CN110571792A - Analysis and evaluation method and system for operation state of power grid regulation and control system - Google Patents
Analysis and evaluation method and system for operation state of power grid regulation and control system Download PDFInfo
- Publication number
- CN110571792A CN110571792A CN201910693859.8A CN201910693859A CN110571792A CN 110571792 A CN110571792 A CN 110571792A CN 201910693859 A CN201910693859 A CN 201910693859A CN 110571792 A CN110571792 A CN 110571792A
- Authority
- CN
- China
- Prior art keywords
- classification
- power grid
- control system
- data
- probability
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/06—Resources, workflows, human or project management; Enterprise or organisation planning; Enterprise or organisation modelling
- G06Q10/063—Operations research, analysis or management
- G06Q10/0639—Performance analysis of employees; Performance analysis of enterprise or organisation operations
- G06Q10/06393—Score-carding, benchmarking or key performance indicator [KPI] analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
- G06Q50/06—Electricity, gas or water supply
-
- H—ELECTRICITY
- H02—GENERATION; CONVERSION OR DISTRIBUTION OF ELECTRIC POWER
- H02J—CIRCUIT ARRANGEMENTS OR SYSTEMS FOR SUPPLYING OR DISTRIBUTING ELECTRIC POWER; SYSTEMS FOR STORING ELECTRIC ENERGY
- H02J3/00—Circuit arrangements for ac mains or ac distribution networks
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y04—INFORMATION OR COMMUNICATION TECHNOLOGIES HAVING AN IMPACT ON OTHER TECHNOLOGY AREAS
- Y04S—SYSTEMS INTEGRATING TECHNOLOGIES RELATED TO POWER NETWORK OPERATION, COMMUNICATION OR INFORMATION TECHNOLOGIES FOR IMPROVING THE ELECTRICAL POWER GENERATION, TRANSMISSION, DISTRIBUTION, MANAGEMENT OR USAGE, i.e. SMART GRIDS
- Y04S10/00—Systems supporting electrical power generation, transmission or distribution
- Y04S10/50—Systems or methods supporting the power network operation or management, involving a certain degree of interaction with the load-side end user applications
Abstract
the invention discloses a method and a system for analyzing and evaluating the running state of a power grid regulation and control system, wherein the method comprises the following steps: acquiring real-time operation data of a power grid regulation and control system; analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system; the state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operation state data of the power grid regulation and control system. According to the invention, the massive operation state data of the power grid regulation and control system is utilized to analyze and evaluate various operation state data in the regulation and control system, and the operation state of the regulation and control system is accurately predicted, so that the work efficiency of an operation and maintenance center and the power grid regulation and control system is obviously improved, and meanwhile, the power grid regulation and control system can be prevented from getting ill.
Description
Technical Field
the invention relates to the field of computer deep learning and electric power system dispatching automation, in particular to a method and a system for analyzing and evaluating the running state of a power grid regulation and control system.
Background
The intelligent power grid regulation and control system is a neural center for intelligent power grid operation, and the operation and maintenance center realizes remote picture retrieval and operation maintenance of various local systems by intensively monitoring basic data of provincial and above intelligent power grid regulation and control system operation, quickly solves system abnormity and faults, timely discovers hidden dangers in the system and provides technical support for stable operation of the intelligent power grid regulation and control system.
At present, a centralized operation and maintenance center monitors power grid regulation and control systems in a plurality of regions every day, and acquires a large amount of monitoring data. The monitoring data specifically includes: application state, process state, node state, link state, I/O, CPU usage, disk usage, memory usage. Since no intelligent analysis and evaluation system exists before, the regulation and control system is often informed of the problem later by a manual mode, such as telephone notification. The operation and maintenance center cannot be troubled by the operation state of the regulation and control system timely and accurately, the efficiency of the operation and maintenance work of the system is reduced by the non-intelligent sensing mode, and the work efficiency of the regulation and control system is also reduced.
disclosure of Invention
in order to solve the defects that the operation state of the regulation and control system cannot be pre-judged and the problems cannot be known in the first time by the operation and maintenance personnel in the prior art, the invention provides an analysis and evaluation method for the operation state of the power grid regulation and control system. The invention provides an analysis and evaluation method for the running state of a power grid regulation and control system, which comprises the following steps:
acquiring real-time operation data of a power grid regulation and control system;
analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system;
The state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operation state data of the power grid regulation and control system.
Preferably, the constructing of the state prediction model includes:
Based on historical operating state data of the power grid regulation and control system, dividing the state corresponding to the operating data of the power grid regulation and control system into two categories, namely a system normal category and a system abnormal category by adopting a gradient lifting tree binary classification algorithm;
Based on historical running state data corresponding to the system abnormity, classifying the specific conditions of the system abnormity by adopting a gradient lifting tree multi-classification algorithm to obtain a multi-classification result;
Performing secondary cross validation on the multi-classification result by adopting a feedforward neural network based on historical running state data corresponding to the system abnormity, and selecting the class with the highest probability as a specific abnormal state in the system abnormity;
Wherein the historical operating state data comprises: run data and corresponding states; the states include: system normal and system abnormal; the system exception includes: downtime, CPU flush, process deadlock, network outage.
Preferably, the historical operating state data based on the power grid regulation and control system is obtained by classifying the states corresponding to the operating data of the power grid regulation and control system into two categories, namely a normal system and an abnormal system, by using a gradient lifting tree classification algorithm, and the method includes the following steps:
dividing historical operation state data of the power grid regulation and control system into a training sample and a verification sample;
Training a training sample by adopting a gradient lifting tree binary classification algorithm based on a first loss function set by the difference between the predicted probability value and the real probability value of the state of the power grid regulation and control system to obtain a binary classification tree;
Calculating the probability that the states corresponding to the running data in the two classification trees are respectively normal and abnormal based on a set probability conversion formula;
determining the state corresponding to the operation data based on the probability that the state corresponding to each operation data is respectively the system normal and the system abnormal;
verifying the trained binary tree by using a verification sample, finishing training when the error meets the requirement, and dividing the corresponding state of the operation data of the power grid regulation and control system into two categories, namely a system normal category and a system abnormal category; otherwise, the training is carried out again.
preferably, the training samples are trained by using a gradient lifting tree binary classification algorithm based on a first loss function set based on a difference between a predicted probability value and a real probability value of the state of the power grid regulation and control system to obtain a binary classification tree, including:
Constructing an optimal prediction probability value function based on minimizing the first loss function;
calculating the prediction probability values of all training samples based on the current optimal prediction probability value function;
obtaining a pseudo residual error of a first loss function based on the predicted probability value and the real probability value of all the training samples;
fitting all the operation data and the corresponding pseudo-residuals in the current training sample into a new classification tree by taking the operation data and the corresponding pseudo-residuals as training data of a next tree;
updating an optimal prediction probability value function based on the current pseudo residual error, performing iterative computation on a new classification tree based on the updated optimal prediction probability value function, ending circulation when the value of the first loss function is minimum, and updating an optimal prediction value function based on the minimum value of the first loss function;
And summing the optimal predicted value functions in all the iteration processes to obtain a two-class tree.
preferably, the first loss function is represented by the following formula:
L(y,f(x))＝ln(1+exp(-2yf(x)))
in the formula: x: training samples in the training set; y: training the real state corresponding to the sample; wherein y is {0,1 }; f (x): training a predicted value of the corresponding running state of the sample; exp: an exponential function with a natural constant e as the base.
preferably, the probability conversion formula includes:
when y is 1, the system normal probability conversion formula is shown as the following formula:
In the formula: p (y ═ 1| x): training the probability that the sample x corresponds to the system normality; f (x): a second classification tree;
when y is 0, the system anomaly probability conversion formula is shown as the following formula:
In the formula: p (y ═ 0| x): the training sample x corresponds to the probability of system anomaly.
Preferably, the classifying the specific condition of the system anomaly by using a gradient lifting tree multi-classification algorithm based on the historical operating state data corresponding to the system anomaly to obtain a multi-classification result includes:
Dividing historical running state data corresponding to the system abnormity into a training set and a verification set;
training the training set by adopting a gradient lifting tree multi-classification algorithm based on a set second loss function to obtain a multi-classification tree;
obtaining corresponding probabilities of sample data classified into various categories based on the multi-classification tree;
converting the maximum probability in the classes corresponding to the sample data into corresponding classes to obtain multi-classification results;
verifying the trained multi-classification tree by using a verification set, finishing training when the error meets the requirement, and classifying the abnormal specific condition of the power grid regulation and control system to obtain a multi-classification result; otherwise, the training is carried out again.
Preferably, the training set is trained by using a gradient lifting tree multi-classification algorithm based on the set second loss function to obtain a multi-classification tree, including:
constructing a classification function based on minimizing the second loss function;
Predicting the probability that all sample nodes in the training set belong to each class based on the classification function;
calculating pseudo-residual values of all classes based on real class and prediction class probability of sample nodes in all training sets;
Fitting a new multi-classification tree based on all the sample nodes in the training set and the pseudo residual error values;
updating a classification function based on the current pseudo residual value, iteratively calculating the sum of values of each leaf node region on the basis of the updated classification function, ending the loop when the minimum value of a second loss function is obtained, and updating the classification function based on the minimum value of the second loss function;
And obtaining a multi-classification tree based on all classification functions in the iteration process.
preferably, the performing secondary cross validation on the multi-classification result by using a feed-forward neural network based on the historical operating state data corresponding to the system anomaly includes:
adding a softmax function into a hidden layer of the feedforward neural network model to generate a classifier;
Obtaining a multi-dimensional vector group based on the operation data in the historical operation state data and a set probability calculation formula;
based on the multi-dimensional vector group and the classifier, obtaining probability distribution formed by probabilities of each group of operation data corresponding to each category;
selecting the category with the highest probability from the probability distribution as a classification result corresponding to the operating data;
and performing secondary cross validation on the multi-classification result based on the classification result corresponding to each group of operation data.
Preferably, each element expression in the multidimensional vector set is represented by the following formula:
in the formula: sigma (z)j: vector sigma (z)jDescription of the jth element; z is a radical ofk: the z-th value of the k-dimensional vector; k: number of categories classified.
Preferably, the probability of each category corresponding to the operation data is calculated according to the following formula:
in the formula: p (s ═ v | x): probability when the classification result s belongs to the v-th class; n: the total number of categories; x': a derivative of the run data x in the training set; x is the number ofT: setting a weight value T for the operation data x in the training set; w is av: the proportion of class v; w is an: the proportion of the nth class.
Based on the same invention concept, the invention also provides an analysis and evaluation system for the running state of the power grid regulation and control system, which comprises:
the acquisition module is used for acquiring real-time operation data of the power grid regulation and control system;
The evaluation module is used for analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system;
the state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operation state data of the power grid regulation and control system.
Preferably, the system further comprises: the construction module is used for constructing a state prediction model;
The building module comprises:
the two classification units are used for classifying states corresponding to the operation data of the power grid regulation and control system into two categories, namely a normal system and an abnormal system, by adopting a gradient lifting tree two classification algorithm based on the historical operation state data of the power grid regulation and control system;
The multi-classification unit is used for classifying the specific conditions of the system abnormity by adopting a gradient lifting tree multi-classification algorithm based on the historical running state data corresponding to the system abnormity to obtain a multi-classification result;
the verification unit is used for performing secondary cross verification on the multi-classification result by adopting a feedforward neural network based on historical running state data corresponding to the system abnormity, and selecting the class with the highest probability as a specific abnormal state in the system abnormity;
Wherein the historical operating state data comprises: run data and corresponding states; the states include: system normal and system abnormal; the system exception includes: downtime, CPU flush, process deadlock, network outage.
compared with the prior art, the invention has the beneficial effects that:
1. According to the technical scheme provided by the invention, the real-time operation data of the power grid regulation and control system is obtained; analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system; the state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operating state data of the power grid regulation and control system, the invention utilizes massive historical operating state data of the power grid regulation and control system, analyzes and evaluates various operating state data in the regulation and control system, accurately predicts the real-time operating state of the regulation and control system, obviously improves the working efficiency of an operation and maintenance center and the power grid regulation and control system, and simultaneously can prevent the power grid regulation and control system from getting ill.
2. according to the technical scheme provided by the invention, a feedforward neural network and a gradient lifting tree algorithm are selected, and online calculation and offline calculation can be provided simultaneously. The method comprises the steps of carrying out off-line calculation on mass historical state data of the power grid regulation and control system, carrying out on-line calculation according to real-time operation state data of the regulation and control system on the basis of an off-line calculation result, analyzing and evaluating the current operation state of the system, and ensuring the accuracy of a final result.
3. according to the technical scheme provided by the invention, the output result is secondarily verified based on the softmax function added to the hidden layer in the feedforward neural network model, so that the method has strong fitting capability, the problem of overlarge difference between the calculated value and the true value is avoided to the maximum extent, and the accuracy of the result is further ensured.
4. according to the technical scheme provided by the invention, the two-classification algorithm based on the gradient lifting tree GBDT can accurately classify the operation state of the power grid regulation and control system into a normal class and an abnormal class according to historical operation state data; the multi-classification algorithm can perform more specific analysis and evaluation on the system abnormity of the large class aiming at the system abnormity condition, and the accuracy of the evaluation on the operation state of the power grid regulation and control system is improved.
5. in the technical scheme provided by the invention, a model based on a feedforward neural network and a sensor can be continuously iterated to generate a new learner, and a sample is continuously trained and iterated; the trained learner can be used for multiple times, and the overall calculation efficiency is obviously improved; the method is suitable for the conditions that the number of power grid regulation and control systems to be monitored is large, the types of monitoring data reflecting the running state of the system are various, and the timeliness and the expansibility of the analysis and evaluation system are greatly improved.
drawings
Fig. 1 is a flowchart of an analysis and evaluation method for an operation state of a power grid regulation and control system according to the present invention;
FIG. 2 is a general service flow chart of an operation state analysis and evaluation method of a power grid regulation and control system according to the present invention;
Fig. 3 is a calculation flow chart of a calculation model of the power grid regulation and control system operation state analysis and evaluation method of the present invention.
Detailed Description
For a better understanding of the present invention, reference is made to the following description taken in conjunction with the accompanying drawings and examples.
Example 1
As shown in fig. 1, the present invention provides a method for analyzing and evaluating an operation state of a regulation and control system based on a deep learning technique, comprising:
Step S1, acquiring real-time operation data of the power grid regulation and control system;
Step S2, analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system;
The state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operation state data of the power grid regulation and control system.
As shown in fig. 2 and fig. 3, the specific implementation process is as follows: the method comprises a data acquisition and verification stage, a state prediction model construction stage, a data analysis stage, a calculation result giving and analysis and evaluation. The method can well predict and analyze the state of the dispatching system, can avoid major problems of the regulating and controlling system to the maximum extent, and better achieves the intellectualization and automation of the system operation monitoring; in addition, the method utilizes the neural network as a technical foundation, can change the configuration of the neurons at any time according to the service requirements and the actual service data, and has strong adaptability to the external environment.
the method carries out analysis modeling according to the operation data of the regulation and control system monitored by the system, takes historical monitoring data as a calculation basis, establishes a corresponding regression type learning task according to continuous data, and finally obtains a prediction analysis result based on classification.
Step S1, acquiring real-time operation data of the power grid regulation and control system;
the real-time operation data of the power grid regulation and control system is obtained through the existing technical means.
step S2, analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system, which specifically comprises the following steps:
1. and acquiring and verifying historical running state data of the regulation and control system. And acquiring the running state data of the regulating and controlling system through the monitoring system.
2. and establishing a state prediction model for regulating and controlling the running state of the system.
3. And (3) performing secondary classification on the state value calculation result by using a gradient lifting tree binary classification algorithm (GBDT binary classification algorithm), namely classifying the operation state of the regulation and control system into two categories, namely normal system and abnormal system.
4. And (3) refining samples in the system abnormal class on the basis of the result of the step (3), and classifying the specific conditions of the system abnormal by applying a gradient lifting tree multi-classification algorithm.
5. And performing probability hit calculation on the output result of each abnormal classification by using a Softmax function, and selecting the maximum hit probability as an evaluation result of the operation state of the power grid regulation and control system.
Step 1: the data acquisition and verification stage specifically comprises the following steps:
and performing data collection and aggregation on the operation parameters of the scheduling control system in a historical database of a monitoring system of the centralized operation and maintenance center.
and after the calculation data is obtained, the format of the calculation data is verified, when the calculation data conforms to the set calculation data format, the calculation data is added to the calculation data set, otherwise, the calculation data is discarded.
The operating parameters include: application state, process state, node state, link state, I/O, CPU usage, disk usage, and memory usage.
Step 2: establishing a neural network model for analyzing and predicting the operation state of the power grid regulation and control system according to the calculation data set, wherein the neural network model comprises the following specific steps:
The feedforward neural network and the perceptron are selected as technical schemes, and layers included in a general neural network are parallel to each other, namely neural structures of an input layer, a hidden layer and an output layer. Neurons in individual nerve cell layers are not interconnected; and the two adjacent nerve cell layers are all connected. The scheme adopts unsupervised learning, which means that only input is given, and then the neural network is enabled to search for the rule in the data. In the scheme, a data set trained by the neural network is all historical data information of the running state of the power grid regulation and control system in a certain place, which is acquired by a monitoring system. Wherein the middle hidden layer selects the softmax function as the activation function.
and step 3: and reading historical data files of the running states of all the power grid regulation and control systems, and adopting a gradient lifting tree GBDT classification algorithm. The GBDT algorithm is invoked using a skleran integration method in the machine learning library. In order to make cross validation and model reuse possible on the training sample set, the model is saved using a joblib function. The concept of GBDT can be explained by a popular example, if an individual is 30 years old, and first fits by 20 years old, the loss is found to be 10 years old, then fits by 6 years old to the remaining loss, the gap is found to be 4 years old, and the third fits by 3 years old to the remaining gap, which is only one year old. If the iteration turns are not finished, iteration can be continued, the fitting years error can be reduced in each iteration, and finally the fitting years are added up to be the output result of the model.
generally, GBDT is a linear regression-type model, which can be abbreviated as:
z＝f(x)＝wTx+b (1)
in the formula: z is a predicted value obtained by model calculation, and x is a calculation sample. w is aTand b is a parameter, when the binary task is performed, the output predicted value needs to be marked as y ═ 0,1, so that the real value z needs to be converted into {0,1}, but the function y ═ 0,1} is not continuous, so that a function which can be approximated to a certain extent needs to be foundy is a monotonic differentiable function, so the log-probability function shown by the following equation is chosen as the replacement function:
The function of equation (2) is to convert the value of z to a value of y close to 0 or 1, thereby achieving two-class classification.
At present, historical data of the operation states of all power grid regulation and control systems are divided into a training set and a verification set, and 80% of the historical data are used as the training set and 20% of the historical data are used as the verification set in step 3. However, because the sample output is not a continuous value but a discrete value, the class output error cannot be directly fitted from the output class, so that the problem is solved by adopting a logistic regression-like log-likelihood loss function, namely, the difference between the predicted probability value and the real probability value of the class is adopted to fit the loss, and the historical data of the operation state of the power grid regulation and control system is classified into two categories:
inputting: training data set T consisting of historical data of running states of all power grid regulation and control systems, wherein T is { (x)1,y1),(x2,y2),......,(xi,yi) I represents the number of samples, I is 1, 2. Selecting log-likelihood as a loss function for expressing the degree of difference between the prediction and the actual data, wherein for binary classification, the loss function formula is as follows:
L(y,f(x))＝ln(1+exp(-2yf(x))) (3)
in formula (1), x and y mean the calculated data x in the training set sample T, and its predicted value is y, where y ═ 0,1 }; exp means an exponential function with a natural constant e as the base.
And (3) outputting: two classification tree F (x)
(1) Initializing, constructing a function f of optimal prediction value for minimizing loss function0(x)：
in equation (4), the meaning of P (y ═ 1| x) is the probability when the calculated data x in the training set sample T, and the predicted value y is 1; in the same manner, P (y ═ 1| x) can be known.
(2) Calculating the value of the negative gradient of the loss function of all training samples in the current model, namely the residual value, for the logarithmic loss function, calculating the residual approximation of the loss function, called pseudo residual, and denoted as rmi：
in the formula (5), m represents the number of iterations, namely the number of generated weak learners, and each iteration generates a new classification tree; m ═ 1,2,3,. No. M; i represents a sample, I ═ 1, 2.
(3) Data to be recordedAs training data for the next tree, a new classification tree, r, is fit tomiI represents the number of samples, I is calculated for sample I1, 2mj：
In the formula (6), RmjMeaning the leaf node regions of the mth tree, where m represents the number of iterations, J represents the number of leaf nodes per tree, J is 1, 2.
updating f (x) to obtain:
from this result, the final classification tree F (x) can be obtained:
Since the difference between the predicted probability value and the true probability value of the class is used to fit the loss, the probability is finally converted into the class, see formula (9) and formula (10). Finally, the output is compared with the probability of the category, and the category is predicted if the probability is high.
The output categories in the method are two major categories, namely system normal and system abnormal:
When y is 1, the system is in a normal state:
When y is set to 0, the system is in an abnormal state:
And 4, step 4: and further making specific analysis prediction on the specific condition of the system abnormity. And (3) extracting training data with a system abnormity result obtained by the binary classification algorithm in the step (3), and selecting 75% of the training data as a training set and 25% of the training data as a verification set in order to more accurately analyze and predict the specific abnormal condition of the system. Using the GBDT multi-classification algorithm, the following is specific:
Inputting: in step 3, the result obtained by the binary classification algorithm is training data with system abnormity, the log likelihood is continuously selected as a loss function, and for the multi-element classification, the log loss function formula is as follows:
in formula (11), K means how many categories are shared, and K is 0, 1. y iskIndicates whether it belongs to the kth category, yk1 denotes yes, 0 denotes no; pk(x) Representing the probability that sample x belongs to the kth class.
and (3) outputting: multi-classification tree F (x)
(1) Initialization f (x):
the formula (12) represents a classification function, K represents a class, and K is 0, 1. k is a radical of0Representing initializing a non-iterated function.
(2) for M1, 2, 3.. a.m, the probability P (x) that all sample nodes belong to each class is calculated:
in equation (13), m represents the number of iterations, i.e., the number of weak learners generated. M ═ 1,2, 3.
(3) for all classes K1, 2
In formula (14), i denotes a sample, and i is 1, 2. Pk(xi) Represents the probability that the ith sample belongs to the class k, 0 ≦ Pk(xi)≤1。Indicating whether the ith sample belongs to the kth class.
(4) for probability pseudo residualFitting a classification tree, calculating the sum of the values of each leaf node region, and calculating the minimum value of the loss function
In the formula (15), the first and second groups,Calculated pseudo-residual values, R, representing the class to which each sample belongsmjmeaning the leaf node regions of the mth tree, where m represents the number of iterations, J represents the number of leaf nodes per tree, J is 1, 2. K represents the total number of categories.
Updating f (x) yields:
(5) Get the final multi-classification tree FMk(x)：
the resulting result can be used as a corresponding probability P for classification into the kth classMk(x)：
Finally, the probabilities are converted into categories as follows:
For the final output class, c (k, k ') is the joint cost when the k-th class is predicted when the real value is k', i.e. the class with the highest probability is the class we predict.
and 5: and (4) carrying out secondary cross validation on the calculation results of the multi-classification in the step (4) by adopting a neural network and the training set and the validation set in the step (4). The softmax function mentioned in step 2 is used as a new classifier. It can "compress" a K-dimensional vector z containing arbitrary real numbers into another K-dimensional real vector σ (z) such that each element ranges between (0,1) and the sum of all elements is 1. For example: z ═ 47,58,92, where 47 represents CPU usage, 58 represents memory usage, and 92 represents disk usage. The following formula is used for each element in the vector z, and σ (z) is obtained as {0.4,0.3,0.3 }:
In the formula (20), σ (z)j: vector sigma (z)jdescription of the jth element; z is a radical ofk: the z-th value of the k-dimensional vector; e represents a natural constant, K represents the dimension of the vector, K is 1, 2. K: the number of categories to be classified;
The multi-classification problem can be completed by using a softmax classifier, the output of a plurality of neurons is mapped to the (0,1) interval, and the multi-classification is understood to be based on probability in the method. Suppose that the output of the fully-connected layer of the last layer of the neural network model is a multi-dimensional vector set:
Representing the elements in a multi-dimensional vector,Means the kth element in the I-th multi-dimensional vector x. First, the vector set logits is converted into a probability distribution using softmax, and the probability P (s ═ v | x) that the sample vector x belongs to the jth class is:
In the formula: p (s ═ v | x): probability when the classification result s belongs to the v-th class; n: the total number of categories; x': a derivative of the run data x in the training set; x is the number ofT: setting a weight value T for the operation data x in the training set; w is av: the proportion of class v; w is an: the proportion of the nth class.
and then taking the classification result of the sample with the maximum probability value. Therefore, the result with the maximum probability is the final result obtained by screening the output results of the neural network, the result means the analysis and prediction result of the operation state of the power grid regulation and control system, and the result obtained in the step 4 is calibrated and verified, so that the operation state of the regulation and control system can be well and accurately analyzed and predicted.
The invention is improved from the following aspects:
1. In the technical scheme provided by the invention, the historical data of the operation state of the existing regulation and control system is subjected to data preprocessing based on the neural network and the GBDT, a state prediction model is established, and the data of the operation state can be accurately and quickly classified and predicted; meanwhile, the neural network can change the neuron configuration according to different requirements, and the GBDT can continuously iterate according to the requirements, so that the accuracy and the rapidity of the analysis and the evaluation of the operating state of the control system are ensured, and the application flexibility and the expandability of a calculation model are ensured.
2. In the technical scheme provided by the invention, the operation state of the power grid regulation and control system can be accurately classified based on the GBDT two-classification algorithm and the multi-classification algorithm; the two-classification algorithm can effectively and accurately analyze the mass historical data, distinguish two conditions of normal operation states and abnormal operation states of the system, divide the specific operation states of the system in more detail on the basis, and realize accurate evaluation of the current operation state of the power grid regulation and control system through analysis and calculation of the mass historical data.
3. In the technical scheme provided by the invention, a model based on the feedforward neural network and the perceptron can continuously generate a new learner, and a sample is continuously trained and iterated; the selected activation function can ensure the accuracy of the output result, and meanwhile, the trained learner can obviously improve the calculation efficiency in the next calculation, thereby greatly reducing the calculation time.
4. in the technical scheme provided by the invention, the output result of the GBDT algorithm is secondarily verified by using the softmax function, so that the accuracy of the final result is ensured, and the accuracy of the analysis and evaluation result of the operation state of the regulation and control system is ensured.
example 2
Based on the same inventive concept, the embodiment of the invention also provides an analysis and evaluation system for the operation state of the power grid regulation and control system, which comprises:
the acquisition module is used for acquiring real-time operation data of the power grid regulation and control system;
The evaluation module is used for analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system;
the state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operation state data of the power grid regulation and control system.
in an embodiment, the system further comprises: the construction module is used for constructing a state prediction model;
The building module comprises:
The two classification units are used for classifying states corresponding to the operation data of the power grid regulation and control system into two categories, namely a normal system and an abnormal system, by adopting a gradient lifting tree two classification algorithm based on the historical operation state data of the power grid regulation and control system;
the multi-classification unit is used for classifying the specific conditions of the system abnormity by adopting a gradient lifting tree multi-classification algorithm based on the historical running state data corresponding to the system abnormity to obtain a multi-classification result;
the verification unit is used for performing secondary cross verification on the multi-classification result by adopting a feedforward neural network based on historical running state data corresponding to the system abnormity, and selecting the class with the highest probability as a specific abnormal state in the system abnormity;
wherein the historical operating state data comprises: run data and corresponding states; the states include: system normal and system abnormal; the system exception includes: downtime, CPU overshoot, process deadlock, network unavailability, and the like.
As will be appreciated by one skilled in the art, embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the present application may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present application may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.
The present application is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the application. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.
these computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.
The present invention is not limited to the above embodiments, and any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention are included in the scope of the claims of the present invention which are filed as the application.
Claims (13)
1. A method for analyzing and evaluating the operation state of a power grid regulation and control system is characterized by comprising the following steps:
acquiring real-time operation data of a power grid regulation and control system;
Analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system;
The state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operation state data of the power grid regulation and control system.
2. the method of claim 1, wherein the constructing of the state prediction model comprises:
Based on historical operating state data of the power grid regulation and control system, dividing the state corresponding to the operating data of the power grid regulation and control system into two categories, namely a system normal category and a system abnormal category by adopting a gradient lifting tree binary classification algorithm;
Based on historical running state data corresponding to the system abnormity, classifying the specific conditions of the system abnormity by adopting a gradient lifting tree multi-classification algorithm to obtain a multi-classification result;
Performing secondary cross validation on the multi-classification result by adopting a feedforward neural network based on historical running state data corresponding to the system abnormity, and selecting the class with the highest probability as a specific abnormal state in the system abnormity;
Wherein the historical operating state data comprises: run data and corresponding states; the states include: system normal and system abnormal; the system exception includes: downtime, CPU flush, process deadlock, network outage.
3. The method of claim 2, wherein the step of classifying the states corresponding to the operation data of the power grid regulation and control system into two categories, namely a normal system and an abnormal system, by using a gradient lifting tree classification algorithm based on the historical operation state data of the power grid regulation and control system comprises the following steps:
dividing historical operation state data of the power grid regulation and control system into a training sample and a verification sample;
Training a training sample by adopting a gradient lifting tree binary classification algorithm based on a first loss function set by the difference between the predicted probability value and the real probability value of the state of the power grid regulation and control system to obtain a binary classification tree;
calculating the probability that the states corresponding to the running data in the two classification trees are respectively normal and abnormal based on a set probability conversion formula;
determining the state corresponding to the operation data based on the probability that the state corresponding to each operation data is respectively the system normal and the system abnormal;
verifying the trained binary tree by using a verification sample, finishing training when the error meets the requirement, and dividing the corresponding state of the operation data of the power grid regulation and control system into two categories, namely a system normal category and a system abnormal category; otherwise, the training is carried out again.
4. the method of claim 3, wherein the training samples are trained by a gradient lifting tree classification algorithm based on a first loss function set based on a difference between a predicted probability value and a real probability value of the state of the power grid regulation system to obtain a two-classification tree, and the method comprises:
constructing an optimal prediction probability value function based on minimizing the first loss function;
Calculating the prediction probability values of all training samples based on the current optimal prediction probability value function;
obtaining a pseudo residual error of a first loss function based on the predicted probability value and the real probability value of all the training samples;
Fitting all the operation data and the corresponding pseudo-residuals in the current training sample into a new classification tree by taking the operation data and the corresponding pseudo-residuals as training data of a next tree;
updating an optimal prediction probability value function based on the current pseudo residual error, performing iterative computation on a new classification tree based on the updated optimal prediction probability value function, ending circulation when the value of the first loss function is minimum, and updating an optimal prediction value function based on the minimum value of the first loss function;
and summing the optimal predicted value functions in all the iteration processes to obtain a two-class tree.
5. the method of claim 3, wherein the first loss function is expressed by:
L(y,f(x))＝ln(1+exp(-2yf(x)))
In the formula: x: training samples in the training set; y: training the real state corresponding to the sample; wherein y is {0,1 }; f (x): training a predicted value of the corresponding running state of the sample; exp: an exponential function with a natural constant e as the base.
6. the method of claim 5, wherein the probability transformation formula comprises:
When y is 1, the system normal probability conversion formula is shown as the following formula:
In the formula: p (y ═ 1| x): training the probability that the sample x corresponds to the system normality; f (x): a second classification tree;
when y is 0, the system anomaly probability conversion formula is shown as the following formula:
in the formula: p (y ═ 0| x): the training sample x corresponds to the probability of system anomaly.
7. The method of claim 2, wherein the classifying the specific condition of the system anomaly by using a gradient lifting tree multi-classification algorithm based on the historical operating state data corresponding to the system anomaly to obtain a multi-classification result comprises:
dividing historical running state data corresponding to the system abnormity into a training set and a verification set;
training the training set by adopting a gradient lifting tree multi-classification algorithm based on a set second loss function to obtain a multi-classification tree;
Obtaining corresponding probabilities of sample data classified into various categories based on the multi-classification tree;
Converting the maximum probability in the classes corresponding to the sample data into corresponding classes to obtain multi-classification results;
verifying the trained multi-classification tree by using a verification set, finishing training when the error meets the requirement, and classifying the abnormal specific condition of the power grid regulation and control system to obtain a multi-classification result; otherwise, the training is carried out again.
8. the method of claim 7, wherein the training set is trained using a gradient lifting tree multi-classification algorithm based on the set second loss function to obtain a multi-classification tree, comprising:
constructing a classification function based on minimizing the second loss function;
predicting the probability that all sample nodes in the training set belong to each class based on the classification function;
calculating pseudo-residual values of all classes based on real class and prediction class probability of sample nodes in all training sets;
fitting a new multi-classification tree based on all the sample nodes in the training set and the pseudo residual error values;
updating a classification function based on the current pseudo residual value, iteratively calculating the sum of values of each leaf node region on the basis of the updated classification function, ending the loop when the minimum value of a second loss function is obtained, and updating the classification function based on the minimum value of the second loss function;
And obtaining a multi-classification tree based on all classification functions in the iteration process.
9. the method of claim 2, wherein the performing a second cross-validation of the multi-classification results using a feed-forward neural network based on historical operating state data corresponding to the system anomaly comprises:
adding a softmax function into a hidden layer of the feedforward neural network model to generate a classifier;
Obtaining a multi-dimensional vector group based on the operation data in the historical operation state data and a set probability calculation formula;
based on the multi-dimensional vector group and the classifier, obtaining probability distribution formed by probabilities of each group of operation data corresponding to each category;
selecting the category with the highest probability from the probability distribution as a classification result corresponding to the operating data;
and performing secondary cross validation on the multi-classification result based on the classification result corresponding to each group of operation data.
10. the method of claim 9, wherein each element within the multidimensional set of vectors is expressed as follows:
in the formula: sigma (z)j: vector sigma (z)jDescription of the jth element; z is a radical ofk: the z-th value of the k-dimensional vector; k: number of categories classified.
11. the method of claim 9, wherein the probability of the operational data corresponding to each category is calculated as follows:
in the formula: p (s ═ v | x): probability when the classification result s belongs to the v-th class; n: the total number of categories; x': a derivative of the run data x in the training set; x is the number ofT: setting a weight value T for the operation data x in the training set; w is av: the proportion of class v; w is an: the proportion of the nth class.
12. An analysis and evaluation system for the operation state of a power grid regulation and control system is characterized by comprising:
the acquisition module is used for acquiring real-time operation data of the power grid regulation and control system;
the evaluation module is used for analyzing and evaluating the operation state of the power grid regulation and control system based on the real-time operation data and a pre-constructed state prediction model to obtain the operation state of the power grid regulation and control system;
the state prediction model is constructed by adopting a feedforward neural network and a gradient lifting tree algorithm based on historical operation state data of the power grid regulation and control system.
13. the system of claim 12, wherein the system further comprises: the construction module is used for constructing a state prediction model;
the building module comprises:
the two classification units are used for classifying states corresponding to the operation data of the power grid regulation and control system into two categories, namely a normal system and an abnormal system, by adopting a gradient lifting tree two classification algorithm based on the historical operation state data of the power grid regulation and control system;
the multi-classification unit is used for classifying the specific conditions of the system abnormity by adopting a gradient lifting tree multi-classification algorithm based on the historical running state data corresponding to the system abnormity to obtain a multi-classification result;
The verification unit is used for performing secondary cross verification on the multi-classification result by adopting a feedforward neural network based on historical running state data corresponding to the system abnormity, and selecting the class with the highest probability as a specific abnormal state in the system abnormity;
wherein the historical operating state data comprises: run data and corresponding states; the states include: system normal and system abnormal; the system exception includes: downtime, CPU flush, process deadlock, network outage.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910693859.8A CN110571792A (en) | 2019-07-29 | 2019-07-29 | Analysis and evaluation method and system for operation state of power grid regulation and control system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910693859.8A CN110571792A (en) | 2019-07-29 | 2019-07-29 | Analysis and evaluation method and system for operation state of power grid regulation and control system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN110571792A true CN110571792A (en) | 2019-12-13 |
Family
ID=68773609
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910693859.8A Pending CN110571792A (en) | 2019-07-29 | 2019-07-29 | Analysis and evaluation method and system for operation state of power grid regulation and control system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110571792A (en) |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111144472A (en) * | 2019-12-23 | 2020-05-12 | 深圳供电局有限公司 | Attack identification method based on GBDT algorithm and photovoltaic grid-connected interface device |
| CN111383128A (en) * | 2020-03-09 | 2020-07-07 | 中国电力科学研究院有限公司 | Method and system for monitoring running state of power grid embedded terminal equipment |
| CN111950928A (en) * | 2020-08-24 | 2020-11-17 | 国网冀北电力有限公司 | Loss reduction method and device for power distribution network, storage medium and computing equipment |
| CN111999580A (en) * | 2020-08-14 | 2020-11-27 | 佳都新太科技股份有限公司 | Detection method and device for subway platform gate, computer equipment and storage medium |
| CN113159517A (en) * | 2021-03-24 | 2021-07-23 | 国网浙江省电力有限公司宁波供电公司 | Three-dimensional visual power grid operation data analysis system |
| CN113449914A (en) * | 2021-06-25 | 2021-09-28 | 国网山东省电力公司梁山县供电公司 | Power system monitoring method and system |
| CN113760629A (en) * | 2020-06-15 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for monitoring running state of application system |
| CN114757104A (en) * | 2022-04-28 | 2022-07-15 | 中国水利水电科学研究院 | Construction method of series gate group water transfer engineering hydraulic real-time regulation model based on data driving |
| CN117312293A (en) * | 2023-11-28 | 2023-12-29 | 福建华电福瑞能源发展有限公司 | Electric power multisource heterogeneous data management and intelligent analysis method and system |
| CN117519054A (en) * | 2023-12-11 | 2024-02-06 | 广州智业节能科技有限公司 | High-efficient cold station control system |
-
2019
- 2019-07-29 CN CN201910693859.8A patent/CN110571792A/en active Pending
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111144472A (en) * | 2019-12-23 | 2020-05-12 | 深圳供电局有限公司 | Attack identification method based on GBDT algorithm and photovoltaic grid-connected interface device |
| CN111383128A (en) * | 2020-03-09 | 2020-07-07 | 中国电力科学研究院有限公司 | Method and system for monitoring running state of power grid embedded terminal equipment |
| CN113760629A (en) * | 2020-06-15 | 2021-12-07 | 北京沃东天骏信息技术有限公司 | Method, device and equipment for monitoring running state of application system |
| CN111999580A (en) * | 2020-08-14 | 2020-11-27 | 佳都新太科技股份有限公司 | Detection method and device for subway platform gate, computer equipment and storage medium |
| CN111999580B (en) * | 2020-08-14 | 2023-12-01 | 佳都科技集团股份有限公司 | Method and device for detecting subway platform gate, computer equipment and storage medium |
| CN111950928A (en) * | 2020-08-24 | 2020-11-17 | 国网冀北电力有限公司 | Loss reduction method and device for power distribution network, storage medium and computing equipment |
| CN111950928B (en) * | 2020-08-24 | 2024-02-06 | 国网冀北电力有限公司 | Loss reduction method and device for power distribution network, storage medium and computing equipment |
| CN113159517A (en) * | 2021-03-24 | 2021-07-23 | 国网浙江省电力有限公司宁波供电公司 | Three-dimensional visual power grid operation data analysis system |
| CN113449914A (en) * | 2021-06-25 | 2021-09-28 | 国网山东省电力公司梁山县供电公司 | Power system monitoring method and system |
| CN114757104A (en) * | 2022-04-28 | 2022-07-15 | 中国水利水电科学研究院 | Construction method of series gate group water transfer engineering hydraulic real-time regulation model based on data driving |
| CN117312293A (en) * | 2023-11-28 | 2023-12-29 | 福建华电福瑞能源发展有限公司 | Electric power multisource heterogeneous data management and intelligent analysis method and system |
| CN117312293B (en) * | 2023-11-28 | 2024-04-05 | 福建华电福瑞能源发展有限公司 | Electric power multisource heterogeneous data management and intelligent analysis method and system |
| CN117519054A (en) * | 2023-12-11 | 2024-02-06 | 广州智业节能科技有限公司 | High-efficient cold station control system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110571792A (en) | Analysis and evaluation method and system for operation state of power grid regulation and control system | |
| CN105589806B (en) | A kind of software defect tendency Forecasting Methodology based on SMOTE+Boosting algorithms | |
| KR101966557B1 (en) | Repairing-part-demand forecasting system and method using big data and machine learning | |
| CN112990556A (en) | User power consumption prediction method based on Prophet-LSTM model | |
| US11650968B2 (en) | Systems and methods for predictive early stopping in neural network training | |
| CN111105332A (en) | Highway intelligent pre-maintenance method and system based on artificial neural network | |
| CN107644297B (en) | Energy-saving calculation and verification method for motor system | |
| CN110757510B (en) | Method and system for predicting remaining life of robot | |
| CN111898867B (en) | Airplane final assembly production line productivity prediction method based on deep neural network | |
| CN114595623A (en) | XGboost algorithm-based unit equipment reference value prediction method and system | |
| CN115098330A (en) | Equipment anomaly detection system and method based on cloud edge collaborative mode | |
| CN112363896A (en) | Log anomaly detection system | |
| CN112308298B (en) | Multi-scenario performance index prediction method and system for semiconductor production line | |
| CN112990587A (en) | Method, system, equipment and medium for accurately predicting power consumption of transformer area | |
| CN114881101A (en) | Power system typical scene associated feature selection method based on bionic search | |
| CN117273440A (en) | Engineering construction Internet of things monitoring and managing system and method based on deep learning | |
| CN110781206A (en) | Method for predicting whether electric energy meter in operation fails or not by learning meter-dismantling and returning failure characteristic rule | |
| Lutska et al. | Forecasting the efficiency of the control system of the technological object on the basis of neural networks | |
| CN113033898A (en) | Electrical load prediction method and system based on K-means clustering and BI-LSTM neural network | |
| CN111984514A (en) | Prophet-bLSTM-DTW-based log anomaly detection method | |
| Wang et al. | Cloud computing and extreme learning machine for a distributed energy consumption forecasting in equipment-manufacturing enterprises | |
| CN116542701A (en) | Carbon price prediction method and system based on CNN-LSTM combination model | |
| CN113743461B (en) | Unmanned aerial vehicle cluster health degree assessment method and device | |
| CN114692507A (en) | Counting data soft measurement modeling method based on stacking Poisson self-encoder network | |
| Chernyshev et al. | Integration of building information modeling and artificial intelligence systems to create a digital twin of the construction site |
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 |