WO2019165692A1 - 碳期货价格预测方法、装置、计算机设备和存储介质 - Google Patents

碳期货价格预测方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2019165692A1
WO2019165692A1 PCT/CN2018/083987 CN2018083987W WO2019165692A1 WO 2019165692 A1 WO2019165692 A1 WO 2019165692A1 CN 2018083987 W CN2018083987 W CN 2018083987W WO 2019165692 A1 WO2019165692 A1 WO 2019165692A1
Authority
WO
WIPO (PCT)
Prior art keywords
carbon
futures
price
futures price
model
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.)
Ceased
Application number
PCT/CN2018/083987
Other languages
English (en)
French (fr)
Inventor
王义文
王健宗
肖京
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.)
Ping An Technology Shenzhen Co Ltd
Original Assignee
Ping An Technology Shenzhen Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Ping An Technology Shenzhen Co Ltd filed Critical Ping An Technology Shenzhen Co Ltd
Publication of WO2019165692A1 publication Critical patent/WO2019165692A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/04—Trading; Exchange, e.g. stocks, commodities, derivatives or currency exchange
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06Q30/0278—Product appraisal
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06Q40/00—Finance; Insurance; Tax strategies; Processing of corporate or income taxes
    • G06Q40/06—Asset management; Financial planning or analysis
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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
    • G06Q50/06—Energy or water supply

Definitions

  • the present application relates to the field of price prediction, and in particular to a carbon futures price prediction method, apparatus, computer device and storage medium.
  • Carbon trading includes the Clean Development Mechanism (CDM), Joint Implementation Mechanism (JI) and International Carbon Emissions Trading (IET), among which the Clean Development Mechanism (CDM) is the most important form of carbon trading in developed and developing countries.
  • CDM Clean Development Mechanism
  • JI Joint Implementation Mechanism
  • IET International Carbon Emissions Trading
  • CDM Clean Development Mechanism
  • the channels for Chinese domestic commercial banks to participate in carbon finance are concentrated in CDM's credit investment financing and its intermediary business.
  • the prior art has a method for predicting the price of carbon futures, but it has the disadvantages of poor effect on long-term prediction, a certain overlap of information reflected by statistical data, and local minimum points in nonlinear modeling.
  • the main purpose of the present application is to provide a carbon futures price prediction method, device, computer device and storage medium with good long-term prediction effect, no data overlap, and small training samples.
  • the present application proposes a carbon futures price prediction method, including: obtaining a current energy futures price of a specified energy; and inputting the current energy futures price to a preset carbon based on a least squares support vector machine In the futures price forecasting model, price forecasting is performed; wherein the kernel function of the carbon futures price forecasting model is an RBF kernel function.
  • the application also provides a carbon futures price forecasting device, comprising: an obtaining unit, configured to obtain a current energy futures price of a specified energy source;
  • a prediction calculation unit configured to input the current energy futures price into a predetermined carbon futures price prediction model based on a least squares support vector machine, and perform price prediction; wherein the carbon futures price prediction model has a kernel function Is the RBF kernel function.
  • the application also provides a computer device comprising a memory and a processor, the memory storing computer readable instructions, the processor implementing the steps of the method described above when the computer readable instructions are executed.
  • the present application also provides a computer non-transitory readable storage medium having stored thereon computer readable instructions that, when executed by a processor, implement the steps of the methods described above.
  • the carbon futures price prediction method, device, computer equipment and storage medium of the present application are based on a least squares support vector machine, select an RBF kernel function, establish a carbon futures price prediction model, and then predict the price of carbon futures, etc.
  • the type constraint replaces the inequality constraint in the standard SVM algorithm; the solution to the quadratic programming problem is directly translated into the linear equations; the price prediction is applicable to the long, medium and short time lengths, reflecting no overlap of information and small training samples. There are no local minimums that affect the final predictions.
  • FIG. 1 is a schematic flow chart of a carbon futures price forecasting method according to an embodiment of the present application
  • step S2 is a schematic diagram of a specific process of step S2 in the carbon futures price prediction method according to an embodiment of the present application
  • FIG. 3 is a schematic diagram of a specific process of step S24 in the foregoing step S2 according to an embodiment of the present application;
  • step S21 is a schematic diagram of a specific process of step S21 in the foregoing step S2 according to an embodiment of the present application;
  • FIG. 5 is a schematic block diagram showing the structure of a carbon futures price forecasting apparatus according to an embodiment of the present application.
  • FIG. 6 is a schematic block diagram showing the structure of a prediction calculation unit according to an embodiment of the present application.
  • FIG. 7 is a schematic block diagram showing the structure of a verification module according to an embodiment of the present application.
  • FIG. 8 is a schematic block diagram showing the structure of a variable selection module according to an embodiment of the present application.
  • FIG. 9 is a schematic block diagram showing the structure of a computer device according to an embodiment of the present application.
  • an embodiment of the present application provides a carbon futures price prediction method, including the steps:
  • the current energy futures price of the specified energy refers to the energy futures price required for the carbon futures price forecast, and the price is the actual price that can be obtained in the current time period.
  • the power futures price is one of the specified energy futures prices. If the current time is the opening time of the power futures market, then the current future futures price is the current futures price obtained, if the current time is When the power futures market is closed, then the current futures price is the price of the power futures market when it was closed.
  • the above designated energy sources include various types, such as electricity, natural gas, carbon emissions, etc., which correspond to electricity futures prices, natural gas futures prices, and carbon futures prices.
  • the above-mentioned least squares support vector machine is an improvement of the support vector machine, and compared with the standard SVM (Support Vector Machine support vector machine) model, the advantages are obvious, specific: (1), using the equality constraint It replaces the inequality constraint in the standard SVM algorithm; (2) transforms the solution to the quadratic programming problem to directly solve the linear equations; (3) is most suitable for the learning environment of small samples.
  • the performance of different kernel functions varies greatly. Among them, the linear kernel function can't handle the nonlinear input value; the high-dimensional kernel parameter of the RBF kernel function is less than the polynomial kernel function, and it is trained in SVM. In the process, the training time required for the polynomial kernel function is much larger than the RBF kernel function; when the Sigmoid function is used, some parameters have error values; therefore, this study mainly uses the RBF kernel function for regression modeling.
  • the modeling method of the carbon futures price prediction model based on the least squares support vector machine includes the following steps: S21: determining a model variable according to a preset rule. In this step, there are many factors affecting the price of carbon futures. For example, the existing energy market includes electricity futures prices, natural gas futures prices, crude oil futures prices, carbon emissions prices, etc., but whether these energy futures prices really will be for carbon futures prices. If there is an impact, then a specific analysis is needed before it can be determined that the determined energy futures price can be used as a model variable. S22. Obtain sample data of the model variable, and divide the sample data into a training set and a test set.
  • the sample data needs to be obtained, and the method for obtaining the sample data includes various methods, such as directly obtaining the third-party transaction platform through the specified interface, or receiving the manual input value of the user.
  • the sample data is obtained by cross-validation.
  • the main idea of the cross-validation method is: assuming that the original training sample data is N (N is a positive integer), and is divided into k (k is a positive integer) disjoint sets ( Generally, it is evenly divided, and each set has N/k samples. Each subset is used to make a verification set. The remaining k-1 subsets are used as training sets, and k models are obtained. The k models are used.
  • the final training result is used as a criterion for parameter selection.
  • the above method can effectively avoid over-learning and under-learning, and the resulting training results are more persuasive.
  • the one carbon futures price forecasting model is in an initial state.
  • the first carbon futures price prediction model based on the least squares support vector machine in the initial state is the most primitive mathematical model, and the model without any training is performed.
  • the above-mentioned second carbon futures price prediction model is a model that has been trained and trained after training sets.
  • the second carbon futures price prediction model is the carbon futures price prediction model.
  • the second carbon futures price prediction model is determined as the carbon futures price prediction model described in the above step S2, that is, the second carbon futures price prediction model can be used in the actual scenario to predict the market carbon futures. price.
  • the step S24 of verifying the validity of the second carbon futures price prediction model by inputting sample data corresponding to the test set to the second carbon futures price prediction model is performed.
  • the second carbon futures price forecasting model uses the root mean square error method, the average absolute percentage error method, and the directional correctness method to compare the predicted price of the test with the actual price through different dimensions, only each dimension. If the expected results are achieved, the second carbon futures price forecasting model will be determined as an effective model, and the second carbon futures price forecasting model will be able to provide more accurate prediction results when it is actually put into use.
  • the step S21 of determining the model variable according to the preset rule includes:
  • the energy price is mainly selected as the main variable for price prediction.
  • the energy price selects the representative natural gas futures price, coal futures price and electricity futures price in Europe.
  • the European market the UK is the largest consumer of natural gas, so the UK natural gas futures price for delivery in December is used.
  • the coal futures price is the Rotterdam coal futures price for December delivery provided by the Intercontinental Exchange.
  • Electricity futures prices are the price of UK electricity futures offered by the Intercontinental Exchange.
  • the crude oil futures price is selected from the Brent crude oil futures price.
  • Table 1 above shows the results of the stationarity test. It can be found that each sequence does not reject the unit root hypothesis at a level of significance of 1% and is non-stationary.
  • the first-order difference sequence rejects the unit root hypothesis at the 1% significance level, which indicates that each time series satisfies I(1).
  • the cointegration test is carried out. According to the cointegration theory, if two sequences satisfy the same condition of single integer order, it is possible to form a low-order single integer variable by linear combination, and if there is a cointegration relationship between two non-stationary sequences of the same order, it indicates that There is a long-term stable equilibrium relationship between the two, so that the problem of pseudo-regression can be avoided.
  • Table 2 above is the co-integration model parameter of energy futures price and carbon futures price.
  • the significance level is 1%.
  • Table 3 above shows the test results of the cointegration relationship between various energy futures prices and carbon futures prices.
  • the test results show that the residual sequence rejects the null hypothesis when the significance level is 1%, that is, their respective residual sequences are It is a stationary sequence. Therefore, there is a long-term stable relationship between carbon futures prices and various energy futures prices, and the impact of coal futures prices, power futures prices, natural gas futures prices and oil futures prices on carbon futures contract prices are positive.
  • Granger causality (Granje causality) test on the selected energy futures price and carbon futures price.
  • Granger causality tests can be performed on each group of sequences to specifically determine the causal relationship between various energy futures prices and carbon futures prices:
  • Table 4 above is a test of causality between coal futures prices and carbon futures prices. It can be seen that there is a one-way causal relationship between carbon futures prices and coal futures prices at a 5% confidence level. Coal futures prices are the Granger reason for carbon futures prices, and carbon futures prices are not the Granger reasons for coal futures prices. That is, changes in coal futures prices can cause changes in carbon futures prices, and there is a significant long-term cointegration relationship between the two. Therefore, the coal futures price can be determined as a model input variable to model the carbon futures price forecast.
  • Table 5 above is a test of causality between natural gas futures prices and carbon futures prices. It can be seen that there is a one-way causal relationship between carbon futures prices and natural gas futures prices at a 12% confidence level. Natural gas futures prices are the Granger reason for carbon futures prices, and carbon futures prices are not the Granger reasons for natural gas futures prices. That is, changes in natural gas futures prices can cause changes in carbon futures prices, and there is a significant long-term cointegration relationship between the two. Therefore, it is possible to determine the natural gas futures price as a model input variable to model the carbon futures price forecast.
  • Table 6 above is a test of causality between the price of electricity gas futures and the price of carbon futures. It can be seen that at the 10% confidence level, there is a two-way Granger causal relationship between carbon futures prices and electricity futures prices, that is, There is an interaction between carbon futures prices and electricity futures prices. According to the Granger causality test, changes in carbon futures prices will cause changes in electricity futures prices, and vice versa. This means that there has been a significant impact between the two markets. This is because the life cycle of power plants is generally long, and the cost of technology renewal is relatively high. Therefore, more emission rights quotas are often purchased to promote the rise of carbon prices. In order to transfer some of the additional costs of purchasing quotas, power plants generally increase electricity prices.
  • the forward electricity price is also affected by the price of natural gas and the spot price. It is said that carbon futures price is an important factor affecting the price of electricity futures, but it is not the only factor, and the price of electricity futures is also a major factor affecting the price of carbon futures. factor. Therefore, the power futures price can be determined as a model input variable to model the carbon futures price forecast.
  • Table 7 above is a test of causality between LPG futures prices and carbon futures prices. It can be seen that there is no significant Granger causality between carbon futures prices and oil futures prices. This shows that during the study period, fluctuations in oil futures prices did not affect the carbon futures market.
  • the sample data of the model variable is acquired, and the specific process of the step S22 of dividing the sample data into the training set and the test set is parameter optimization, and the optimal parameter is obtained. group.
  • w is the weight vector and b is the paranoid value, which is a nonlinear mapping from the input space to the high-dimensional feature space.
  • the least squares support vector machine optimization goal can be expressed as:
  • Equation (7) can be transformed into:
  • K(x i ,x) represents a nonlinear mapping from the input space to the high-dimensional feature space
  • ⁇ i is a Lagrangian multiplier
  • b is a paranoid value
  • x i is a related independent variable
  • ⁇ i and b It can be obtained by solving the linear equation of equation (5).
  • the kernel function, the mapping function and the feature space are one-to-one correspondence. It is determined that the kernel function K(x i , x j ) implicitly determines the mapping function. And feature space F.
  • the use of the kernel function enables the support vector machine to obtain powerful nonlinear processing capabilities, and avoids complex calculations in high-dimensional feature space, effectively overcoming the dimensionality disaster problem.
  • x i is the associated independent variable
  • x j is the carbon futures price
  • ⁇ is the nuclear width
  • the regularization parameter and the kernel width ⁇ of the RBF function directly affect the learning ability and generalization ability of the least squares support vector machine. Therefore, the selection method of its regularization parameter and kernel width ⁇ The research is very important. On the other hand, there is no certain relationship between the kernel function ⁇ and the regularization parameter in the performance impact of the least squares support vector machine, which makes the effective selection of the kernel function ⁇ and the regularization parameter become the focus.
  • the regularization parameter of the least squares support vector machine and the kernel width of the RBF kernel function are selected based on the cross-validation method.
  • the method further includes: S3: acquiring a time length of the prediction period; S4, if the prediction period belongs to a short-term or medium-term prediction period, calling the predicted time series analysis model to perform the second Sub-carbon futures forecast; S5, the first predicted price predicted by the carbon futures price prediction model based on the least squares support vector machine and the second predicted price predicted based on the time series analysis model are subjected to a preset weighted average operation The final carbon futures forecast prices.
  • the time series analysis model can accurately predict the carbon futures price in the short or medium term, it can be used as part of the carbon futures price forecast, for example, based on the minimum
  • the first predicted price predicted by the carbon futures price prediction model of the two-squares support vector machine is multiplied by the first percentage, and then the second predicted price predicted based on the time series analysis model is multiplied by the second percentage to obtain a Comprehensive carbon futures forecast prices and so on.
  • step S2 the method further includes: S6, re-acquiring the current energy futures price of the specified energy source from different energy trading platforms respectively; S7, repeating the energy futures prices obtained on different trading platforms respectively In the process of step S2, a plurality of carbon futures predicted prices are obtained; S8, the average predicted carbon futures prices are averaged to obtain the final carbon futures prices.
  • the carbon futures price prediction method in the embodiment of the present application based on a least squares support vector machine, selects an RBF kernel function, establishes a carbon futures price prediction model, and then predicts the price of carbon futures, and replaces the standard SVM algorithm with an equality constraint.
  • an embodiment of the present application further provides a carbon futures price forecasting apparatus, including:
  • the obtaining unit 10 is configured to obtain a current energy futures price of the specified energy source
  • a prediction calculation unit 20 configured to input the current energy futures price into a predetermined carbon futures price prediction model based on a least squares support vector machine, and perform price prediction; wherein, the core of the carbon futures price prediction model
  • the function is an RBF kernel function.
  • the current energy futures price of the specified energy source refers to the energy futures price required for the carbon futures price forecast, and the price is the actual price that can be obtained in the current time period.
  • the power futures price is one of the specified energy futures prices. If the current time is the opening time of the power futures market, then the current future futures price is the current futures price obtained, if the current time is When the power futures market is closed, then the current futures price is the price of the power futures market when it was closed.
  • the above designated energy sources include various types, such as electricity, natural gas, carbon emissions, etc., which correspond to electricity futures prices, natural gas futures prices, and carbon futures prices.
  • the above-mentioned least squares support vector machine is an improvement of the support vector machine, and compared with the standard SVM model, the advantages are obvious, and specific: (1), using the equality constraint instead of the inequality in the standard SVM algorithm. Constraint; (2), to solve the quadratic programming problem into a direct solution to the linear equations; (3), the most suitable for small sample learning environment.
  • the performance of different kernel functions varies greatly.
  • the linear kernel function can't handle the nonlinear input value;
  • the high-dimensional kernel parameter of the RBF kernel function is less than the polynomial kernel function, and it is trained in SVM.
  • the training time required for the polynomial kernel function is much larger than the RBF kernel function; when the Sigmoid function is used, some parameters have error values; therefore, this study mainly uses the RBF kernel function for regression modeling.
  • the foregoing prediction calculation unit 20 includes:
  • the variable selection module 21 is configured to determine a model variable according to a preset rule.
  • the existing energy market includes electricity futures prices, natural gas futures prices, crude oil futures prices, carbon emissions prices, etc., but whether these energy futures prices will actually affect carbon futures prices, then A specific analysis is needed before it can be determined that the determined energy futures price can be used as a model variable.
  • the obtaining module 22 is configured to acquire sample data of the model variable, and divide the sample data into a training set and a test set.
  • the sample data needs to be obtained, and the method for obtaining the sample data includes a plurality of methods, such as directly obtaining the third-party transaction platform through a specified interface, or receiving a manual input value of the user.
  • the sample data is obtained by cross-validation.
  • the main idea of the cross-validation method is: assuming that the original training sample data is N (N is a positive integer), and is divided into k (k is a positive integer) disjoint sets ( Generally, it is evenly divided, and each set has N/k samples. Each subset is used to make a verification set, and the remaining k-1 subsets are used as training sets.
  • the training module 23 is configured to input the sample data corresponding to the training set into a preset first carbon futures price prediction model based on a least squares support vector machine, and obtain a trained second carbon futures price prediction model.
  • the first carbon futures price forecasting model is in an initial state.
  • the first carbon futures price prediction model in the initial state is the most primitive mathematical model, and is a model without any training.
  • the above-mentioned second carbon futures price prediction model is a model that has been trained and trained after training sets.
  • the verification module 24 is configured to input the sample data corresponding to the test set into the second carbon futures price prediction model to verify the validity of the second carbon futures price prediction model.
  • the accurate test here is that the price of carbon futures is not 100% accurate, but allows a certain deviation, but it is expected Within the scope, only if the predicted carbon futures price is within the expected range, the second carbon futures price forecasting model will be determined as an effective model.
  • the model determining module 25 is configured to determine, when the verification module verifies the validity model of the second carbon futures price prediction model, the second carbon futures price prediction model as the carbon futures price prediction model.
  • the second carbon futures price prediction model is determined as the carbon futures price prediction model described in the above prediction calculation unit 20, that is, the second carbon futures price prediction model can be used in the actual scenario for real carbon futures price .
  • the verification module 24 includes:
  • a calculation price sub-module 241 configured to substitute the sample data corresponding to the test set into the second carbon futures price prediction model to obtain a predicted price
  • a multi-dimensional calculation sub-module 242 configured to calculate a magnitude of the error between the predicted price and the actual price by using a root mean square error method, and calculate a degree of deviation between the predicted price and the actual price by using an average absolute percentage error method, and a direction of passing
  • the correctness method calculates the direction correctness rate of the predicted price
  • the determining sub-module 243 determines that the second carbon futures price forecasting model is an effective model if the error magnitude, the degree of deviation, and the direction correctness rate all meet the preset requirements.
  • the root mean square error method, the average absolute percentage error method, and the directional correctness method are used, and the predicted price of the test is compared with the actual price through different dimensions, and only if each dimension achieves the expected effect is determined.
  • the second carbon futures price forecasting model is an effective model, and the second carbon futures price forecasting model can provide more accurate forecasting results when it is actually put into use.
  • variable selection module 21 includes:
  • the obtaining submodule 211 is configured to obtain multiple types of energy futures prices within a specified time period
  • the first test sub-module 212 is configured to check the cointegration relationship between the carbon futures price and various energy futures prices, and select a cointegration relationship with the carbon futures price as a positive energy futures price;
  • a second test sub-module 213, configured to perform a Granger causality test on the selected energy futures price and the carbon futures price
  • the determining sub-module 214 is configured to select an energy futures price that reaches a preset significant requirement with respect to the Granger causal relationship of the carbon futures price, and determines the selected energy futures price as the model variable.
  • the specific implementation process is substantially the same as that in the embodiment in the carbon futures price prediction method described above, and is not described herein.
  • From the results of Granger causality test there is a certain degree of causal relationship between carbon futures prices and coal futures prices, natural gas futures prices and electricity futures prices.
  • the price of electricity futures is the main factor affecting the price of carbon futures.
  • the Granger causal relationship between carbon futures prices and oil futures prices is not obvious. This is because there are many factors affecting oil prices.
  • the carbon market is the same as the oil market. The main oil fluctuations are driven by events, so the transmission mechanism of the two markets has not been effectively established.
  • the sample data of the model variable is obtained, and the specific process of dividing the sample data into the training set and the test set 2 is parameter optimization, and the optimal parameter group is obtained.
  • w is the weight vector and b is the paranoid value, which is a nonlinear mapping from the input space to the high-dimensional feature space.
  • the least squares support vector machine optimization goal can be expressed as:
  • Equation (7) can be transformed into:
  • K(x i ,x) represents a nonlinear mapping from the input space to the high-dimensional feature space
  • ⁇ i is a Lagrangian multiplier
  • b is a paranoid value
  • x i is a related independent variable
  • ⁇ i and b It can be obtained by solving the linear equation of equation (5).
  • the kernel function, the mapping function and the feature space are one-to-one correspondence. It is determined that the kernel function K(x i , x j ) implicitly determines the mapping function. And feature space F.
  • the use of the kernel function enables the support vector machine to obtain powerful nonlinear processing capabilities, and avoids complex calculations in high-dimensional feature space, effectively overcoming the dimensionality disaster problem.
  • x i is the associated independent variable
  • x j is the carbon futures price
  • ⁇ is the nuclear width
  • the carbon futures price forecasting device further includes:
  • a calling unit configured to: if the prediction period belongs to a short-term or medium-term prediction period, invoke a predicted time series analysis model to perform a second carbon futures prediction;
  • a weighting calculation unit configured to perform a preset weighted average operation on the first predicted price predicted by the carbon futures price prediction model based on the least squares support vector machine and the second predicted price predicted based on the time series analysis model, to obtain a final Carbon futures forecast prices.
  • the time series analysis model can predict the carbon futures price in the short or medium term more accurately, it can be used as part of the carbon futures price forecast.
  • the carbon futures price forecasting model based on the least squares support vector machine will be used. The predicted first predicted price is multiplied by the first percentage, and then the second predicted price predicted based on the time series analysis model is multiplied by the second percentage to obtain a comprehensive carbon futures predicted price and the like.
  • the carbon futures price forecasting device further includes:
  • a separately calculating unit configured to use the foregoing prediction calculation unit 20 to calculate the energy futures prices obtained by different trading platforms, and calculate a plurality of carbon futures predicted prices;
  • the average calculation unit is used to average all the predicted carbon futures prices to obtain the final carbon futures price.
  • the carbon futures price forecasting device of the embodiment of the present application selects an RBF kernel function based on a least squares support vector machine, establishes a carbon futures price prediction model, and then predicts the price of the carbon futures, and replaces the standard SVM algorithm with an equality constraint.
  • the computer device may be a server, and its internal structure may be as shown in FIG. 9.
  • the computer device includes a processor, memory, network interface, and database connected by a system bus. Among them, the processor designed by the computer is used to provide calculation and control capabilities.
  • the memory of the computer device includes a non-volatile storage medium, an internal memory.
  • the non-volatile storage medium stores an operating system, computer readable instructions, and a database.
  • the memory provides an environment for the operation of operating systems and computer readable instructions in a non-volatile storage medium.
  • the database of the computer equipment is used to store data such as a carbon futures price prediction model.
  • the network interface of the computer device is used to communicate with an external terminal via a network connection.
  • the computer readable instructions when executed, perform the flow of an embodiment of the methods described above. It will be understood by those skilled in the art that the structure shown in FIG. 9 is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation of the computer device to which the present application is applied.
  • An embodiment of the present application also provides a computer non-volatile readable storage medium having stored thereon computer readable instructions that, when executed, perform the processes of the embodiments of the methods described above.
  • the above description is only the preferred embodiment of the present application, and is not intended to limit the scope of the patent application, and the equivalent structure or equivalent process transformations made by the specification and the drawings of the present application, or directly or indirectly applied to other related The technical field is equally included in the scope of patent protection of the present application.

Landscapes

  • Business, Economics & Management (AREA)
  • Engineering & Computer Science (AREA)
  • Strategic Management (AREA)
  • Economics (AREA)
  • Development Economics (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • Physics & Mathematics (AREA)
  • Marketing (AREA)
  • General Business, Economics & Management (AREA)
  • General Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • Human Resources & Organizations (AREA)
  • Game Theory and Decision Science (AREA)
  • Entrepreneurship & Innovation (AREA)
  • Technology Law (AREA)
  • Operations Research (AREA)
  • Tourism & Hospitality (AREA)
  • Health & Medical Sciences (AREA)
  • General Health & Medical Sciences (AREA)
  • Public Health (AREA)
  • Water Supply & Treatment (AREA)
  • Primary Health Care (AREA)
  • Quality & Reliability (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

本申请揭示了一种碳期货价格预测方法、装置、计算机设备和存储介质，方法包括：获取指定能源当前的能源期货价格；将当前的能源期货价格输入到基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测，碳期货价格预测模型的核函数是RBF核函数。本申请适用于各时间长度的价格预测，训练样本较小，不会出现局部极小点。

Description

碳期货价格预测方法、装置、计算机设备和存储介质
本申请要求于2018年2月27日提交中国专利局、申请号为201810162631.1，发明名称为“碳期货价格预测方法、装置、计算机设备和存储介质”的中国专利申请的优先权，其全部内容通过引用结合在本申请中。
技术领域
本申请涉及价格预测领域，特别是涉及到一种碳期货价格预测方法、装置、计算机设备和存储介质。
背景技术
作为新兴的金融市场，碳排放权交易市场在近些年发展迅猛。碳交易包括清洁发展机制(CDM)、联合实施机制(JI)和国际碳排放权交易(IET)三种形式，其中清洁发展机制(CDM)是发达国家和发展中国家进行碳交易的最主要形式，目前中国国内商业银行参与碳金融的渠道集中于CDM的信贷投资融资及其中间业务。现有技术存在对碳期货价格预测的方法，但是其具有对于长期预测的效果较差、统计数据所反映的信息有一定的重叠、以及在非线性建模中容易出现局部极小点等缺点。
所以提供一种新的碳期货价格的预测方法是亟需解决的问题。
技术问题
本申请的主要目的为提供一种长期预测效果好、数据反映无重叠、训练样本小的碳期货价格预测方法、装置、计算机设备和存储介质。
技术解决方案
为了实现上述申请目的，本申请提出一种碳期货价格预测方法，包括：获取指定能源当前的能源期货价格；将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF核函数。
本申请还提供一种碳期货价格预测装置，包括：获取单元，用于获取指定能源当前的能源期货价格；
预测计算单元，用于将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF核函数。
本申请还提供一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，所述处理器执行所述计算机可读指令时实现上述方法的步骤。
本申请还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，所述计算机可读指令被处理器执行时实现上述的方法的步骤。
有益效果
本申请的碳期货价格预测方法、装置、计算机设备和存储介质，基于最小二乘支持向量机，选择RBF核函数，建立碳期货价格预测的模型，然后对碳期货的价格进行预测，其用等式约束代替标准SVM算法中的不等式约束；将求解二次规划问题转化为直接求解线性方程组；适用于长、中、短期的各时间长度 的价格预测，反映信息无重叠，训练样本较小，还不会出现局部极小点而影响最终的预测结果。
附图说明
图1为本申请一实施例的碳期货价格预测方法的流程示意图；
图2为本申请一实施例的上述碳期货价格预测方法中步骤S2的具体流程示意图；
图3为本申请一实施例的上述步骤S2中的步骤S24的具体流程示意图；
图4为本申请一实施例的上述步骤S2中的步骤S21的具体流程示意图；
图5为本申请一实施例的碳期货价格预测装置的结构示意框图；
图6为本申请一实施例的预测计算单元的结构示意框图；
图7为本申请一实施例的验证模块的结构示意框图；
图8为本申请一实施例的变量选择模块的结构示意框图；
图9为本申请一实施例的计算机设备的结构示意框图。
本发明的最佳实施方式
应当理解，此处所描述的具体实施例仅仅用以解释本申请，并不用于限定本申请。
参照图1，本申请实施例提供一种碳期货价格预测方法，包括步骤：
S1、获取指定能源当前的能源期货价格；
S2、将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF(Radial Basis Function径向基函数)核函数。
如上述步骤S1所述，上述指定能源当前的能源期货价格是指在碳期货价格预测时需要的能源期货价格，其价格为当前时间段内可以获取到的实际价格。在一具体实施例中，比如电力期货价格是指定的能源期货价格之一，如果当前时间为电力期货市场开放时间，那么上述当前的期货价格即为当前时刻获取的电力期货价格，如果当前时间为电力期货市场关闭时间，那么上述当前的期货价格即为电力期货市场前一次关闭时的价格。上述指定能源包括多种，比如电力、天然气、碳排放等，其对应的分别为电力期货价格、天然气期货嘉价格和碳期货价格。
如上述步骤S2所述，上述最小二乘支持向量机是支持向量机的改进，与标准SVM(Support Vector Machine支持向量机)模型比较，其优势明显，具体的：(1)、用等式约束代替标准SVM算法中的不等式约束；(2)、将求解二次规划问题转化为直接求解线性方程组；(3)、最适合于小样本的学习环境。在最小二乘支持向量机中，不同核函数，其性能差异很大，其中，线性核函数不能处理非线性输入值；RBF核函数的高维核参数量少于多项式核函数，同时在SVM训练过程中，采用多项式核函数所需训练时间远大于RBF核函数；采用Sigmoid函数时，某些参数存在错误值；因此，本研究主要采用RBF核函数进行回归建模。
参照图2，本申请实施例中，上述基于最小二乘支持向量机的碳期货价格预测模型的建模方法，包括步骤：S21、根据预设规则确定模型变量。本步骤中，影响碳期货价格的因素很多，比如，现有的能源市场包括电力期货价格、天然气期货价格、原油期货价格、碳排放价格等，但是这些能源期货价格是否真的都会对碳期货价格产生影响，则需要具体分析，然后才能确定，确定下来的能源期货价格既可以作为模型的变量。S22、获取所述模型变量的样本数据，并将所述样本数据分为训练集和测试集。本步骤中，模型变量确定后，即需要获取到样本数据，获取样本数据的方法包括多种，如通过指定接口直接到第三方交易平台获取，或者接收用户的手动输入值等。本实施例中，样本数据采用交叉验证法获取，交叉验证法的主要思想为：假定原始训练样本数据为N(N为正整数)，被分成k(k为正整数)个不相交的集合(一般是均分，每个集合有N/k个样本)，每一个子集都用来作一次验证集，其余的k-1个子集作为训练集，将得到k个模型，用该k个模型的最终训练结果作为参数选择的标准。实际应用中，常取k＝10，即通常所说的10步交叉验证法。上述方法可以有效避免过学习以及欠学习的情况发生，最后得到的训练结果比较具有说服性。
S23、将所述训练集对应的样本数据输入至预设的基于最小二乘支持向量机的第一碳期货价格预测模型中进行训练，得到训练后的第二碳期货价格预测模型；所述第一碳期货价格预测模型是处于初始状态。本步骤中，上述基于最小二乘支持向量机的处于初始状态的第一碳期货价格预测模型即为最原始的数学模型，其没有进行任何训练的模型。上述第二碳期货价格预测模型即为已经经过训练集训练学习后的模型。
S24、将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中验证所述第二碳期货价格预测模型的有效性。本步骤中，即为通过上述的测试集测试第二碳期货价格预测模型是否可以准确测试出碳期货的价格，该准确测试出碳期货的价格不是百分之百的准确，而是允许有一定的偏差，只要在预期的范围之内即可，只有预测的碳期货价格在预期范围内，才会判定第二碳期货价格预测模型为有效模型。
S25、若验证通过，则确定所述第二碳期货价格预测模型为所述碳期货价格预测模型。本步骤中，将第二碳期货价格预测模型确定为上述步骤S2中所述的碳期货价格预测模型，即，可以将第二碳期货价格预测模型拿到实际场景中进行使用，预测市场碳期货价格。
参照图3，在一具体实施例中，上述将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中验证所述第二碳期货价格预测模型的有效性的步骤S24，包括：
S241、将所述测试集对应的样本数据代入到所述第二碳期货价格预测模型中进行计算，得到预测价格；S242、通过均方根误差法计算所述预测价格与实际价格的误差大小，通过平均绝对百分比误差法计算所述预测价格与所述实际价格的偏差程度，以及通过方向正确性法计算所述预测价格的方向正确率；S243、若所述误差大小、偏差程度以及方向正确率均符合预设要求，则判定所述第二碳期货价格预测模 型为有效模型。如上述步骤S241、S242和S243所述，其使用了均方根误差法、平均绝对百分比误差法和方向正确性法，通过不同的维度对测试的预测价格与实际价格进行比较，只有每一个维度均达到预期效果才会判定第二碳期货价格预测模型为有效模型，提高第二碳期货价格预测模型在真正投入使用时，能够提供更加准确的预测结果。
参照图4，在另一具体实施例中，上述根据预设规则确定模型变量的步骤S21，包括：
S211、获取指定时间段内的多种类型的能源期货价格。在本步骤中，选择1年内的期货交易价格数据，主要从洲际交易所(Intercontinental Exchange)搜集碳期货及其影响因素价格历史数据，删去无统计数据，或者数据频度不符的变量。本步骤中主要通过避开第一阶段配额过松导致的价格失常数据以最小化政策及市场监管的影响。本实施例主要选取能源价格作为主要变量，用于价格预测。能源价格则选取欧洲具有代表性的天然气期货价格、煤炭期货价格以及电力期货价格。在欧洲市场上，英国是天然气最大的消费国，所以采用12月份交割的英国天然气期货价格。煤炭期货价格是由洲际交易所(Intercontinental Exchange)提供的12月份交割的Rotterdam煤炭期货价格。电力期货价格则是气候交易所(Intercontinental Exchange)提供的英国电力期货的价格。原油期货价格选取的是布伦特原油期货价格。
S212、检验碳期货价格与各种所述能源期货价格的协整关系，并选取出与碳期货价格的协整关系为正向的能源期货价格。本步骤中，影响碳期货价格波动因素的选择将直接决定模型的预测效果，因此根据定量分析严格筛选影响因素。首先考虑到将要检验碳价与各能源价格之间的协整关系，所以先采用ADF方法对各个序列进行平稳性检验。
表1
Figure PCTCN2018083987-appb-000001
上述表1为平稳性检验结果，通过观察可以发现各序列在1％的显著性水平下不拒绝单位根假设，是非平稳的。而其一阶差分序列在1％的显著性水平下拒绝了单位根假设，这说明各时间序列满足I(1)。其次进行协整检验。根据协整理论，如果两个序列满足单整阶数相同的条件，就有可能通过线性组合构 成低阶单整变量，而且如果两个同阶非平稳序列之间存在协整关系，就表明二者之间存在长期稳定的均衡关系，从而可以避免伪回归的问题。从表1中ADF(单位根检验)单整检验的结果得知，碳价与各种能源价格之间都是一阶单整的，则有可能存在某种平稳的线性组合。这里采用EG(Engle-Granger)两步法对其进行协整检验，以分析它们之间的协整关系。首先，进行线性回归。用普通最小二乘法(OLS)估计能源期货价格和碳期货价格之间的方程，并计算非均衡误差。
能源期货价格与碳期货价格协整模型如下：
EUA t＝α+βE t+γAR(1)+μ t(t＝1,2,3…)
其中E＝煤期货价格，天然气期货价格，电力期货价格，原油期货价格，AR(1)可消除一阶自相关，μ t满足全部基本假定。
表2
Figure PCTCN2018083987-appb-000002
注：小括号内为相应的概率。
上述表2是能源期货价格与碳期货价格协整模型参数。
检验残差是否是平稳序列。ADF检验的结果如表3所示。
表3
Figure PCTCN2018083987-appb-000003
注：显著性水平为1％。
上述表3为各种能源期货价格与碳期货价格协整关系检验结果，检验结果显示，残差序列在显著性水平为1％的情况下均拒绝了原假设，即它们各自的残差序列都是平稳序列。因此碳期货价格和各种能源期货价格之间都存在一种长期稳定的关系，并且煤期货价格、电力期货价格、天然气期货价格以及石油期货价格对碳期货合约价格的影响都是正向的。
S213、将选取出的能源期货价格与碳期货价格进行Granger因果(格兰杰因果关系)检验。本步骤中，由于各能源期货价格与碳期货价格之间存在显著的协整关系，即可以对各组序列进行Granger因果检验，具体判断各种能源期货价格和碳期货价格之间的因果关系：
表4
原假设 F统计量 P值
煤期货价格不是碳期货价格的Granger原因 3.51947 0.0311
碳期货价格不是煤期货价格的Granger原因 0.1963 0.8219
上述表4为煤期货价格和碳期货价格之间的因果关系检验表，可以看出，在5％的置信水平下，碳期货价格和煤期货价格之间存在单向的因果关系。煤期货价格是碳期货价格的Granger原因，而碳期货价格不是煤期货价格的Granger原因。即煤期货价格的变化能够引起碳期货价格的变化，并且两者之间存在显著的长期协整关系。所以可以确定煤期货价格为模型输入变量，来对碳期货价格预测进行建模。
表5
原假设 F统计量 P值
天然气期货价格不是碳期货价格的Granger原因 2.55016 0.1116
碳期货价格不是天然气期货价格的Granger原因 0.80554 0.3703
上述表5为天然气期货价格和碳期货价格之间的因果关系检验表，可以看出，在12％的置信水平下，碳期货价格和天然气期货价格之间存在单向的因果关系。天然气期货价格是碳期货价格的Granger原因，而碳期货价格不是天然气期货价格的Granger原因。即天然气期货价格的变化能够引起碳期货价格的变化，并且两者之间存在显著的长期协整关系。所以可以确定天然气期货价格为模型输入变量，来对碳期货价格预测进行建模。
表6
原假设 F统计量 P值
电力期货价格不是碳期货价格的Granger原因 3.00125 0.0516
碳期货价格不是电力期货价格的Granger原因 1.91802 0.0991
上述表6为电力气期货价格和碳期货价格之间的因果关系检验表，可以看出，在10％置信水平下，碳期货价格和电力期货价格之间存在双向的Granger因果关系，也就是说碳期货价格和电力期货价格之间是相互影响的。根据Granger因果检验，碳期货价格的变动会引起电力期货价格的变动，反之亦然。也就是说两个市场之间已经有了显著的影响。这是因为电厂生命周期一般比较长，技术更新的成本较高，所以往往会购买较多的排放权配额，推动碳价上升，电厂为了转移部分购买配额的额外费用，一般也会增加电价。同时，远期电价也受到天然气价格和即期电价的影响，也就说碳期货价格虽然是影响电力期货价格的重要因素，但不是唯一的因素，而电力期货价格也是影响碳期货价格的一个主要因素。所以可以确定电力期货价格为模型输入变量，来对碳期货价格预测进行建模。
表7
原假设 F统计量 P值
布伦特原油期货价格不是碳期货价格的Granger原因 1.13309 0.3415
碳期货价格不是布伦特原油期货价格的Granger原因 0.87351 0.4804
上述表7为石油气期货价格和碳期货价格之间的因果关系检验表，可以看出，碳期货价格和石油期货价格之间不存在显著的Granger因果关系。这说明，在研究期内，石油期货价格的波动并未影响到碳期货市场。
从期货价格出发，研究发现各种能源期货价格与碳期货价格之间的存在显著的互动关系。总体而言，各种能源期货价格都与碳期货价格之间存在长期均衡的协整关系。从Granger因果关系检验结果来看，碳期货价格与煤期货价格、天然气期货价格以及电力期货价格都存在一定程度的因果关系。尤其是与电力期货价格存在双向的因果关系，可以得知电力期货价格是影响碳期货价格的主要因素。然而，碳期货价格与石油期货价格之间的Granger因果关系并不明显。这是因为影响石油价格的因素众多，同时碳市场和石油市场一样，其价格波动主要石油是由事件驱动，因此两个市场的传导机制还没有有效的建立。
S214、将与所述碳期货价格的Granger因果关系达到预设的显著要求的能源期货价格确定为所述模型变量。本步骤中，上述煤期货价格、天然气期货价格以及电力期货价格与碳期货价格都存在一定程度的Granger因果关系，所以实施例中，可以确定使用煤期货价格、天然气期货价格以及电力期货价格作为基于最小二乘支持向量机的碳期货价格预测模型的模型变量。
在本实施例中，在确定好模型变量之后，获取所述模型变量的样本数据，并将所述样本数据分为训 练集和测试集的步骤S22的具体过程为参数寻优，得到最优参数组。
对于给定的数据样本(x i,y i),i＝1,2,...,n，x i∈R d是与预测量密切相关的影响因素，x i为相关的自变量，y i能源价格，为碳交易价格，对非线性期货价格预测模型，回归函数变为：
Figure PCTCN2018083987-appb-000004
其中，w为权值向量，b是偏执值，是从输入空间到高维特征空间的非线性映射。最小二乘支持向量机优化目标可表示为：
Figure PCTCN2018083987-appb-000005
Figure PCTCN2018083987-appb-000006
其中，ξ i为误差，ξ∈R n*1，C代表的是在线性不可分的情况下，对分类错误的惩罚程度；式(7)可转化为：
Figure PCTCN2018083987-appb-000007
其中，α i(i＝1,…,n)是拉格朗日乘子。根据KKT优化条件：
Figure PCTCN2018083987-appb-000008
Figure PCTCN2018083987-appb-000009
Figure PCTCN2018083987-appb-000010
Figure PCTCN2018083987-appb-000011
消去w和ξ i，则式(4)的解为：
Figure PCTCN2018083987-appb-000012
其中α＝[α 1,α 2,…,α n] T，Q＝[1,1,…,1] T，Y＝[y 1,y 2,…,y n] T，I为单位矩阵，K(x i,x j)为合适的核函数，为了简化计算，使用原空间的核函数来进行计算。
从而得到最优参数组。
然后建立预测模型，其中，最小二乘支持向量机的数学模型为：
Figure PCTCN2018083987-appb-000013
其中，K(x i,x)表示从输入空间到高维特征空间的非线性映射，α i是拉格朗日乘子，b是偏执值，x i是 相关的自变量，α i和b可由解式(5)的线性方程求出。
核函数、映射函数以及特征空间是一一对应的，确定了核函数K(x i,x j)就隐含地确定了映射函数
Figure PCTCN2018083987-appb-000014
和特征空间F。核函数的使用使支持向量机获得了强有力的非线性处理能力，并且避免了在高维特征空间上的复杂计算，有效的克服了维数灾难问题。
上述RBF核函数的表达式为：
Figure PCTCN2018083987-appb-000015
其中，x i是相关的自变量，x j是碳期货价格，σ是核宽度。
在最小二乘支持向量机中，正则化参数和RBF函数的核宽度σ直接影响到最小二乘支持向量机的学习能力和泛化能力，因此，对其正则化参数和核宽度σ的选择方法的研究十分重要。另一方面，在对最小二乘支持向量机的性能影响上，核函数σ和正则化参数之间并无一定的联系，这就使得核函数σ和正则化参数的有效选择成为了焦点。本实施例中，基于交叉验证法选择所述最小二乘支持向量机的正则化参数和RBF核函数的核宽度。
在本实施例中，在步骤S2之后还包括：S3、获取预测周期的时间长度；S4、若所述预测周期属于短期或中期的预测周期，则调用预测的时间序列分析法模型，进行第二次碳期货预测；S5、将通过基于最小二乘支持向量机的碳期货价格预测模型预测的第一预测价格以及基于时间序列分析法模型预测的第二预测价格进行预设的加权平均运算，得到最终的碳期货预测价格。
如上述步骤S3、S4和S5所述，因为基于时间序列分析法模型可以较为准确地预测短期或中期的碳期货价格，所以，可以将其作为碳期货价格预测的一部分，比如，将过基于最小二乘支持向量机的碳期货价格预测模型预测的第一预测价格乘以第一百分比，然后加上基于时间序列分析法模型预测的第二预测价格乘以第二百分比，得到一个综合的碳期货预测价格等。
在另一实施例中，在步骤S2之后还包括：S6、分别从不同的能源交易平台再次获取到指定能源当前的能源期货价格；S7、将在不同交易平台获取到的能源期货价格分别重复上述步骤S2的过程，得到多个碳期货预测价格；S8、将全部预测的碳期货价格进行平均计算得到最终的碳期货价格。
本申请实施例的碳期货价格预测方法，基于最小二乘支持向量机，选择RBF核函数，建立碳期货价格预测的模型，然后对碳期货的价格进行预测，其用等式约束代替标准SVM算法中的不等式约束；将求解二次规划问题转化为直接求解线性方程组；适用于长、中、短期的各时间长度的价格预测，反映信息无重叠，训练样本较小，还不会出现局部极小点而影响最终的预测结果。
参照图5，本申请实施例中还提供一种碳期货价格预测装置，包括：
获取单元10，用于获取指定能源当前的能源期货价格；
预测计算单元20，用于将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF核函数。
在上述获取单元10中，上述指定能源当前的能源期货价格是指在碳期货价格预测时需要的能源期货价格，其价格为当前时间段内可以获取到的实际价格。在一具体实施例中，比如电力期货价格是指定的能源期货价格之一，如果当前时间为电力期货市场开放时间，那么上述当前的期货价格即为当前时刻获取的电力期货价格，如果当前时间为电力期货市场关闭时间，那么上述当前的期货价格即为电力期货市场前一次关闭时的价格。上述指定能源包括多种，比如电力、天然气、碳排放等，其对应的分别为电力期货价格、天然气期货嘉价格和碳期货价格。
在上述预测计算单元20中，上述最小二乘支持向量机是支持向量机的改进，与标准SVM模型比较，其优势明显，具体的：(1)、用等式约束代替标准SVM算法中的不等式约束；(2)、将求解二次规划问题转化为直接求解线性方程组；(3)、最适合于小样本的学习环境。在最小二乘支持向量机中，不同核函数，其性能差异很大，其中，线性核函数不能处理非线性输入值；RBF核函数的高维核参数量少于多项式核函数，同时在SVM训练过程中，采用多项式核函数所需训练时间远大于RBF核函数；采用Sigmoid函数时，某些参数存在错误值；因此，本研究主要采用RBF核函数进行回归建模。
参照图6，本实施例中，上述预测计算单元20，包括：
变量选择模块21，用于根据预设规则确定模型变量。
影响碳期货价格的因素很多，比如，现有的能源市场包括电力期货价格、天然气期货价格、原油期货价格、碳排放价格等，但是这些能源期货价格是否真的都会对碳期货价格产生影响，则需要具体分析，然后才能确定，确定下来的能源期货价格既可以作为模型的变量。
获取划分模块22，用于获取所述模型变量的样本数据，并将所述样本数据分为训练集和测试集。
模型变量确定后，即需要获取到样本数据，获取样本数据的方法包括多种，如通过指定接口直接到第三方交易平台获取，或者接收用户的手动输入值等。本实施例中，样本数据采用交叉验证法获取，交叉验证法的主要思想为：假定原始训练样本数据为N(N为正整数)，被分成k(k为正整数)个不相交的集合(一般是均分，每个集合有N/k个样本)，每一个子集都用来作一次验证集，其余的k-1个子集作为训练集，这样将得到k个模型，用这k个模型的最终训练结果作为参数选择的标准。实际应用中，常取k＝10，即通常所说的10步交叉验证法。上述方法可以有效避免过学习以及欠学习的情况发生，最后得到的训练结果比较具有说服性。
训练模块23，用于将所述训练集对应的样本数据输入至预设的基于最小二乘支持向量机的第一碳期货价格预测模型中进行训练，得到训练后的第二碳期货价格预测模型；所述第一碳期货价格预测模型是处于初始状态。
上述处于初始状态的第一碳期货价格预测模型即为最原始的数学模型，是没有进行过任何训练的模型。上述第二碳期货价格预测模型即为已经经过训练集训练学习后的模型。
验证模块24，用于将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中验证所述 第二碳期货价格预测模型的有效性。
通过上述的测试集测试第二碳期货价格预测模型是否可以准确测试出碳期货的价格，这里的准确测试出碳期货的价格不是百分之百的准确，而是允许有一定的偏差，但是要在预期的范围之内，只有预测的碳期货价格在预期范围内，才会判定第二碳期货价格预测模型为有效模型。
模型确定模块25，用于若验证模块验证第二碳期货价格预测模型的有效性模型，则确定所述第二碳期货价格预测模型为所述碳期货价格预测模型。
将第二碳期货价格预测模型确定为上述预测计算单元20中所述的碳期货价格预测模型，即，可以将第二碳期货价格预测模型拿到实际场景中进行使用，真正的进行碳期货价格。
参照图7，上述验证模块24，包括：
计算价格子模块241，用于将所述测试集对应的样本数据代入到所述第二碳期货价格预测模型中进行计算，得到预测价格；
多维度计算子模块242，用于通过均方根误差法计算所述预测价格与实际价格的误差大小，通过平均绝对百分比误差法计算所述预测价格与所述实际价格的偏差程度，以及通过方向正确性法计算所述预测价格的方向正确率；
判定子模块243，若所述误差大小、偏差程度以及方向正确率均符合预设要求，则判定所述第二碳期货价格预测模型为有效模型。
本实施例中，使用了均方根误差法、平均绝对百分比误差法和方向正确性法，通过不同的维度对测试的预测价格与实际价格进行比较，只有每一个维度均达到预期效果才会判定第二碳期货价格预测模型为有效模型，提高第二碳期货价格预测模型在真正投入使用时，能够提供更加准确的预测结果。
参照图8，本实施例中，上述变量选择模块21，包括：
获取子模块211，用于获取指定时间段内的多种类型的能源期货价格；
第一检验子模块212，用于检验碳期货价格与各种所述能源期货价格的协整关系，并选取出与碳期货价格的协整关系为正向的能源期货价格；
第二检验子模块213，用于将选取出的能源期货价格与碳期货价格进行格兰杰因果检验；
确定子模块214，用于选取出与所述碳期货价格的格兰杰因果关系达到预设的显著要求的能源期货价格，并将该选取出的能源期货价格确定为所述模型变量。
其具体的执行过程与上述的碳期货价格预测方法中的实施例中的实施方式大致相同，在此不在赘述。各种能源期货价格都与碳期货价格之间存在长期均衡的协整关系。从Granger因果关系检验结果来看，碳期货价格与煤期货价格、天然气期货价格以及电力期货价格都存在一定程度的因果关系。尤其是与电力期货价格存在双向的因果关系，可以得知电力期货价格是影响碳期货价格的主要因素。然而，碳期货价格与石油期货价格之间的Granger因果关系并不明显。这是因为影响石油价格的因素众多，同时碳市 场和石油市场一样，其价格波动主要石油是由事件驱动，因此两个市场的传导机制还没有有效的建立。
在本实施例中，在确定好模型变量的之后，获取所述模型变量的样本数据，并将所述样本数据分为训练集和测试集2的具体过程为参数寻优，得到最优参数组。
对于给定的数据样本(x i,y i),i＝1,2,...,n，x i∈R d是与预测量密切相关的影响因素，x i为相关的自变量，y i能源价格，为碳交易价格，对非线性期货价格预测模型，回归函数变为：
Figure PCTCN2018083987-appb-000016
其中，w为权值向量，b是偏执值，是从输入空间到高维特征空间的非线性映射。最小二乘支持向量机优化目标可表示为：
Figure PCTCN2018083987-appb-000017
Figure PCTCN2018083987-appb-000018
其中，ξ i为误差，ξ∈R n*1，C代表的是在线性不可分的情况下，对分类错误的惩罚程度；式(7)可转化为：
Figure PCTCN2018083987-appb-000019
其中，α i(i＝1,…,n)是拉格朗日乘子。根据KKT优化条件：
Figure PCTCN2018083987-appb-000020
Figure PCTCN2018083987-appb-000021
Figure PCTCN2018083987-appb-000022
Figure PCTCN2018083987-appb-000023
消去w和ξ i，则式(4)的解为：
Figure PCTCN2018083987-appb-000024
其中α＝[α 1,α 2,…,α n] T，Q＝[1,1,…,1] T，Y＝[y 1,y 2,…,y n] T，I为单位矩阵，K(x i,x j)为合适的核函数，为了简化计算，使用原空间的核函数来进行计算。
从而得到最优参数组。
然后建立预测模型，其中，最小二乘支持向量机的数学模型为：
Figure PCTCN2018083987-appb-000025
其中，K(x i,x)表示从输入空间到高维特征空间的非线性映射，α i是拉格朗日乘子，b是偏执值，x i是相关的自变量，α i和b可由解式(5)的线性方程求出。
核函数、映射函数以及特征空间是一一对应的，确定了核函数K(x i,x j)就隐含地确定了映射函数
Figure PCTCN2018083987-appb-000026
和特征空间F。核函数的使用使支持向量机获得了强有力的非线性处理能力，并且避免了在高维特征空间上的复杂计算，有效的克服了维数灾难问题。
上述RBF核函数的表达式为：
K(x i，x j)＝exp{-|x i-x j| 2/σ 2}}      (7)
其中，x i是相关的自变量，x j是碳期货价格，σ是核宽度。
在本实施例中，上述碳期货价格预测装置还包括：
获取周期单元，用于获取预测周期的时间长度；
调用单元，用于若所述预测周期属于短期或中期的预测周期，则调用预测的时间序列分析法模型，进行第二次碳期货预测；
加权计算单元，用于将通过基于最小二乘支持向量机的碳期货价格预测模型预测的第一预测价格以及基于时间序列分析法模型预测的第二预测价格进行预设的加权平均运算，得到最终的碳期货预测价格。因为基于时间序列分析法模型可以较为准确地预测短期或中期的碳期货价格，所以，可以将其作为碳期货价格预测的一部分，比如，将过基于最小二乘支持向量机的碳期货价格预测模型预测的第一预测价格乘以第一百分比，然后加上基于时间序列分析法模型预测的第二预测价格乘以第二百分比，得到一个综合的碳期货预测价格等。
在另一实施例中，上述碳期货价格预测装置还包括：
分别获取单元，用于分别从不同的能源交易平台再次获取到指定能源当前的能源期货价格；
分别计算单元，用于将在不同交易平台获取到的能源期货价格分别利用上述预测计算单元20，计算得到多个碳期货预测价格；
平均计算单元，用于将全部预测的碳期货价格进行平均计算得到最终的碳期货价格。
本申请实施例的碳期货价格预测装置，基于最小二乘支持向量机，选择RBF核函数，建立碳期货价格预测的模型，然后对碳期货的价格进行预测，其用等式约束代替标准SVM算法中的不等式约束；将求解二次规划问题转化为直接求解线性方程组；适用于长、中、短期的各时间长度的价格预测，反映信息无重叠，训练样本较小，还不会出现局部极小点而影响最终的预测结果。
参照图9，本申请实施例中还提供一种计算机设备，该计算机设备可以是服务器，其内部结构可以如图9所示。该计算机设备包括通过系统总线连接的处理器、存储器、网络接口和数据库。其中，该计 算机设计的处理器用于提供计算和控制能力。该计算机设备的存储器包括非易失性存储介质、内存储器。该非易失性存储介质存储有操作系统、计算机可读指令和数据库。该内存器为非易失性存储介质中的操作系统和计算机可读指令的运行提供环境。该计算机设备的数据库用于存储碳期货价格预测模型等数据。该计算机设备的网络接口用于与外部的终端通过网络连接通信。该计算机可读指令在执行时，执行如上述各方法的实施例的流程。本领域技术人员可以理解，图9中示出的结构，仅仅是与本申请方案相关的部分结构的框图，并不构成对本申请方案所应用于其上的计算机设备的限定。
本申请一实施例还提供一种计算机非易失性可读存储介质，其上存储有计算机可读指令，该计算机可读指令在执行时，执行如上述各方法的实施例的流程。以上所述仅为本申请的优选实施例，并非因此限制本申请的专利范围，凡是利用本申请说明书及附图内容所作的等效结构或等效流程变换，或直接或间接运用在其他相关的技术领域，均同理包括在本申请的专利保护范围内。

Claims (20)

  1. 一种碳期货价格预测方法，其特征在于，包括：
    获取指定能源当前的能源期货价格；
    将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF核函数。
  2. 根据权利要求1所述的碳期货价格预测方法，其特征在于，所述基于最小二乘支持向量机的碳期货价格预测模型的建模方法，包括：根据预设规则确定模型变量；
    获取所述模型变量的样本数据，并将所述样本数据分为训练集和测试集；
    将所述训练集对应的样本数据输入至预设的基于最小二乘支持向量机的第一碳期货价格预测模型中进行训练，得到训练后的第二碳期货价格预测模型；所述第一碳期货价格预测模型是处于初始状态；
    将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中，验证所述第二碳期货价格预测模型的有效性；
    若验证通过，则确定所述第二碳期货价格预测模型为所述碳期货价格预测模型。
  3. 根据权利要求2所述的碳期货价格预测方法，其特征在于，所述将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中验证所述第二碳期货价格预测模型的有效性的步骤，包括：
    将所述测试集对应的样本数据代入到所述第二碳期货价格预测模型中进行计算，得到预测价格；
    通过均方根误差法计算所述预测价格与实际价格的误差大小，通过平均绝对百分比误差法计算所述预测价格与所述实际价格的偏差程度，以及通过方向正确性法计算所述预测价格的方向正确率；
    若所述误差大小、偏差程度以及方向正确率均符合预设要求，则判定所述第二碳期货价格预测模型为有效模型。
  4. 根据权利要求2所述的碳期货价格预测方法，其特征在于，所述根据预设规则确定模型变量的步骤，包括：
    获取指定时间段内的多种类型的能源期货价格；
    检验碳期货价格与各种所述能源期货价格的协整关系，并选取出与碳期货价格的协整关系为正向的能源期货价格；
    将选取出的能源期货价格与碳期货价格进行格兰杰因果检验；
    选取出与所述碳期货价格的格兰杰因果关系达到预设的显著要求的能源期货价格，并将该选取出的能源期货价格确定为所述模型变量。
  5. 根据权利要求1所述的碳期货价格预测方法，其特征在于，所述最小二乘支持向量机的数学模型为：
    Figure PCTCN2018083987-appb-100001
    其中，K(x i,x)表示从输入空间到高维特征空间的非线性映射，α i是拉格朗日乘子，b是偏执值，x i是相关的自变量。
  6. 根据权利要求5所述的碳期货价格预测方法，其特征在于，所述RBF核函数的表达式为：
    K(x i,x j)＝exp{-|x i-x j| 2/σ 2}}
    其中，x i是相关的自变量，x j是碳期货价格，σ是核宽度。
  7. 一种碳期货价格预测装置，其特征在于，包括：
    获取单元，用于获取指定能源当前的能源期货价格；
    预测计算单元，用于将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF核函数。
  8. 根据权利要求7所述的碳期货价格预测装置，其特征在于，所述预测计算单元，包括：
    变量选择模块，用于根据预设规则确定模型变量；
    获取划分模块，用于获取所述模型变量的样本数据，并将所述样本数据分为训练集和测试集；
    训练模块，用于将所述训练集对应的样本数据输入至预设的基于最小二乘支持向量机的第一碳期货价格预测模型中进行训练，得到训练后的第二碳期货价格预测模型；所述第一碳期货价格预测模型是处于初始状态；
    验证模块，用于将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中，验证所述第二碳期货价格预测模型的有效性；
    模型确定模块，用于若验证模块验证第二碳期货价格预测模型的有效性模型，则确定所述第二碳期货价格预测模型为所述碳期货价格预测模型。
  9. 根据权利要求8所述的碳期货价格预测装置，其特征在于，所述验证模块，包括：
    计算价格子模块，用于将所述测试集对应的样本数据代入到所述第二碳期货价格预测模型中进行计算，得到预测价格；
    多维度计算子模块，用于通过均方根误差法计算所述预测价格与实际价格的误差大小，通过平均绝对百分比误差法计算所述预测价格与所述实际价格的偏差程度，以及通过方向正确性法计算所述预测价格的方向正确率；
    判定子模块，用于若所述误差大小、偏差程度以及方向正确率均符合预设要求，则判定所述第二碳期货价格预测模型为有效模型。
  10. 根据权利要求8所述的碳期货价格预测装置，其特征在于，所述变量选择模块，包括：
    获取子模块，用于获取指定时间段内的多种类型的能源期货价格；
    第一检验子模块，用于检验碳期货价格与各种所述能源期货价格的协整关系，并选取出与碳期货价格的协整关系为正向的能源期货价格；
    第二检验子模块，用于将选取出的能源期货价格与碳期货价格进行格兰杰因果检验；
    确定子模块，用于选取出与所述碳期货价格的格兰杰因果关系达到预设的显著要求的能源期货价格，并将该选取出的能源期货价格确定为所述模型变量。
  11. 根据权利要求7所述的碳期货价格预测装置，其特征在于，所述最小二乘支持向量机的数学模型为：
    Figure PCTCN2018083987-appb-100002
    其中，K(x i,x)表示从输入空间到高维特征空间的非线性映射，α i是拉格朗日乘子，b是偏执值，x i是相关的自变量。
  12. 根据权利要求11所述的碳期货价格预测装置，其特征在于，所述RBF核函数的表达式为：
    K(x i,x j)＝exp{-|x i-x j| 2/σ 2}}
    其中，x i是相关的自变量，x j是碳期货价格，σ是核宽度。
  13. 一种计算机设备，包括存储器和处理器，所述存储器存储有计算机可读指令，其特征在于，所述处理器执行所述计算机可读指令时实现碳期货价格预测方法，该碳期货价格预测方法，包括：
    获取指定能源当前的能源期货价格；
    将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF核函数。
  14. 根据权利要求13的计算机设备，其特征在于，所述基于最小二乘支持向量机的碳期货价格预测模型的建模方法，包括：根据预设规则确定模型变量；
    获取所述模型变量的样本数据，并将所述样本数据分为训练集和测试集；
    将所述训练集对应的样本数据输入至预设的基于最小二乘支持向量机的第一碳期货价格预测模型中进行训练，得到训练后的第二碳期货价格预测模型；所述第一碳期货价格预测模型是处于初始状态；
    将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中，验证所述第二碳期货价格预测模型的有效性；若验证通过，则确定所述第二碳期货价格预测模型为所述碳期货价格预测模型。
  15. 根据权利要求14所述的计算机设备，其特征在于，所述将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中验证所述第二碳期货价格预测模型的有效性的步骤，包括：
    将所述测试集对应的样本数据代入到所述第二碳期货价格预测模型中进行计算，得到预测价格；
    通过均方根误差法计算所述预测价格与实际价格的误差大小，通过平均绝对百分比误差法计算所述预测价格与所述实际价格的偏差程度，以及通过方向正确性法计算所述预测价格的方向正确率；
    若所述误差大小、偏差程度以及方向正确率均符合预设要求，则判定所述第二碳期货价格预测模型为有效模型。
  16. 根据权利要求14所述的计算机设备，其特征在于，所述根据预设规则确定模型变量的步骤，包括：获取指定时间段内的多种类型的能源期货价格；
    检验碳期货价格与各种所述能源期货价格的协整关系，并选取出与碳期货价格的协整关系为正向的能源期货价格；将选取出的能源期货价格与碳期货价格进行格兰杰因果检验；
    选取出与所述碳期货价格的格兰杰因果关系达到预设的显著要求的能源期货价格，并将该选取出的能源期货价格确定为所述模型变量
  17. 一种计算机非易失性可读存储介质，其上存储有计算机可读指令，其特征在于，所述计算机可读指令被处理器执行时实现碳期货价格预测方法，该碳期货价格预测方法，包括：
    获取指定能源当前的能源期货价格；
    将所述当前的能源期货价格输入到预设的基于最小二乘支持向量机的碳期货价格预测模型中，进行价格预测；其中，所述碳期货价格预测模型的核函数是RBF核函数。
  18. 根据权利要求17的计算机非易失性可读存储介质，其特征在于，所述处理器建立基于最小二乘支持向量机的碳期货价格预测模型的建模方法，包括：根据预设规则确定模型变量；
    获取所述模型变量的样本数据，并将所述样本数据分为训练集和测试集；
    将所述训练集对应的样本数据输入至预设的基于最小二乘支持向量机的第一碳期货价格预测模型中进行训练，得到训练后的第二碳期货价格预测模型；所述第一碳期货价格预测模型是处于初始状态；
    将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中，验证所述第二碳期货价格预测模型的有效性；若验证通过，则确定所述第二碳期货价格预测模型为所述碳期货价格预测模型。
  19. 根据权利要求18所述的计算机非易失性可读存储介质，其特征在于，所述处理器将所述测试集对应的样本数据输入至所述第二碳期货价格预测模型中验证所述第二碳期货价格预测模型的有效性的步骤，包括：将所述测试集对应的样本数据代入到所述第二碳期货价格预测模型中进行计算，得到预测价格；
    通过均方根误差法计算所述预测价格与实际价格的误差大小，通过平均绝对百分比误差法计算所述预测价格与所述实际价格的偏差程度，以及通过方向正确性法计算所述预测价格的方向正确率；
    若所述误差大小、偏差程度以及方向正确率均符合预设要求，则判定所述第二碳期货价格预测模型为有效模型。
  20. 根据权利要求18所述的计算机非易失性可读存储介质，其特征在于，所述处理器根据预设规则确定模型变量的步骤，包括：获取指定时间段内的多种类型的能源期货价格；
    检验碳期货价格与各种所述能源期货价格的协整关系，并选取出与碳期货价格的协整关系为正向的 能源期货价格；将选取出的能源期货价格与碳期货价格进行格兰杰因果检验；
    选取出与所述碳期货价格的格兰杰因果关系达到预设的显著要求的能源期货价格，并将该选取出的能源期货价格确定为所述模型变量。
PCT/CN2018/083987 2018-02-27 2018-04-20 碳期货价格预测方法、装置、计算机设备和存储介质 Ceased WO2019165692A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201810162631.1A CN108197762A (zh) 2018-02-27 2018-02-27 碳期货价格预测方法、装置、计算机设备和存储介质
CN201810162631.1 2018-02-27

Publications (1)

Publication Number Publication Date
WO2019165692A1 true WO2019165692A1 (zh) 2019-09-06

Family

ID=62593946

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2018/083987 Ceased WO2019165692A1 (zh) 2018-02-27 2018-04-20 碳期货价格预测方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN108197762A (zh)
WO (1) WO2019165692A1 (zh)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733437A (zh) * 2020-12-29 2021-04-30 哈尔滨工业大学 一种厂网河一体化模型参数优化方法

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110288482B (zh) * 2019-07-02 2021-03-30 欧冶云商股份有限公司 钢厂期货交易方法及系统
CN112862413A (zh) * 2021-03-30 2021-05-28 普洛斯科技(重庆)有限公司 承运订单的校验方法及装置、电子设备、存储介质
CN114493054A (zh) * 2022-04-18 2022-05-13 广东埃文低碳科技股份有限公司 一种基于大数据的碳交易市场智能分析方法及系统

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102982229A (zh) * 2012-09-06 2013-03-20 淮阴工学院 一种基于神经网络的多品种商品价格预测的数据预处理方法
CA2847044A1 (en) * 2013-03-15 2014-09-15 Open Access Technology International, Inc. Systems and methods for tracking greenhouse gas emissions
CN105160204A (zh) * 2015-10-28 2015-12-16 河海大学 一种碳排放价格组合预测方法
CN105787599A (zh) * 2016-03-02 2016-07-20 上海源庐加佳信息科技有限公司 一种甲醇期货价格预测方法

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102982229A (zh) * 2012-09-06 2013-03-20 淮阴工学院 一种基于神经网络的多品种商品价格预测的数据预处理方法
CA2847044A1 (en) * 2013-03-15 2014-09-15 Open Access Technology International, Inc. Systems and methods for tracking greenhouse gas emissions
CN105160204A (zh) * 2015-10-28 2015-12-16 河海大学 一种碳排放价格组合预测方法
CN105787599A (zh) * 2016-03-02 2016-07-20 上海源庐加佳信息科技有限公司 一种甲醇期货价格预测方法

Non-Patent Citations (2)

* Cited by examiner, † Cited by third party
Title
MING-TANG TSAI ET AL.: "Application of Radial Basis Function Neural Network for Carbon Price forecasting", APPLIED MECHANICS AND MATERIALS, 30 June 2014 (2014-06-30), ISSN: 1662-7482 *
ZHU, ET AL: "Bookmark and Share Email this article Print this article Carbon price prediction based on integration of GMDH, particle swarm optimization and least squares support vector machines", SYSTEMS ENGINEERING-THEORY & PRACTICE, vol. 31, no. 12, 15 December 2011 (2011-12-15), pages 2264 - 2271, ISSN: 1000-6788 *

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112733437A (zh) * 2020-12-29 2021-04-30 哈尔滨工业大学 一种厂网河一体化模型参数优化方法

Also Published As

Publication number Publication date
CN108197762A (zh) 2018-06-22

Similar Documents

Publication Publication Date Title
US20220092346A1 (en) Computer-based systems, computing components and computing objects configured to implement dynamic outlier bias reduction in machine learning models
WO2022060411A1 (en) Computer-based systems, computing components and computing objects configured to implement dynamic outlier bias reduction in machine learning models
Đozić et al. Application of artificial neural networks for testing long-term energy policy targets
CN110826774B (zh) 母线负荷预测方法、装置、计算机设备及存储介质
Nozhati et al. Stochastic optimal control methodologies in risk-informed community resilience planning
WO2019165692A1 (zh) 碳期货价格预测方法、装置、计算机设备和存储介质
CN113469570A (zh) 信息质量评价模型构建方法、装置、设备及存储介质
CN116091118A (zh) 电价预测方法、装置、设备、介质及产品
Köhnen et al. The potential of deep learning to reduce complexity in energy system modeling
Kavaklioglu Principal components based robust vector autoregression prediction of Turkey’s electricity consumption: K. Kavaklioglu
CN113313562A (zh) 产品数据处理方法、装置、计算机设备和存储介质
Talebi et al. Graph neural networks for efficient ac power flow prediction in power grids
CN112800674A (zh) 一种基于极限学习机的电力系统元件可靠性参数计算方法
CN119323007A (zh) 基于Mamba架构的霍克斯过程建模方法、装置、设备及介质
CN117372070A (zh) 地产市场趋势的确定方法、装置、计算机设备及存储介质
CN118552238A (zh) 基于售电量预测的决策方法、决策装置和决策系统
US20160110641A1 (en) Determining a level of risk for making a change using a neuro fuzzy expert system
Volodina et al. Propagating uncertainty in a network of energy models
Oshodi et al. Using an adaptive neuro-fuzzy inference system for tender price index forecasting: A univariate approach
Koopman et al. Modified efficient importance sampling for partially non‐Gaussian state space models
CN119850190B (zh) 抽水蓄能设备检修策略评价方法、装置、计算机设备、可读存储介质和程序产品
Georgieva et al. Software reliability assessment via fuzzy logic model
Mantri A data-driven approach to predict the load profile at an electric vehicle charging station
US20250088563A1 (en) Counterfactual analysis of user models for prescriptive output
Kagrecha et al. Adaptive Crowdsourcing Via Self-Supervised Learning

Legal Events

Date Code Title Description
NENP Non-entry into the national phase

Ref country code: DE

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 21/12/2020)

122 Ep: pct application non-entry in european phase

Ref document number: 18907568

Country of ref document: EP

Kind code of ref document: A1