CN116050615A - Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation - Google Patents

Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation Download PDF

Info

Publication number
CN116050615A
CN116050615A CN202310032596.2A CN202310032596A CN116050615A CN 116050615 A CN116050615 A CN 116050615A CN 202310032596 A CN202310032596 A CN 202310032596A CN 116050615 A CN116050615 A CN 116050615A
Authority
CN
China
Prior art keywords
data
chaotic mapping
output
memristor
formula
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
CN202310032596.2A
Other languages
Chinese (zh)
Inventor
盛春阳
安昊
卢晓
王海霞
张治国
聂君
宋诗斌
孙巧巧
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.)
Shandong University of Science and Technology
Original Assignee
Shandong University of Science and Technology
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 Shandong University of Science and Technology filed Critical Shandong University of Science and Technology
Priority to CN202310032596.2A priority Critical patent/CN116050615A/en
Publication of CN116050615A publication Critical patent/CN116050615A/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
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F30/00—Computer-aided design [CAD]
    • G06F30/20—Design optimisation, verification or simulation
    • G06F30/27—Design optimisation, verification or simulation using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N7/00—Computing arrangements based on specific mathematical models
    • G06N7/08—Computing arrangements based on specific mathematical models using chaos models or non-linear system models
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
    • G06Q50/04—Manufacturing
    • Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
    • Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
    • Y02P—CLIMATE CHANGE MITIGATION TECHNOLOGIES IN THE PRODUCTION OR PROCESSING OF GOODS
    • Y02P90/00—Enabling technologies with a potential contribution to greenhouse gas [GHG] emissions mitigation
    • Y02P90/30—Computing systems specially adapted for manufacturing

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Evolutionary Computation (AREA)
  • Human Resources & Organizations (AREA)
  • Marketing (AREA)
  • Artificial Intelligence (AREA)
  • Tourism & Hospitality (AREA)
  • Software Systems (AREA)
  • General Business, Economics & Management (AREA)
  • General Engineering & Computer Science (AREA)
  • Operations Research (AREA)
  • Geometry (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Primary Health Care (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Medical Informatics (AREA)
  • Game Theory and Decision Science (AREA)
  • Computer Hardware Design (AREA)
  • Manufacturing & Machinery (AREA)
  • Development Economics (AREA)
  • Nonlinear Science (AREA)
  • Algebra (AREA)
  • Computational Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Mathematical Analysis (AREA)
  • Mathematical Optimization (AREA)
  • Pure & Applied Mathematics (AREA)
  • Computing Systems (AREA)
  • Mathematical Physics (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention discloses a flow prediction method of a byproduct gas system of a steel enterprise based on hyperchaotic reserve tank integrated calculation, which takes different generated and consumed user flow data of the existing byproduct gas system of the steel enterprise site as experimental objects, firstly, dividing the original flow data of a certain user at equal intervals according to a sliding time window to obtain a data sample; secondly, selecting a proper memristor chaotic mapping system according to the characteristics of the data sample, and identifying parameters of the system to enable the system to have hyperchaotic characteristics; thirdly, designing an integrated calculation network model of the reserve pool, calculating the output state of the reserve pool based on the memristive chaotic mapping system, and simplifying the state; finally, identifying weight parameters of the network model based on a Bayesian linear regression method to complete prediction; the invention effectively identifies the dynamic contained in the byproduct gas flow data of the iron and steel enterprises, accurately predicts the flow of the byproduct gas system, and further assists the on-site dispatching expert to carry out gas balance adjustment work.

Description

Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation
Technical Field
The invention belongs to the technical field of information, and particularly relates to a flow prediction method for a byproduct gas system of a steel enterprise based on hyperchaotic reserve tank integrated calculation.
Background
The national 'double carbon' strategy advocates a green, environment-friendly and low-carbon life style, and is imperative for iron and steel enterprises to save energy, reduce emission and reduce consumption. The byproduct gas generated in the steel production process has the characteristics of large volume and insufficient utilization, and whether the byproduct gas can be reasonably utilized directly influences the energy saving and consumption reduction effects of enterprises or not, so that the optimized dispatching of a byproduct gas system is very necessary. The byproduct gas is an important secondary energy source generated in iron making, coking and steelmaking processes of iron and steel enterprises, is an important component part of an energy system of the iron and steel enterprises, and because the generation of the byproduct gas is closely related to the smelting process and the consumption of the byproduct gas is closely related to the generation processes of hot rolling, cold rolling and the like, the byproduct gas system has the characteristics of multiple situations, coupling, nonlinearity and uncertainty, and the establishment of an optimal scheduling model of the byproduct gas system is extremely difficult. The operation rule of each generation and consumption link of the byproduct gas system is known and mastered, the internal operation dynamics of each generation and consumption link is excavated, the flow of each generation and consumption link is accurately predicted, and the method has important value, and can effectively assist a dispatcher to make a dispatching plan.
At present, a great deal of research on identification and prediction of a byproduct gas flow data model of an iron and steel enterprise mainly regards real-time monitoring data of byproduct gas as a time sequence, and then a data-driven model is established by a method based on the time sequence prediction, and common methods such as a neural network, a support vector machine, gaussian process regression and the like are adopted. However, these methods often suffer from the following disadvantages: (1) The method has a good identification effect on data with a quasi-periodic operation rule, and tends to have effect deviation on internal dynamic complex data; (2) Even if the parameters are regulated, the method is only suitable for a certain data sequence or a certain class of data sequence, and the generalization capability is weak; (3) Industrial data are generally noisy, and the above-described methods have insufficient capability to cope with noise.
In the current practical application, different methods are often selected according to different data characteristics, such as a machine learning method with cycle-like characteristics, for model identification and prediction of byproduct gas system data of iron and steel enterprises, and a simple model is established for prediction by combining non-periodic data with factors such as a production plan, a maintenance plan or a technical improvement project. The method has the following defects: (1) Although the model is easier to build, the reference value for optimizing the scheduling is relatively low, and the scheduling personnel can rely on manual experience to perform accurate pre-judgment; (2) The dynamic mining of the inside of complex data is insufficient, and even if accurate production plans and maintenance plans exist in most cases, the byproduct gas operation process is influenced by other uncertainties, so that the mining of the uncertainties is more valuable.
Disclosure of Invention
Aiming at the problems in the prior art, the invention provides a flow prediction method for a byproduct gas system of a steel enterprise based on hyperchaotic reserve tank integrated calculation, which has reasonable design, overcomes the defects in the prior art and has good effect.
In order to achieve the aim of the invention, the invention adopts the following technical scheme:
a method for predicting the flow of byproduct gas systems of iron and steel enterprises based on hyperchaotic reserve pool integrated calculation comprises the following steps:
s1, reading required byproduct gas flow operation data from a field real-time database of an iron and steel enterprise, preprocessing the real-time data, intercepting original data into data fragments according to an equal time interval by utilizing a sliding time window method, and constructing a data sample required by model identification;
s2, selecting a proper memristor chaotic mapping system according to the characteristics of the data sample;
s3, taking the memristor chaotic mapping system as a reserve tank unit, establishing a reserve tank integrated calculation network model, calculating the output state of the reserve tank unit, and simplifying the output state;
s4, integrating and calculating the input of the network model output layer by taking the simplified output state in the S3 as a storage pool, taking a target predicted value as output, and identifying weight parameters of the network model based on a Bayesian linear regression method;
s5, predicting and verifying the established network model by using the test sample.
Further, in step S1, flow operation data of a user of the byproduct gas system is read from a steel enterprise site real-time database, and the data sequence is recorded as { u } * (1),u * (2),…,u * (k) …, the raw data is normalized, the normalization formula is as follows:
Figure BDA0004047488960000021
where u (k) is the data point corresponding to u * (k) Is used for the normalization of the data in the (c),
Figure BDA0004047488960000022
maximum data of all data points, < ->
Figure BDA0004047488960000023
The data with the minimum value in all data points is taken;
all the original data are mapped between [0,1] through normalization, and then byproduct gas system flow operation data are divided into data fragments with finite time intervals by utilizing a sliding time window, wherein the data fragments are shown in a formula (2):
Figure BDA0004047488960000024
wherein L is the length of the input samples, n is the number of samples, s i ＝{u i ,t i -i-th sample in the set of samples; taking 70% of data samples in the sample data set as training samples, taking the rest 30% as test samples, and recording the number of samples in the training sample set as n 1 The number of samples in the test sample set is n 2 。
Further, in step S2, for given user flow data, a memristor chaotic mapping system suitable for the user characteristics is selected by an experimental method as a pool unit for generating a pool output state;
four reported memristor chaotic mapping systems are taken as research objects, and mathematical models of the four memristor chaotic mapping systems are as follows:
Figure BDA0004047488960000025
wherein ,
Figure BDA0004047488960000031
Figure BDA0004047488960000032
and />
Figure BDA0004047488960000033
Representing the internal state of the memristor, with variable F being the input layer parameter, b 1 Is an intermediate layer parameter;
Figure BDA0004047488960000034
/>
wherein ,
Figure BDA0004047488960000035
and />
Figure BDA0004047488960000036
Representing internal state variables of memristors, a 2 To input layer parameters, b 2 Is an intermediate layer parameter;
Figure BDA0004047488960000037
wherein ,
Figure BDA0004047488960000038
and />
Figure BDA0004047488960000039
Representing internal state variables of the memristor, wherein lambda is an input layer parameter, and a is an intermediate layer parameter;
Figure BDA00040474889600000310
wherein ,
Figure BDA00040474889600000311
and />
Figure BDA00040474889600000312
Representing internal state variables of the memristor, wherein μ is an input layer parameter, and b is an intermediate layer parameter;
firstly, setting parameter value ranges of the four memristive chaotic mapping systems according to experience, and then substituting experimental data into repeated circulation to determine optimal parameters so that each memristive chaotic mapping system has hyperchaotic characteristics; secondly, inputting data of a user training sample to a memristor chaotic mapping system shown in a formula (3), enabling the memristor chaotic mapping system to generate a reserve pool state, taking the output state of the reserve pool as input, taking an output weight parameter as an unknown quantity, taking a target predicted value as output, and then utilizing a least square method to identify the output weight parameter so as to establish a prediction model; thirdly, the memristor chaotic mapping system shown in the test data sample test formula (3) is utilized to adapt to the user flow data, a prediction error is selected as an adaptability evaluation standard, a root mean square error is selected as a prediction error, and a root mean square error expression is:
Figure BDA00040474889600000313
wherein ,n2 To test the number of sample data, y i As a predicted value, t i Is a target value;
and finally, respectively using the testing method for the memristor chaotic mapping systems shown in the formulas (4), (5) and (6), comparing the performance differences of the four types of memristor chaotic mapping systems, and finally determining the memristor chaotic mapping system suitable for the user data.
Further, in step S3, the suitable memristive chaotic mapping system selected in step S2 is used as a pool unit, when the system shown in formula (6) is most suitable as a pool calculation unit, the parameter μ in (6) is used as a variable input parameter of the system, and b is used as a system parameter; because one memristor chaotic mapping system has only 1 input inlet mu, and the dimension of an input sample is L, L data in an output sample are required to be transmitted to the L memristor chaotic mapping systems, a reserve pool integrated calculation network model is established, and L groups of reserve pool neuron states are generated by the L input sample data;
since mu is only in a certain value range mu min <μ<μ max In the system, the hyper-chaos characteristic is generated, so that the following formula is adopted to map the input data into the value range:
μ(i)＝μ min +(μ max -μ min )u(i) (8)
wherein u (i) is the ith data point in the input sample normalized in the step 1, and μ (i) is the data transmitted to the memristor chaotic mapping system;
assuming that the dimension of the memristor chaotic mapping system is n×n, the output state of the finally obtained storage pool is an l×n matrix, and l×n states are obtained, as shown in the following formula:
Figure BDA0004047488960000041
and selecting an effective reserve pool state every M states, wherein the M size is determined according to the dimension N of the memristor chaotic mapping system and the dimension L of an input sample, ensuring that each group of N state values keep M states, M is less than N/2, and finally keeping the number of the reserve pool effective states as L multiplied by M, taking the states as input, taking a target predicted value as output, and solving a network output weight by using a Bayesian linear regression-based method.
Further, in step S4, a training sample set is given
Figure BDA0004047488960000042
wherein si ＝{u i ,t i According to step S3, based on input u i The calculated reservoir output status is noted as X i Output weight of the reserve pool integrated calculation network model is marked as W out Target output t i The description is as follows:
t i ＝W out X i +ε i (10)
wherein ,εi Is observed noise, using zero-mean Gaussian white noise
Figure BDA0004047488960000043
To describe it;
solving for W using Bayesian principle out Will be to W out Solution to (c) translates to W out Solving for posterior distribution, i.e. solving for p (W out |S)：
Figure BDA0004047488960000044
Wherein p (S) = ≡p (s|w) out )p(W out )dW out Is a normalization factor;
the probability distribution of the target output is described as:
Figure BDA0004047488960000045
wherein ,
Figure BDA0004047488960000046
is a super parameter related to the output distribution; if the data samples are independent from each other, the likelihood function is expressed as:
Figure BDA0004047488960000051
for a priori distribution of weight parameters, the description is made using gaussian distribution with little knowledge of the prior information:
Figure BDA0004047488960000052
wherein α is a hyper-parameter related to the variance of the weight distribution;
the posterior distribution of the weight parameters is described as follows, according to equations (11), (13) and (14):
Figure BDA0004047488960000053
solving the optimal weight parameters
Figure BDA0004047488960000054
That is, solving W when formula (15) is maximized out Solving the maximum value of the formula (15) by adopting a maximum likelihood estimation method, and enabling:
Figure BDA0004047488960000055
the problem is simplified into a solution to the minimum value of the formula (16), the posterior distribution of the weight parameters is assumed to be Gaussian distribution, and Taylor series expansion is utilized to obtain:
Figure BDA0004047488960000056
wherein A is M (W out ) Is a Hessian matrix of (2); the posterior distribution of the weight parameters is expressed as:
Figure BDA0004047488960000057
wherein ,
Figure BDA0004047488960000058
the super parameters alpha and beta of two positions in the formula (16) are utilized to calculate the functional relation between the weight parameter and the super parameters by utilizing the Bayes principle, and then the gradient descent method is adopted to estimate W out,α and β.
Further, in step S5, a set of test samples is given
Figure BDA0004047488960000059
wherein ,si '＝{u i ′,t i ' firstly, calculating the output state X of the reserve tank based on a memristive chaotic mapping system i ' then, using Bayesian reasoning to obtain the distribution of the output of the integrated calculation network model:
p(t i ′|X i ′,S')＝∫p(t i ′|X i ′,W out )p(W out |S')dW out (19)
substituting the formula (12) and the formula (18) into the formula (19) yields:
Figure BDA0004047488960000061
the formula (20) is arranged into a Gaussian distribution:
Figure BDA0004047488960000062
wherein ,yMP To predict the mean, i.e. the predicted value,
Figure BDA0004047488960000063
to predict the variance of the distribution.
The invention has the following beneficial effects:
when the flow of the byproduct gas system is predicted, the fact that the flow data of the byproduct gas contains complex dynamics is fully considered, the complex dynamics hidden in the data is extracted by adopting the chaotic mapping system, different chaotic mapping systems are selected according to different data characteristics, and the generalization capability of the model is improved; the industrial data is dynamic and complex, so that an input sample with higher dimensionality is always required to be constructed, if the chaotic mapping system is used as a calculation unit in the storage pool, a large number of states are generated in the storage pool, and difficulty is brought to the subsequent network weight parameter identification; because the industrial data generally has noise, the invention adopts the parameter identification method based on Bayesian linear regression to effectively improve the identification effect of the weight parameters of the network in order to reduce the noise influence. The invention can fully utilize the flow data of the existing byproduct gas system, effectively predict the change condition of the flow of the byproduct gas system after the current time, and thereby provide online decision support for balance adjustment of the byproduct gas system.
Drawings
FIG. 1 is a flow prediction flow chart of a byproduct gas system provided by the invention;
FIG. 2 is a schematic diagram of an integrated computing network model of a hyperchaotic reserve tank designed in accordance with the invention;
FIG. 3 is a graph of gas flow data trend of a byproduct gas system of a certain iron and steel enterprise;
wherein, (a) a gas generation flow data trend graph and (b) a gas consumption flow data trend graph;
Detailed Description
The following description of the embodiments of the invention will be given with reference to the accompanying drawings and examples:
the gas flow data of the auxiliary plant gas systems of different iron and steel enterprises show different data dynamic states, as shown in fig. 3 (a) and (b), flow data monitoring curves of gas generation and consumption of the byproduct gas systems of certain iron and steel enterprises can be seen, the data have obvious difference, and other user flow data also have own uniqueness. The present invention will be described with reference to the blast furnace gas generation flow rate data shown in fig. 3. From the trend of the data in the graph, the implicit dynamic in the data is very complex, and the data cannot be effectively predicted by simply relying on the blast furnace ironmaking work plan and the manual experience of a dispatching expert, so that a scientific byproduct gas flow data prediction model must be established to predict the generation and consumption flow change trend of the byproduct gas. According to the method flow shown in fig. 1, the specific implementation steps of the invention are as follows:
s1, preparing data, and obtaining n data samples;
reading flow operation data of a user of a byproduct gas system from a field real-time database of an iron and steel enterprise, wherein a data sequence is recorded as { u } * (1),u * (2),…,u * (k) …, the raw data is normalized, the normalization formula is as follows:
Figure BDA0004047488960000071
where u (k) is the data point corresponding to u * (k) Is used for the normalization of the data in the (c),
Figure BDA0004047488960000072
maximum data of all data points, < ->
Figure BDA0004047488960000073
Get from all data pointsData with the smallest value;
all the original data are mapped between [0,1] through normalization, and then byproduct gas system flow operation data are divided into data fragments with finite time intervals by utilizing a sliding time window, wherein the data fragments are shown in a formula (2):
Figure BDA0004047488960000074
wherein L is the length of the input samples, n is the number of samples, s i ＝{u i ,t i -i-th sample in the set of samples; taking 70% of data samples in the sample data set as training samples, taking the rest 30% as test samples, and recording the number of samples in the training sample set as n 1 The number of samples in the test sample set is n 2 。
S2, selecting a proper memristor chaotic mapping system according to the characteristics of the data sample;
for given user flow data, selecting a memristor chaotic mapping system suitable for the user characteristics as a reserve pool unit by an experimental method, and generating a reserve pool output state;
four reported memristor chaotic mapping systems are taken as research objects, and mathematical models of the four memristor chaotic mapping systems are as follows:
Figure BDA0004047488960000075
wherein ,
Figure BDA0004047488960000076
Figure BDA0004047488960000077
and />
Figure BDA0004047488960000078
Representing the internal state of the memristor, with variable F being the input layer parameter, b 1 Is an intermediate layer parameter;
Figure BDA0004047488960000079
/>
wherein ,
Figure BDA00040474889600000710
and />
Figure BDA00040474889600000711
Representing internal state variables of memristors, a 2 To input layer parameters, b 2 Is an intermediate layer parameter;
Figure BDA00040474889600000712
wherein ,
Figure BDA0004047488960000081
and />
Figure BDA0004047488960000082
Representing internal state variables of the memristor, wherein lambda is an input layer parameter, and a is an intermediate layer parameter;
Figure BDA0004047488960000083
wherein ,
Figure BDA0004047488960000084
and />
Figure BDA0004047488960000085
Representing internal state variables of the memristor, wherein μ is an input layer parameter, and b is an intermediate layer parameter;
firstly, setting parameter value ranges of the four memristive chaotic mapping systems according to experience, and then substituting experimental data into repeated circulation to determine optimal parameters so that each memristive chaotic mapping system has hyperchaotic characteristics; secondly, inputting data of a user training sample to a memristor chaotic mapping system shown in a formula (3), enabling the memristor chaotic mapping system to generate a reserve pool state, taking the output state of the reserve pool as input, taking an output weight parameter as an unknown quantity, taking a target predicted value as output, and then utilizing a least square method to identify the output weight parameter so as to establish a prediction model; thirdly, the memristor chaotic mapping system shown in the test data sample test formula (3) is utilized to test the adaptability of the memristor chaotic mapping system to the user flow data, a prediction error is selected as an adaptability evaluation standard, a root mean square error is selected as a prediction error, and a root mean square error RMSE expression is:
Figure BDA0004047488960000086
wherein ,n2 To test the number of sample data, y i As a predicted value, t i Is a target value;
and finally, respectively using the testing method for the memristor chaotic mapping systems shown in the formulas (4), (5) and (6), comparing the performance differences of the four types of memristor chaotic mapping systems, and finally determining the memristor chaotic mapping system suitable for the user data. Remarks: the prediction model of the part is simpler, is only used for testing the chaotic mapping system suitable for the user, has low precision, and can distinguish the mining capability of different chaotic mapping systems on the group of data.
S3, establishing a reserve pool integrated computing network model and outputting a state;
taking the memristive chaotic mapping system selected to be suitable in the step S2 as a reserve pool unit, and taking the parameter mu in the formula (6) as a variable input parameter of the system and b as a system parameter on the assumption that the system shown in the formula (6) is selected as the most suitable reserve pool calculation unit for the flow data of the user; because one memristor chaotic mapping system has only 1 input inlet mu, and the dimension of an input sample is L, L data in an output sample is required to be transmitted to the L memristor chaotic mapping systems, a reserve pool integrated calculation network model is established, and as shown in fig. 2, L groups of reserve pool neuron states are generated by the L input sample data;
since mu is only in a certain value range mu min <μ<μ max In the system, the hyper-chaos characteristic is generated, so that the following formula is adopted to map the input data into the value range:
μ(i)＝μ min +(μ max -μ min )u(i) (8)
wherein u (i) is the ith data point in the input sample normalized in the step 1, and μ (i) is the data transmitted to the memristor chaotic mapping system; since the value range of u (i) is [0,1]Therefore, the value of μ (i) must fall within μ min <μ<μ max Within the value range, the memristive chaotic system can be ensured to have the hyperchaotic characteristic.
Assuming that the dimension of the memristive chaotic mapping system is n×n, each memristive chaotic system generates N states, the dimension of an input sample is L, and the output state of the finally obtained storage pool is an l×n matrix, namely l×n states are obtained. If the data dynamic is complex, the value of L is often larger, so that the integral value of L multiplied by N is larger, the output weight to be regulated is more, and the system may have certain redundancy, therefore, the invention proposes a strategy for simplifying the output state of the reserve pool. Let l×n pool states obtained be in the form shown in the following matrix:
Figure BDA0004047488960000091
selecting states of a reserve pool by adopting an M-step skip method, namely selecting an effective reserve pool state every M states, determining M according to the dimension N of a memristor chaotic mapping system and the dimension L of an input sample, ensuring that M states are reserved for each group of N state values, wherein the number of the reserved effective states of the reserve pool is L multiplied by M finally when M is less than N/2, taking the states as input, taking a target predicted value as output, and solving a network output weight by using a Bayesian linear regression-based method.
S4, identifying weight parameters of the integrated calculation network model based on a Bayesian linear regression method;
given training sample set
Figure BDA0004047488960000092
wherein si ＝{u i ,t i According to step S3, based on input u i The calculated reservoir output status is noted as X i Output weight of the reserve pool integrated calculation network model is marked as W out Target output t i The description is as follows:
t i ＝W out X i +ε i (10)
wherein ,εi Is observed noise, using zero-mean Gaussian white noise
Figure BDA0004047488960000093
To describe it;
the purpose of the parameter solving is to calculate the unknown parameter W out The present invention uses bayesian principles to solve for unknown parameters, taking into account the fact that the data is noisy and the system of linear regression equations may be ambiguous. Will be to W out Solution to (c) translates to W out Solution of posterior distribution of (a), i.e. p (W) out S), further using bayesian principles:
Figure BDA0004047488960000094
wherein p (S) = ≡p (s|w) out )p(W out )dW out Is a normalization factor; in this way, the problem can be translated into a solution to the product of the likelihood function and the a priori distribution of the output weight parameters.
For equation (10), the output weight W of the unknown parameter out And noise epsilon i Variance of (1) at a given pool state X i In the case of (a), the probability distribution of the target output is described as:
Figure BDA0004047488960000101
wherein ,
Figure BDA0004047488960000102
is a super parameter related to the output distribution; if the data samples are independent from each other, the likelihood function is expressed as:
Figure BDA0004047488960000103
for a priori distribution of weight parameters, the description is made using gaussian distribution with little knowledge of the prior information:
Figure BDA0004047488960000104
/>
wherein α is a hyper-parameter related to the variance of the weight distribution;
the posterior distribution of the weight parameters is described as follows, according to equations (11), (13) and (14):
Figure BDA0004047488960000105
solving the optimal weight parameters
Figure BDA0004047488960000106
That is, solving W when formula (15) is maximized out Solving the maximum value of the formula (15) by adopting a maximum likelihood estimation method, and enabling:
Figure BDA0004047488960000107
the problem is reduced to a solution to the minimum of equation (16), and for further simplification, the posterior distribution of the weight parameters is directly assumed to be gaussian distribution, and the Taylor series expansion is utilized to obtain:
Figure BDA0004047488960000108
wherein A is M (W out ) Is a Hessian matrix of (2); the posterior distribution of the weight parameters is expressed as:
Figure BDA0004047488960000109
wherein ,
Figure BDA00040474889600001010
however, since the equation (16) contains the super-parameters α and β at two positions, the Bayesian principle is further utilized to calculate the functional relationship between the weight parameter and the super-parameters, and then the gradient descent method is adopted to estimate W out,α and β.
S5, predicting and verifying the established network model by using the test sample.
Given a test sample set
Figure BDA00040474889600001011
wherein ,si '＝{u i ′,t i ' firstly, calculating the output state X of the reserve tank based on a memristive chaotic mapping system i ' then, using Bayesian reasoning to obtain the distribution of the output of the integrated calculation network model:
p(t i ′|X i ′,S')＝∫p(t i ′|X i ′,W out )p(W out |S')dW out (19)
wherein ,p(ti ′|X i ′,W out) and p(Wout The expression of S) can be obtained from the expression (12) and the expression (18), and substituted into the expression (19):
Figure BDA0004047488960000111
the arrangement of formula (20) into a gaussian distribution is:
Figure BDA0004047488960000112
wherein ,yMP To predict the mean, i.e. the predicted value,
Figure BDA0004047488960000113
to predict the variance of the distribution.
So far, the prediction of the gas flow of the byproduct gas system can be completed, and then the on-site dispatching expert is assisted to carry out gas balance adjustment work.
It should be understood that the above description is not intended to limit the invention to the particular embodiments disclosed, but to limit the invention to the particular embodiments disclosed, and that the invention is not limited to the particular embodiments disclosed, but is intended to cover modifications, adaptations, additions and alternatives falling within the spirit and scope of the invention.

Claims (6)

1. The method for predicting the flow of the byproduct gas system of the iron and steel enterprise based on the hyperchaotic reserve pool integrated calculation is characterized by comprising the following steps of:
s1, reading required byproduct gas flow operation data from a field real-time database of an iron and steel enterprise, preprocessing the real-time data, intercepting original data into data fragments according to an equal time interval by utilizing a sliding time window method, and constructing a data sample required by model identification;
s2, selecting a proper memristor chaotic mapping system according to the characteristics of the data sample;
s3, taking the memristor chaotic mapping system as a reserve tank unit, establishing a reserve tank integrated calculation network model, calculating the output state of the reserve tank unit, and simplifying the output state;
s4, integrating and calculating the input of the network model output layer by taking the simplified output state in the S3 as a storage pool, taking a target predicted value as output, and identifying weight parameters of the network model based on a Bayesian linear regression method;
s5, predicting and verifying the established network model by using the test sample.
2. A according to claim 1A method for predicting flow of byproduct gas system of iron and steel enterprises based on hyperchaotic reserve pool integrated calculation is characterized in that in step S1, flow operation data of a certain user of byproduct gas system is read from an on-site real-time database of iron and steel enterprises, and the data sequence is recorded as { u } * (1),u * (2),…,u * (k) …, the raw data is normalized, the normalization formula is as follows:
Figure FDA0004047488950000011
where u (k) is the data point corresponding to u * (k) Is used for the normalization of the data in the (c),
Figure FDA0004047488950000012
the data with the largest value in all data points is taken,
Figure FDA0004047488950000013
the data with the minimum value in all data points is taken;
all the original data are mapped between [0,1] through normalization, and then byproduct gas system flow operation data are divided into data fragments with finite time intervals by utilizing a sliding time window, wherein the data fragments are shown in a formula (2):
Figure FDA0004047488950000014
wherein L is the length of the input samples, n is the number of samples, s i ＝{u i ,t i -i-th sample in the set of samples; taking 70% of data samples in the sample data set as training samples, taking the rest 30% as test samples, and recording the number of samples in the training sample set as n 1 The number of samples in the test sample set is n 2 。
3. The method for predicting the byproduct gas system flow of the iron and steel enterprise based on the hyperchaotic reserve tank integrated calculation, which is characterized in that in the step S2, a memristive chaotic mapping system suitable for the characteristics of a given user is selected as a reserve tank unit by an experimental method for generating the output state of the reserve tank according to the given user flow data;
four reported memristor chaotic mapping systems are taken as research objects, and mathematical models of the four memristor chaotic mapping systems are as follows:
Figure FDA0004047488950000021
wherein ,
Figure FDA0004047488950000022
Figure FDA0004047488950000023
and />
Figure FDA0004047488950000024
Representing the internal state of the memristor, with variable F being the input layer parameter, b 1 Is an intermediate layer parameter; />
Figure FDA0004047488950000025
wherein ,
Figure FDA0004047488950000026
and />
Figure FDA0004047488950000027
Representing internal state variables of memristors, a 2 To input layer parameters, b 2 Is an intermediate layer parameter;
Figure FDA0004047488950000028
wherein ,
Figure FDA0004047488950000029
and />
Figure FDA00040474889500000210
Representing internal state variables of the memristor, wherein lambda is an input layer parameter, and a is an intermediate layer parameter;
Figure FDA00040474889500000211
wherein ,
Figure FDA00040474889500000212
and />
Figure FDA00040474889500000213
Representing internal state variables of the memristor, wherein μ is an input layer parameter, and b is an intermediate layer parameter;
firstly, setting parameter value ranges of the four memristive chaotic mapping systems according to experience, and then substituting experimental data into repeated circulation to determine optimal parameters so that each memristive chaotic mapping system has hyperchaotic characteristics; secondly, inputting data of a user training sample to a memristor chaotic mapping system shown in a formula (3), enabling the memristor chaotic mapping system to generate a reserve pool state, taking the output state of the reserve pool as input, taking an output weight parameter as an unknown quantity, taking a target predicted value as output, and then utilizing a least square method to identify the output weight parameter so as to establish a prediction model; thirdly, the memristor chaotic mapping system shown in the test data sample test formula (3) is utilized to adapt to the user flow data, a prediction error is selected as an adaptability evaluation standard, a root mean square error is selected as a prediction error, and a root mean square error expression is:
Figure FDA00040474889500000214
wherein ,n2 To test the number of sample data, y i As a predicted value, t i Is a target value;
and finally, respectively using the testing method for the memristor chaotic mapping systems shown in the formulas (4), (5) and (6), comparing the performance differences of the four types of memristor chaotic mapping systems, and finally determining the memristor chaotic mapping system suitable for the user data.
4. The method for predicting the byproduct gas system flow of the iron and steel enterprise based on the hyperchaotic reserve tank integrated calculation according to claim 3, wherein in the step S3, the proper memristive chaotic mapping system selected in the step S2 is used as a reserve tank unit, when the system shown in the formula (6) is most suitable as a reserve tank calculation unit, the parameter mu in the formula (6) is used as a variable input parameter of the system, and b is used as a system parameter; because one memristor chaotic mapping system has only 1 input inlet, and the dimension of an input sample is L, L data in an output sample are required to be transmitted to the L memristor chaotic mapping systems, a reserve pool integrated calculation network model is established, and L groups of reserve pool neuron states are generated by the L input sample data;
since mu is only in a certain value range mu min <μ<μ max In the system, the hyper-chaos characteristic is generated, so that the following formula is adopted to map the input data into the value range:
μ(i)＝μ min +(μ max -μ min )u(i) (8)
wherein u (i) is the ith data point in the input sample normalized in the step 1, and μ (i) is the data transmitted to the memristor chaotic mapping system;
assuming that the dimension of the memristor chaotic mapping system is n×n, the output state of the finally obtained storage pool is an l×n matrix, and l×n states are obtained, as shown in the following formula:
Figure FDA0004047488950000031
and selecting an effective reserve pool state every M states, wherein the M size is determined according to the dimension N of the memristor chaotic mapping system and the dimension L of an input sample, ensuring that each group of N state values keep M states, M is less than N/2, and finally keeping the number of the reserve pool effective states as L multiplied by M, taking the states as input, taking a target predicted value as output, and solving a network output weight by using a Bayesian linear regression-based method.
5. The method for predicting the byproduct gas system flow of the iron and steel enterprise based on the hyperchaotic reserve tank integrated calculation as set forth in claim 4, wherein in step S4, a training sample set s= { S is given 1 ,s 2 ,…,s n1}, wherein si ＝{u i ,t i According to step S3, based on input u i The calculated reservoir output status is noted as X i Output weight of the reserve pool integrated calculation network model is marked as W out Target output t i The description is as follows:
t i ＝W out X i +ε i (10)
wherein ,εi Is observed noise, using zero-mean Gaussian white noise
Figure FDA0004047488950000032
To describe it;
solving for W using Bayesian principle out Will be to W out Solution to (c) translates to W out Solving for posterior distribution, i.e. solving for p (W out |S)：
Figure FDA0004047488950000033
Wherein p (S) = ≡p (s|w) out )p(W out )dW out Is a normalization factor;
the probability distribution of the target output is described as:
Figure FDA0004047488950000041
wherein ,
Figure FDA0004047488950000042
is a super parameter related to the output distribution; if the data samples are independent from each other, the likelihood function is expressed as:
Figure FDA0004047488950000043
for a priori distribution of weight parameters, the description is made using gaussian distribution with little knowledge of the prior information:
Figure FDA0004047488950000044
wherein α is a hyper-parameter related to the variance of the weight distribution;
the posterior distribution of the weight parameters is described as follows, according to equations (11), (13) and (14):
Figure FDA0004047488950000045
solving the optimal weight parameters
Figure FDA0004047488950000046
That is, solving W when formula (15) is maximized out Solving the maximum value of the formula (15) by adopting a maximum likelihood estimation method, and enabling:
Figure FDA0004047488950000047
the posterior distribution of the weight parameters is assumed to be Gaussian distribution, and Taylor series expansion is utilized to obtain:
Figure FDA0004047488950000048
wherein A is M (W out ) Is a Hessian matrix of (2); the posterior distribution of the weight parameters is expressed as:
Figure FDA0004047488950000049
wherein ,
Figure FDA00040474889500000410
the super parameters alpha and beta of two positions in the formula (16) are utilized to calculate the functional relation between the weight parameter and the super parameters by utilizing the Bayes principle, and then the gradient descent method is adopted to estimate W out,α and β.
6. The method for predicting the byproduct gas system flow of the iron and steel enterprise based on the hyperchaotic reserve tank integrated calculation according to claim 5, wherein in step S5, a set of test samples is given
Figure FDA0004047488950000051
wherein ,si ′＝{u i ′,t i ' firstly, calculating the output state X of the reserve tank based on a memristive chaotic mapping system i ' then, using Bayesian reasoning to obtain the distribution of the output of the integrated calculation network model:
p(t i ′|X i ′,S′)＝∫p(t i ′|X i ′,W out )p(W out |S′)dW out (19)
substituting the formula (12) and the formula (18) into the formula (19) yields:
Figure FDA0004047488950000052
the formula (20) is arranged into a Gaussian distribution:
Figure FDA0004047488950000053
wherein ,yMP To predict the mean, i.e. the predicted value,
Figure FDA0004047488950000054
to predict the variance of the distribution. />
CN202310032596.2A 2023-01-10 2023-01-10 Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation Pending CN116050615A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202310032596.2A CN116050615A (en) 2023-01-10 2023-01-10 Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202310032596.2A CN116050615A (en) 2023-01-10 2023-01-10 Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation

Publications (1)

Publication Number Publication Date
CN116050615A true CN116050615A (en) 2023-05-02

Family

ID=86118408

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202310032596.2A Pending CN116050615A (en) 2023-01-10 2023-01-10 Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation

Country Status (1)

Country Link
CN (1) CN116050615A (en)

Similar Documents

Publication Publication Date Title
CN111353656B (en) Steel enterprise oxygen load prediction method based on production plan
CN108009667A (en) A kind of energy demand total amount and structure prediction system
CN109242188B (en) Long-term interval prediction and structure learning method for steel gas system
CN107239849A (en) One kind is based on improved exponential smoothing gray model Methods of electric load forecasting
CN106650944B (en) Knowledge-based converter gas scheduling method for metallurgical enterprises
US20200219027A1 (en) Knowledge transfer-based modeling method for blast furnace gas scheduling systems
CN110598958B (en) Ladle hierarchical management analysis method and system
CN107368125A (en) A kind of blast furnace temperature control system and method based on CBR Yu the parallel mixed inferences of RBR
CN114611845B (en) Method and device for predicting carbon emission, electronic device, and medium
CN110428053A (en) A kind of steam produces the dynamic prediction method of consumption
CN110097929A (en) A kind of blast furnace molten iron silicon content on-line prediction method
CN114841410A (en) Heat exchange station load prediction method and system based on combination strategy
CN104408317A (en) Metallurgy enterprise gas flow interval predicting method based on Bootstrap echo state network integration
CN115688989A (en) Carbon footprint prediction method and terminal based on Stacking ensemble learning
Matino et al. Application of echo state neural networks to forecast blast furnace gas production: pave the way to off-gas optimized management
CN115310685A (en) Hydroelectric generating set full life cycle cost calculation method based on parameter correction
CN111680452A (en) Power grid engineering accurate investment decision simulation method based on full-factor data mining
Han et al. Stackelberg–Nash game approach for constrained robust optimization with fuzzy variables
CN116050615A (en) Flow prediction method for byproduct gas system of iron and steel enterprise based on hyperchaotic reserve tank integrated calculation
Yang et al. China's Energy Demand Forecasting Based on the Hybrid PSO-LSSVR Model.
CN117113086A (en) Energy storage unit load prediction method, system, electronic equipment and medium
CN116384254A (en) Error correction method and system for predicting molten steel temperature of LF refining furnace
CN116823008A (en) Park energy utilization efficiency evaluation method, system, equipment and storage medium
CN115081687A (en) Knowledge-guided multi-source information fusion-based blast furnace gas prediction method
CN114004065A (en) Transformer substation engineering multi-objective optimization method based on intelligent algorithm and environmental constraints

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