain corresponding model index feature coding sets, and training an enterprise electricity utilization behavior analysis model based on a sequencing network by adopting the model index feature coding sets and the monitoring information;
and 4, analyzing the power utilization behaviors of the enterprises according to the power utilization behavior analysis models of the enterprises and the benchmark enterprises screened from the training data set.
The invention further comprises the following preferred embodiments:
preferably, in step 1, a power utilization characteristic index set comprising power utilization load level indexes, power utilization specification indexes and power utilization interaction capacity indexes of an enterprise is constructed, and all indexes in the power utilization characteristic index set are sequentially spliced to obtain a model index characteristic vector; and scoring the power utilization behavior of each index according to expert experience so as to construct supervision information.
Preferably, step 2 specifically comprises:
step 2.1, constructing a feature vector set of a training data set: taking the model index feature vector of each enterprise as a feature vector of a training data set, and collecting the model index feature vectors of a plurality of enterprises to form a feature vector set;
step 2.2, constructing supervision information of a training data set: pairing and sampling the enterprises in the training data set, endowing each pair of enterprises with a label representing the primary evaluation result of the power utilization behavior among the enterprises, and forming supervision information by the label representing the primary evaluation result of the power utilization behavior among different enterprises;
wherein, the value mode of the label is as follows:
the power utilization behavior scores of the enterprises obtained by accumulating the power utilization behavior scores of the indexes in the step 1 are used for preliminarily evaluating the power utilization behaviors among different enterprises;
comparing the initial evaluation scores of the first enterprise and the second enterprise in each pair of enterprises, and if the first enterprise is higher than the second enterprise by a set score, marking the label as 1 to indicate that the power utilization behavior of the first enterprise is better; if the second enterprise is higher than the first enterprise by a set point, the label is marked as-1, and the power utilization behavior of the second enterprise is better; otherwise, the label is marked as 0, which indicates that the electricity consumption behaviors of the first enterprise and the second enterprise are the same.
Preferably, step 3 specifically comprises:
step 3.1, respectively carrying out multilayer perceptron processing on model index feature vectors a of a plurality of enterprises in the training data set to obtain a model index feature coding set b:
b=mlp(a,w)
wherein w is each index feature weight set in the parameter a of the multilayer perceptron mlp;
step 3.2, constructing a sequencing network model, and constructing a loss function based on the sequencing network model, the supervision information and the symmetry of the sequencing relation;
3.3, constructing a random gradient descent optimization model based on the loss function so as to learn the weights in the model index feature coding set;
and 3.4, constructing and training an enterprise power utilization behavior analysis model by adopting a random gradient descent optimization method according to the model index feature coding set and the supervision information.
Preferably, in step 3.2, the ranking network model is constructed as expressed in the following formula,
in the formula:
P ij representing the probability that the enterprise i is considered to be better than the enterprise j in power utilization behavior evaluation by the sequencing network model;
representing that the power utilization behavior evaluation of the enterprise i is better than that of the enterprise j;
σ is a hyper parameter constant;
s i =s(b i ) Representing the enterprise power utilization behavior evaluation value given to enterprise i by the enterprise power utilization behavior analysis model;
s j =s(b j ) And representing the enterprise electricity utilization behavior evaluation value given to the enterprise j by the enterprise electricity utilization behavior analysis model.
Preferably, in step 3.2, based on the ranking network model, the supervision information, and the symmetry of the ranking relationship, a loss function is constructed, specifically as follows:
constructing a cross entropy loss function, expressed by the following formula,
simplifying the cross entropy loss function formula to obtain:
in the formula:
C ij a value representing a loss function for business i and business j;
S ij is of the formula [ -1,0,1]The relation of the enterprise electricity utilization behaviors between the enterprise i and the enterprise j obtained according to the supervision information is shown;
defining a pairing set I according to the symmetry of the ordering relationship, namely the symmetry of the relatively good relationship between the first enterprise and the second enterprise, and requiring that the first enterprise has better power utilization behavior than the second enterprise;
a loss function C of set I is defined, the formula being:
in the formula, satisfy s i >s j 。
Preferably, step 3.3, a stochastic gradient descent optimization model is constructed based on the loss function, expressed in the following formula,
in the formula: w is a k Representing the kth weight of a feature weight set w in the model index feature coding set b;
η represents the learning rate;
λ i is a coefficient for business i.
Preferably, step 3.4, an enterprise electricity utilization behavior analysis model is constructed based on the stochastic gradient descent optimization model, and is expressed by the following formula,
s i ,s j =f(a i ,a j ;w) (4)
in the formula: w is a characteristic weight set in the model index characteristic coding set b obtained in the step 3.3;
f(a i ,a j (ii) a w) represents model indexes according to power utilization behaviors of enterprise i and enterprise jFeature vector a i ,a j Outputting a function of the evaluation value;
s i =s(b i ) Representing the enterprise power utilization behavior evaluation value given to enterprise i by the enterprise power utilization behavior analysis model;
s j =s(b j ) And representing the enterprise electricity utilization behavior evaluation value given to the enterprise j by the enterprise electricity utilization behavior analysis model.
Preferably, in step 4, a reference enterprise is introduced, and the evaluation value output by the enterprise to be analyzed by the enterprise electricity consumption behavior analysis model is integrated to an evaluation score of 0-100, specifically as follows:
step 4.1, obtaining the evaluation value of each enterprise in the training data set by adopting the enterprise power utilization behavior analysis model, sequencing all the enterprises in the training data set in an ascending order according to the evaluation value, dividing the first enterprise definition evaluation into 0 score, dividing the last enterprise definition evaluation into 100 scores, and sampling 100 enterprises at average intervals as the reference enterprise b 1 -b 100 The model index feature set is A = { a = 1 ,a 2 ,……a 100 And defining the electricity utilization behavior evaluation score S 1 -S 100 Sequentially comprises {1,2, \8230;, 100};
step 4.2, assuming that the model index characteristic vector of the enterprise to be evaluated is a;
firstly, model index characteristic vectors a of an enterprise to be evaluated and each reference enterprise b in a characteristic set A of reference enterprises are sequentially compared i Vector a of i A pair of the power utilization behavior evaluation values s and s is formed and input into the enterprise power utilization behavior analysis model in the step 3.4 i Based on this, a pair of adjacent reference enterprises b is found i And b i+1 Such that s is located at s i And s i+1 Obtaining the evaluation score interval of the enterprise to be evaluated;
obtaining weight according to the evaluation value, and dividing the reference enterprise b i And b i+1 Evaluation score of (S) i And S i+1 Weighting to obtain an electricity utilization behavior evaluation score S of the enterprise to be evaluated:
S=S i+1 *(s-s i )/(s i+1 -s i )+S i *(s i+1 -s)/(s i+1 -s i )。
the invention also provides an enterprise electricity consumption behavior analysis system based on the multilayer perceptron and the sequencing network, which comprises:
the model index characteristic vector construction and index scoring module is used for constructing a power utilization characteristic index set to obtain a model index characteristic vector and scoring each index by power utilization behaviors;
the system comprises a training data set construction module, a power consumption behavior evaluation module and a power consumption behavior evaluation module, wherein the training data set construction module is used for constructing a training data set, and comprises a feature vector set formed by model index feature vectors of a plurality of enterprises and monitoring information which is obtained based on the power consumption behavior scores of all indexes and represents the primary evaluation results of the power consumption behaviors among different enterprises;
the enterprise power consumption behavior analysis model training module is used for respectively carrying out multilayer perceptron processing on model index feature vectors of a plurality of enterprises in the training data set to obtain corresponding model index feature coding sets, and training the enterprise power consumption behavior analysis model based on the sequencing network by adopting the model index feature coding sets and the monitoring information;
and the power utilization behavior analysis module of the enterprise is used for analyzing the power utilization behavior of the enterprise according to the power utilization behavior analysis model of the enterprise and the benchmark enterprise screened out from the training data set.
Compared with the prior art, the invention has the beneficial effects that:
the method extracts the power utilization characteristics of the enterprise by a data-based classification method, converts the data requirements on the power utilization behaviors of the enterprise from quantification to qualitative, and greatly increases the expandability of the model; secondly, by using the characteristic vector of the model index and the supervision information, an enterprise electricity consumption behavior analysis model is constructed and trained by a method of a multilayer perceptron and a sequencing network, and compared with methods such as logistic regression, numerical regression, linear regression and the like, the method has the advantages that:
the invention relates to an enterprise power consumption behavior analysis model based on a multilayer perceptron and a sequencing network, wherein index scoring in step 1 only needs to help an expert to preliminarily evaluate power consumption behaviors among different enterprises to obtain supervision information, specific index scoring is not used for model training, for example, 400 points of a first enterprise and 250 points of a second enterprise, and index scoring is not used for model training. The accurate and quantitative power utilization behavior evaluation score of the training sample is not required, and only the qualitative evaluation of the power utilization behavior of each pair of enterprises in the training data set is required, so that the cost of data annotation is greatly reduced.
The enterprise power utilization behavior analysis model can overcome the defect that a traditional expert experience model has strong subjectivity, can better utilize multidimensional electric big data, ensures the accuracy of enterprise power utilization characteristic extraction, and overcomes the subjectivity and inaccuracy of an expert manual evaluation method.
Drawings
FIG. 1 is a flow chart of an enterprise electricity utilization behavior analysis method based on a multilayer perceptron and a sequencing network.
Detailed Description
The invention is further described below with reference to the accompanying drawings. The following examples are only for illustrating the technical solutions of the present invention more clearly, and the protection scope of the present invention is not limited thereby.
As shown in fig. 1, the present invention provides a method for analyzing the electricity consumption behavior of an enterprise based on a multi-layer perceptron and a ranking network, and in a further preferred but non-limiting embodiment of the invention, the method comprises the following steps 1-4:
step 1, constructing a power utilization characteristic index set to obtain a model index characteristic vector, and scoring each index by power utilization behavior;
further preferably, a power utilization characteristic index set comprising power utilization load level indexes, power utilization standard indexes and power utilization interaction capacity indexes of enterprises is constructed, and all indexes in the power utilization characteristic index set are sequentially spliced to obtain model index characteristic vectors; and scoring the power utilization behavior of each index according to expert experience so as to construct supervision information. The method comprises the following specific steps:
1) Extracting the power load level indexes of the enterprise, including the power consumption growth rate index and the power consumption stability index of the enterprise;
the electricity consumption increase rate index comprises an index of the same-proportion increase rate of the electricity consumption of an enterprise in the last 12 months compared with the electricity consumption of the enterprise in the last 12 months, and the electricity consumption behavior scoring calculation method comprises the following steps:
the power consumption of the enterprise increases proportionally by the same ratio of 12 months and the same ratio = u,
u >0.8, score: 100 minutes;
u = [0.4,0.8], score: 50 minutes;
u = (0, 0.4), score: 30 minutes;
the power utilization stability indexes comprise a level index of power utilization fluctuation of an enterprise in a local peer in the last 12 months and a power utilization difference index of the enterprise in the last 12 months, and the power utilization behavior scoring calculation method comprises the following steps:
the fluctuation rate of the electricity consumption of the enterprise in the last 12 months = v, the electricity quantity difference of the enterprise in the last 12 months = w,
v <0.2, score: 100 minutes;
v = [0.2,0.6], score: 60 minutes;
v >0.6, score 0;
w <0.2, score 100;
w = [0.2,0.7], score 60;
w >0.7, score 0;
2) Extracting power utilization standard indexes of enterprises, including power stealing behavior indexes and default power utilization behavior indexes;
the electricity stealing behavior indexes comprise an electricity stealing frequency index of about 24 months and an electricity stealing default total cost index of about 24 months, and the electricity stealing behavior scoring calculation method comprises the following steps:
the electricity stealing times in the last 24 months = q, the total cost of electricity stealing default in the last 24 months = r ten thousand yuan,
q <2, score: 100 minutes;
q = [2,10], score; 40 minutes;
q >10, score: 0 minute;
r <1, score: 100 minutes;
r = [1,10], score: 40 minutes;
r >10, score: 0 minute;
the default electricity consumption behavior indexes comprise an index of default electricity consumption times in nearly 24 months and an index of total default electricity consumption cost in nearly 24 months, and the power consumption behavior scoring calculation method comprises the following steps:
the number of times of default electricity utilization = s in the last 24 months, the total cost of default electricity utilization = t in the last 24 months,
s <2, score: 100 minutes;
s = [2,10], score; 40 minutes;
s >10, score: 0 minute;
t <1, score: 100 minutes;
t = [1,10], score: 40 minutes;
t >10, score: and 0 minute.
3) Extracting power utilization interaction capacity indexes of enterprises, including capacity indexes of the enterprises and operators for carrying out equipment capacity expansion in an interaction mode;
the capacity indexes of the enterprise and the operator for carrying out equipment capacity expansion in an interactive mode comprise a capacity increasing time index accumulated in about 24 months and a capacity reducing time index accumulated in about 24 months, and the power consumption behavior scoring calculation method comprises the following steps:
the cumulative volume increase times z1 in the last 24 months, the cumulative volume reduction times z2 in the last 24 months,
z1>8, score: 100 minutes;
z1= [4,8], score: 70 minutes;
z1<4, score: 20 min;
z2<4, score: 100 minutes;
z2= [4,10], score: 70 minutes;
z2>10, score: and 0 point.
Splicing the extracted enterprise electricity load level, the extracted enterprise electricity consumption behavior mode and the extracted enterprise electricity consumption interaction capacity indexes according to the index sequence to form a model index feature vector; (that is, the model index feature vector includes the same-proportion growth rate u, fluctuation rate v, difference degree w, etc., and does not include index score)
The obtained scores are used for subsequently constructing supervision information.
The above-mentioned contents are a series of steps of fixed flow, weight and the like in the neural network, and are implemented in the following multi-layer perceptron.
Step 2, constructing a training data set which comprises a feature vector set formed by model index feature vectors of a plurality of enterprises and supervision information which is obtained based on the scores of the power utilization behaviors of all indexes and represents the primary evaluation results of the power utilization behaviors among different enterprises, and specifically comprising the following steps of:
step 2.1, constructing a feature vector set of a training data set: taking the model index feature vector of each enterprise as a feature vector of a training data set, and collecting the model index feature vectors of a plurality of enterprises to form a feature vector set;
step 2.2, constructing supervision information of a training data set: pairing and sampling the enterprises in the training data set, endowing each pair of enterprises with a label representing the primary evaluation result of the power utilization behavior among the enterprises, and forming supervision information by the label representing the primary evaluation result of the power utilization behavior among different enterprises;
wherein, the value mode of label does:
accumulating the power utilization behavior scores of the indexes in the step 1 to obtain an enterprise power utilization behavior initial evaluation score, and using the enterprise power utilization behavior initial evaluation score to initially evaluate power utilization behaviors among different enterprises;
comparing the initial evaluation scores of the first enterprise and the second enterprise in each pair of enterprises, and if the first enterprise is higher than the second enterprise by a set value, such as 150 points, marking the label as 1, which indicates that the power utilization behavior of the first enterprise is better; if the second enterprise is 150 points higher than the first enterprise, the label is marked as-1, and the power utilization behavior of the second enterprise is better; otherwise, the label is marked as 0, which indicates that the electricity consumption behaviors of the first enterprise and the second enterprise are the same.
That is, if the preliminary evaluation score difference is less than 150, it is marked as 0.
Step 3, respectively carrying out multilayer perceptron processing on model index feature vectors of a plurality of enterprises in the training data set to obtain corresponding model index feature coding sets, and training an enterprise electricity utilization behavior analysis model based on a sequencing network by adopting the model index feature coding sets and the monitoring information;
the method specifically comprises the following steps:
step 3.1, respectively carrying out multilayer perceptron processing on model index feature vectors a of a plurality of enterprises in the training data set to obtain a model index feature coding set b:
b=mlp(a,w)
wherein w is each index feature weight set in the parameter a of the multilayer perceptron mlp;
and correspondingly outputting a model index feature coding set b by the model index feature vector a of each enterprise. And the value of b is input into the enterprise power consumption behavior analysis model, and finally the enterprise power consumption behavior evaluation value s is obtained, and the loss function value C can be obtained from s.
Step 3.1 is an automated process, inputting a, i.e. automatically learning and outputting b, which is common knowledge in the field.
Step 3.1.1, constructing a plurality of neurons as input layers of a multilayer perceptron mlp, and sequentially inputting model index characteristic vectors a of enterprises in a training data set into each neuron; the number of the neurons is set manually, the specific number is adjusted according to the effect of the model, and the specific number is irrelevant to the number of enterprises in the training data set and the number of index features in the model index feature vector a.
Step 3.1.2, constructing a plurality of hidden layers, and encoding input eigenvectors, namely, firstly multiplying a matrix by the input eigenvectors to obtain a new vector as an encoding result, wherein the matrix value is a part of the parameter weight of the multilayer perceptron and can be obtained by training and learning, each bit of the new vector corresponds to a neuron, and a Tanh function is used for each neuron as an activation function to determine whether the current neuron outputs the result;
in order to reduce the calculation cost of feature coding, a Tanh function is used as an activation function, and the formula is shown as follows;
where x is the value of the vector bit corresponding to the neuron and e is a natural constant.
Step 3.1.3, constructing a plurality of neurons as output layers of the multilayer perceptron, and sequentially outputting feature coding results of feature vectors, wherein the obtained feature coding result is a high-dimensional vector b which is used as a feature transformation result of the electricity consumption behavior model index feature vectors, and the formula is as follows:
b=mlp(a,w)
wherein w is the parameter weight of the multilayer perceptron mlp;
the encoding in the present invention means: carrying out weight assignment on each index in the vector;
step 3.2, constructing a sequencing network model, and constructing a loss function based on the sequencing network model, the supervision information and the symmetry of the sequencing relation;
1) The ranking network model is constructed as represented by the following formula,
in the formula:
P ij representing the probability that the enterprise i is considered to be better than the enterprise j in power utilization behavior evaluation by the sequencing network model;
the expression assumes that the power utilization behavior evaluation of the enterprise i is better than that of the enterprise j;
σ is a hyper-parameter constant;
s i =s(b i ) Representing an enterprise power utilization behavior evaluation value given to an enterprise i by the enterprise power utilization behavior analysis model;
s j =s(b j ) And representing the enterprise electricity utilization behavior evaluation value given to the enterprise j by the enterprise electricity utilization behavior analysis model.
2) Constructing a loss function based on the sequencing network model, the supervision information and the symmetry of the sequencing relation, wherein the loss function specifically comprises the following steps:
constructing a cross entropy loss function, expressed by the following formula,
the formula is simplified to obtain:
in the formula:
C ij a value representing a loss function for business i and business j;
S ij is of the formula [ -1,0,1]The relation of the enterprise electricity utilization behaviors between the enterprise i and the enterprise j obtained according to the supervision information is shown;
defining a pairing set I according to the symmetry of the ordering relationship, namely the symmetry of the relatively good relationship between the first enterprise and the second enterprise, and requiring that the first enterprise has better power utilization behavior than the second enterprise;
a loss function C of set I is defined, the formula being:
in the formula, satisfy s i >s j 。
3.3, constructing a random gradient descent optimization model based on the loss function so as to learn the weights in the model index feature coding set;
and constructing a random gradient descent optimization model based on the loss function, and expressing the random gradient descent optimization model by the following formula,
in the formula: w is a k Index of expression modelCharacterizing the kth weight of a feature weight set w in the coding set b;
η represents the learning rate;
λ i is a coefficient for business i;
the value of the loss function C can be calculated by step 3.2 using the feature encoding result b.
And 3.4, constructing and training an enterprise power utilization behavior analysis model by adopting a random gradient descent optimization method according to the model index feature coding set and the supervision information.
An enterprise electricity utilization behavior analysis model is constructed based on the stochastic gradient descent optimization model and is expressed by the following formula,
s i ,s j =f(a i ,a j ;w) (4)
in the formula: w is a characteristic weight set in the model index characteristic coding set b obtained in the step 3.3;
f(a i ,a j (ii) a w) represents a model index feature vector a according to the power consumption behaviors of the enterprise i and the enterprise j i ,a j Outputting a function of the evaluation value;
s i =s(b i ) Representing the enterprise power utilization behavior analysis model to the enterprise power utilization behavior evaluation value of the enterprise i, wherein the model input is a vector b i Coding the result for the characteristics of the enterprise i;
s j =s(b j ) Representing the enterprise power utilization behavior analysis model to the enterprise power utilization behavior evaluation value of the enterprise j, wherein the model input is a vector b j And coding the result for the characteristics of the enterprise j.
And 4, analyzing the power utilization behaviors of the enterprises according to the power utilization behavior analysis models of the enterprises and the benchmark enterprises screened from the training data set.
In the invention, when the power consumption behavior of an actual enterprise is analyzed, interpretability is considered, the standard score is made by percentage, and the whole real number range is output by using the power consumption behavior analysis model of the enterprise and cannot be directly fitted to the standard score, so that the evaluation value of the output of the neural network is regulated to 0-100 through the step 4. For this purpose, if a power consumption behavior evaluation score of 0-100 is to be obtained, a benchmark enterprise must be used, and this requirement cannot be met by comparing the relative quality of the power consumption behaviors of the two enterprises.
In the step 4, a reference enterprise is introduced, and the evaluation values output by the enterprise to be analyzed through the enterprise power utilization behavior analysis model are integrated to 0-100 evaluation scores, wherein the evaluation scores are as follows:
step 4.1, obtaining the evaluation value of each enterprise in the training data set by adopting the enterprise power utilization behavior analysis model, sequencing all the enterprises in the training data set in an ascending order according to the evaluation value, dividing the first enterprise definition evaluation into 0 score, dividing the last enterprise definition evaluation into 100 scores, and sampling 100 enterprises at average intervals as the reference enterprise b 1 -b 100 The set of model index features is A = { a = 1 ,a 2 ,……a 100 And defining its electricity consumption behavior evaluation score S 1 -S 100 Sequentially comprises {1,2, \8230;, 100};
step 4.2, assuming that the model index characteristic vector of the enterprise to be evaluated is a;
firstly, model index feature vectors a of an enterprise to be evaluated and each reference enterprise b in a feature set A of reference enterprises are sequentially combined i Vector a of i A pair of the power utilization behavior evaluation values s and s is formed and input into the enterprise power utilization behavior analysis model in the step 3.4 i The method comprises the steps of substituting the enterprise to be evaluated and each reference enterprise into the model to operate once, and operating for 100 times in total to obtain s and s 1 -s 100 Based on this, a pair of adjacent reference enterprises b is found i And b i+1 Such that s is located at s i And s i+1 Obtaining the evaluation score interval of the enterprise to be evaluated;
obtaining weight according to the evaluation value, and dividing the reference enterprise b i And b i+1 Evaluation score of (S) i And S i+1 Weighting to obtain an electricity utilization behavior evaluation score S of the enterprise to be evaluated:
S=S i+1 *(s-s i )/(s i+1 -s i )+S i *(s i+1 -s)/(s i+1 -s i )。
assuming that i =1, the power consumption behavior evaluation value of the reference enterprise 1 is-20.0, the evaluation is 1, the evaluation value of the reference enterprise 2 is 10.0, the evaluation is 2, and the evaluation value of the enterprise to be evaluated is-5.0, then the section where the evaluation score of the enterprise to be evaluated is located is between 1 and 2;
the power utilization behavior evaluation of the enterprise to be evaluated is divided into a weighted combination of evaluation scores of a reference enterprise 1 and a reference enterprise 2, the weight of the weighted combination is determined according to the power utilization behavior evaluation value of the enterprise, and the method specifically comprises the following steps: 2 (-5.0- (-20.0))/(10.0- (-20.0)) +1 (10.0- (-5.0))/(10.0- (-20.0)) =1.5.
The invention relates to an enterprise electricity consumption behavior analysis system based on a multilayer perceptron and a sequencing network, which runs the enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network, and the system comprises:
the model index characteristic vector construction and index grading module is used for constructing a power utilization characteristic index set to obtain a model index characteristic vector and grading each index in a power utilization behavior;
the system comprises a training data set construction module, a data analysis module and a data analysis module, wherein the training data set construction module is used for constructing a training data set, and comprises a feature vector set formed by model index feature vectors of a plurality of enterprises and monitoring information which is obtained based on the power utilization behavior scores of all indexes and represents the primary evaluation results of the power utilization behaviors among different enterprises;
the enterprise power consumption behavior analysis model training module is used for respectively carrying out multilayer perceptron processing on model index feature vectors of a plurality of enterprises in the training data set to obtain corresponding model index feature coding sets, and training an enterprise power consumption behavior analysis model based on a sequencing network by adopting the model index feature coding sets and the monitoring information;
and the power utilization behavior analysis module of the enterprise is used for analyzing the power utilization behavior of the enterprise according to the power utilization behavior analysis model of the enterprise and the benchmark enterprise screened out from the training data set.
Compared with the prior art, the invention has the beneficial effects that:
the invention adopts a sorting method to convert the dimensionality of the electric power data in an indeterminate way into a qualitative problem, and the method also obviously improves the accuracy of the marked data. Aiming at the problems that the multi-dimension of the electric power data is not easy to utilize and weak supervised learning is difficult to realize, the multi-layer perceptron is adopted to process the electric power data, so that the problems are met; the invention adopts a multilayer perceptron and a sequencing network to construct an enterprise power consumption behavior analysis model, realizes automatic enterprise power consumption behavior analysis, greatly reduces the cost compared with the evaluation of artificial experts, and has good expansibility, thereby providing paid power data products and services such as financial wind control, customer recommendation, data sharing and the like for financial institutions better.
The method extracts the power utilization characteristics of the enterprise by a data-based classification method, converts the data requirements on the power utilization behaviors of the enterprise from quantification to qualitative, and greatly increases the expandability of the model; secondly, by using the characteristic vector of the model index and the supervision information, an enterprise electricity consumption behavior analysis model is constructed and trained by a method of a multilayer perceptron and a sequencing network, and compared with methods such as logistic regression, numerical regression, linear regression and the like, the method has the advantages that:
the invention relates to an enterprise power consumption behavior analysis model based on a multilayer perceptron and a sequencing network, wherein index scoring in step 1 only needs to help an expert to preliminarily evaluate power consumption behaviors among different enterprises to obtain supervision information, specific index scoring is not used for model training, for example, 400 points of a first enterprise and 250 points of a second enterprise, and index scoring is not used for model training. The accurate and quantitative power utilization behavior evaluation score of the training sample is not required, and only the qualitative evaluation of the power utilization behavior of each pair of enterprises in the training data set is required, so that the cost of data annotation is greatly reduced.
The enterprise power utilization behavior analysis model can overcome the defect that a traditional expert experience model has strong subjectivity, can better utilize multidimensional electric big data, ensures the accuracy of enterprise power utilization characteristic extraction, and overcomes the subjectivity and inaccuracy of an expert manual evaluation method.
The present applicant has described and illustrated embodiments of the present invention in detail with reference to the accompanying drawings, but it should be understood by those skilled in the art that the above embodiments are only preferred embodiments of the present invention, and the detailed description is only for the purpose of helping the reader to better understand the spirit of the present invention, and not for the purpose of limiting the scope of the present invention, and on the contrary, any modifications or modifications based on the spirit of the present invention should fall within the scope of the present invention.
Claims (10)
1. An enterprise electricity utilization behavior analysis method based on a multilayer perceptron and a sequencing network is characterized in that:
the method comprises the following steps:
step 1, constructing a power utilization characteristic index set to obtain a model index characteristic vector, and scoring each index by power utilization behavior;
step 2, constructing a training data set, wherein the training data set comprises a feature vector set formed by model index feature vectors of a plurality of enterprises and monitoring information which is obtained based on the power utilization behavior scores of all indexes and represents the primary evaluation results of the power utilization behaviors among different enterprises;
step 3, respectively carrying out multilayer perceptron processing on model index feature vectors of a plurality of enterprises in the training data set to obtain corresponding model index feature coding sets, and training an enterprise power utilization behavior analysis model based on a sequencing network by adopting the model index feature coding sets and the monitoring information;
and 4, analyzing the power utilization behaviors of the enterprises according to the power utilization behavior analysis models of the enterprises and the benchmark enterprises screened from the training data set.
2. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network of claim 1, characterized in that:
in the step 1, a power utilization characteristic index set comprising power utilization load level indexes, power utilization standard indexes and power utilization interaction capacity indexes of enterprises is constructed, and all indexes in the power utilization characteristic index set are sequentially spliced to obtain a model index characteristic vector; and scoring the power utilization behavior of each index according to expert experience so as to construct supervision information.
3. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network according to claim 1, characterized in that:
the step 2 specifically comprises the following steps:
step 2.1, constructing a feature vector set of a training data set: taking the model index feature vector of each enterprise as a feature vector of a training data set, and collecting the model index feature vectors of a plurality of enterprises to form a feature vector set;
step 2.2, constructing supervision information of a training data set: carrying out pairing sampling on enterprises in the training data set, endowing each pair of enterprises with a label representing the initial evaluation result of the power consumption behavior among the enterprises, and forming supervision information by the label representing the initial evaluation result of the power consumption behavior among different enterprises;
wherein, the value mode of the label is as follows:
accumulating the power utilization behavior scores of the indexes in the step 1 to obtain an enterprise power utilization behavior initial evaluation score, and using the enterprise power utilization behavior initial evaluation score to initially evaluate power utilization behaviors among different enterprises;
comparing the initial evaluation scores of the first enterprise and the second enterprise in each pair of enterprises, and if the first enterprise is higher than the second enterprise by a set score, marking the label as 1 to indicate that the power utilization behavior of the first enterprise is better; if the second enterprise is higher than the first enterprise by a set point, the label is marked as-1, and the power utilization behavior of the second enterprise is better; otherwise, the label is marked as 0, which indicates that the electricity consumption behaviors of the first enterprise and the second enterprise are the same.
4. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network according to claim 1, characterized in that:
the step 3 specifically comprises the following steps:
step 3.1, respectively carrying out multilayer perceptron processing on model index feature vectors a of a plurality of enterprises in the training data set to obtain a model index feature coding set b:
b=mlp(a,w)
w is each index feature weight set in the parameter a of the multilayer perceptron mlp;
step 3.2, constructing a sequencing network model, and constructing a loss function based on the sequencing network model, the supervision information and the symmetry of the sequencing relation;
3.3, constructing a random gradient descent optimization model based on the loss function so as to learn the weights in the model index feature coding set;
and 3.4, constructing and training an enterprise power utilization behavior analysis model by adopting a random gradient descent optimization method according to the model index feature coding set and the supervision information.
5. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network according to claim 4, characterized in that:
in step 3.2, a ranking network model is constructed and expressed by the following formula,
in the formula:
P ij representing the probability that the enterprise i is considered to be better than the enterprise j in power utilization behavior evaluation by the sequencing network model;
the expression assumes that the power utilization behavior evaluation of the enterprise i is better than that of the enterprise j;
σ is a hyper parameter constant;
s i =s(b i ) Representing the enterprise power utilization behavior evaluation value given to enterprise i by the enterprise power utilization behavior analysis model;
s j =s(b j ) And representing the enterprise electricity utilization behavior evaluation value given to the enterprise j by the enterprise electricity utilization behavior analysis model.
6. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and sequencing network of claim 4, characterized in that:
in step 3.2, based on the sequencing network model, the supervision information and the symmetry of the sequencing relationship, a loss function is constructed, which specifically comprises the following steps:
constructing a cross entropy loss function, expressed by the following formula,
simplifying the cross entropy loss function formula to obtain:
in the formula:
C ij a value representing a loss function for business i and business j;
S ij are of the formula [ -1,0,1]The relation of the enterprise electricity utilization behaviors between the enterprise i and the enterprise j obtained according to the supervision information is shown;
defining a pairing set I according to the symmetry of the ordering relationship, namely the symmetry of the relatively good relationship between the first enterprise and the second enterprise, and requiring that the first enterprise has better power utilization behavior than the second enterprise;
a loss function C of set I is defined, the formula being:
in the formula, satisfy s i >s j 。
7. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and sequencing network of claim 4, characterized in that:
step 3.3, constructing a random gradient descent optimization model based on the loss function, and expressing the random gradient descent optimization model by the following formula,
in the formula: w is a k Representing the kth weight of a feature weight set w in the model index feature coding set b;
η represents the learning rate;
λ i is a coefficient for business i.
8. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network according to claim 4, characterized in that:
and 3.4, constructing an enterprise electricity utilization behavior analysis model based on the stochastic gradient descent optimization model, and expressing the model by the following formula,
s i ,s j =f(a i ,a j ;w) (4)
in the formula: w is a feature weight set in the model index feature coding set b obtained in the step 3.3;
f(a i ,a j (ii) a w) represents a model index feature vector a according to the power consumption behaviors of the enterprise i and the enterprise j i ,a j Outputting a function of the evaluation value;
s i =s(b i ) Representing an enterprise power utilization behavior evaluation value given to an enterprise i by the enterprise power utilization behavior analysis model;
s j =s(b j ) And representing the enterprise electricity utilization behavior evaluation value given to the enterprise j by the enterprise electricity utilization behavior analysis model.
9. The enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network of claim 1, characterized in that:
in the step 4, a reference enterprise is introduced, and the evaluation value output by the enterprise to be analyzed through the enterprise power utilization behavior analysis model is integrated to an evaluation score of 0-100, specifically as follows:
step 4.1, obtaining the evaluation value of each enterprise in the training data set by adopting an enterprise power utilization behavior analysis model, sequencing all the enterprises in the training data set in an ascending order according to the evaluation values, dividing the first enterprise definition evaluation into 0, dividing the last enterprise definition evaluation into 100, and sampling 100 enterprises at average intervals as a reference enterprise b 1 -b 100 The set of model index features is A = { a = 1 ,a 2 ,……a 100 And defining its electricity consumption behavior evaluation score S 1 -S 100 Sequentially comprises {1,2, \8230;, 100};
step 4.2, assuming that the model index characteristic vector of the enterprise to be evaluated is a;
firstly, model index characteristic vectors a of an enterprise to be evaluated and each reference enterprise b in a characteristic set A of reference enterprises are sequentially compared i Vector a of i Forming a pair to be input into an enterprise electricity consumption behavior analysis model to obtain corresponding electricity consumption behavior evaluation values s and s i Based on this, a pair of adjacent reference enterprises b is found i And b i+1 Such that s is located at s i And s i+1 Get the weight according to the evaluation value, and get the reference enterprise b i And b i+1 Evaluation score of (S) i And S i+1 Weighting to obtain an electricity utilization behavior evaluation score S of the enterprise to be evaluated:
S=S i+1 *(s-s i )/(s i+1 -s i )+S i *(s i+1 -s)/(s i+1 -s i )。
10. the enterprise electricity consumption behavior analysis system based on the multilayer perceptron and the sequencing network runs the enterprise electricity consumption behavior analysis method based on the multilayer perceptron and the sequencing network according to any one of claims 1 to 9, and is characterized in that:
the system comprises:
the model index characteristic vector construction and index grading module is used for constructing a power utilization characteristic index set to obtain a model index characteristic vector and grading each index in a power utilization behavior;
the system comprises a training data set construction module, a data analysis module and a data analysis module, wherein the training data set construction module is used for constructing a training data set, and comprises a feature vector set formed by model index feature vectors of a plurality of enterprises and monitoring information which is obtained based on the power utilization behavior scores of all indexes and represents the primary evaluation results of the power utilization behaviors among different enterprises;
the enterprise power consumption behavior analysis model training module is used for respectively carrying out multilayer perceptron processing on model index feature vectors of a plurality of enterprises in the training data set to obtain corresponding model index feature coding sets, and training the enterprise power consumption behavior analysis model based on the sequencing network by adopting the model index feature coding sets and the monitoring information;
and the power utilization behavior analysis module of the enterprise is used for analyzing the power utilization behaviors of the enterprise according to the power utilization behavior analysis model of the enterprise and the benchmark enterprise screened out from the training data set.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210737510.1A CN115310999B (en) | 2022-06-27 | 2022-06-27 | Enterprise electricity behavior analysis method and system based on multi-layer perceptron and sequencing network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202210737510.1A CN115310999B (en) | 2022-06-27 | 2022-06-27 | Enterprise electricity behavior analysis method and system based on multi-layer perceptron and sequencing network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN115310999A true CN115310999A (en) | 2022-11-08 |
CN115310999B CN115310999B (en) | 2024-02-02 |
Family
ID=83856032
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202210737510.1A Active CN115310999B (en) | 2022-06-27 | 2022-06-27 | Enterprise electricity behavior analysis method and system based on multi-layer perceptron and sequencing network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN115310999B (en) |
Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190130247A1 (en) * | 2017-10-31 | 2019-05-02 | General Electric Company | Multi-task feature selection neural networks |
US20190130257A1 (en) * | 2017-10-27 | 2019-05-02 | Sentient Technologies (Barbados) Limited | Beyond Shared Hierarchies: Deep Multitask Learning Through Soft Layer Ordering |
US20190378017A1 (en) * | 2018-06-12 | 2019-12-12 | Sun-Yuan Kung | System and method for implementing a neural network |
WO2020000399A1 (en) * | 2018-06-29 | 2020-01-02 | 东莞理工学院 | Multi-agent deep reinforcement learning proxy method based on intelligent grid |
US20200074353A1 (en) * | 2018-09-04 | 2020-03-05 | Didi Research America, Llc | System and method for ride order dispatching and vehicle repositioning |
CN112633421A (en) * | 2021-03-09 | 2021-04-09 | 国网江西综合能源服务有限公司 | Method and device for detecting abnormal electricity utilization behavior of user |
US20210248450A1 (en) * | 2020-02-07 | 2021-08-12 | Google Llc | Sorting attention neural networks |
CN113988459A (en) * | 2021-11-11 | 2022-01-28 | 国网江苏省电力有限公司苏州供电分公司 | Small and medium-sized enterprise growth assessment method and system based on electric power marketing data |
CN114004530A (en) * | 2021-11-11 | 2022-02-01 | 国网江苏省电力有限公司苏州供电分公司 | Enterprise power credit score modeling method and system based on sequencing support vector machine |
CN114219606A (en) * | 2021-11-11 | 2022-03-22 | 国网四川省电力公司信息通信公司 | Power data-based power consumption enterprise credit evaluation method and system |
-
2022
- 2022-06-27 CN CN202210737510.1A patent/CN115310999B/en active Active
Patent Citations (10)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20190130257A1 (en) * | 2017-10-27 | 2019-05-02 | Sentient Technologies (Barbados) Limited | Beyond Shared Hierarchies: Deep Multitask Learning Through Soft Layer Ordering |
US20190130247A1 (en) * | 2017-10-31 | 2019-05-02 | General Electric Company | Multi-task feature selection neural networks |
US20190378017A1 (en) * | 2018-06-12 | 2019-12-12 | Sun-Yuan Kung | System and method for implementing a neural network |
WO2020000399A1 (en) * | 2018-06-29 | 2020-01-02 | 东莞理工学院 | Multi-agent deep reinforcement learning proxy method based on intelligent grid |
US20200074353A1 (en) * | 2018-09-04 | 2020-03-05 | Didi Research America, Llc | System and method for ride order dispatching and vehicle repositioning |
US20210248450A1 (en) * | 2020-02-07 | 2021-08-12 | Google Llc | Sorting attention neural networks |
CN112633421A (en) * | 2021-03-09 | 2021-04-09 | 国网江西综合能源服务有限公司 | Method and device for detecting abnormal electricity utilization behavior of user |
CN113988459A (en) * | 2021-11-11 | 2022-01-28 | 国网江苏省电力有限公司苏州供电分公司 | Small and medium-sized enterprise growth assessment method and system based on electric power marketing data |
CN114004530A (en) * | 2021-11-11 | 2022-02-01 | 国网江苏省电力有限公司苏州供电分公司 | Enterprise power credit score modeling method and system based on sequencing support vector machine |
CN114219606A (en) * | 2021-11-11 | 2022-03-22 | 国网四川省电力公司信息通信公司 | Power data-based power consumption enterprise credit evaluation method and system |
Non-Patent Citations (2)
Title |
---|
张莹莹;钱胜胜;方全;徐常胜;: "基于多模态知识感知注意力机制的问答方法", 计算机研究与发展, no. 05, pages 1037 - 1045 * |
赵阳;胡诗尧;杨书强;冯成;王毅;: "售电市场环境下基于数据驱动的用户用电行为分析", 电力需求侧管理, no. 04, pages 45 - 50 * |
Also Published As
Publication number | Publication date |
---|---|
CN115310999B (en) | 2024-02-02 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111080408B (en) | Order information processing method based on deep reinforcement learning | |
CN110866819A (en) | Automatic credit scoring card generation method based on meta-learning | |
CN110956273A (en) | Credit scoring method and system integrating multiple machine learning models | |
CN105427138A (en) | Neural network model-based product market share analysis method and system | |
CN110837523A (en) | High-confidence reconstruction quality and false-transient-reduction quantitative evaluation method based on cascade neural network | |
CN111079941B (en) | Credit information processing method, credit information processing system, terminal and storage medium | |
CN117151870B (en) | Portrait behavior analysis method and system based on guest group | |
CN111178585A (en) | Fault reporting amount prediction method based on multi-algorithm model fusion | |
CN112700324A (en) | User loan default prediction method based on combination of Catboost and restricted Boltzmann machine | |
CN115470962A (en) | LightGBM-based enterprise confidence loss risk prediction model construction method | |
CN114764682B (en) | Rice safety risk assessment method based on multi-machine learning algorithm fusion | |
CN113837266B (en) | Software defect prediction method based on feature extraction and Stacking ensemble learning | |
CN111950195A (en) | Engineering progress prediction method based on portrait system and depth regression model | |
CN111260490A (en) | Rapid claims settlement method and system based on tree model for car insurance | |
CN113935413A (en) | Distribution network wave recording file waveform identification method based on convolutional neural network | |
CN117372144A (en) | Wind control strategy intelligent method and system applied to small sample scene | |
CN112163731A (en) | Special transformer user electric charge recovery risk identification method based on weighted random forest | |
CN115201394B (en) | Multi-component transformer oil chromatography online monitoring method and related device | |
CN114004530B (en) | Enterprise electric power credit modeling method and system based on ordering support vector machine | |
CN115310999B (en) | Enterprise electricity behavior analysis method and system based on multi-layer perceptron and sequencing network | |
CN115292381A (en) | Virtual currency mining behavior identification method based on extreme gradient lifting algorithm | |
CN114266337A (en) | Intelligent tobacco leaf grading model based on residual error network and grading method using model | |
CN114757495A (en) | Membership value quantitative evaluation method based on logistic regression | |
CN114298472A (en) | Method and system for evaluating images of upstream and downstream enterprises of digital factory | |
CN113191569A (en) | Enterprise management method and system based on big data |
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 | ||
GR01 | Patent grant | ||
GR01 | Patent grant |