CN105320991A - Factor-based steel enterprise process energy consumption prediction method - Google Patents

Abstract

The invention provides a factor-based steel enterprise process energy consumption prediction method. The method comprises the following steps: the step 1) preparing data: obtaining steel enterprise process energy consumption history actual data from a real-time database, and importing planning data; the step 2) displaying the planning data, wherein the planning data serves as an influence factor; the step 3) obtaining history energy consumption actual data according to the history actual data, and determining calculation sample data; the step 4) obtaining fixed unit consumption value according to the calculation sample; and the step 5) carrying out predication by adopting a BP neural network mode or a support vector machine mode. The method can predicate process medium energy consumption and ton steel comprehensive energy consumption variation trends in a future period of time accurately, so that scheduling personnel are allowed to obtain ton steel comprehensive energy consumption with reference to the process medium energy consumption variation trend, complexity of the original method is reduced, accuracy of the ton steel comprehensive energy consumption is improved, and time consumption period is reduced.

Description

Based on iron and steel enterprise's process energy consumption Forecasting Methodology of factor

Technical field

The present invention relates to metallurgical automation field, particularly, relate to the iron and steel enterprise's process energy consumption Forecasting Methodology based on factor.

Background technology

Iron and steel enterprise's energy consumption prediction and balance studies adopt the accurately advanced trend of Forecasting Methodology to energy demand to provide correct estimation, according to utility consumption history data, consider the industrial structure of each operation, energy resource consumption feature, short-term is carried out to process energy consumption and Enterprise Integrated energy consumption, medium and long term is predicted, population equilibrium and adjustment are carried out to the use of enterprise energy, the feature of the further investigation energy consumption of iron and steel enterprises and mechanism, form comprehensive Energy Consumption Evaluation, analysis system, for energy management optimization provides effective decision support, for energy cost budget provides foundation accurately and reliably, realize reasonably utilizing high efficiency of energy.

Current way is, all secondary factories below general headquarters monthly report and submit energy environment protection portion the energy consumption budget of our factory, Ministry of Energy carries out overall balance and adjustment according to the situation of each factory, result after adjustment is assigned to secondary factory on the one hand, form energy consumption budget on the one hand and offer treasurer's department, whole process is manual carries out, and monthly three days consuming time, efficiency was low.

Summary of the invention

For the defect of above-mentioned prior art, the technical problem to be solved in the present invention is a kind of iron and steel enterprise's process energy consumption Forecasting Methodology based on factor, each secondary operation medium energy-output ratio can be predicted more exactly, obtain a ton steel comprehensive energy consumption, thus reduce the complexity of original way and improve the accuracy of prediction ton steel comprehensive energy consumption, shorten the cycle consuming time.

According to a kind of iron and steel enterprise's process energy consumption Forecasting Methodology based on factor provided by the invention, comprise the steps:

Step 1: prepare data, obtain iron and steel enterprise's process energy consumption history real data from real-time data base, and from importing planning data;

Step 2: show planning data, wherein, using planning data as influence factor;

Step 3: according to history real data, obtains history energy consumption real data, and determines to calculate sample data;

Step 4: fix unit consumption value according to calculating sample acquisition;

Step 5: adopt BP neural network fashion or support vector machine mode to predict;

I () adopts BP neural network fashion to predict, specifically comprise the steps:

Step 501: first initialization BP network structure, sets the neuron number of the input layer of BP network, hidden layer, output layer;

Step 502: according to neural network structure determination total number of particles, this total number of particles is the summation of all weights and threshold;

Step 503: the fitness calculating each particle in all particles;

Step 504: the fitness of more each particle compares with the fitness of the desired positions lived through, if better, then using the fitness of described particle as current desired positions P _i(t);

Step 505: the fitness of desired positions that the fitness of each particle and the overall situation experience is compared, if better, then using the fitness of described particle as current overall desired positions P _g(t) ∈ { P ₀(t), P ₁(t) ..., P _i(t) }, wherein, P _it () represents the desired positions that the i-th+1 particle is current;

Step 506: the position X upgrading each particle according to formula (1) and formula (2) _ijand speed v _ij;

v _ij(t+1)＝v _ij(t)+c ₁*rand()*(p _ij(t)-x _ij(t))+c ₂*rand()*(p _gj(t)-x _ij(t))(1)

x _ij(t+1)＝x _ij(t)+v _ij(t+1)(2)

Wherein: subscript " j " represents the jth dimension of particle, and " i " represents particle i, and t represents t generation, c ₁, c ₂for aceleration pulse, usually between 0 ~ 2, rand () is that (0,1) is uniformly distributed separate random function;

V _ij(t+1) represent particle i t+1 for time speed, v _ij(t) represent particle i t for time upgrade after speed, p _ij(t) represent particle i t for time desired positions, x _ij(t) represent particle i t for time position, x _ij(t+1) represent particle i t+1 for time position, p _gj(t) represent extreme point particle t for time desired positions, g represents extreme point particle subscript

Can obtain from the above-mentioned particle evolution equation be made up of formula (1) and formula (2), c ₁particle is regulated to fly to the step-length in self desired positions direction, c ₂regulate the step-length that particle flies to overall desired positions;

Step 507: expect threshold value if reach default fitness or reach a default maximum algebraically, then the search performed by step 504 to step 506 stops, and exports global optimum position, otherwise, return step 504 and continue to perform;

(ii) adopt support vector machine mode to predict, specifically comprise the steps:

Step 5A: the particle populations of initialization population PSO;

Step 5B: sample data is computed weighted and obtains new feature value, the new feature value formed after utilizing weighting carries out modeling, and utilize the detection model obtained to carry out crosscheck, obtain crosscheck value, this crosscheck value is with regard to fitness value;

Step 5C: according to calculating the fitness value that obtains, uses the speed of population PSO and location updating formula to upgrade particle, upgrades individual history optimal location and ideal adaptation angle value simultaneously, and global optimum position and overall fitness value;

Step 5D: the operation repeating step 5B and step 5C, until meet the iterations or algorithm just stopping operation time that preset.

Preferably, in step 503, using training error precision E as fitness function to obtain fitness:

$E=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i^d-y_i)}^2$

Wherein, E represents fitness, and N represents sum, represent desired output, y _irepresent true output valve.

Compared with prior art, the present invention has following beneficial effect:

The present invention can predict operation medium energy consumption and the ton steel comprehensive energy consumption variation tendency of following a period of time more accurately, make dispatcher can with reference to operation medium energy consumption variation tendency, obtain a ton steel comprehensive energy consumption, thus reduce the complexity of original way and improve the accuracy of prediction ton steel comprehensive energy consumption, shorten the cycle consuming time.

The invention solves the technical matters how accurately obtaining process energy consumption in iron and steel manufacture field, what it adopted is based on the computing method of support vector machine or BP network and the technological means utilizing forecast model to realize, and can obtain the technique effect improving process energy consumption prediction and calculation speed thus.

Accompanying drawing explanation

By reading the detailed description done non-limiting example with reference to the following drawings, other features, objects and advantages of the present invention will become more obvious:

Fig. 1 is process energy consumption forecasting process.

Embodiment

Below in conjunction with specific embodiment, the present invention is described in detail.Following examples will contribute to those skilled in the art and understand the present invention further, but not limit the present invention in any form.It should be pointed out that to those skilled in the art, without departing from the inventive concept of the premise, some distortion and improvement can also be made.These all belong to protection scope of the present invention.

Step one: prepare data, obtain history real data from real-time data base, imports planning data by hand;

Step 2: show planning data and influence factor;

Step 3: based on history real data, uses BP network to build factor Model A;

Step 4: based on history real data, uses support vector machine to build factor Model B, specific as follows:

(1) in modeling process first using historical data as training sample, and sample data to be processed, to improve the precision of training result;

(2) by the conversion of kernel function, nonlinear problem is carried out linearization, be converted into the form solving quadratic programming problem;

(3) adopt mathematic programming methods to solve above problem, select support vector, obtain forecast model.

Step 5: step 3 or step 4 are predicted operation medium energy consumption according to the modeling system of step 2 influence factor, thus by calculating a ton steel comprehensive energy consumption.

The present invention can predict operation medium energy consumption and the ton steel comprehensive energy consumption variation tendency of following a period of time more accurately, energy centre managerial personnel can according to the variation tendency of operation medium energy consumption and ton steel comprehensive energy consumption, in conjunction with the full factory ton steel comprehensive energy consumption target that the beginning of the year formulates, regulate and control each secondary factory operation medium energy resource consumption, the Energy Sources Equilibrium reaching full factory utilizes; Realize Energy Sources Equilibrium, reduce energy dissipation, improve energy utilization rate.Compared with manually, improve precision, shorten the cycle consuming time.

Below the present invention is more specifically described.

Embodiment 1: plum steel steel making working procedure energy consumption is predicted

Fig. 1 is the process energy consumption prediction process flow diagram of the embodiment of the present invention 1.

Embodiment

Shown in Figure 1, described in embodiments of the invention 1, iron and steel enterprise's steel making working procedure energy consumption Forecasting Methodology realizes in accordance with the following steps:

Step one, reads steel making working procedure energy consumption history real data and steel making working procedure actual output.By reading steel making working procedure energy consumption history real data and steel making working procedure actual output from the real-time data base in on-the-spot energy resource system.

Step 2, importing and needs the relevant steel making working procedure scheduled production data of prediction as influence factor, obtaining steel making working procedure scheduled production data in the mode by importing from outside.

Step 3, the steel making working procedure energy consumption history real data got according to step one and steel making working procedure actual output, selected moon number is as total sample number to manually, and wherein the steel making working procedure energy consumption history real data of the same moon and steel making working procedure actual output data are as a sample.

Step 4, according to the process energy consumption history real data of each month legitimate reading as sample, using the actual output of each month as particle, with the algorithm of the inside of BP neural network, by repetition training with prediction particle result as predicting the outcome, to legitimate reading and comparing of predicting the outcome, finally obtain searching out optimum solution in less solution space.

The method of specific implementation is:

A. first initialization BP network structure, sets the neuron number of the input layer of BP network, hidden layer, output layer;

B. according to neural network structure determination total number of particles, this total number of particles is the summation of all weights and threshold;

C. the fitness of each particle in all particles is calculated;

D. the fitness of more each particle compares with the fitness of desired positions lived through, if better, then using the fitness of described particle as current desired positions P _i(t);

E. the fitness of desired positions that the fitness of each particle and the overall situation experience is compared, if better, then using the fitness of described particle as current overall desired positions P _g(t) ∈ { P ₀(t), P ₁(t) ..., P _i(t) }, wherein, P _it () represents the desired positions that the i-th+1 particle is current;

F. the position X of each particle is upgraded according to formula (1) and formula (2) _ijand speed v _ij;

v _ij(t+1)＝v _ij(t)+c ₁*rand()*(p _ij(t)-x _ij(t))+c ₂*rand()*(p _gj(t)-x _ij(t))(1)

x _ij(t+1)＝x _ij(t)+v _ij(t+1)(2)

Wherein: subscript " j " represents the jth dimension of particle, and " i " represents particle i, and t represents t generation, c ₁, c ₂for aceleration pulse, usually between 0 ~ 2, rand () is that (0,1) is uniformly distributed separate random function v _ij(t+1) represent particle i t+1 for time speed, v _ij(t) represent particle i t for time upgrade after speed, p _ij(t) represent particle i t for time desired positions, x _ij(t) represent particle i t for time position, x _ij(t+1) represent particle i t+1 for time position, p _gj(t) represent extreme point particle t for time desired positions, g represents extreme point particle subscript;

Can obtain from the above-mentioned particle evolution equation be made up of formula (1) and formula (2), c ₁particle is regulated to fly to the step-length in self desired positions direction, c ₂regulate the step-length that particle flies to overall desired positions;

If G. reach default fitness to expect threshold value or reach a default maximum algebraically, then the search performed by step 504 to step 506 stops, and exports global optimum position, otherwise, return step 504 and continue to perform;

Step 5, utilizes the influence factor that step 4 analysis obtains, structure training sample set S.

$S=\{(X_j^{t-1},H_j^{t-1}),H_j^t),j=1\·\·\·N\}$

Wherein, with composition model input amendment, for model output sample, represent actual output, represent history process energy consumption, represent the current month process energy consumption of prediction.

Step 6, adopts BP algorithm to build steel making working procedure energy consumption forecast model according to training sample set S; Namely based on steel-making plan yield data, BP modeling method is adopted to set up the model embodying relation between steel making working procedure energy consumption and planning data, steel making working procedure energy consumption forecast model.

Concrete grammar is:

A. G-P algorithm is utilized to determine the Embedded dimensions m of each major influence factors;

B. BP neural network is utilized to obtain training sample set $S_i=\left\{(x_j,y_j)\right|j=\mathrm{1,2},\·\·\·,n_i-m_i\}:$

$\left[\begin{array}{c}{x}_1\\ x_2\\ .\\ .\\ .\\ x_{\mathrm{ni}-\mathrm{mi}}\end{array}\right]=\left[\begin{array}{cccc}x\left(1\right)& x\left(2\right)& ...& x\left(m_i\right)\\ x\left(2\right)& x\left(3\right)& ...& x(m_i+1)\\ .& .& .& .\\ .& .& .& .\\ .& .& .& .\\ x(n-m_i)& x(n_i-m_i+1)& ...& x(n_i-1)\end{array}\right],\left[\begin{array}{c}{y}_1\\ y_2\\ .\\ .\\ .\\ y_{\mathrm{ni}-\mathrm{mi}}\end{array}\right]=\left[\begin{array}{c}x(m_i+1)\\ x(m_i+2)\\ .\\ .\\ .\\ x\left(n_i\right)\end{array}\right];$

(wherein, x _j∈ R ^mirepresent the input of influence factor time series predicting model, y _j∈ R represents the output of model.)

Step 7, according to the steel making working procedure energy consumption forecast model that BP neural network algorithm obtains, with steel making working procedure scheduled production for influence factor, with actual achievement output and actual achievement process energy consumption value for sample, obtains the steel making working procedure energy consumption predicted value in scheduled production corresponding month.

The present invention is the field data based on iron and steel enterprise, can be applicable to the energy centre of iron and steel enterprise, alleviate the prediction work of the staff of energy centre, shorten consuming time, the energy service condition in faster more convenient grasp next month or in the future more months, knows in advance and arranges the outsourcing of the short energy.

Here applies of the present invention predicting the outcome:

Table 1 steel making working procedure predicts the outcome table

Note:

Table 2 data item implication

Above specific embodiments of the invention are described.It is to be appreciated that the present invention is not limited to above-mentioned particular implementation, those skilled in the art can make various distortion or amendment within the scope of the claims, and this does not affect flesh and blood of the present invention.

Claims (2)

1., based on iron and steel enterprise's process energy consumption Forecasting Methodology of factor, it is characterized in that, comprise the steps:

Step 1: prepare data, obtain iron and steel enterprise's process energy consumption history real data from real-time data base, and from importing planning data;

Step 2: show planning data, wherein, using planning data as influence factor;

Step 3: according to history real data, obtains history energy consumption real data, and determines to calculate sample data;

Step 4: fix unit consumption value according to calculating sample acquisition;

Step 5: adopt BP neural network fashion or support vector machine mode to predict;

I () adopts BP neural network fashion to predict, specifically comprise the steps:

Step 501: first initialization BP network structure, sets the neuron number of the input layer of BP network, hidden layer, output layer;

Step 502: according to neural network structure determination total number of particles, this total number of particles is the summation of all weights and threshold;

Step 503: the fitness calculating each particle in all particles;

Step 504: the fitness of more each particle compares with the fitness of the desired positions lived through, if better, then using the fitness of described particle as current desired positions P _i(t);

Step 505: the fitness of desired positions that the fitness of each particle and the overall situation experience is compared, if better, then using the fitness of described particle as current overall desired positions P _g(t) ∈ { P ₀(t), P ₁(t) ..., P _i(t) }, wherein, P _it () represents the desired positions that the i-th+1 particle is current;

Step 506: the position X upgrading each particle according to formula (1) and formula (2) _ijand speed v _ij;

v _ij(t+1)＝v _ij(t)+c ₁*rand()*(p _ij(t)-x _ij(t))+c ₂*rand()*(p _gj(t)-x _ij(t))(1)

x _ij(t+1)＝x _ij(t)+v _ij(t+1)(2)

Wherein: subscript " j " represents the jth dimension of particle, and " i " represents particle i, and t represents t generation, c ₁, c ₂for aceleration pulse, usually between 0 ~ 2, rand () is that (0,1) is uniformly distributed separate random function;

V _ij(t+1) represent particle i t+1 for time speed, v _ij(t) represent particle i t for time upgrade after speed, p _ij(t) represent particle i t for time desired positions, x _ij(t) represent particle i t for time position, x _ij(t+1) represent particle i t+1 for time position, p _gj(t) represent extreme point particle t for time desired positions, g represents extreme point particle subscript;

Can obtain from the above-mentioned particle evolution equation be made up of formula (1) and formula (2), c ₁particle is regulated to fly to the step-length in self desired positions direction, c ₂regulate the step-length that particle flies to overall desired positions;

Step 507: expect threshold value if reach default fitness or reach a default maximum algebraically, then the search performed by step 504 to step 506 stops, and exports global optimum position, otherwise, return step 504 and continue to perform;

(ii) adopt support vector machine mode to predict, specifically comprise the steps:

Step 5A: the particle populations of initialization population PSO;

Step 5B: sample data is computed weighted and obtains new feature value, the new feature value formed after utilizing weighting carries out modeling, and utilize the detection model obtained to carry out crosscheck, obtain crosscheck value, this crosscheck value is with regard to fitness value;

Step 5C: according to calculating the fitness value that obtains, uses the speed of population PSO and location updating formula to upgrade particle, upgrades individual history optimal location and ideal adaptation angle value simultaneously, and global optimum position and overall fitness value;

Step 5D: the operation repeating step 5B and step 5C, until meet the iterations or algorithm just stopping operation time that preset.

2. the iron and steel enterprise's process energy consumption Forecasting Methodology based on factor according to claim 1, is characterized in that, in step 503, using training error precision E as fitness function to obtain fitness:

$E=\frac{1}{N}\underset{i=1}{\overset{N}{\mathrm{\Σ}}}{(y_i^d-y_i)}^2$

Wherein, E represents fitness, and N represents sum, represent desired output, y _irepresent true output valve.