CN115879361A - Intelligent determination method for operation, maintenance, overhaul and debugging cost of primary equipment - Google Patents
Intelligent determination method for operation, maintenance, overhaul and debugging cost of primary equipment Download PDFInfo
- Publication number
- CN115879361A CN115879361A CN202211112092.3A CN202211112092A CN115879361A CN 115879361 A CN115879361 A CN 115879361A CN 202211112092 A CN202211112092 A CN 202211112092A CN 115879361 A CN115879361 A CN 115879361A
- Authority
- CN
- China
- Prior art keywords
- overhaul
- output
- data
- clustering
- debugging
- 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
Links
Images
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention belongs to the technical field of intelligent cost determination, and particularly relates to an intelligent method for determining the operation, maintenance and debugging cost of primary equipment; the method comprises the following steps of S1, identifying influence factors, and deeply analyzing the difference of the level of overhaul and debugging expenses; s2, collecting historical data, including overhaul and debugging costs of different devices and internal and external related influence elements, and preprocessing the data, including data correction and continuity check, and processing abnormal values and missing values; s3, clustering analysis of different scenes is achieved by combining a clustering analysis method, and the maintenance cost levels of the equipment with the same type and the same type are classified to generate a training data set; s4, constructing a neural network model and training by utilizing a training data set to obtain a prediction model; the method solves the actual problem that the maintenance and debugging cost level is difficult to accurately estimate by the conventional method, and improves the high-efficiency fine management level of operation and maintenance of the power grid enterprise.
Description
Technical Field
The invention belongs to the technical field of intelligent cost determination, and particularly relates to an intelligent method for determining the operation, maintenance, overhaul and debugging cost of primary equipment.
Background
With the continuous promotion of the investment scale of the power grid and the complex influence of policy innovation, the cost input scale has certain influence on enterprise operation and benefit guarantee; the overhaul and debugging cost is influenced by various aspects such as natural environment, social economy, technical development and the like, in order to meet the requirements of power grid operation management and detailed development under the new situation and realize cost differentiation determination, system identification must be carried out aiming at the influence factors of the cost level, main influence factors are clarified, the multi-service rule characteristics and the service cost lean management requirement are deeply explored, the cost collection and deviation influence mechanism is analyzed, the multi-service cost attributes and the fluctuation characteristic deep analysis under the new situation are promoted, the service and region difference characteristics are further considered, and the cost optimization adjustment analysis technology is provided.
In recent years, the investment of the power grid in China is increased year by year, and the scale of the power grid assets is remarkably increased; along with the development of national economy, the capital construction investment scale of a power grid is gradually expanded, the contradiction between the development of high-strength investment, the increase of cost rigidity and the gradual increase of electric quantity speed and the difficulty in benefit increase is increasingly prominent, the difficulty in keeping steady operation and completing profit targets is increased, and higher requirements are put forward for power grid investment decisions; in the future, the newly increased investment scale of the power grid enterprise is strictly restricted and is comprehensively influenced by complex factors such as macroscopic economy downlink and the like, the electric quantity acceleration maintaining pressure is increased, and the power grid service yield level and the income risk are increased; the high-speed investment does not accord with the current power grid enterprise operation practice, and the investment strategy needs to be readjusted to adapt to the change of new situation; in the face of such huge investment scale, how to comply with the development direction of resource conservation and environmental friendliness formulated in China is to meet the power development requirement, optimize the cost input scale of a power grid enterprise and improve the lean level of power grid investment management becomes a vital part in the power grid enterprise operation management; therefore, the method for intelligently determining the maintenance and debugging cost of the primary equipment is necessary to solve the actual problem that the maintenance and debugging cost level is difficult to accurately estimate by the conventional method.
Disclosure of Invention
The invention aims to overcome the defects of the prior art, and provides an intelligent determination method for primary equipment operation maintenance, overhaul and debugging cost, which is used for solving the actual problem that the overhaul and debugging cost level is difficult to estimate accurately in the conventional method and improving the efficient and fine management level of operation maintenance and overhaul of a power grid enterprise.
The purpose of the invention is realized by the following steps: an intelligent determination method for operation, maintenance, overhaul and debugging cost of primary equipment comprises the following steps:
s1, identifying influence factors, and deeply analyzing the difference of the level of overhaul and debugging expenses;
s2, collecting historical data including overhaul and debugging costs of different devices and internal and external related influence factors, and preprocessing the data, including data correction and continuity check, and abnormal value and missing value processing;
s3, clustering analysis of different scenes is achieved by combining a clustering analysis method, and the maintenance cost levels of the same type of equipment under the same type of conditions are classified to generate a training data set;
and S4, constructing a neural network model and training by utilizing a training data set to obtain a prediction model.
The step S1 is used for identifying influence factors, and the deep analysis of the difference of the maintenance and debugging expense levels comprises the following steps of identifying the influence factors based on system dynamics:
(1) And (3) system analysis: the research object and the research target are determined, and a good foundation is laid for the construction of a system model: looking up literature data and rules and regulations, and determining variable elements contained in a research system through a brainstorming method, a Delphi method and the like;
(2) Model construction: describing the system structure, dividing subsystems according to the overall and local structures of the system, determining the internal mechanism and the mutual relation among the subsystems, and analyzing the sub-factors and the mutual relation among the subsystems; and combining the two steps, and simultaneously considering the interrelation of the sub-factors belonging to different subsystems to construct an influence factor general relationship diagram.
(3) The influencing factors are as follows: the influence relates to a plurality of types, the factors of different types are mutually influenced and correlated to form complex link combination, the system dynamics concept is combined, the overhaul and debugging cost is taken as a core object, and an information identification result is combined.
Step S2, collecting historical data, including overhaul and debugging costs of different devices and internal and external related influence elements, preprocessing the data, including data correction and continuity check, and processing abnormal values and missing values, wherein the step S comprises the following steps:
initialization: determining fuzzy factor m, cluster number c and maximum iteration number t max Or a convergence threshold epsilon and satisfies: c is more than 1 and less than or equal to n, epsilon is more than 0, and the fuzzy factor m is (1.5,2.5)]The clustering algorithm performs best in intervals, m is 2, and a missing value X is initialized M Initializing the clustering center V 0 ；
(1) Calculating (updating) the membership matrix U
The membership degree matrix U meets the condition:
(2) Updating the clustering center V
(3) If | | | v (r+1) -v (r) If | < epsilon or the maximum number of iterations is reached, the algorithm stops, otherwise go to (4).
(4) And calculating a filling value of the missing data.
(5) Let r = r +1 go to (1).
The step S3, in combination with the cluster analysis method, implements cluster analysis of different scenarios, and classifying the overhaul cost levels of the same type of equipment under the same type of conditions to generate a training data set includes:
aiming at all sample data searching, a particle swarm improvement algorithm is applied to find the cluster number with the minimum error square sum J value and the corresponding initial cluster center; after the initial clustering center is determined, finally determining the category and each category center of the sample data according to an improved fuzzy clustering algorithm; carrying out clustering analysis on the influence factor data of the expense levels under different conditions by adopting a fuzzy C-means clustering algorithm;
the sample set is D = { x1, x2, …, xn } represents influence factor data of n samples, fuzzy clustering is carried out on D = { x1, x2, …, xn } to obtain C clusters of C1, C2, …, cc, P = { P1, P2, …, pc } represents a cluster center set of all subsets, wherein U = (U) (x 1, x2, …, xn) } represents a cluster center set of all subsets ij ) Represents a matrix of degrees of membership u ij Is used to indicate the membership of the sample xi to its subset Cj, and satisfies:
the objective function of the dissimilarity index of the clustering algorithm is defined as:
wherein J represents the sum of squared distances between the instance data and the cluster center; d is a radical of ij The represented ith data point and the jth clustering center are distance measurement functions, and the distribution condition of the clustering centers adopts Euclidean distance d ij ＝||x i -p j ||；
Constructing a new objective function as follows, and enabling the objective function of the non-similarity index to reach the minimum requirement:
λ j (j =1,2, …, n) is lagrange multiplier of n constraint formulas, m is used for determining smoothing factor of membership degree matrix U fuzzy level, the fuzzy level of U is in positive correlation with m value, derivation is carried out on all input parameters, and the objective function of the non-similarity index reaches minimumThe requirements are as follows:
an iterative calculation mode is always adopted until a target condition is met, and at the moment, the target function J obtains a minimum value to complete optimization; and continuously adjusting the clustering center and the membership degree in the iteration process until the iteration condition is met.
S4, constructing a neural network model and training by using a training data set to obtain a prediction model comprises the following steps:
an intelligent prediction model based on a BP neural network is constructed, activation functions of nodes of all layers in the network are S-shaped functions, and input of i nodes of a first layer in the network is recorded as net i Output is recorded as o i The output of the kth node of the output layer is y k Then, the input of the jth node of the middle layer is:
o j ＝f(net j )，
the error of the network is the difference between the expected output and the actual output, and then
The output layer has i neurons, and the square error of the actual output and the expected output is:
since the BP algorithm corrects the weight according to the negative gradient of the error E, the modification of the weight is expressed as;
W m+1 ＝w m +Δw m ＝w m -λg m ，
where m represents the number of iterations,
wherein the step size of λ learning;
because it is the output layer, this time
Is the actual output value, according to e k The definition of (c) and the square error can be found:
According to the formula
The following can be obtained: />
According to the formula
The following can be obtained: />
Obtaining:
let the learning error sigma of the output layer k ＝e k f′(net k ) Obtaining:
weight modifier delta w of hidden layer neural unit kj ：
According to the above formula
And o j ＝f(net j ) The following can be obtained:
because of the variation of the weights of the hidden layers, it is inherent to consider the effect of the previous layer on it:
according to
Can know and be->
In turn according to
The following can be obtained:
handle
Belt type device for picking up or picking up a person>
To obtain:
let the learning error of the hidden layer:
the BP neural network comprises: an input layer, an intermediate hidden layer and an output layer; the input vector for the input layer is X = (X1, X2, …, xi), and the output vector for the intermediate hidden layer is: y = (Y1, Y2, …, yj), output vector of output layer: the actual failure rate corresponding to the moment to be predicted, O = (O1, O2, …, ok); the output vector is the overhaul cost of the equipment to be predicted, which is obtained by using a prediction model to calculate; the vector of expected outputs is:
wherein it is present>
A target cost level vector; by adjusting the weight of the BP neural network, the output vector of the output layer continuously approaches the expected output vector, and the training of the BP neural network is completed.
The invention has the beneficial effects that: according to the intelligent determination method for the operation, maintenance, overhaul and debugging costs of the primary equipment, influence factors are identified through the step S1, and the level difference of the overhaul and debugging costs is deeply analyzed; s2, collecting historical data including overhaul and debugging costs of different devices and internal and external related influence factors, and preprocessing the data, including data correction and continuity check, and abnormal value and missing value processing; s3, clustering analysis of different scenes is achieved by combining a clustering analysis method, and the maintenance cost levels of the same type of equipment under the same type of conditions are classified to generate a training data set; s4, constructing a neural network model and training by using a training data set to obtain a prediction model; the method solves the actual problem that the maintenance and debugging cost level is difficult to accurately estimate in the conventional method, and improves the high-efficiency fine management level of operation and maintenance of the power grid enterprise.
Drawings
FIG. 1 is a flow chart of the method for intelligently determining the debugging cost of primary equipment operation and maintenance.
FIG. 2 is a schematic diagram of a total relationship diagram of influencing factors of the intelligent determination method for the primary equipment operation maintenance and debugging cost.
Detailed Description
The invention is further described below with reference to the accompanying drawings.
As shown in fig. 1, the present invention comprises the steps of:
s1, identifying influence factors, and deeply analyzing the level difference of overhaul and debugging expenses;
s2, collecting historical data including overhaul and debugging costs of different devices and internal and external related influence factors, and preprocessing the data, including data correction and continuity check, and abnormal value and missing value processing;
s3, clustering analysis of different scenes is achieved by combining a clustering analysis method, and the maintenance cost levels of the same type of equipment under the same type of conditions are classified to generate a training data set;
and S4, constructing a neural network model and training by using the training data set to obtain a prediction model.
System Dynamics (SD) was originally created in 1956 by professor jayw forrester of the university of labor and technology in massachusetts, and is a cross and comprehensive discipline for specially recognizing systems and solving system problems. The initial system dynamics theory is mainly applied to the field of enterprise management, and then has great advantages in the aspect of researching complex systems, so that the system dynamics theory can help researchers to provide good insights and countermeasures, the application range of the system dynamics theory is gradually expanded, and great contributions are made to the fields of social economy, biology, medical treatment, environmental protection and the like. In the project management field, system dynamics was initially applied to engineering project management in 1980, and CooperKG quantitatively analyzed a cause of cost over budget for large military shipbuilding projects using a system dynamics model, which is also considered as one of the most successful applications of system dynamics in the project management field.
The theory of system dynamics considers that the internal structures of the systems have certain connectivity, the internal conduction relationship is determined, each system has a structure, different system structures have different system functions, the system is defined as an aggregate with certain functions by organically combining interacting and mutually distinguishing elements, and a feedback loop is adopted to describe the structure of the system. A complex system is an overall feedback structure consisting of various subsystems and sub-factors of the subsystems, and the intersection and interaction of all feedback loops form the overall function and behavior of the complex system. Based on the feedback characteristics, the source of the problem can be found from within the system architecture.
The basic tool of system dynamics is a causal graph, in which a number of factors are connected by causal arrows to form a causal chain. Causality is positive and negative, represented by "+" and "-" on the arrows, respectively, meaning that causal increases have either a promoting or inhibiting effect on the outcome. The causal loop is formed by connecting more than two causal relationship chains end to end, and when a certain element in the loop is strengthened, the causal relationship of the whole loop is strengthened, so that the causal loop is called as a positive causal loop, and otherwise, the causal loop is called as a negative causal loop. When the property of the variable and the inflow and outflow directions of the system energy are definite, the flow direction of the system energy can be shown in a graph so as to clarify the feedback form and the control law of the system, and the graph is called as a flow graph.
The primary equipment overhaul debugging relates to various resource consumption and factor influences, the information can be regarded as a complex system, and the system dynamic model is suitable for analysis and establishment of a relatively perfect influence factor system.
The step S1 is used for identifying influence factors, and deep analysis of the difference of the overhaul and debugging cost levels comprises the following steps of identifying the influence factors based on system dynamics:
(1) And (3) system analysis: the research object and the research target are determined, and a good foundation is laid for the construction of a system model: looking up literature data and rules and regulations, and determining variable elements contained in a research system through a brainstorming method, a Delphi method and the like;
(2) Constructing a model: describing the system structure, dividing subsystems according to the overall and local structures of the system, determining the internal mechanism and the mutual relation among the subsystems, and analyzing the sub-factors and the mutual relation among the subsystems; the above two steps are combined, and the mutual relations of the sub-factors belonging to different subsystems are considered at the same time, so as to construct an influence factor general relation graph, as shown in the attached figure 2.
The identification of influencing factors is already done at this point, but the principles of system dynamics can play more than this role. If the influence of the inflow of energy on the system needs to be further analyzed, the running condition of the system in a certain time can be simulated through specific software by means of computer technology. The operation characteristics of the system can be improved through a manual control operation process, and the way of improving the system through the result analysis of the experiment of the model on the computer is provided for the reference of a decision maker.
(3) The influencing factors are as follows: the influence relates to a plurality of types, the factors of different types are mutually influenced and correlated to form complex link combination, the system dynamics concept is combined, the overhaul and debugging cost is taken as a core object, and an information identification result is combined.
Step S2, collecting historical data, including overhaul and debugging costs of different devices and internal and external related influence factors, preprocessing the data, including data correction and continuity check, and processing abnormal values and missing values, including:
initialization: determining a fuzzy factor m, a cluster number c and a maximum iteration number t max Or a convergence threshold epsilon and satisfies: c is more than 1 and less than or equal to n, epsilon is more than 0, and the blurring factor m is in the range of 1.5,2.5]The clustering algorithm is best in performance during interval, m is 2, and a missing value X is initialized M Initializing the clustering center V 0 ；
(1) Calculating (updating) membership matrix U
The membership matrix U satisfies the condition:
(2) Updating the clustering center V
(3) If | | | v (r+1) -v (r) If | < epsilon or the maximum number of iterations is reached, the algorithm stops, otherwise go to (4).
(4) And calculating a filling value of the missing data.
(5) Let r = r +1 go to (1).
The step S3, in combination with the cluster analysis method, implements cluster analysis in different scenarios, and classifying the overhaul cost levels of the same type of equipment under the same type of conditions to generate a training data set includes:
aiming at all sample data searching, a particle swarm improvement algorithm is applied to find the cluster number with the minimum error square sum J value and the corresponding initial cluster center; after the initial clustering center is determined, finally determining the category and each category center of the sample data according to an improved fuzzy clustering algorithm; carrying out clustering analysis on the influence factor data of the cost level under different conditions by adopting a fuzzy C-means clustering algorithm;
the sample set is D = { x1, x2, …, xn } represents influence factor data of n samples, fuzzy clustering is carried out on D = { x1, x2, …, xn } to obtain C numbers of C1, C2, …, cc, P = { P1, P2, …, xn } cluster represents a cluster center set of all subsets, wherein U Pc = (U Pc) (U1, P2, …, xn) cluster represents a cluster center set of all subsets ij ) Representing a matrix of degrees of membership u ij Is used to indicate the membership of the sample xi to its subset Cj, and satisfies:
the objective function of the non-similarity index of the clustering algorithm is defined as:
wherein J represents the sum of squares of distances between the instance data and the cluster center; d ij The represented ith data point and the jth clustering center are distance measurement functions, and the distribution condition of the clustering centers adopts Euclidean distance d ij ＝||x i -p j ||；
Constructing a new objective function as follows, and enabling the objective function of the non-similarity index to reach the minimum requirement:
λ j (j =1,2, …, n) are lagrange multipliers of n constraint formulas, m is used for determining a smoothing factor of a membership matrix U fuzzy level, the fuzzy level of U is in positive correlation with the value m, and derivation is performed on all input parameters, so that a necessary condition for minimizing an objective function of the non-similarity index is as follows:
an iterative calculation mode is always adopted until a target condition is met, and at the moment, the target function J obtains a minimum value to complete optimization; and continuously adjusting the clustering center and the membership degree in the iteration process until the iteration condition is met.
S4, constructing a neural network model and training by using a training data set to obtain a prediction model comprises the following steps:
an intelligent prediction model based on a BP neural network is constructed, the activation functions of all layers of nodes in the network are S-shaped functions, and the input of the first layer of i nodes in the network is recorded as net i Output is recorded as o i The output of the kth node of the output layer is y k Then, the input of the jth node of the middle layer is:
o j ＝f(net j )，
error of networkFor the difference between the desired output and the actual output, then there is
The output layer has i neurons, and the square error of the actual output and the expected output is:
since the BP algorithm corrects the weight according to the negative gradient of the error E, the modification of the weight is expressed as;
W m+1 ＝w m +Δw m ＝w m -λg m ，
where m represents the number of iterations,
wherein the step size of λ learning; />
Because it is the output layer, this time
Is the actual output value, according to e k The definition of (2) and the square error can be found:
According to the formula
The following can be obtained: />
According to the formula
The following can be obtained: />
Obtaining:
let the learning error sigma of the output layer k ＝e k f′(net k ) Obtaining:
weight modifier delta w of hidden layer neural unit kj ：
According to the above formula
And o j ＝f(net j ) The following can be obtained:
because of the variation of the weights of the hidden layers, it is inherent to consider the effect of the previous layer on it:
according to
Can know and be->
In turn according to
The following can be obtained:
handle
Belt type device for picking up or picking up a person>
To obtain:
let the learning error of the hidden layer:
constructing a BP neural network prediction model, and optimizing model parameters by using a genetic algorithm so as to obtain better initial weight and threshold; the BP neural network comprises: an input layer, an intermediate hidden layer and an output layer; the input vector for the input layer is X = (X1, X2, …, xi), and the output vector for the intermediate hidden layer is: y = (Y1, Y2, …, yj), output vector of output layer: o = (O1, O2, …, ok); wherein the output isThe vector is the maintenance cost of the equipment to be predicted, which is obtained by using a prediction model for calculation; the vector of expected outputs is:
wherein it is present>
A target cost level vector; by adjusting the weight of the BP neural network, the output vector of the output layer continuously approaches the expected output vector, and the training of the BP neural network is completed.
The algorithm steps of the BP neural network comprise: initializing a network, selecting random numbers in a (-1,1) interval to assign values to each connection weight of the network, and setting an error and the maximum iteration times of the network;
training a network model:
the output of the intermediate hidden layer is: y is j ＝f 1 (net j )；
Wherein j =1,2, …, m,
the output of the output layer is: o is k ＝f 2 (net k )；
Wherein j =1,2, …, l,
function f 1 、f 2 The transfer functions of (a) are Sigmoid functions:
f′(x)＝f(x)[1-f(x)]；
calculating an error function for each layer based on the desired output and the actual output:
for the input layer, the error function is:
error calculation and adjustment of network weights using error back-propagation:
the error signal is derived as:
by continuously adjusting the weights, the error is continuously reduced:
calculating a global error:
according to whether the global error reaches the accuracy of initial setting or whether the training times reach the maximum iteration times of the initial setting, and the algorithm is ended; otherwise, the error function of each layer, the network weight adjustment and the global error calculation are continuously calculated.
Before the BP neural network is trained by utilizing the training data set, an improved particle swarm algorithm is applied to optimize a BP neural network prediction model, so that better initial weight and threshold are obtained. The optimization principle of the improved particle swarm optimization is as follows:
1. determining a solution space of the optimization problem, namely: and solving the value ranges of the parameters and different parameters. Meanwhile, the size N of the population, namely the number of alternative solutions of each generation, is also determined. It is noted that the alternative solutions of each generation may be repeated;
2. n candidate solutions are randomly drawn in the solution space and each candidate solution is encoded. There are three common encoding methods: binary encoding, floating point encoding, and sign encoding. The binary coding is simple, but the global searching capability of the continuous function is not high; the floating point coding precision is high, but the coding is difficult; the symbol coding is suitable for solving the special knowledge;
3. and solving the fitness of each individual in the population. The fitness is generally an objective function in the optimization problem or a function obtained by transforming the objective function;
4. and judging whether to stop generating the next generation population according to the fitness of the current population. The general judgment conditions include: the algorithm comprises maximum iteration times, a fitness threshold value and a fitness change threshold value of two adjacent generations. If the termination condition is met, decoding the individual corresponding to the optimal fitness value in the current population to obtain the optimal solution of the problem; if the termination condition is not met, entering the step 5 for heredity;
5. and generating a next generation population by three operations of selection, crossing and variation according to the fitness of the current population. Wherein, the 'selection' is to select excellent individuals from parents to be directly inherited, and the commonly used 'selection' operation comprises the following steps: roulette selection, random competition, optimal selection and the like; "crossover" is the exchange of the codes of two individuals that mate in some way, resulting in two new individuals. Common "crossover" operations include: single-point crossing, multi-point crossing, even crossing and the like; "mutation" refers to a change in the sign of certain bits of the encoded individual. Common "mutation" operations are: basic potential variation, boundary variation, non-uniform variation, and the like;
6. after the genetic step is finished, 3 is returned.
The specific implementation process for optimizing the model parameters by applying the improved particle swarm optimization is as follows:
the principle of the PSO algorithm for optimizing the neural network is as follows: assuming that p particles are scattered in n-dimensional space, the velocity of the ith particle at the current iteration number k is
Is positioned at->
And the fitness value F is a training error of the network, and an individual optimal value pibestk and a global optimal value pgbestk of the whole particle swarm in the current iteration process are recorded. The position, velocity of the next iteration of particles is given by the following equation until a particle is found that satisfies the condition.
According to the intelligent determination method for the operation, maintenance, overhaul and debugging costs of the primary equipment, influence factors are identified through the step S1, and the level difference of the overhaul and debugging costs is deeply analyzed; s2, collecting historical data including overhaul and debugging costs of different devices and internal and external related influence factors, and preprocessing the data, including data correction and continuity check, and abnormal value and missing value processing; s3, clustering analysis of different scenes is achieved by combining a clustering analysis method, and the maintenance cost levels of the same type of equipment under the same type of conditions are classified to generate a training data set; s4, constructing a neural network model and training by utilizing a training data set to obtain a prediction model; the method solves the actual problem that the maintenance and debugging cost level is difficult to accurately estimate in the conventional method, and improves the high-efficiency fine management level of operation and maintenance of the power grid enterprise.
Claims (5)
1. An intelligent determination method for primary equipment operation maintenance, overhaul and debugging cost is characterized by comprising the following steps:
s1, identifying influence factors, and deeply analyzing the level difference of overhaul and debugging expenses;
s2, collecting historical data, including overhaul and debugging costs of different devices and internal and external related influence elements, and preprocessing the data, including data correction and continuity check, and processing abnormal values and missing values;
s3, clustering analysis of different scenes is achieved by combining a clustering analysis method, and the maintenance cost levels of the equipment with the same type and the same type are classified to generate a training data set;
and S4, constructing a neural network model and training by utilizing a training data set to obtain a prediction model.
2. The method for intelligently determining the maintenance, overhaul and debugging cost of primary equipment according to claim 1, wherein the step S1 of identifying the influencing factors, and the deep analysis of the difference of the overhaul and debugging cost levels comprises identifying the influencing factors based on system dynamics:
(1) And (3) system analysis: the research object and the research target are determined, and a good foundation is laid for the construction of a system model: looking up literature data and rules and regulations, and determining variable elements contained in a research system through a brainstorming method, a Delphi method and the like;
(2) Constructing a model: describing a system structure, dividing subsystems according to the overall and local structures of the system, determining the internal mechanism and the mutual relation among the subsystems, and analyzing the subfactors and the mutual relation among the subsystems; combining the two steps, and simultaneously considering the interrelation of the sub-factors belonging to different subsystems to construct a total relationship diagram of the influencing factors;
(3) The influencing factors are as follows: the influence relates to a plurality of types, the factors of different types are mutually influenced and correlated to form complex link combination, the system dynamics concept is combined, the overhaul and debugging cost is taken as a core object, and an information identification result is combined.
3. The method as claimed in claim 1, wherein the step S2 of collecting historical data, including the overhaul and debugging costs of different devices and internal and external related influence elements, pre-processing the data, including data modification and continuity check, and processing abnormal values and missing values includes:
initialization: determining fuzzy factor m, cluster number c and maximum iteration number t max Or a convergence threshold epsilon and satisfies: c is more than 1 and less than or equal to n, epsilon is more than 0, and the blurring factor m is in the range of 1.5,2.5]The clustering algorithm is best in performance during interval, m is 2, and a missing value X is initialized M Initializing the clustering center V 0 ；
(1) Calculating (updating) membership matrix U
The membership degree matrix U meets the condition:
(2) Updating the clustering center V
(3) If | | | v (r+1) -v (r) I < epsilon toImmediately or when the maximum iteration number is reached, stopping the algorithm, and otherwise, turning to the step (4);
(4) Calculating a filling value of missing data;
(5) Let r = r +1 go to (1).
4. The method as claimed in claim 1, wherein the step S3 of combining with a cluster analysis method to implement cluster analysis of different scenarios, and classifying the overhaul cost levels of the same type of equipment under the same type of conditions to generate the training data set comprises:
aiming at all sample data searching, a particle swarm improvement algorithm is applied to find the cluster number with the minimum error square sum J value and the corresponding initial cluster center; after the initial clustering center is determined, finally determining the category and each category center of the sample data according to an improved fuzzy clustering algorithm; carrying out clustering analysis on the influence factor data of the cost level under different conditions by adopting a fuzzy C-means clustering algorithm;
the sample set is D = { x1, x2, …, xn } represents influence factor data of n samples, fuzzy clustering is carried out on D = { x1, x2, …, xn } to obtain C clusters of C1, C2, …, cc, P = { P1, P2, …, pc } represents a cluster center set of all subsets, wherein U = (U) (x 1, x2, …, xn) } represents a cluster center set of all subsets ij ) Representing a matrix of degrees of membership u ij Is used to indicate the membership of the sample xi to its subset Cj, satisfying:
the objective function of the non-similarity index of the clustering algorithm is defined as:
wherein J represents the sum of squares of distances between the instance data and the cluster center; d ij The represented ith data point and the jth clustering center are distance measurement functions, and the distribution condition of the clustering centers adopts Euclidean distance d ij ＝||x i -p j ||；
Constructing a new objective function as follows, and enabling the objective function of the non-similarity index to reach the minimum requirement:
λ j (j =1,2, …, n) are lagrange multipliers of n constraint formulas, m is used for determining a smoothing factor of a membership matrix U fuzzy level, the fuzzy level of U is in positive correlation with the value m, derivation is performed on all input parameters, and the necessary condition for minimizing the objective function of the non-similarity index is as follows:
an iterative calculation mode is always adopted until a target condition is met, and at the moment, the target function J obtains a minimum value to complete optimization; and continuously adjusting the clustering center and the membership degree in the iteration process until the iteration condition is met.
5. The method of claim 1, wherein the step S4 of constructing a neural network model and training the neural network model with a training data set to obtain a prediction model comprises:
constructing an intelligent prediction model based on a BP neural network, and activating functions of nodes of each layer in the networkThe numbers are all sigmoid functions, and the input to the first layer i-node in the network is denoted as net i Output is recorded as o i The output of the kth node of the output layer is y k Then, the input of the jth node of the middle layer is:
o j ＝f(net j )，
the error of the network is the difference between the expected output and the actual output, and then
The output layer has i neurons, and the square error of the actual output and the expected output is:
since the BP algorithm corrects the weight according to the negative gradient of the error E, the modification of the weight is expressed as;
W m+1 ＝w m +Δw m ＝w m -λg m ，
where m represents the number of iterations,
step size for where λ learning;
because it is the output layer, this time
Is the actual output value, according to e k The definition of (2) and the square error can be found:
According to the formula
The following can be obtained: />
According to the formula
The following can be obtained: />
Obtaining:
let the learning error sigma of the output layer k ＝e k f′(net k ) Obtaining:
weight modifier delta w of hidden layer neural unit kj ：
According to the above formula
And o j ＝f(net j ) The following can be obtained:
since it is the change in the weights of the hidden layer that is sought, it is inherent to consider the effect of the previous layer on it:
according to
Can know and be->
According to>
The following can be obtained:
handle
Belt type device for picking up or picking up a person>
To obtain:
let the learning error of the hidden layer:
the BP neural network comprises: an input layer, an intermediate hidden layer and an output layer; the input vector for the input layer is X = (X1, X2, …, xi), and the output vector for the intermediate hidden layer is: y = (Y1, Y2, …, yj), output vector of output layer: the actual failure rate corresponding to the moment to be predicted, O = (O1, O2, …, ok); the output vector is the overhaul cost of the equipment to be predicted, which is obtained by using a prediction model to calculate; the vector of expected outputs is:
wherein it is present>
A target cost level vector; by adjusting the weight of the BP neural network, the output vector of the output layer continuously approaches the expected output vector, and the training of the BP neural network is completed. />
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211112092.3A CN115879361A (en) | 2022-09-13 | 2022-09-13 | Intelligent determination method for operation, maintenance, overhaul and debugging cost of primary equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211112092.3A CN115879361A (en) | 2022-09-13 | 2022-09-13 | Intelligent determination method for operation, maintenance, overhaul and debugging cost of primary equipment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115879361A true CN115879361A (en) | 2023-03-31 |
Family
ID=85769840
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211112092.3A Pending CN115879361A (en) | 2022-09-13 | 2022-09-13 | Intelligent determination method for operation, maintenance, overhaul and debugging cost of primary equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115879361A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117557113A (en) * | 2023-10-10 | 2024-02-13 | 国网湖北省电力有限公司经济技术研究院 | Power grid operation and maintenance scheme planning method and system considering equipment characteristics |
-
2022
- 2022-09-13 CN CN202211112092.3A patent/CN115879361A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117557113A (en) * | 2023-10-10 | 2024-02-13 | 国网湖北省电力有限公司经济技术研究院 | Power grid operation and maintenance scheme planning method and system considering equipment characteristics |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Jia et al. | An optimized RBF neural network algorithm based on partial least squares and genetic algorithm for classification of small sample | |
| Obayashi et al. | Multi-objective design exploration for aerodynamic configurations | |
| CN110544011B (en) | Intelligent system combat effectiveness evaluation and optimization method | |
| Khayyam et al. | A novel hybrid machine learning algorithm for limited and big data modeling with application in industry 4.0 | |
| Deng et al. | A novel parallel hybrid intelligence optimization algorithm for a function approximation problem | |
| CN106600059A (en) | Intelligent power grid short-term load predication method based on improved RBF neural network | |
| Jadav et al. | Optimizing weights of artificial neural networks using genetic algorithms | |
| CN101782743A (en) | Neural network modeling method and system | |
| Qiao et al. | A self-organizing RBF neural network based on distance concentration immune algorithm | |
| Li et al. | A bilevel learning model and algorithm for self-organizing feed-forward neural networks for pattern classification | |
| CN113988358A (en) | Carbon emission index prediction and treatment method based on transfer reinforcement learning | |
| CN115879361A (en) | Intelligent determination method for operation, maintenance, overhaul and debugging cost of primary equipment | |
| CN106845696B (en) | Intelligent optimization water resource configuration method | |
| Chatterjee et al. | Artificial neural network and the financial markets: A survey | |
| Zhu et al. | An Efficient Hybrid Feature Selection Method Using the Artificial Immune Algorithm for High‐Dimensional Data | |
| CN111563614A (en) | Load prediction method based on adaptive neural network and TLBO algorithm | |
| Rajkumar et al. | Weather forecasting using fuzzy neural network (FNN) and hierarchy particle swarm optimization algorithm (HPSO) | |
| CN115713144A (en) | Short-term wind speed multi-step prediction method based on combined CGRU model | |
| Ying et al. | A novel optimization algorithm for BP neural network based on RS-MEA | |
| Wang et al. | Optimizing financial engineering time indicator using bionics computation algorithm and neural network deep learning | |
| CN113656707A (en) | Financing product recommendation method, system, storage medium and equipment | |
| Zhang et al. | A fast evolutionary knowledge transfer search for multiscale deep neural architecture | |
| Skakalina | Investigation of intelligent technologies for formation forecasting models | |
| Hou et al. | Wind power forecasting method of large-scale wind turbine clusters based on DBSCAN clustering and an enhanced hunter-prey optimization algorithm | |
| Hsu et al. | Reinforcement evolutionary learning using data mining algorithm with TSK-type fuzzy controllers |
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 |