CN114169763A - Measuring instrument demand prediction method, system, computing device and storage medium - Google Patents

Measuring instrument demand prediction method, system, computing device and storage medium Download PDF

Info

Publication number
CN114169763A
CN114169763A CN202111495653.8A CN202111495653A CN114169763A CN 114169763 A CN114169763 A CN 114169763A CN 202111495653 A CN202111495653 A CN 202111495653A CN 114169763 A CN114169763 A CN 114169763A
Authority
CN
China
Prior art keywords
data
demand
algorithm
metering
prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202111495653.8A
Other languages
Chinese (zh)
Inventor
张羽
舒永生
赵莉
欧习洋
黄磊
王奕
周游
李刚
孙恺霞
邓红梅
吕梁
贺业梅
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
State Grid Chongqing Electric Power Co Marketing Service Center
State Grid Corp of China SGCC
Shandong Luruan Digital Technology Co Ltd
Original Assignee
State Grid Chongqing Electric Power Co Marketing Service Center
State Grid Corp of China SGCC
Shandong Luruan Digital Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by State Grid Chongqing Electric Power Co Marketing Service Center, State Grid Corp of China SGCC, Shandong Luruan Digital Technology Co Ltd filed Critical State Grid Chongqing Electric Power Co Marketing Service Center
Priority to CN202111495653.8A priority Critical patent/CN114169763A/en
Publication of CN114169763A publication Critical patent/CN114169763A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • 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
    • G06Q10/0631—Resource planning, allocation, distributing or scheduling for enterprises or organisations
    • G06Q10/06315—Needs-based resource requirements planning or analysis
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/048—Activation functions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Systems or methods specially adapted for specific business sectors, e.g. utilities or tourism
    • G06Q50/06—Electricity, gas or water supply
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P80/00—Climate change mitigation technologies for sector-wide applications
    • Y02P80/10—Efficient use of energy, e.g. using compressed air or pressurized fluid as energy carrier

Abstract

The invention discloses a demand forecasting method, a system, a computing device and a storage medium for a metering device, belongs to the technical field of electric energy demand forecasting and solves the technical problems of low forecasting accuracy and higher time cost of the existing metering device; if the installation amount is irrelevant to external factors, forecasting is carried out through an SRU + Bayesian network algorithm, the demand amount of the 'failure first-aid repair' installation type metering device is accurately forecasted, and a forecasting result is output; the invention provides scientific support for optimizing the inventory management of electric power materials and reducing the enterprise cost.

Description

Measuring instrument demand prediction method, system, computing device and storage medium
Technical Field
The invention relates to the technical field of demand prediction of measuring instruments, in particular to the technical field of demand prediction methods, systems, computing devices and storage media of measuring instruments.
Background
The demand prediction of the metering appliance refers to accurately predicting the demand of the electric energy metering appliance in a certain period of time in the future of the power company through a prediction algorithm implanted by a system, and obtaining reliable demand data which can be used for the power company to make a scientific material purchasing plan; with the development of social economy and the advance of reform of power enterprises, the demand of electric energy metering devices increases year by year, and in order to ensure the quick response of the demand of power users and avoid material waste, a purchasing management mode of the metering devices represented by 'material consignment sale' is created; under the operation of the mode, before a user generates a demand of a metering device, a power company needs to obtain the demand of the metering device of the user in a future period through a demand prediction technology, formulate a purchasing plan according to predicted demand data, and provide the metering device in material inventory, so that a demand prediction algorithm for the metering device needs to be constructed urgently, and accurate and reliable data basis is provided for the power company to formulate a scientific purchasing plan and purchasing management mode of the metering device.
Currently, common methods for predicting the demand of a measuring instrument are mainly divided into two types: single algorithm prediction and combined algorithm prediction, wherein the single algorithm prediction comprises univariate prediction and multivariate prediction, and the univariate prediction method mostly adopts an algorithm based on statistics, so that the method has higher requirements on the stability of historical data, otherwise, the prediction error is large, and the prediction accuracy is low; the multivariate prediction method adopts an algorithm based on machine learning or deep learning, and usually has good prediction performance only for specific characteristics; the combined algorithm prediction combines two or more than two complementary algorithms, so that the prediction precision of the algorithms is improved, and the operation time cost is correspondingly improved due to the complex structure.
Disclosure of Invention
The invention aims to: in order to solve the technical problems of low prediction accuracy and high time cost of the measuring instrument demand prediction method, the invention provides a measuring instrument demand prediction method, a system, a computing device and a storage medium, and the prediction accuracy of an algorithm are improved on the basis of not consuming excessive computing resources and computing time.
The technical scheme adopted by the invention is as follows: a method for predicting demand of a metering device comprises the following steps:
step 1: the data acquisition unit acquires historical data of installation quantity of the measuring instruments of various models and external data of various influencing factors in corresponding periods, a data processor calculates chi-square values of two variables, if the chi-square value is less than 1, the installation quantity is related to the external factors, the installation types of the measuring instruments are in a 'business expansion new installation and transformation rotation' type, and the step 2 is executed; if the chi-square value is greater than or equal to 1, determining that the installation amount is irrelevant to external factors, and executing a step 3, wherein the installation amount is a fault first-aid repair installation type;
step 2: forecasting by adopting an SRU + RBF algorithm, forecasting the demand of the metering appliance by using historical data of the installation quantity/demand of the 'industry expansion new installation, transformation rotation' installation type metering appliance in the current period and data of the influence factors of the installation quantity of the installation type metering appliance, and outputting a forecasting result;
the SRU + RBF algorithm is used for predicting, namely a combined algorithm of a Simple circulation Unit (SRU) and a Radial basis function neural network (RBF), and the demand of the installation type metering appliances of the transformation rotation and the business expansion new installation is calculated according to the demand data of the installation type metering appliances of the transformation rotation and the business expansion new installation in the current period and by combining the external factor information related to the business characteristics;
and step 3: and predicting by adopting an SRU + Bayesian network algorithm, acquiring the attribute data of the metering appliance with faults in the power field in the past year, processing to obtain high-correlation characteristics/attributes of the faults of the metering appliance, acquiring the attribute data of the metering appliance in the power field according to the characteristics/attributes to predict the demand of the metering appliance, and outputting a prediction result. And predicting by the SRU + Bayesian network algorithm, and predicting the faults of the metering devices in operation to obtain the types and the number of the metering devices with faults, and calculating the demand of the 'fault first-aid repair' installation type metering devices.
Step 2, predicting by the SRU + RBF algorithm, predicting the demand of the 'transformation rotation, industry expansion new installation' installation type metering devices, and calculating the demand P of a certain type of metering devicean(ii) a The specific demand forecasting steps are as follows:
a1: collecting historical order data of a measuring instrument, and preprocessing the data;
a1.1: acquiring historical orders of 'transformation rotation' and 'business expansion new installation' installation type metering devices in an MDS system or an SG186 system, acquiring historical installation quantity of each type of metering device in each week/month/quarter, and forming a historical data matrix by taking different types as row directions and different week/month/quarter directions as column directions;
a1.2: in the historical data matrix, each column of data forms a column vector
Figure BDA0003399997360000031
The minimum value data and the maximum value data of each row respectively form a column vector
Figure BDA0003399997360000032
Carrying out normalization pretreatment on the data in the matrix, and normalizing the original data to [0, 1%]Get each column vector of the normalized data
Figure BDA0003399997360000033
The normalization formula is as follows:
Figure BDA0003399997360000034
normalizing the column vector
Figure BDA0003399997360000035
Sequentially splicing to form a normalized historical data matrix X';
a2: performing an SRU algorithm on the historical installation quantity time sequence data to obtain the preliminary prediction demand of the metering device;
a2.1: the first 4/5 columns of data of the historical data matrix X 'are used as a training set X'1With the remaining data as test set X'2；
A2.2: the method comprises the following steps of building an SRU network, setting the number of network layers to be 3, setting the number of SRU neurons in the first layer to be 8, then, setting the number of the neurons in the second layer to be decreased progressively according to integral multiple of 2, selecting a network initialization method to be uniform distribution initialization, setting activation functions to be a tanh function and a Sigmoid function, and setting a function expression to be:
Figure BDA0003399997360000036
Figure BDA0003399997360000037
the number of neural iterations is 100, the batch processing parameter is 8, the number of neuron jitters is 5, and the training set X 'is brought'1Training the whole network, and performing parameter optimization by using an Adam optimization algorithm to obtain a parameter matrix W, Wf、Wr、bf、br；
A2.3: input test set X'2For the current time input data XtCalculating the forgetting information degree f of the networktThe calculation formula is as follows:
ft＝σ(WfXt+bf) (4)
a2.4: determining information to be updated
Figure BDA0003399997360000041
A2.5: determining a degree r of update informationt＝σ(WrXt+br)；
A2.6: according to the cell state C at the last momentt-1Degree of forgetting information ftInformation required to be updated
Figure BDA0003399997360000042
Calculating the current newly formed cell state CtThe calculation formula is as follows:
Figure BDA0003399997360000043
a2.7: according to the current cell state CtDegree of update rtCurrent input XtCalculating the output h at the current momenttThe calculation formula is as follows:
ht＝rttanh(ct)+(1-rt)Xt (6)
thus, the internal information transfer of the SRU is only for the current time input XtParticipating in operation, outputting h at the current momenttIs not dependent on the previous time output ht-1The structure enables the SRU to complete independent parallel operation;
a2.8: obtaining the initial demand prediction results of metering devices of different models through an SRU algorithm: y is1,Y2,Y3,...；
A3: analyzing and acquiring external influence factor information of the demand of the transformation rotation and business expansion new installation type metering device, and forecasting the final demand P of the installation type metering device by using a RBF network algorithm in combination with the preliminary demand forecasting result obtained in the step A2.8an。
Step A3 final demand PanThe prediction steps are as follows:
a3.1: the demand sources of the metering appliances with installation types of 'transformation rotation' and 'industry new installation' are mainly as follows: the method comprises the following steps of low-voltage non-resident new installation, low-voltage batch new installation, high-voltage new installation, household division, temporary power utilization, low-voltage non-resident capacity increase, low-voltage capacity increase, household combination and periodic alternate execution; the external influencing factors are mainly as follows: seasonal influence factors, regional GDP growth rate, building completion area, power consumption prediction data, service cycle of a measuring instrument, cost price of the measuring instrument and household application data of individual households;
a3.2: acquiring external influence factor data, and combining demand data preliminarily predicted by an SRU algorithm of a certain type of metering appliance to form a data set X ═ X1,x2,...,xnTaking the samples as training samples, and randomly selecting r different samples from the training samples as initial centers;
a3.3: carrying out normalization processing on sample data in the data set, wherein the formula is as follows:
Figure BDA0003399997360000051
in the formula (7), xiSample value representing a training sample in data set X, min (X) representing the smallest sample value in data set X, max (X) representing the largest sample value in data set X, X'iNormalizing the processed sample value of a certain sample in the data set X;
a3.4: calculating the distance between the sample and the center, and finding out the minimum distance dminThe formula is as follows:
dmin＝min||X-ci(k)|| (8)
in the formula (8), ci(k) Representing the center of the currently obtained training sample by computing each training sample X in the data set XiAnd training sample center ci(k) And obtaining the smallest difference among them, obtaining the smallest distance d of the sample from the centermin；
A3.5: using averaging
Figure BDA0003399997360000052
Center of adjustment ciJudging whether to continue according to whether all the training samples are learned and the central distribution does not change any more, if the two items are not met, the k value is plus 1, returning to the step A3.4 to recalculate dminAnd further adjust ciUntil the training samples are learned and the central distribution is not changed any more, obtaining a final RBF network center;
a3.6: according to the formula
Figure BDA0003399997360000053
Determining a normalization constant delta of a Gaussian function;
a3.7: according to the RBF neural network structure of the input layer node number 8, the output layer node number 1 and the hidden layer node number 8, the activation function of the hidden layer is a Gaussian function, namely the output function of the hidden layer node
Figure BDA0003399997360000054
The output function can be calculated on the basis of the step A3.2 to the step A3.6;
a3.8: calculating nodes from each hidden layer to each output layer by using least square principleWeight value w ofijOn the basis of the step A3.7, the output layer is the weighted sum of the output functions of all nodes of the hidden layer:
Figure BDA0003399997360000055
obtaining the required quantity P of the measuring instrument of the model on the installation type of' transformation rotation and business expansion new installationaj＝yj；
A3.9: repeating the steps A3.2-A3.8 to obtain the required quantity P of various types of metering devices of the installation type of' transformation rotation, industry expansion new installationan。
Step 3, predicting the SRU + Bayesian network algorithm, and predicting the demand of the 'failure repair' installation type metering device specifically comprises the following steps:
b1: acquiring attribute data of a measuring instrument which has failed from a power field: acquiring success rate, online rate, use time, specification and model, manufacturers, production batches, communication flow, communication protocols, channel types, times of uploading important items, times of clock out-of-tolerance and addresses;
b2: preprocessing the attribute data acquired in the step B1;
b2.1: data cleaning is carried out on the data collected in the step B1, error data and repeated data are deleted, and missing data are supplemented;
b2.2: based on the step B2.1, the chi-square test is adopted to mainly carry out correlation analysis on the nominal attributes of the specification model, the production manager, the production batch and the communication protocol and the faults of the metering device, and the correlation between the numerical attributes of the acquisition success rate, the online rate, the service time and the communication flow and the faults of the metering device is mainly considered by calculating the Pearson correlation coefficient, so that the attributes related to the faults of the metering device are obtained, wherein the correlation includes the following steps: acquiring success rate, online rate, specification and model, service time, manufacturer, production batch, times of uploading important items and times of time synchronization out-of-tolerance;
b3: judging the influence degree of each attribute on the fault of the metering device by adopting a reliefF algorithm, thereby extracting high-correlation characteristic data of the fault of the metering device; the method comprises the following specific steps:
b3.1: based on step B2.2, the attribute data is normalized, with the following formula:
Figure BDA0003399997360000061
in the formula (10), X' is normalized data, XminFor the minimum value of each attribute data, XmaxA maximum value for each attribute data;
b3.2: constructing a training set { (x'1,y1),(x′2,y2),...,(x′n,yn) Get the tuple x randomly selectedi；
B3.3: at tuple xiFinding k adjacent neighbors x in similar samplesi,hThen k neighbor x are found out from the heterogeneous samplei,m；
B3.4: calculating a sample R and a neighboring sample HjThe difference diff (A, R, H) over the attribute Aj) The formula is as follows:
Figure BDA0003399997360000071
in formula (11), R < A >]Represents the value of the sample R on the attribute A, Hj[A]Represents a neighbor sample HjThe value of the attribute A is represented by the following equation (11), where the attribute A is of continuous type, the difference diff (A, R, H)j) Passing through type
Figure BDA0003399997360000072
The calculation is performed if the attribute A is discrete and the sample R and the neighboring sample HjIf the values on the attribute A are equal, the difference diff (A, R, H)j) 0; if the attribute A is discrete, and the sample R and the neighboring sample HjIf the values on the attribute A are not equal, the difference diff (A, R, H)j)＝1；
B3.5: calculating the influence weight w (A) of the attribute A on the fault of the metering device, wherein the formula is as follows:
Figure BDA0003399997360000073
h in formula (12)jhRepresenting similar neighbor samples, HjmRepresenting heterogeneous neighbor samples, both homogeneous and heterogeneous neighbor samples being k neighbor samples obtained by m times of random sampling among homogeneous/heterogeneous samples of random sample R,
Figure BDA0003399997360000074
and
Figure BDA0003399997360000075
expressing the accumulation of the value difference of k homogeneous/heterogeneous neighbor samples and the random sample R on the attribute A;
b3.6: repeating the steps B3.4 and B3.5 to obtain the influence weights w corresponding to all the attributes, performing normalization processing on the influence weights w, sorting the influence weights w, setting a threshold value T, and if w (A) > T, extracting corresponding attribute data as high-correlation characteristic data of the faults of the metering device;
b3.7: finally, selecting acquisition success rate, online rate, service time, manufacturers, specification models, production batches and communication flow as high-correlation characteristic data of the faults of the metering devices;
b4: predicting the stable state of the metering appliance based on the high-correlation characteristic data obtained in the step B3.7;
b5: based on the steady state of the data collected by the measuring instrument and the prediction result of the steady state of the data transmitted by the transmission network obtained in the step B4, and the characteristic data in the step B3.7: establishing a Bayesian network by mainly using time, manufacturers and specifications;
b6: repeating the steps B1-B5 to obtain the fault prediction result of the measuring instrument and the specification model and the number of the measuring instruments with faults, counting the number of the faults according to different models, and calculating the demand P of the 'fault first-aid repair' installation type measuring instrumentbn。
Step B4, predicting the steady state of the measuring instrument, specifically comprising the steps of:
b4.1: obtaining the time sequence related features in the high correlation features: acquiring success rate and online rate, and forming two rows of time sequence data;
b4.2: and taking the two rows of time sequence data as input data of the SRU algorithm, executing the operation step of obtaining the preliminary prediction demand of the metering appliance by the SRU algorithm, and respectively obtaining the prediction results of the SRU algorithm on the stable state of the data collected by the metering appliance and the stable state of the data transmitted by the network.
The specific steps of constructing the bayesian network described in step B5 are as follows:
b5.1: and B3.7, based on the prediction results of the stable state of the data collected by the measuring instrument and the stable state of the data transmitted by the network obtained in the step B4.2 and the non-time sequence related characteristic data in the step B3.7: mainly using time, manufacturers and specification models as nodes, establishing an undirected graph containing all the nodes, wherein the nodes in the graph represent characteristic data, and edges among the nodes represent the relation among different characteristics;
b5.2: adopting a grading search method to construct and iteratively optimize a Bayesian network topology structure, and defining a network structure evaluation function as follows:
ScoreBDe(N|D)＝lnP(N)P(D|N)＝lnP(N)+lnP(D|N) (13)
in the formula (13), N is a network structure, and D is training data;
b5.3: inputting node data X ═ V into undirected graph1,V2,...,VnDefining the maximum father node number mu, the node order rho and a data set D, and initializing a father node set;
b5.4: adding each node variable into a father node set in sequence, calculating corresponding network scores, comparing the network scores with old scores, if the new scores are larger than the old scores and the maximum father node number is not reached, taking the node variable as a father node, and adding corresponding edges;
b5.5: combining the variables of other nodes, repeating the step B5.4, and combining domain expert knowledge to adjust to obtain a Bayesian network topological structure;
b5.6: the stable state and transmission network of the data acquired by the measuring instrument based on the step B4.2The prediction result of the stable state of the correlation data, and the non-time-sequence related characteristic data in the step B3.7: determining the use time, the manufacturer and the specification model: data tuple set D ═ x1,x2,...,xn) The category set Y ═ {0,1}, where Y ═ 0 indicates that the meter has not failed, and Y ═ 1 indicates that the meter has failed;
b5.7: determining relevance of edges in undirected graph
Figure BDA0003399997360000091
Wherein the maximum likelihood estimation learning is adopted to obtain
Figure BDA0003399997360000092
Value of (a) corresponds to YiTraining a Bayesian network according to the conditional probability table.
A meter demand forecasting system comprising:
the data acquisition module receives sample data of the faults of the metering device, performs data preprocessing, and extracts high correlation characteristics related to the faults of the metering device;
the system comprises a fault first-aid repair installation type demand prediction algorithm building module, an SRU algorithm, a Bayesian network topology structure building and iterative optimization algorithm building module, a Bayesian network topology structure prediction module, a Bayesian network topology structure optimization module, a fault first-aid repair installation type demand prediction module, a transmission network data prediction module and a fault first-aid repair installation type demand prediction module, wherein the SRU algorithm is used for predicting the stable state of the terminal acquisition data and the stable state of the transmission network data according to the attribute characteristics, and the fault first-aid repair installation type demand prediction module is used for building and iteratively optimizing the Bayesian network topology structure by a scoring search method based on other high-correlation attribute characteristics and the stable state prediction result; based on the Bayesian network topological structure, parameter learning is carried out on the Bayesian network structure, and a terminal fault prediction algorithm is trained;
the prediction module predicts the installation type requirement of the first-aid repair based on the fault prediction algorithm;
the data acquisition module is used for acquiring historical order data of the industrial expansion new installation type electric energy meter and external influence factor information data related to business characteristics, and performing data preprocessing;
the algorithm construction module for predicting the demand of the business expansion new installation and transformation alternate installation type is used for primarily predicting the demand by utilizing an SRU algorithm according to the processed historical time sequence data to obtain the respective primary demand of the metering appliances of multiple models, and training the RBF neural network structure based on the primary demand data and the information of each external influence factor required by the business expansion new installation and transformation alternate installation type metering appliances to obtain the prediction algorithm for predicting the demand of the business expansion new installation and transformation alternate installation type metering appliances;
the system comprises a prediction module for predicting the demand of industry expansion new installation and transformation alternate installation type metering devices, wherein the demand of the industry expansion new installation and transformation alternate installation type metering devices is predicted based on a demand prediction algorithm of the industry expansion new installation and transformation alternate installation type metering devices;
and the counting and counting module counts the demand of each type of metering appliance obtained by the prediction module for the demand of the installation type of the fault first-aid repair, the prediction module for the demand of the installation type of the service expansion and transformation alternate installation type according to different types, and calculates the demand of the metering appliances of different types in a certain period of time in the future.
A computing device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, the processor implementing the method of demand forecasting for a metrology appliance when executing the program.
A storage medium having stored thereon a computer program which, when executed by a processor, implements the method of demand prediction for a metrology appliance.
In summary, due to the adoption of the technical scheme, the invention has the beneficial effects that:
according to the method, different demand prediction algorithms are constructed for the metering devices under different demand conditions in a demand decomposition mode, so that the prediction accuracy is ensured;
by using the combined prediction algorithm, time sequence data, characteristic data, influence factor data and the like related to demand prediction of the metering device can be learned, the information comprehensiveness and the algorithm fault tolerance capability of the prediction algorithm are improved, and the prediction accuracy of the algorithm can be further improved;
in the constructed prediction algorithm, the SRU algorithm is used for processing time sequence data related to demand prediction, and according to the outstanding capability of the SRU algorithm in processing time sequence problems, the calculation efficiency is improved on the basis of smooth information transfer, and the prediction timeliness in short-term prediction is ensured; the invention processes the external influence factor information related to demand forecast by using the RBF neural network algorithm, the RBF network can greatly accelerate the learning speed and avoid the problem of local minimum, and the precision, robustness and adaptability of the forecast are further effectively improved;
according to the forecast demand of the measuring instruments of various models, the invention can combine the existing stock quantity, the quantity to be detected, the quantity to be delivered and the like to make the replenishment strategy of the stock, form the purchasing plan of the measuring instruments in the period, and provide scientific support for optimizing the stock management of electric power materials and reducing the enterprise cost.
Drawings
The invention will now be described, by way of example, with reference to the accompanying drawings, in which:
FIG. 1 is a schematic diagram of a demand decomposition-based metering appliance demand prediction algorithm construction of the present invention;
FIG. 2 is a flow chart of the demand forecasting of the "transformation rotation, business expansion new installation" type metering device of the present invention;
FIG. 3 is a diagram illustrating an SRU information transfer process according to the present invention;
fig. 4 is a flow chart of demand prediction of the "breakdown rush repair" installation type metering device of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. The components of embodiments of the present invention generally described and illustrated in the figures herein may be arranged and designed in a wide variety of different configurations.
Thus, the following detailed description of the embodiments of the present invention, presented in the figures, is not intended to limit the scope of the invention, as claimed, but is merely representative of selected embodiments of the invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example 1
As shown in fig. 1, the present embodiment provides a demand forecasting method for a metering device, including the following steps:
step 1: acquiring historical data of installation quantity of each type of measuring instrument and external data of each influencing factor in a corresponding period, calculating chi-square values of two variables through chi-square test, if the chi-square values are smaller and close to 0, the installation quantity is related to the external factors, the installation type of the measuring instrument is in a 'industry expansion new installation and transformation rotation' type, and executing the step 2; if the chi-square value is larger, determining that the installation amount is irrelevant to external factors and is a fault first-aid repair installation type, and executing the step 3;
step 2: forecasting by adopting an SRU + RBF algorithm, forecasting the demand of the metering appliance by using historical data of the installation quantity/demand of the 'industry expansion new installation, transformation rotation' installation type metering appliance in the current period and data of the influence factors of the installation quantity of the installation type metering appliance, and outputting a forecasting result;
specifically, the SRU + RBF algorithm is used for predicting, namely a combination algorithm of a simple cycle unit (Simp l e ReBurent Un it, SRU) + a radial basis function neural network (Rad i a l Aas i s FunBt i on, RBF), and calculating the demand of the installation type metering appliance for 'transformation rotation' and 'business expansion new installation' according to the current demand data of the installation type metering appliance for 'transformation rotation' and 'business expansion new installation' and the external factor information related to the service characteristics;
and step 3: and predicting by adopting an SRU + Bayesian network algorithm, acquiring the attribute data of the metering appliance with faults in the power field in the past year, processing to obtain high-correlation characteristics/attributes of the faults of the metering appliance, acquiring the attribute data of the metering appliance in the power field according to the characteristics/attributes to predict the demand of the metering appliance, and outputting a prediction result.
Specifically, the SRU + Bayesian network algorithm is used for predicting, the type and the number of the metering appliances with faults are obtained by predicting the faults of the metering appliances in operation, and the demand of the 'fault first-aid repair' installation type metering appliances is calculated.
Example 2
Based on the example 1, as shown in fig. 2 to 3, the SRU + RBF algorithm predicts the demand of the "transformation rotation, industry expansion new installation type metering device" by the general process:
a1: collecting historical order data of a measuring instrument, and preprocessing the data;
a1.1: acquiring historical orders of 'transformation rotation' and 'business expansion new installation' installation type metering devices in an MDS system or an SG186 system, acquiring historical installation quantity of each type of metering device in each week/month/quarter, and forming a historical data matrix by taking different types as row directions and different week/month/quarter directions as column directions;
a1.2: in the historical data matrix, each column of data forms a column vector
Figure BDA0003399997360000131
The minimum value data and the maximum value data of each row respectively form a column vector
Figure BDA0003399997360000132
Carrying out normalization pretreatment on the data in the matrix, and normalizing the original data to [0, 1%]Get each column vector of the normalized data
Figure BDA0003399997360000133
The normalization formula is as follows:
Figure BDA0003399997360000134
normalizing the column vector
Figure BDA0003399997360000135
And sequentially splicing to form a normalized historical data matrix X'.
A2: performing an SRU algorithm on the historical installation quantity time sequence data to obtain the preliminary prediction demand of the metering device;
a2.1: the first 4/5 columns of data of the historical data matrix X 'are used as a training set X'1With the remaining data as test set X'2；
A2.2: the method comprises the following steps of building an SRU network, setting the number of network layers to be 3, setting the number of SRU neurons in the first layer to be 8, then, setting the number of the neurons in the second layer to be decreased progressively according to integral multiple of 2, selecting a network initialization method to be uniform distribution initialization, setting activation functions to be a tanh function and a Sigmoid function, and setting a function expression to be:
Figure BDA0003399997360000136
Figure BDA0003399997360000137
the number of neural iterations is 100, the batch processing parameter is 8, the number of neuron jitters is 5, and the training set X 'is brought'1Training the whole network, and performing parameter optimization by using an Adam optimization algorithm to obtain a parameter matrix W, Wf、Wr、bf、br；
A2.3: input test set X'2For the current time input data XtCalculating the forgetting information degree f of the networktThe calculation formula is as follows:
ft＝σ(WfXt+bf)
a2.4: determining information to be updated
Figure BDA0003399997360000138
A2.5: determining a degree r of update informationt＝σ(WrXt+br)；
A2.6: cell shape according to last momentState Ct-1Degree of forgetting information ftInformation required to be updated
Figure BDA0003399997360000141
Calculating the current newly formed cell state CtThe calculation formula is as follows:
Figure BDA0003399997360000142
a2.7: according to the current cell state CtDegree of update rtCurrent input XtCalculating the output h at the current momenttThe calculation formula is as follows:
ht＝rttanh(ct)+(1-rt)Xt
thus, the internal information transfer of the SRU is only for the current time input XtParticipating in operation, outputting h at the current momenttIs not dependent on the previous time output ht-1As shown in fig. 3, the structure enables the SRU to complete independent parallel operations, thereby greatly improving the computation efficiency.
A2.8: obtaining the initial demand prediction results of metering devices of different models through an SRU algorithm: y is1,Y2,Y3,...
A3: analyzing and acquiring external influence factor information of the demand of the transformation rotation and business expansion new installation type metering device, and forecasting the final demand P of the installation type metering device by using a RBF network algorithm in combination with the preliminary demand forecasting result obtained in the step A2.8an；
A3.1: in combination with actual business, the main sources of requirements for installing metering devices with the installation types of 'transformation rotation' and 'industry expansion new installation' are as follows: the low-voltage non-resident new installation, the low-voltage batch new installation, the high-voltage new installation, the household division, the temporary power consumption, the low-voltage non-resident capacity increase, the low-voltage capacity increase, the household combination, the periodic cycle alternate execution and the like, and according to the business experience, the external influence factors of the demand of the metering device with the installation type of 'transformation alternate, industry expansion new installation' can be analyzed as follows: seasonal influence factors, regional GDP growth rate, building completion area, power consumption prediction data, service cycle of a measuring instrument, cost price of the measuring instrument, individual household application data and the like;
a3.2: acquiring external influence factor data, and combining demand data preliminarily predicted by an SRU algorithm of a certain type of metering appliance to form a data set X ═ X1,x2,...,xnTaking the samples as training samples, and randomly selecting r different samples from the training samples as initial centers;
a3.3: carrying out normalization processing on sample data in the data set, wherein the formula is as follows:
Figure BDA0003399997360000143
a3.4: calculating the distance between the sample and the center, and finding out the minimum distance dminThe formula is as follows:
dmin＝min||X-ci(k)||
a3.5: using averaging
Figure BDA0003399997360000151
Center of adjustment ciJudging whether to continue according to whether all the training samples are learned and the central distribution does not change any more, if the two items are not met, the k value is plus 1, returning to the step A3.4 to recalculate dminAnd further adjust ciUntil the training samples are learned and the central distribution is not changed any more, obtaining a final RBF network center;
a3.6: according to the formula
Figure BDA0003399997360000152
Determining a normalization constant delta of a Gaussian function;
a3.7: the RBF neural network structure constructed by the invention is as follows: the number of nodes of the input layer is 8, the number of nodes of the output layer is 1, the number of nodes of the hidden layer is 8, and the activation function of the hidden layer is a Gaussian function, namely the output function of the nodes of the hidden layer
Figure BDA0003399997360000153
The output function can be calculated on the basis of the step A3.2 to the step A3.6;
a3.8: calculating the weight w from each hidden layer node to each output layer node by using the least square principleijOn the basis of the step A3.7, the output layer is the weighted sum of the output functions of all nodes of the hidden layer:
Figure BDA0003399997360000154
obtaining the required quantity P of the measuring instrument of the model on the installation type of' transformation rotation and business expansion new installationaj＝yj；
A3.9: repeating the steps A3.2-A3.8 to obtain the required quantity P of various types of metering devices of the installation type of' transformation rotation, industry expansion new installationan。
Example 3
On the basis of the embodiment 1, the SRU + Bayesian network algorithm is adopted for prediction, the type and the number of the metering appliances with faults are obtained by predicting the faults of the metering appliances in operation, and the required quantity P of the 'fault first-aid repair' installation type metering appliances is calculatedbn。
Specifically, as shown in fig. 4, the general process of demand prediction of the SRU + bayesian network combination algorithm for the "failure repair" installation type metering device is as follows:
b1: acquiring attribute data of a measuring instrument which has failed from a power field: acquiring success rate, online rate, use time, specification and model, manufacturers, production batches, communication flow, communication protocols, channel types, times of uploading important items, times of clock out-of-tolerance and addresses;
b2: preprocessing the attribute data acquired in the step B1;
b2.1: data cleaning is carried out on the data collected in the step B1, error data and repeated data are deleted, and missing data are supplemented;
b2.2: based on the step B2.1, the chi-square test is adopted to carry out correlation analysis on the nominal attributes such as specification models, production managers, production batches, communication rules and the like and the faults of the measuring instruments, and the correlation between the numerical attributes such as acquisition success rate, online rate, service time, communication flow and the like and the faults of the measuring instruments is investigated by calculating the Pearson correlation coefficient, so that the obtained attributes related to the faults of the measuring instruments comprise: acquiring success rate, online rate, specification and model, service time, manufacturer, production batch, times of uploading important items and times of time synchronization out-of-tolerance;
b3: in order to ensure the generalization capability of a fault prediction algorithm, the influence degree of each attribute on the fault of the metering device is judged by adopting a re l i efF algorithm, so that high-correlation characteristic data of the fault of the metering device is extracted;
b3.1: based on step B2.2, the attribute data is normalized, with the following formula:
Figure BDA0003399997360000161
wherein X' is normalized data, XminFor the minimum value of each attribute data, XmaxA maximum value for each attribute data;
b3.2: constructing a training set { (x'1,y1),(x′2,y2),...,(x′n,yn) Get the tuple x randomly selectedi；
B3.3: at tuple xiFinding k adjacent neighbors x in similar samplesi,hThen k neighbor x are found out from the heterogeneous samplei,m；
B3.4: calculating a sample R and a neighboring sample HjThe difference diff (A, R, H) over the attribute Aj) The formula is as follows:
Figure BDA0003399997360000171
b3.5: calculating the influence weight w (A) of the attribute A on the fault of the metering device, wherein the formula is as follows:
Figure BDA0003399997360000172
b3.6: repeating the steps B3.4 and B3.5 to obtain the influence weights w corresponding to all the attributes, performing normalization processing on the influence weights w, sorting the influence weights w, setting a threshold value T, and if w (A) > T, extracting corresponding attribute data as high-correlation characteristic data of the faults of the metering device;
b3.7: according to the invention, the acquisition success rate, the online rate, the use time, the manufacturer, the specification and the model, the production batch and the communication flow are finally selected as high-correlation characteristic data of the faults of the metering device;
b4: predicting the stable state of the metering appliance based on the high-correlation characteristic data obtained in the step B3.7;
b4.1: obtaining the time sequence related features in the high correlation features: acquiring success rate and online rate, and forming two rows of time sequence data;
b4.2: taking the two rows of time sequence data as input data of an SRU algorithm, executing the steps A1.2-A2.7, and respectively obtaining the prediction results of the SRU algorithm on the stable state of the data collected by the metering device and the stable state of the data transmitted by the network;
b5: and B3.7, based on the prediction results of the stable state of the data collected by the measuring instrument and the stable state of the data transmitted by the network obtained in the step B4.2 and the non-time sequence related characteristic data in the step B3.7: constructing a Bayesian network by using time, manufacturers, specifications and models and the like;
b5.1: and B3.7, based on the prediction results of the stable state of the data collected by the measuring instrument and the stable state of the data transmitted by the network obtained in the step B4.2 and the non-time sequence related characteristic data in the step B3.7: using time, manufacturers, specification models and the like as nodes, establishing an undirected graph containing all the nodes, wherein the nodes in the graph represent characteristic data, and edges among the nodes represent the relation among different characteristics;
b5.2: adopting a grading search method to construct and iteratively optimize a Bayesian network topology structure, and defining a network structure evaluation function as follows:
ScoreBDe(N|D)＝lnP(N)P(D|N)＝lnP(N)+lnP(D|N)
wherein N is a network structure and D is training data;
b5.3: inputting node data X ═ V into undirected graph1,V2,...,VnDefining the maximum father node number mu, the node order rho and a data set D, and initializing a father node set;
b5.4: adding each node variable into a father node set in sequence, calculating corresponding network scores, comparing the network scores with old scores, if the new scores are larger than the old scores and the maximum father node number is not reached, taking the node variable as a father node, and adding corresponding edges;
b5.5: combining the variables of other nodes, repeating the step B5.4, and combining domain expert knowledge to adjust to obtain a Bayesian network topological structure;
b5.6: and B3.7, based on the prediction results of the stable state of the data collected by the measuring instrument and the stable state of the data transmitted by the network obtained in the step B4.2 and the non-time sequence related characteristic data in the step B3.7: determining the use time, the manufacturer, the specification and the model, and the like: data tuple set D ═ x1,x2,...,xn) The category set Y ═ {0,1}, where Y ═ 0 indicates that the meter has not failed, and Y ═ 1 indicates that the meter has failed;
b5.7: determining relevance of edges in undirected graph
Figure BDA0003399997360000181
Wherein the maximum likelihood estimation learning is adopted to obtain
Figure BDA0003399997360000182
Value of (a) corresponds to YiTraining a Bayesian network by using the conditional probability table;
b6: repeating the steps B1-B5 to obtain the fault prediction result of the measuring instrument and the specification model and the number of the measuring instruments with faults, counting the number of the faults according to different models, and calculating the demand P of the 'fault first-aid repair' installation type measuring instrumentbn；
C: summarizing the predicted demand of the measuring instruments according to different models by using two algorithms to obtain the demand P of each model of measuring instrument in the future time periodn＝Pan+PbnIn which P isnIndicating the predicted demand, P, for a particular type of measuring appliancean、PbnRepresenting the respective predicted demand of the two algorithms for a certain type of metering appliance;
in this embodiment, relevant data is collected, and the above-described algorithm is used to verify the prediction result of the cycle demand of the measuring instrument, and the experimental results are shown in the following table:
TABLE 1 Performance test Condition for demand prediction Algorithm
Prediction value Actual value Prediction accuracy%
Type A 6026 5811 96.3
Type A 4396 4227 96.0
Type B 5141 4929 95.7
Type D 2666 2506 93.6
Type E 6320 6030 95.2
As can be seen from Table 1, the forecasting accuracy of the measuring instrument demand forecasting method constructed by the invention is basically over 95%, and experiments show that the method has a high algorithm training speed, and the forecasting accuracy of the algorithm are improved on the basis of not consuming excessive computing resources and computing time.
D: according to the predicted demand of the metering devices of various models, the current stock quantity, the quantity to be detected, the quantity to be delivered and the like can be combined to make a replenishment strategy of the stock, so that a purchasing plan of the metering devices in a period is formed, and scientific support is provided for optimizing the stock management of electric power materials and reducing the enterprise cost.

Claims (9)

1. A method for predicting a demand of a measuring instrument, comprising the steps of:
step 1: the data acquisition unit acquires historical data of installation quantity of the measuring instruments of various models and external data of various influencing factors in corresponding periods, the data processor calculates chi-square values of two variables, and if the chi-square values are smaller than 1, the step 2 is executed; if the chi-square value is greater than or equal to 1, executing the step 3;
step 2: forecasting by adopting an SRU + RBF algorithm, forecasting the demand of the metering appliance by using historical data of the installation quantity/demand of the 'industry expansion new installation, transformation rotation' installation type metering appliance in the current period and data of the influence factors of the installation quantity of the installation type metering appliance, and outputting a forecasting result;
and step 3: and predicting by adopting an SRU + Bayesian network algorithm, acquiring the attribute data of the metering appliance with faults in the power field in the past year, processing to obtain high-correlation characteristics/attributes of the faults of the metering appliance, acquiring the attribute data of the metering appliance in the power field according to the characteristics/attributes to predict the demand of the metering appliance, and outputting a prediction result.
2. The demand forecasting method for measuring instruments as claimed in claim 1, wherein the SRU + RBF algorithm in step 2 predicts demand of measuring instruments of modified rotation and industry expansion new installation type, and calculates demand quantity P of measuring instruments of a certain modelan(ii) a The specific demand forecasting steps are as follows:
a1: collecting historical order data of a measuring instrument, and preprocessing the data;
a1.1: acquiring historical orders of 'transformation rotation' and 'business expansion new installation' installation type metering devices in an MDS system or an SG186 system, acquiring historical installation quantity of each type of metering device in each week/month/quarter, and forming a historical data matrix by taking different types as row directions and different week/month/quarter directions as column directions;
a1.2: in the historical data matrix, each column of data forms a column vector
Figure FDA0003399997350000011
The minimum value data and the maximum value data of each row respectively form a column vector
Figure FDA0003399997350000012
Carrying out normalization pretreatment on the data in the matrix, and normalizing the original data to [0, 1%]Get each column vector of the normalized data
Figure FDA0003399997350000013
The normalization formula is as follows:
Figure FDA0003399997350000014
the normalized column(Vector)
Figure FDA0003399997350000021
Sequentially splicing to form a normalized historical data matrix X';
a2: performing an SRU algorithm on the historical installation quantity time sequence data to obtain the preliminary prediction demand of the metering device;
a2.1: the first 4/5 columns of data of the historical data matrix X 'are used as a training set X'1With the remaining data as test set X'2；
A2.2: the method comprises the following steps of building an SRU network, setting the number of network layers to be 3, setting the number of SRU neurons in the first layer to be 8, then, setting the number of the neurons in the second layer to be decreased progressively according to integral multiple of 2, selecting a network initialization method to be uniform distribution initialization, setting activation functions to be a tanh function and a Sigmoid function, and setting a function expression to be:
Figure FDA0003399997350000022
Figure FDA0003399997350000023
the number of neural iterations is 100, the batch processing parameter is 8, the number of neuron jitters is 5, and the training set X 'is brought'1Training the whole network, and performing parameter optimization by using an Adam optimization algorithm to obtain a parameter matrix W, Wf、Wr、bf、br；
A2.3: input test set X'2For the current time input data XtCalculating the forgetting information degree f of the networktThe calculation formula is as follows:
ft＝σ(WfXt+bf) (4)
a2.4: determining information to be updated
Figure FDA0003399997350000024
A2.5: determining a degree r of update informationt＝σ(WrXt+br)；
A2.6: according to the cell state C at the last momentt-1Degree of forgetting information ftInformation required to be updated
Figure FDA0003399997350000025
Calculating the current newly formed cell state CtThe calculation formula is as follows:
Figure FDA0003399997350000026
a2.7: according to the current cell state CtDegree of update rtCurrent input XtCalculating the output h at the current momenttThe calculation formula is as follows:
ht＝rttanh(ct)+(1-rt)Xt (6)
thus, the internal information transfer of the SRU is only for the current time input XtParticipating in operation, outputting h at the current momenttIs not dependent on the previous time output ht-1The structure enables the SRU to complete independent parallel operation;
a2.8: obtaining the initial demand prediction results of metering devices of different models through an SRU algorithm: y is1,Y2,Y3,...；
A3: analyzing and acquiring external influence factor information of the demand of the transformation rotation and business expansion new installation type metering device, and forecasting the final demand P of the installation type metering device by using a RBF network algorithm in combination with the preliminary demand forecasting result obtained in the step A2.8an。
3. The method for predicting the demand of a measuring instrument according to claim 2, wherein the final demand P in step a3anThe prediction steps are as follows:
a3.1: the demand sources of the metering appliances with installation types of 'transformation rotation' and 'industry new installation' are mainly as follows: the method comprises the following steps of low-voltage non-resident new installation, low-voltage batch new installation, high-voltage new installation, household division, temporary power utilization, low-voltage non-resident capacity increase, low-voltage capacity increase, household combination and periodic alternate execution; the external influencing factors are mainly as follows: seasonal influence factors, regional GDP growth rate, building completion area, power consumption prediction data, service cycle of a measuring instrument, cost price of the measuring instrument and household application data of individual households;
a3.2: acquiring external influence factor data, and combining demand data preliminarily predicted by an SRU algorithm of a certain type of metering appliance to form a data set X ═ X1,x2,...,xnTaking the samples as training samples, and randomly selecting r different samples from the training samples as initial centers;
a3.3: carrying out normalization processing on sample data in the data set, wherein the formula is as follows:
Figure FDA0003399997350000031
in the formula (7), xiSample value representing a training sample in data set X, min (X) representing the smallest sample value in data set X, max (X) representing the largest sample value in data set X, X'iNormalizing the processed sample value of a certain sample in the data set X;
a3.4: calculating the distance between the sample and the center, and finding out the minimum distance dminThe formula is as follows:
dmin＝min||X-ci(k)|| (8)
in the formula (8), ci(k) Representing the center of the currently obtained training sample by computing each training sample X in the data set XiAnd training sample center ci(k) And obtaining the smallest difference among them, obtaining the smallest distance d of the sample from the centermin；
A3.5: using averaging
Figure FDA0003399997350000041
Center of adjustment ciJudging whether to continue according to whether all the training samples are learned and the central distribution does not change any more, if the two items are not met, the k value is plus 1, returning to the step A3.4 to recalculate dminAnd further adjust ciUntil the training samples are learned and the central distribution is not changed any more, obtaining a final RBF network center;
a3.6: according to the formula
Figure FDA0003399997350000042
Determining a normalization constant delta of a Gaussian function;
a3.7: according to the RBF neural network structure of the input layer node number 8, the output layer node number 1 and the hidden layer node number 8, the activation function of the hidden layer is a Gaussian function, namely the output function of the hidden layer node
Figure FDA0003399997350000043
The output function can be calculated on the basis of the step A3.2 to the step A3.6;
a3.8: calculating the weight w from each hidden layer node to each output layer node by using the least square principleijOn the basis of step a37, the output layer is a weighted sum of output functions of nodes of the hidden layer:
Figure FDA0003399997350000044
obtaining the required quantity P of the measuring instrument of the model on the installation type of' transformation rotation and business expansion new installationaj＝yj；
A3.9: repeating the steps A3.2-A3.8 to obtain the required quantity P of various types of metering devices of the installation type of' transformation rotation, industry expansion new installationan。
4. The demand forecasting method for the metering device, as claimed in claim 1, wherein the SRU + bayesian network algorithm predicts in step 3, and the specific steps for forecasting the demand of the "breakdown rush repair" installation type metering device are as follows:
b1: acquiring attribute data of a measuring instrument which has failed from a power field: acquiring success rate, online rate, use time, specification and model, manufacturers, production batches, communication flow, communication protocols, channel types, times of uploading important items, times of clock out-of-tolerance and addresses;
b2: preprocessing the attribute data acquired in the step B1;
b2.1: data cleaning is carried out on the data collected in the step B1, error data and repeated data are deleted, and missing data are supplemented;
b2.2: based on the step B2.1, the chi-square test is adopted to mainly carry out correlation analysis on the nominal attributes of the specification model, the production manager, the production batch and the communication protocol and the faults of the metering device, and the correlation between the numerical attributes of the acquisition success rate, the online rate, the service time and the communication flow and the faults of the metering device is mainly considered by calculating the Pearson correlation coefficient, so that the attributes related to the faults of the metering device are obtained, wherein the correlation includes the following steps: acquiring success rate, online rate, specification and model, service time, manufacturer, production batch, times of uploading important items and times of time synchronization out-of-tolerance;
b3: judging the influence degree of each attribute on the fault of the metering device by adopting a reliefF algorithm, thereby extracting high-correlation characteristic data of the fault of the metering device; the method comprises the following specific steps:
b3.1: based on step B22, the attribute data is normalized according to the following formula:
Figure FDA0003399997350000051
in the formula (10), X' is normalized data, XminFor the minimum value of each attribute data, XmaxA maximum value for each attribute data;
b3.2: constructing a training set { (x'1,y1),(x′2,y2),...,(x′n,yn) Get the tuple x randomly selectedi；
B3.3: at tuple xiFinding k adjacent neighbors x in similar samplesi,hThen k neighbor x are found out from the heterogeneous samplei,m；
B3.4: calculating a sample R and a neighboring sample HjThe difference diff (A, R, H) over the attribute Aj) The formula is as follows:
Figure FDA0003399997350000052
in formula (11), R < A >]Represents the value of the sample R on the attribute A, Hj[A]Represents a neighbor sample HjThe value of the attribute A is represented by the following equation (11), where the attribute A is of continuous type, the difference diff (A, R, H)j) Passing through type
Figure FDA0003399997350000061
The calculation is performed if the attribute A is discrete and the sample R and the neighboring sample HjIf the values on the attribute A are equal, the difference diff (A, R, H)j) 0; if the attribute A is discrete, and the sample R and the neighboring sample HjIf the values on the attribute A are not equal, the difference diff (A, R, H)j)＝1；
B3.5: calculating the influence weight w (A) of the attribute A on the fault of the metering device, wherein the formula is as follows:
Figure FDA0003399997350000062
h in formula (12)jhRepresenting similar neighbor samples, HjmRepresenting heterogeneous neighbor samples, both homogeneous and heterogeneous neighbor samples being k neighbor samples obtained by m times of random sampling among homogeneous/heterogeneous samples of random sample R,
Figure FDA0003399997350000063
and
Figure FDA0003399997350000064
representing k homogeneous/heterogeneous neighborsAccumulating the value difference of the sample and the random sample R on the attribute A;
b3.6: repeating the steps B3.4 and B3.5 to obtain the influence weights w corresponding to all the attributes, performing normalization processing on the influence weights w, sorting the influence weights w, setting a threshold value T, and if w (A) > T, extracting corresponding attribute data as high-correlation characteristic data of the faults of the metering device;
b3.7: finally, selecting acquisition success rate, online rate, service time, manufacturers, specification models, production batches and communication flow as high-correlation characteristic data of the faults of the metering devices;
b4: predicting the stable state of the metering appliance based on the high-correlation characteristic data obtained in the step B3.7;
b5: based on the steady state of the data collected by the measuring instrument and the prediction result of the steady state of the data transmitted by the transmission network obtained in the step B4, and the characteristic data in the step B3.7: establishing a Bayesian network by mainly using time, manufacturers and specifications;
b6: repeating the steps B1-B5 to obtain the fault prediction result of the measuring instrument and the specification model and the number of the measuring instruments with faults, counting the number of the faults according to different models, and calculating the demand P of the 'fault first-aid repair' installation type measuring instrumentbn。
5. The method for predicting the demand of a measuring instrument according to claim 4, wherein the step B4 predicts the steady state of the measuring instrument by the following specific steps:
b4.1: obtaining the time sequence related features in the high correlation features: acquiring success rate and online rate, and forming two rows of time sequence data;
b4.2: and taking the two rows of time sequence data as input data of the SRU algorithm, executing the operation step of obtaining the preliminary prediction demand of the metering appliance by the SRU algorithm, and respectively obtaining the prediction results of the SRU algorithm on the stable state of the data collected by the metering appliance and the stable state of the data transmitted by the network.
6. The demand forecasting method for a measuring instrument according to claim 5, wherein the specific steps of constructing the Bayesian network in step B5 are as follows:
b5.1: and B4.2, based on the prediction results of the stable state of the data collected by the measuring instrument and the stable state of the data transmitted by the transmission network obtained in the step B, and the characteristic data in the step B3.7: mainly using time, manufacturers and specification models as nodes, establishing an undirected graph containing all the nodes, wherein the nodes in the graph represent characteristic data, and edges among the nodes represent the relation among different characteristics;
b5.2: adopting a grading search method to construct and iteratively optimize a Bayesian network topology structure, and defining a network structure evaluation function as follows:
ScoreBDe(N|D)＝lnP(N)P(D|N)＝lnP(N)+lnP(D|N) (13)
in the formula (13), N is a network structure, and D is training data;
b5.3: inputting node data X ═ V into undirected graph1,V2,...,VnDefining the maximum father node number mu, the node order rho and a data set D, and initializing a father node set;
b5.4: adding each node variable into a father node set in sequence, calculating corresponding network scores, comparing the network scores with old scores, if the new scores are larger than the old scores and the maximum father node number is not reached, taking the node variable as a father node, and adding corresponding edges;
b5.5: combining the variables of other nodes, repeating the step B5.4, and combining domain expert knowledge to adjust to obtain a Bayesian network topological structure;
b5.6: and B4.2, based on the prediction results of the stable state of the data collected by the measuring instrument and the stable state of the data transmitted by the transmission network obtained in the step B, and the characteristic data in the step B3.7: determining the use time, the manufacturer and the specification model: data tuple set D ═ x1,x2,...,xn) The category set Y ═ {0,1}, where Y ═ 0 indicates that the meter has not failed, and Y ═ 1 indicates that the meter has failed;
b5.7: determining relevance of edges in undirected graph
Figure FDA0003399997350000081
Wherein the maximum likelihood estimation learning is adopted to obtain
Figure FDA0003399997350000082
Value of (a) corresponds to YiTraining a Bayesian network according to the conditional probability table.
7. A meter demand prediction system, comprising:
the data acquisition module receives sample data of the faults of the metering device, performs data preprocessing, and extracts high correlation characteristics related to the faults of the metering device;
the system comprises a fault first-aid repair installation type demand prediction algorithm building module, an SRU algorithm, a Bayesian network topology structure building and iterative optimization algorithm building module, a Bayesian network topology structure prediction module, a Bayesian network topology structure optimization module, a fault first-aid repair installation type demand prediction module, a transmission network data prediction module and a fault first-aid repair installation type demand prediction module, wherein the SRU algorithm is used for predicting the stable state of the terminal acquisition data and the stable state of the transmission network data according to the attribute characteristics, and the fault first-aid repair installation type demand prediction module is used for building and iteratively optimizing the Bayesian network topology structure by a scoring search method based on other high-correlation attribute characteristics and the stable state prediction result; based on the Bayesian network topological structure, parameter learning is carried out on the Bayesian network structure, and a terminal fault prediction algorithm is trained;
the prediction module predicts the installation type requirement of the first-aid repair based on the fault prediction algorithm;
the data acquisition module is used for acquiring historical order data of the industrial expansion new installation type electric energy meter and external influence factor information data related to business characteristics, and performing data preprocessing;
the algorithm construction module for predicting the demand of the business expansion new installation and transformation alternate installation type is used for primarily predicting the demand by utilizing an SRU algorithm according to the processed historical time sequence data to obtain the respective primary demand of the metering appliances of multiple models, and training the RBF neural network structure based on the primary demand data and the information of each external influence factor required by the business expansion new installation and transformation alternate installation type metering appliances to obtain the prediction algorithm for predicting the demand of the business expansion new installation and transformation alternate installation type metering appliances;
the system comprises a prediction module for predicting the demand of industry expansion new installation and transformation alternate installation type metering devices, wherein the demand of the industry expansion new installation and transformation alternate installation type metering devices is predicted based on a demand prediction algorithm of the industry expansion new installation and transformation alternate installation type metering devices;
and the counting and counting module counts the demand of each type of metering appliance obtained by the prediction module for the demand of the installation type of the fault first-aid repair, the prediction module for the demand of the installation type of the service expansion and transformation alternate installation type according to different types, and calculates the demand of the metering appliances of different types in a certain period of time in the future.
8. A computing device comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the method for predicting a demand of a metrology tool when executing the program.
9. A storage medium having stored thereon a computer program which, when executed by a processor, implements the method of predicting a demand of a metrology appliance.
CN202111495653.8A 2021-12-08 2021-12-08 Measuring instrument demand prediction method, system, computing device and storage medium Pending CN114169763A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202111495653.8A CN114169763A (en) 2021-12-08 2021-12-08 Measuring instrument demand prediction method, system, computing device and storage medium

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202111495653.8A CN114169763A (en) 2021-12-08 2021-12-08 Measuring instrument demand prediction method, system, computing device and storage medium

Publications (1)

Publication Number Publication Date
CN114169763A true CN114169763A (en) 2022-03-11

Family

ID=80484667

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202111495653.8A Pending CN114169763A (en) 2021-12-08 2021-12-08 Measuring instrument demand prediction method, system, computing device and storage medium

Country Status (1)

Country Link
CN (1) CN114169763A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116070781A (en) * 2023-03-06 2023-05-05 南方电网数字电网研究院有限公司 Electric energy metering equipment demand prediction method and device and computer equipment

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN116070781A (en) * 2023-03-06 2023-05-05 南方电网数字电网研究院有限公司 Electric energy metering equipment demand prediction method and device and computer equipment

Similar Documents

Publication Publication Date Title
CN109784806B (en) Supply chain control method, system and storage medium
Tang et al. Machine learning approach to uncovering residential energy consumption patterns based on socioeconomic and smart meter data
CN111783953A (en) 24-point power load value 7-day prediction method based on optimized LSTM network
WO2019200480A1 (en) Method and system for model auto-selection using an ensemble of machine learning models
CN111612275B (en) Method and device for predicting load of regional user
Dudas et al. Integration of data mining and multi-objective optimisation for decision support in production systems development
CN112734135B (en) Power load prediction method, intelligent terminal and computer readable storage medium
Gajowniczek et al. Simulation study on clustering approaches for short-term electricity forecasting
CN112734128A (en) 7-day power load peak value prediction method based on optimized RBF
CN112381673B (en) Park electricity utilization information analysis method and device based on digital twin
Wang et al. Short-term industrial load forecasting based on ensemble hidden Markov model
CN116596044B (en) Power generation load prediction model training method and device based on multi-source data
CN114861788A (en) Load abnormity detection method and system based on DBSCAN clustering
CN110826237A (en) Bayesian belief network-based wind power equipment reliability analysis method and device
CN114817571B (en) Dynamic knowledge graph-based achievement quoted quantity prediction method, medium and equipment
Hatami-Marbini et al. The role of multiplier bounds in fuzzy data envelopment analysis
Shi et al. Handling uncertainty in financial decision making: a clustering estimation of distribution algorithm with simplified simulation
CN114169763A (en) Measuring instrument demand prediction method, system, computing device and storage medium
CN114091776A (en) K-means-based multi-branch AGCNN short-term power load prediction method
CN117060408A (en) New energy power generation prediction method and system
May et al. Multi-variate time-series for time constraint adherence prediction in complex job shops
CN116470491A (en) Photovoltaic power probability prediction method and system based on copula function
CN115115119A (en) OA-GRU short-term power load prediction method based on grey correlation
CN113762591A (en) Short-term electric quantity prediction method and system based on GRU and multi-core SVM counterstudy
Sunny et al. A Comprehensive Review of the Load Forecasting Techniques Using Single and Hybrid Predictive Models

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