CN109242569A - A kind of molybdenum concentrate Long-term Market price analysis and prediction technique and system - Google Patents
A kind of molybdenum concentrate Long-term Market price analysis and prediction technique and system Download PDFInfo
- Publication number
- CN109242569A CN109242569A CN201811068367.1A CN201811068367A CN109242569A CN 109242569 A CN109242569 A CN 109242569A CN 201811068367 A CN201811068367 A CN 201811068367A CN 109242569 A CN109242569 A CN 109242569A
- Authority
- CN
- China
- Prior art keywords
- time series
- price
- molybdenum concentrate
- prediction
- molybdenum
- 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.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q30/00—Commerce
- G06Q30/02—Marketing; Price estimation or determination; Fundraising
- G06Q30/0201—Market modelling; Market analysis; Collecting market data
- G06Q30/0206—Price or cost determination based on market factors
Abstract
The present invention discloses a kind of analysis of the molybdenum concentrate market price and prediction technique and system, for analyzing and predicting the Long-term Market price of molybdenum concentrate, for molybdenum ore enterprise solve the problems, such as in long-term production decision foundation is provided, first, using molybdenum concentrate market price time series as object, carry out it is steady non-stationary, it is linear non-linear, complexity analyzed, obtain the substantive characteristics of molybdenum concentrate market price time series data；Then, on the basis of the substantive characteristics of molybdenum concentrate market price time series data analyzes result, molybdenum concentrate market price sequence is decomposed, selects different prediction models to carry out classification prediction to it according to the result of decomposition；Finally, the prediction result of classification prediction is integrated, final molybdenum concentrate market price forecasts result is obtained.The prediction technique is scientific and effective, precision of prediction is high, has universality, can be used for the market price forecasts of other similar products, is with a wide range of applications.
Description
Technical field
The invention belongs to system engineering and price expectation technical field, in particular to a kind of molybdenum concentrate Long-term Market price
Analysis and prediction technique and system.
Background technique
Molybdenum concentrate is the base stock of steel and iron industry production, is the indispensable resource of modern industrial production.Molybdenum ore is at me
The rich reserves of state are located at the first in the world for a long time, are the Dominant Mineral Resources in China, in the globalization strategy of China's mining industry
In consequence, economy and resource security to China have important influence.However for a long time, China pays attention to export
It earns foreign exchange, having ignored using the mineral resources of advantage influences and control world market, specifically shows as long-term high intensity
Exploitation, yield and export volume are significantly increased, so that the relation between supply and demand in market is unbalance, therefore, carry out to the molybdenum concentrate market price
Medium-term and long-term prediction is conducive to the adjustment of bargh's production plan and the formulation of National Mining resource macro policy, to improvement
The pattern of the molybdenum concentrate demand-supply relation and Mineral Resources in China in global mining is of great significance.
Currently, analysis and prediction to China's molybdenum concentrate market price, predominantly short-term price expectation and medium-term and long-term valence
Lattice trend analysis generally uses AR (p) model, exponential smoothing and artificial intelligence since the variation fluctuation of molybdenum concentrate short period price is smaller
The method of energy, which can be realized, more accurately to be predicted；However, for medium-term and long-term price, main analysis and prediction method is
The method of econometric model, such methods are analyzed by the influence factor to molybdenum concentrate price, inquire into it to molybdenum concentrate
The influence in market, the final market price trend for influencing result and being reflected as molybdenum concentrate.It is more comprehensive in the ratio that influence factor considers
When, econometric model method can obtain more accurate prediction result, but in the actual operation process, due to molybdenum concentrate
Market system is excessively huge, and Variable Factors in need of consideration are difficult to bring whole system, molybdenum concentrate market economy structure into
Equation is difficult to set up, and is become difficult so that carrying out prediction using the method for econometric model, therefore to the medium-term and long-term of molybdenum concentrate
Price realization is more accurately predicted highly difficult.
Summary of the invention
In order to overcome the disadvantages of the above prior art, the purpose of the present invention is to provide a kind of molybdenum concentrate Long-term Market valences
Case analysis and prediction technique and system, are analyzed and predicted, when by price as sample using molybdenum concentrate historical price
Between sequence substantive characteristics identified and analyzed and the decomposition to Time Series of Random Macro-price, it is long-term to excavate molybdenum concentrate Time Series of Random Macro-price
The potential information contained recycles different prediction models to predict respectively these information, prediction result is integrated,
It can be avoided using the various Variable Factors constructed required for econometric model method, and to Variable Factors and molybdenum concentrate city
The analysis of correlativity, simplifies the complexity of molybdenum concentrate Long-term Market price expectation between the price of field, meanwhile, relative to biography
The time series analysis and prediction technique of system, this method depth have excavated the generation system of molybdenum concentrate market price behind, according to
The characteristics of generation system, selects suitable prediction technique, can effectively improve the prediction essence of molybdenum concentrate Long-term Market price
Degree, it is with the obvious advantage.
To achieve the goals above, the technical solution adopted by the present invention is that:
A kind of molybdenum concentrate Long-term Market price analysis and prediction technique, by dividing molybdenum concentrate Time Series of Random Macro-price
The feature that molybdenum concentrate price generates system is excavated in analysis, using Time Series technology by these character separations, is then directed to this
A little features carry out classification prediction and integrate.
Therefore, the analysis and prediction technique are divided into three parts, and first part is molybdenum concentrate Time Series of Random Macro-price data
Analysis is the differentiation procedure that system is generated to its price；Second part is the decomposition of molybdenum concentrate Time Series of Random Macro-price data, i.e.,
It is the separation for generating system features to molybdenum concentrate price, obtains the intrinsic mode function of Time Series of Random Macro-price；Part III is valence
The classification of lattice time series intrinsic mode function is predicted and is integrated.
In the first portion, the stationarity, linear/non-linear, complexity of molybdenum concentrate Time Series of Random Macro-price are analyzed, is analyzed
Step are as follows:
A. using the method and unit root test method of molybdenum concentrate Time Series of Random Macro-price figure and autocorrelogram come qualitative and quantitative
Judge molybdenum concentrate Time Series of Random Macro-price stationarity；
B. on the basis of step a, suitable AR (p) linear model is selected to intend molybdenum concentrate Time Series of Random Macro-price
It closes, is then tested using BDS method of inspection to the independence of residual error, judge whether molybdenum concentrate price is generated by linear system；
C. using the complexity of the relevant dimension analysis molybdenum concentrate Time Series of Random Macro-price in fractal theory, molybdenum concentrate valence is judged
The complexity of lattice generation system.Relevant dimension calculation method uses G-P algorithm, calculates step are as follows:
C1. the lag period τ of molybdenum concentrate Time Series of Random Macro-price is determined using correlation analysis；
C2. with Embedded dimensions k to molybdenum concentrate Time Series of Random Macro-price x (n), n=1,2, when 3 ..., N, N are molybdenum concentrate price
Between sequence length, carry out phase space reconfiguration, phase space phase point after reconstruct is Xi(n)={ xi(n),xi(n+τ),…,xi(n+
(k-1) τ) }, i=1,2 ..., K, K are the sum of the phase point in the phase space of reconstruct, are calculated by formula K=N- (m-1) τ；
I indicates i-th of phase point in phase space；
C3. a radius of neighbourhood r is given, is calculated correlation integral C (k, r), calculation formula is as follows,
Wherein, θ is function # (x),
‖Xi-Xj‖ is the distance between phase space phase point；
C4. logarithm is taken to correlation integral C (k, r) and r respectively, the pass of molybdenum concentrate Time Series of Random Macro-price is asked using following equation
Join dimension D,
D. to step a, the result of b, c carry out comprehensive analysis, obtain molybdenum concentrate Time Series of Random Macro-price substantive characteristics and its
Corresponding price data generates the characteristics of system.
In the second portion, molybdenum concentrate Time Series of Random Macro-price x (n) is decomposed, the isolation used is EMD point
Solution, specific the step of including are as follows:
A. extreme point is taken to molybdenum concentrate Time Series of Random Macro-price x (n), structurally lower envelope line, seeks the mean value of upper and lower envelope
me；
B. former molybdenum concentrate Time Series of Random Macro-price x (n) is subtracted into mean value me, obtains a new time series h (n)；
C. repeat step a, b, until time series h (n) meet condition zero crossing number it is equal with extreme point number or
It is zero that maximum, which differs one and time series h (n) maximum and the mean value of the envelope up and down of minimum construction, is remembered at this time
Time series h (n) is imf；
D. imf is subtracted with molybdenum concentrate time series x (n), obtains new time series x1(n)；
E. by time series x1(n) repeat the above steps a, b, c, d, until cannot decompose again, thus obtains molybdenum essence
Mine all intrinsic mode function imf and a residual error function res.
Part III is the predicted portions of molybdenum concentrate Time Series of Random Macro-price, mainly in combination with above-mentioned analysis and decomposition result,
The intrinsic mode function imf that the suitable classification prediction model of building obtains decomposition is predicted, then to classification prediction result
The step of being integrated, specifically including are as follows:
A. the analysis for generating the evaluation result of system to molybdenum concentrate Time Series of Random Macro-price according to first part, by second part
It decomposes obtained intrinsic mode function imf and residual error res and is divided into two classes according to its timing diagram, one kind is simple stationary time sequence
Column, it is another kind of for complicated steady Nonlinear Time Series；
B. two class data step a obtained divide, and construct training set x (t), y (t) and forecast set x respectively1(t1)、
y1(t1), the prediction step of two class data is consistent；
C. to simple stationary time series ingredient, ARMA (p, q) model is constructed respectively and is predicted, the p value selected
It is determined with q value according to the auto-correlation and partial correlation figure of each time series, prediction steps specifically:
C1. the value of the parameter p, q of ARMA (p, q) model are determined using data set x (t)；
C2. ARMA (p, q) model is constructed, estimates the unknown parameter in model；
C3. the validity for examining ARMA (p, q) model, if do not passed through, return step b2；
C4. ARMA (p, q) model is tested, and constructs multiple model of fit, therefrom select optimal model.
D. to complicated steady Nonlinear Time Series ingredient, LSTM deep neural network is constructed respectively and is predicted,
The prediction steps of LSTM deep neural network specifically:
D1. the LSTM number of unit inputted according to the list entries tectonic network of training set, while determining output number；
D2. LSTM deep neural network model is constructed, and network is trained using training set y (t)；
D3. test set y is used1(t1) trained LSTM neural network is tested, and according to test result to prediction
Model optimizes, until reaching higher precision of prediction.
E. to step c, the prediction result of d is integrated using three layers of BP neural network, and it is shown that specific step is as follows:
E1. by molybdenum concentrate Time Series of Random Macro-price x (n), according to c, the prediction step in Step d is divided into training set and test
Collection；
E2. by c, input value of the prediction result as BP neural network in Step d, molybdenum concentrate Time Series of Random Macro-price is divided
Obtained training set is trained BP neural network as output valve；
E3. using trained BP neural network to c, the test set prediction result in Step d is integrated, and will be integrated
Obtained sequence is compared as final prediction result, by it with the test set of molybdenum concentrate Time Series of Random Macro-price.
The present invention also provides corresponding systems, comprising:
Molybdenum concentrate Time Series of Random Macro-price data analysis module analyzes molybdenum concentrate Time Series of Random Macro-price data, excavates molybdenum concentrate
The feature of price generation system；
Molybdenum concentrate Time Series of Random Macro-price data decomposing module is obtained using Time Series technology by the character separation
To the intrinsic mode function of Time Series of Random Macro-price；
Classification prediction and integration module, predict intrinsic mode function, and integrate to classification prediction result.
Wherein, the molybdenum concentrate Time Series of Random Macro-price data analysis module, the step of executing the first part.
The molybdenum concentrate Time Series of Random Macro-price data decomposing module, the step of executing the second part.
The step of classification is predicted and integration module, executes the Part III.
The above-mentioned technical method that the present invention uses has the effect that compared with traditional econometric model, of the invention
Requirement of the analysis of proposition with prediction technique, system to data is lower, reduces the acquisition difficulty of master data, but to data
Analysis it is more comprehensive；Compared with general Time Series Forecasting Methods, analysis proposed by the present invention and prediction technique, system clock synchronization
Between sequence generation system carried out the excavation and analysis of depth, and be classified prediction according to the feature of time series script,
It is thorough to consider, precision of prediction is higher.
Detailed description of the invention
Fig. 1 is the general flow chart of the price analysis of molybdenum concentrate Long-term Market and prediction technique in present example.
Fig. 2 is molybdenum concentrate market price time series in present example.
Fig. 3 is the autocorrelogram of molybdenum concentrate Time Series of Random Macro-price in present example.
Fig. 4 is the partial correlation figure of molybdenum concentrate Time Series of Random Macro-price in present example.
Fig. 5 is the structure chart of the single neuron in present example in LSTM deep neural network.
Specific embodiment
The present invention provides a kind of molybdenum concentrate Long-term Market price analysis and prediction techniques, below in conjunction with of the invention real
The attached drawing in example is applied, the technical method in the embodiment of the present invention is described.
Embodiment provides long-term Chinese market price analysis and prediction technique in the molybdenum concentrate of one kind 45%, is used to more
Technical method of the invention is clearly described, refering to what is shown in Fig. 1, analysis proposed by the present invention and prediction technique include following three
Part:
P1. first part is the analysis of molybdenum concentrate Time Series of Random Macro-price data, is the discrimination that system is generated to its price
Process；
P2. second part is the decomposition of molybdenum concentrate Time Series of Random Macro-price data, is to generate system spy to molybdenum concentrate price
The separation of sign obtains the intrinsic mode function of Time Series of Random Macro-price；
P3. Part III is the classification prediction of Time Series of Random Macro-price intrinsic mode function and integrates.
Above three part can be respectively an independent module, to form corresponding analysis and forecast system.
In the first portion, molybdenum concentrate market historical price data is collected, the time span of data is no less than 10 years, and will
It forms the Time Series of Random Macro-price as unit of the moon, as shown in Fig. 2, then analyzing the steady of molybdenum concentrate Time Series of Random Macro-price
Property, linear/non-linear, complexity, analytical procedure are as follows:
A. using the method and unit root test method of molybdenum concentrate Time Series of Random Macro-price figure and auto-correlation coefficient figure come qualitative and
The stationarity of quantitative judge molybdenum concentrate Time Series of Random Macro-price, specific steps are as follows:
A1. molybdenum concentrate Time Series of Random Macro-price figure is observed, it can be found that there is no a certain for the Time Series of Random Macro-price
Regular price fluctuates up and down, therefore can tentatively judge molybdenum concentrate Time Series of Random Macro-price for nonstationary time series；
The delay order of sequence is set as 30, finds out its auto-correlation coefficient, and make by a2. Time Series of Random Macro-price according to fig. 2
Autocorrelogram, as shown in figure 3, auto-correlation coefficient is not reduced to zero, in conjunction with step a1Judgement, can further judge sequence
It is classified as non-stationary Time Series of Random Macro-price；
A3. the unit root null hypothesis for designing molybdenum concentrate Time Series of Random Macro-price, when using the ADF method of inspection to the price
Between sequence test, inspection result is that the sequence does not have a unit root, therefore molybdenum concentrate time price series are non-stationary valence
Lattice time series；
A4. step a1, a2, a3 are integrated, it can be deduced that molybdenum concentrate Time Series of Random Macro-price is nonstationary time series.
B. on the basis of step a, judge whether molybdenum concentrate Time Series of Random Macro-price is generated by linear system, it is specific to judge
Step are as follows:
B1. the PARCOR coefficients of molybdenum concentrate Time Series of Random Macro-price are found out, and make partial autocorrelation figure, as shown in Figure 4；
B2. according to Fig. 3 and Fig. 4, it can be observed that the auto-correlation coefficient of molybdenum concentrate Time Series of Random Macro-price slowly reduces, partially
In two times of standard deviations that related coefficient is quickly decreased to partial correlation coefficient after postponing 3 ranks, it is possible thereby to select AR (3) linear mould
Type is fitted the sequence；
B3. the fitting result obtained according to AR (3) linear model, therefrom extracts residual error item, sets to the independence of residual error
Count null hypothesis；
B4. it is tested using the BDS method of inspection to the hypothesis of step b3, inspection result refuses null hypothesis, shows molybdenum essence
Mine Time Series of Random Macro-price is non-linear price time series, i.e., there are nonlinear organizations for molybdenum concentrate Time Series of Random Macro-price generation system
System.
C. using the complexity of the relevant dimension analysis molybdenum concentrate Time Series of Random Macro-price in fractal theory, molybdenum concentrate valence is judged
The complexity of lattice generation system.Relevant dimension calculation method uses G-P algorithm, calculates step are as follows:
C1. the lag period τ of molybdenum concentrate Time Series of Random Macro-price is determined using correlation analysis；
C2. with Embedded dimensions k to molybdenum concentrate Time Series of Random Macro-price x (n), n=1,2, when 3 ..., N, N are molybdenum concentrate price
Between sequence length, carry out phase space reconfiguration, phase space phase point after reconstruct is Xi(n)={ xi(n),xi(n+τ),…,xi(n+
(k-1) τ) }, i=1,2 ..., K, K are the sum of the phase point in the phase space of reconstruct, it is calculated by formula K=N- (m-1) τ,
I indicates i-th of phase point in phase space；
C3. a radius of neighbourhood r is given, is calculated correlation integral C (k, r), calculation formula is as follows,
Wherein, θ is function # (x),
‖Xi-Xj‖ is the distance between phase space phase point；
C4. logarithm is taken to correlation integral C (k, r) and r respectively, the pass of molybdenum concentrate Time Series of Random Macro-price is asked using following equation
Join dimension D,
Thus the D value acquired is 1.3, it can be considered that the generation system of molybdenum concentrate Time Series of Random Macro-price is more complex.
D. to step a, the result of b, c carry out comprehensive analysis, obtain molybdenum concentrate Time Series of Random Macro-price substantive characteristics and its
It is non-stationary, the medium complication system containing non-linear component that corresponding price data, which generates system,.
In the second portion, molybdenum concentrate Time Series of Random Macro-price x (n) is decomposed, the isolation used is EMD point
Solution, specific the step of including are as follows:
A. extreme point is taken to molybdenum concentrate Time Series of Random Macro-price x (n), structurally lower envelope line, seeks the mean value of upper and lower envelope
me；
B. former molybdenum concentrate Time Series of Random Macro-price x (n) is subtracted into mean value me, obtains a new time series h (n)；
C. repeat step a, b, until time series h (n) meet condition zero crossing number it is equal with extreme point number or
It is zero that maximum, which differs one and time series h (n) maximum and the mean value of the envelope up and down of minimum construction, is remembered at this time
Time series h (n) is imf；
D. imf is subtracted with molybdenum concentrate time series x (n), obtains new time series x1(n)；
E. by time series x1(n) repeat the above steps a, b, c, d, until cannot decompose again, thus obtains molybdenum essence
Mine all intrinsic mode function imf and a residual error function res.
Part III is the predicted portions of molybdenum concentrate Time Series of Random Macro-price, and mainly in combination with step 3,4 analysis and decomposition is tied
Fruit, the intrinsic mode function imf obtained to decomposition carry out classification prediction, then integrate to classification prediction result, specific to wrap
Containing the step of are as follows:
A. the evaluation result for generating system to molybdenum concentrate Time Series of Random Macro-price according to first part, second part is decomposed
To intrinsic mode function imf and residual error res according to its timing diagram be divided into two classes, one kind is simple stationary linear time sequence
Column, it is another kind of for complicated steady Nonlinear Time Series；
B. two class data step a obtained divide, and construct training set x (t), y (t) and forecast set x respectively1(t1)、
y1(t1), the prediction step of two class data is consistent；
C. to simple stationary time series ingredient, ARMA (p, q) model is constructed respectively and is predicted, the p value selected
It is determined with q value according to the auto-correlation and partial correlation figure of each time series, prediction steps specifically:
C1. the value of the parameter p, q of ARMA (p, q) model are determined using data set x (t)；
C2. ARMA (p, q) model is constructed, estimates the unknown parameter in model；
C3. the validity for examining ARMA (p, q) model, if do not passed through, return step b2；
C4. ARMA (p, q) model is tested, and constructs multiple model of fit, therefrom select optimal model.
D. to complicated steady Nonlinear Time Series ingredient, LSTM deep neural network is constructed respectively and is predicted,
The neuronal structure of LSTM neural network is as shown in figure 5, x in figuretIt is neuron number according to input, htIt is neuron number according to output,
ft、it、otRespectively correspond the forgetting door, input gate and output door of inside neurons, the prediction step of LSTM deep neural network
Suddenly specifically:
D1. the LSTM number of unit inputted according to the list entries tectonic network of training set, while determining output number；
D2. LSTM deep neural network model is constructed, and network is trained using training set y (t)；
D3. test set y is used1(t1) trained LSTM neural network is tested, and according to test result to prediction
Model optimizes, until reaching higher precision of prediction.
E. to step c, the prediction result of d is integrated using three layers of BP neural network, and it is shown that specific step is as follows:
E1. by molybdenum concentrate Time Series of Random Macro-price x (n), according to c, the prediction step in Step d is divided into training set and test
Collection；
E2. by c, input value of the prediction result as BP neural network in Step d, molybdenum concentrate Time Series of Random Macro-price is divided
Obtained training set is trained BP neural network as output valve；
E3. using trained BP neural network to c, the test set prediction result in Step d is integrated, and will be integrated
Obtained sequence is compared as final prediction result, by it with the test set of molybdenum concentrate Time Series of Random Macro-price.
To sum up, molybdenum concentrate market price analysis of the present invention and prediction technique and system, for analyzing and predicting molybdenum concentrate
Long-term Market price, for molybdenum ore enterprise solve the problems, such as in long-term production decision foundation, fundamental analysis and forecast reason are provided
Are as follows: by collecting the historical price of molybdenum concentrate, historical price data are converted, are obtained monthly, molybdenum concentrate market is average
Price constitutes molybdenum concentrate market price time series；Firstly, using molybdenum concentrate market price time series as object, carry out it is steady
Non-stationary, it is linear non-linear, complexity analyzed, obtain the substantive characteristics of molybdenum concentrate market price time series data；So
Afterwards, on the basis of the substantive characteristics of molybdenum concentrate market price time series data analyzes result, to molybdenum concentrate market price sequence
Column are decomposed, and select different prediction models to carry out classification prediction to it according to the result of decomposition；Finally, by classification prediction
Prediction result is integrated, and final molybdenum concentrate market price forecasts result is obtained.The prediction technique is scientific and effective, precision of prediction
Height has universality, can be used for the market price forecasts of other similar products, is with a wide range of applications.
Claims (9)
1. a kind of molybdenum concentrate Long-term Market price analysis and prediction technique, which is characterized in that by molybdenum concentrate price time
The analysis of sequence, the feature for excavating molybdenum concentrate price generation system are obtained using Time Series technology by these character separations
To the intrinsic mode function of Time Series of Random Macro-price, then intrinsic mode function is predicted, and classification prediction result is carried out
It is integrated.
2. the price analysis of molybdenum concentrate Long-term Market and prediction technique according to claim 1, which is characterized in that described to molybdenum
The content of concentrate Time Series of Random Macro-price analysis is stationarity, linear/non-linear, the complexity of molybdenum concentrate Time Series of Random Macro-price, point
Analyse step are as follows:
A. it is qualitatively and quantitatively commented using the method and unit root test method of molybdenum concentrate Time Series of Random Macro-price figure and autocorrelogram
Sentence the stationarity of molybdenum concentrate Time Series of Random Macro-price；
B. on the basis of step a, suitable AR (p) linear model is selected to be fitted molybdenum concentrate Time Series of Random Macro-price, so
It is tested afterwards using BDS method of inspection to the independence of residual error, judges whether molybdenum concentrate price is generated by linear system；
C. using the complexity of the relevant dimension analysis molybdenum concentrate Time Series of Random Macro-price in fractal theory, it is raw to judge molybdenum concentrate price
At the complexity of system, relevant dimension calculation method uses G-P algorithm, calculates step are as follows:
C1. the lag period τ of molybdenum concentrate Time Series of Random Macro-price is determined using correlation analysis；
It c2. is molybdenum concentrate price time sequence to molybdenum concentrate Time Series of Random Macro-price x (n), n=1,2,3 ..., N, N with Embedded dimensions k
Column length carries out phase space reconfiguration, and the phase space phase point after reconstruct is Xi(n)={ xi(n),xi(n+τ),…,xi(n+(k-1)
τ) }, i=1,2 ..., K, K are the sum of the phase point in the phase space of reconstruct, are calculated by formula K=N- (m-1) τ；I is indicated
I-th of phase point in phase space；
C3. a radius of neighbourhood r is given, is calculated correlation integral C (k, r), calculation formula is as follows,
Wherein, θ is function # (x),
‖Xi-Xj‖ is the distance between phase space phase point；
C4. logarithm is taken to correlation integral C (k, r) and r respectively, the correlation dimension of molybdenum concentrate Time Series of Random Macro-price is sought using following equation
Number D,
D. to step a, the result of b, c carry out comprehensive analysis, obtain the substantive characteristics and its correspondence of molybdenum concentrate Time Series of Random Macro-price
Price data generate system the characteristics of.
3. the price analysis of molybdenum concentrate Long-term Market and prediction technique according to claim 2, which is characterized in that utilize the time
Sequence decomposition technique separation characteristic is decomposed to molybdenum concentrate Time Series of Random Macro-price x (n), and the isolation used is EMD point
Solution, specific the step of including are as follows:
A. extreme point is taken to molybdenum concentrate Time Series of Random Macro-price x (n), structurally lower envelope line, seeks the mean value me of upper and lower envelope；
B. former molybdenum concentrate Time Series of Random Macro-price x (n) is subtracted into mean value me, obtains a new time series h (n)；
C. step a, b are repeated, until to meet condition zero crossing number equal or maximum with extreme point number by time series h (n)
The mean value of the envelope up and down of maximum and the minimum construction of difference one and time series h (n) is zero, remembers the time at this time
Sequences h (n) is imf；
D. imf is subtracted with molybdenum concentrate time series x (n), obtains new time series x1(n)；
E. by time series x1(n) repeat the above steps a, b, c, d, and until cannot decompose again, it is all thus to obtain molybdenum concentrate
Intrinsic mode function imf and a residual error function res.
4. the price analysis of molybdenum concentrate Long-term Market and prediction technique according to claim 3, which is characterized in that described to this
Sign mode function is predicted, and carries out integrated being binding analysis and decomposition result, building classification prediction to classification prediction result
The intrinsic mode function imf that model obtains decomposition is predicted, is then integrated to classification prediction result, specific steps
Are as follows:
A. according to the analysis for the evaluation result for generating system to molybdenum concentrate Time Series of Random Macro-price, the intrinsic mode letter that decomposition is obtained
Number imf and residual error res is divided into two classes according to its timing diagram, and one kind is simple stationary time series, another kind of for the flat of complexity
Steady Nonlinear Time Series；
B. two class data step a obtained divide, and construct training set x (t), y (t) and forecast set x respectively1(t1)、y1
(t1), the prediction step of two class data is consistent；
C. to simple stationary time series ingredient, ARMA (p, q) model is constructed respectively and is predicted, the p value and q value selected
It is determined according to the auto-correlation of each time series and partial correlation figure, prediction steps specifically:
C1. the value of the parameter p, q of ARMA (p, q) model are determined using data set x (t)；
C2. ARMA (p, q) model is constructed, estimates the unknown parameter in model；
C3. the validity for examining ARMA (p, q) model, if do not passed through, return step b2；
C4. ARMA (p, q) model is tested, and constructs multiple model of fit, therefrom select optimal model.
D. to complicated steady Nonlinear Time Series ingredient, LSTM deep neural network is constructed respectively and is predicted that LSTM is deep
Spend the prediction steps of neural network specifically:
D1. the LSTM number of unit inputted according to the list entries tectonic network of training set, while determining output number；
D2. LSTM deep neural network model is constructed, and network is trained using training set y (t)；
D3. test set y is used1(t1) trained LSTM neural network is tested, and according to test result to prediction model
It optimizes, until reaching higher precision of prediction；
E. to step c, the prediction result of d is integrated using three layers of BP neural network, and it is shown that specific step is as follows:
E1. by molybdenum concentrate Time Series of Random Macro-price x (n), according to c, the prediction step in Step d is divided into training set and test set；
E2. by c, input value of the prediction result as BP neural network in Step d, molybdenum concentrate Time Series of Random Macro-price divides to obtain
Training set as output valve, BP neural network is trained；
E3. using trained BP neural network to c, the test set prediction result in Step d is integrated, and will be integrated
Sequence as final prediction result, it is compared with the test set of molybdenum concentrate Time Series of Random Macro-price.
5. the price analysis of molybdenum concentrate Long-term Market and prediction technique according to claim 1, which is characterized in that pass through collection
The historical price of molybdenum concentrate converts historical price data, obtains molybdenum concentrate market average price monthly, described in composition
Molybdenum concentrate market price time series.
6. a kind of molybdenum concentrate Long-term Market price analysis and forecasting system characterized by comprising
Molybdenum concentrate Time Series of Random Macro-price data analysis module analyzes molybdenum concentrate Time Series of Random Macro-price data, excavates molybdenum concentrate price
The feature of generation system；
Molybdenum concentrate Time Series of Random Macro-price data decomposing module obtains valence using Time Series technology by the character separation
The intrinsic mode function of lattice time series；
Classification prediction and integration module, predict intrinsic mode function, and integrate to classification prediction result.
7. the price analysis of molybdenum concentrate Long-term Market and forecasting system according to claim 6, which is characterized in that the molybdenum essence
Mine Time Series of Random Macro-price data analysis module, executes following analytical procedure:
A. it is qualitatively and quantitatively commented using the method and unit root test method of molybdenum concentrate Time Series of Random Macro-price figure and autocorrelogram
Sentence the stationarity of molybdenum concentrate Time Series of Random Macro-price；
B. on the basis of step a, suitable AR (p) linear model is selected to be fitted molybdenum concentrate Time Series of Random Macro-price, so
It is tested afterwards using BDS method of inspection to the independence of residual error, judges whether molybdenum concentrate price is generated by linear system；
C. using the complexity of the relevant dimension analysis molybdenum concentrate Time Series of Random Macro-price in fractal theory, it is raw to judge molybdenum concentrate price
At the complexity of system, relevant dimension calculation method uses G-P algorithm, calculates step are as follows:
C1. the lag period τ of molybdenum concentrate Time Series of Random Macro-price is determined using correlation analysis；
It c2. is molybdenum concentrate price time sequence to molybdenum concentrate Time Series of Random Macro-price x (n), n=1,2,3 ..., N, N with Embedded dimensions k
Column length carries out phase space reconfiguration, and the phase space phase point after reconstruct is Xi(n)={ xi(n),xi(n+τ),…,xi(n+(k-1)
τ) }, i=1,2 ..., K, K are the sum of the phase point in the phase space of reconstruct, are calculated by formula K=N- (m-1) τ；I is indicated
I-th of phase point in phase space；
C3. a radius of neighbourhood r is given, is calculated correlation integral C (k, r), calculation formula is as follows,
Wherein, θ is function # (x),
‖Xi-Xj‖ is the distance between phase space phase point；
C4. logarithm is taken to correlation integral C (k, r) and r respectively, the correlation dimension of molybdenum concentrate Time Series of Random Macro-price is sought using following equation
Number D,
D. to step a, the result of b, c carry out comprehensive analysis, obtain the substantive characteristics and its correspondence of molybdenum concentrate Time Series of Random Macro-price
Price data generate system the characteristics of.
8. the price analysis of molybdenum concentrate Long-term Market and forecasting system according to claim 7, which is characterized in that the molybdenum essence
Mine Time Series of Random Macro-price data decomposing module executes following steps:
A. extreme point is taken to molybdenum concentrate Time Series of Random Macro-price x (n), structurally lower envelope line, seeks the mean value me of upper and lower envelope；
B. former molybdenum concentrate Time Series of Random Macro-price x (n) is subtracted into mean value me, obtains a new time series h (n)；
C. step a, b are repeated, until to meet condition zero crossing number equal or maximum with extreme point number by time series h (n)
The mean value of the envelope up and down of maximum and the minimum construction of difference one and time series h (n) is zero, remembers the time at this time
Sequences h (n) is imf；
D. imf is subtracted with molybdenum concentrate time series x (n), obtains new time series x1(n)；
E. by time series x1(n) repeat the above steps a, b, c, d, and until cannot decompose again, it is all thus to obtain molybdenum concentrate
Intrinsic mode function imf and a residual error function res.
9. the price analysis of molybdenum concentrate Long-term Market and forecasting system according to claim 8, which is characterized in that the classification
Prediction executes following steps with integration module:
A. according to the analysis for the evaluation result for generating system to molybdenum concentrate Time Series of Random Macro-price, the intrinsic mode letter that decomposition is obtained
Number imf and residual error res is divided into two classes according to its timing diagram, and one kind is simple stationary time series, another kind of for the flat of complexity
Steady Nonlinear Time Series；
B. two class data step a obtained divide, and construct training set x (t), y (t) and forecast set x respectively1(t1)、y1
(t1), the prediction step of two class data is consistent；
C. to simple stationary time series ingredient, ARMA (p, q) model is constructed respectively and is predicted, the p value and q value selected
It is determined according to the auto-correlation of each time series and partial correlation figure, prediction steps specifically:
C1. the value of the parameter p, q of ARMA (p, q) model are determined using data set x (t)；
C2. ARMA (p, q) model is constructed, estimates the unknown parameter in model；
C3. the validity for examining ARMA (p, q) model, if do not passed through, return step b2；
C4. ARMA (p, q) model is tested, and constructs multiple model of fit, therefrom select optimal model；
D. to complicated steady Nonlinear Time Series ingredient, LSTM deep neural network is constructed respectively and is predicted that LSTM is deep
Spend the prediction steps of neural network specifically:
D1. the LSTM number of unit inputted according to the list entries tectonic network of training set, while determining output number；
D2. LSTM deep neural network model is constructed, and network is trained using training set y (t)；
D3. test set y is used1(t1) trained LSTM neural network is tested, and according to test result to prediction model
It optimizes, until reaching higher precision of prediction；
E. to step c, the prediction result of d is integrated using three layers of BP neural network, and it is shown that specific step is as follows:
E1. by molybdenum concentrate Time Series of Random Macro-price x (n), according to c, the prediction step in Step d is divided into training set and test set；
E2. by c, input value of the prediction result as BP neural network in Step d, molybdenum concentrate Time Series of Random Macro-price divides to obtain
Training set as output valve, BP neural network is trained；
E3. using trained BP neural network to c, the test set prediction result in Step d is integrated, and will be integrated
Sequence as final prediction result, it is compared with the test set of molybdenum concentrate Time Series of Random Macro-price.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811068367.1A CN109242569B (en) | 2018-09-13 | 2018-09-13 | Method and system for analyzing and predicting medium-term and long-term market prices of molybdenum concentrate |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811068367.1A CN109242569B (en) | 2018-09-13 | 2018-09-13 | Method and system for analyzing and predicting medium-term and long-term market prices of molybdenum concentrate |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109242569A true CN109242569A (en) | 2019-01-18 |
| CN109242569B CN109242569B (en) | 2022-02-22 |
Family
ID=65058569
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811068367.1A Active CN109242569B (en) | 2018-09-13 | 2018-09-13 | Method and system for analyzing and predicting medium-term and long-term market prices of molybdenum concentrate |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109242569B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110568359A (en) * | 2019-09-04 | 2019-12-13 | 太原理工大学 | lithium battery residual life prediction method |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105354644A (en) * | 2015-11-29 | 2016-02-24 | 北京化工大学 | Financial time series prediction method based on integrated empirical mode decomposition and 1-norm support vector machine quantile regression |
| CN107102969A (en) * | 2017-04-28 | 2017-08-29 | 湘潭大学 | The Forecasting Methodology and system of a kind of time series data |
| CN107239854A (en) * | 2017-05-22 | 2017-10-10 | 华北电力大学 | Load forecasting method based on EMD GRA MPSO LSSVM models |
| US20180060771A1 (en) * | 2016-08-25 | 2018-03-01 | AXI System, Inc. | Systems and methods for creating volume/market weighted average price benchmark indices for fresh foods |
| CN107886351A (en) * | 2017-10-18 | 2018-04-06 | 中国地质大学（武汉） | A kind of forecast of crude oil price method and system based on CEEMD PSO BP models and error compensation |
| CN108154264A (en) * | 2017-12-21 | 2018-06-12 | 北京工业大学 | A kind of Prediction of Stock Index method based on ARMA-LSTM models |
| CN108921701A (en) * | 2018-07-19 | 2018-11-30 | 安徽磐众信息科技有限公司 | A kind of stock index futures price expectation method based on EDM algorithm and BP neural network |
-
2018
- 2018-09-13 CN CN201811068367.1A patent/CN109242569B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN105354644A (en) * | 2015-11-29 | 2016-02-24 | 北京化工大学 | Financial time series prediction method based on integrated empirical mode decomposition and 1-norm support vector machine quantile regression |
| US20180060771A1 (en) * | 2016-08-25 | 2018-03-01 | AXI System, Inc. | Systems and methods for creating volume/market weighted average price benchmark indices for fresh foods |
| CN107102969A (en) * | 2017-04-28 | 2017-08-29 | 湘潭大学 | The Forecasting Methodology and system of a kind of time series data |
| CN107239854A (en) * | 2017-05-22 | 2017-10-10 | 华北电力大学 | Load forecasting method based on EMD GRA MPSO LSSVM models |
| CN107886351A (en) * | 2017-10-18 | 2018-04-06 | 中国地质大学（武汉） | A kind of forecast of crude oil price method and system based on CEEMD PSO BP models and error compensation |
| CN108154264A (en) * | 2017-12-21 | 2018-06-12 | 北京工业大学 | A kind of Prediction of Stock Index method based on ARMA-LSTM models |
| CN108921701A (en) * | 2018-07-19 | 2018-11-30 | 安徽磐众信息科技有限公司 | A kind of stock index futures price expectation method based on EDM algorithm and BP neural network |
Non-Patent Citations (1)
| Title |
|---|
| 包洪洁: "基于集成经验模态分解的城市住宅价格周期波动研究 ——以杭州市为例", 《中国优秀硕士学位论文全文数据库经济与管理科学辑》 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110568359A (en) * | 2019-09-04 | 2019-12-13 | 太原理工大学 | lithium battery residual life prediction method |
| CN110568359B (en) * | 2019-09-04 | 2021-11-23 | 太原理工大学 | Lithium battery residual life prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109242569B (en) | 2022-02-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886498B (en) | EMD-GRU short-term power load prediction method based on feature selection | |
| Wang et al. | An improved ARIMA model for precipitation simulations | |
| Yu et al. | A special event-based K-nearest neighbor model for short-term traffic state prediction | |
| Bai et al. | Compound autoregressive network for prediction of multivariate time series | |
| CN107944612B (en) | Bus net load prediction method based on ARIMA and phase space reconstruction SVR | |
| CN103268519A (en) | Electric power system short-term load forecast method and device based on improved Lyapunov exponent | |
| CN105976077A (en) | Power transmission and transformation project cost dynamic control target calculating system and calculating method | |
| CN106656357B (en) | Power frequency communication channel state evaluation system and method | |
| CN111144281B (en) | Urban rail transit OD passenger flow estimation method based on machine learning | |
| CN106156434A (en) | Sliding window time difference Gaussian process regression modeling method based on the low and deep structure of local time | |
| Xiao et al. | The chaotic characteristic of the carbon-monoxide utilization ratio in the blast furnace | |
| CN114169645A (en) | Short-term load prediction method for smart power grid | |
| CN112434890A (en) | Prediction method of tunnel settlement time sequence based on CEEMDAN-BilSTM | |
| CN104751254A (en) | Line loss rate prediction method based on non-isometric weighted grey model and fuzzy clustering sorting | |
| CN109242569A (en) | A kind of molybdenum concentrate Long-term Market price analysis and prediction technique and system | |
| Tyass et al. | Wind speed prediction based on statistical and deep learning models | |
| Liu et al. | Prediction of water inrush through coal floors based on data mining classification technique | |
| Li et al. | A multi-factor combination prediction model of carbon emissions based on improved CEEMDAN | |
| CN105354644A (en) | Financial time series prediction method based on integrated empirical mode decomposition and 1-norm support vector machine quantile regression | |
| CN105787265A (en) | Atomic spinning top random error modeling method based on comprehensive integration weighting method | |
| CN117114176A (en) | Land utilization change prediction method and system based on data analysis and machine learning | |
| Xiao et al. | The short-time traffic flow prediction at ramp junction based on wavelet neural network | |
| Shiwakoti et al. | Time Series Analysis of Electricity Demand Forecasting Using Seasonal ARIMA and an Exponential Smoothing Model | |
| CN115222113B (en) | CNN-LSTM-based railway area freight volume short-term prediction method | |
| Shang | The Influence of Multidimensional Gray Theory on Higher Education Cooperation Projects under the Mobile Communication System |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |