Product oil supply and demand early warning method and system

Abstract

The invention provides a supply and demand early warning method and system for product oil. Screening first data, consistent data and lagged data from the first feature pool, the consistent feature pool and the lagged feature pool through a time difference test method; according to the method, the supply and demand early warning data and the business comprehensive index are constructed by performing feature priority measurement on the advanced feature pool to screen features and construct the business comprehensive index, the supply and demand early warning data qualitatively indicate the supply and demand state and predict the direction of the price arrival rate, and the business comprehensive index quantitatively predicts the business interval where the price arrival rate is located. Through combined application of the two methods, a current state can be represented, a future supply and demand change direction can be predicted, an important reference is provided for making a production and operation plan, making a sales strategy and optimizing inventory operation, the problem of error transmission is eliminated, and resource waste caused by incapability of supply or greater supply than demand due to sudden demand change of the product oil is prevented.

Description

Technical Field

[0001] The present invention belongs to the field of energy early warning, and in particular relates to a method and system for early warning of supply and demand of refined oil. Background Art

[0002] In 2020, there will be a surplus of about 62.6 million tons of refined oil in China. In 2025, the domestic refined oil production capacity will reach about 1 billion tons. In the future, the increase in domestic refined oil supply will significantly exceed the increase in demand, and the contradiction of resource surplus will be further aggravated. From the perspective of supply and demand, domestic refined oil has entered a period of stock competition, and the competition among the main players has shown a trend of one rising and the other falling. The sales volume and market share that companies have expanded in the market all come from the parts lost by their competitors. Therefore, it is imperative to build early warning data on the supply and demand of refined oil to guide companies to adjust production load in a timely manner and make reasonable inventory and sales plans, which is of great significance for improving quality and efficiency.

[0003] The existing methods mainly predict the supply and demand of refined oil products first, and then judge the supply and demand tension. From the practical results, before 2014, it was not obvious that the supply and demand period had not entered. Simply using the supply and demand ratio could determine whether the supply and demand matched. However, since 2014, it has clearly entered the stage of oversupply, and simply judging by the supply and demand ratio can no longer represent the status. In addition, predicting the supply and demand of refined oil products first, and then judging the supply and demand is prone to error transmission. Summary of the invention

[0004] In view of the above problems, the present invention proposes a method and system for early warning of supply and demand of refined oil. It can characterize the current state and predict the direction of future supply and demand changes, providing an important reference for making production and operation plans, formulating sales strategies, and optimizing inventory operations, eliminating the problem of error transmission, and preventing sudden changes in demand for refined oil from causing a shortage of supply or oversupply, resulting in a waste of resources.

[0005] The technical solution of the present invention is: a method for early warning of supply and demand of refined oil, comprising:

[0006] S101: Obtaining refined oil supply and demand data, and establishing a supply and demand early warning database based on the characteristics of the oil-using industry;

[0007] S102: dividing the supply and demand early warning database into a leading feature pool, a consistent feature pool and a lagging feature pool through time difference analysis, BB peak-valley method and KL information method;

[0008] S103: Filtering supply and demand warning data according to the preceding feature pool, the consistent feature pool and the lagging feature pool; wherein the preceding data, the consistent data and the lagging data are filtered from the preceding feature pool, the consistent feature pool and the lagging feature pool by a time difference inspection method;

[0009] S104: Filtering features by measuring the feature priority of the preceding feature pool, constructing a comprehensive prosperity index, and predicting the price availability rate by using the comprehensive prosperity index;

[0010] S105: The supply and demand warning data and the comprehensive prosperity index are combined to represent the warning situation of the supply and demand of refined oil.

[0011] Furthermore, the feature data in the leading feature pool, the consistent feature pool and the lagging feature pool are screened by the time difference test method, including: screening leading features, consistent features and lagging features that have a significant time difference relationship with the availability rate of gasoline and diesel prices.

[0012] Furthermore, after dividing the warning feature pool into a leading feature pool, a consistent feature pool and a lagging feature pool by using time difference analysis, BB peak-valley method and KL information method, the method further includes:

[0013] The standard change rate Sij of each feature is obtained by the American composite index method, including: obtaining the symmetric change rate of the feature;

[0014] The symmetric change rate is standardized, and the average change rate of each group of features is calculated based on the standardized symmetric change rate.

[0015] Furthermore, the method of obtaining each characteristic standard change rate Sij by the American composite index method also includes: obtaining an initial composite index, obtaining an index trend according to the composite index and adjusting the trend, and calculating a composite feature.

[0016] Furthermore, the method further comprises:

[0017] The importance of each feature is determined by principal component analysis, including:

[0018] The weight is taken based on the inverse of the KL information to obtain the standardized average change rate Rj of each feature group.

[0019] Furthermore, the method further comprises:

[0020] According to the features screened by the preceding feature pool, the consistent feature pool and the lagging feature pool, a principal component score Pj of each feature combination is obtained by a rolling principal component analysis method.

[0021] Furthermore, the method further comprises:

[0022] The standardized average change rate Vj is calculated based on the standardized average change Rj of each feature group and the principal component score Pj of the feature group, and the synthesis index lj is calculated based on the standardized average change rate Vj.

[0023] Furthermore, the method further comprises:

[0024] According to the composite index lj, the divided leading data, consistent data and lagged data are selected by BB peak-valley method, KL information method and time difference correlation coefficient method;

[0025] The selected leading data, consistent data and lagging data are used as supply and demand warning data;

[0026] The supply and demand warning situation of refined oil is characterized according to the supply and demand warning data.

[0027] Based on the same inventive concept, the present invention also provides a supply and demand early warning system for refined oil, comprising:

[0028] The acquisition module is used to obtain the supply and demand data of refined oil products and establish a supply and demand early warning database based on the characteristics of the oil-using industry;

[0029] A division module is used to divide the warning feature pool into a leading feature pool, a consistent feature pool and a lagging feature pool through time difference analysis, BB peak-valley method and KL information method;

[0030] A screening module, used to screen the leading features according to the leading feature pool to construct supply and demand warning data; wherein the leading data, consistent data and lagging data are screened from the leading feature pool, the consistent feature pool and the lagging feature pool by a time difference inspection method;

[0031] A prediction module, used for constructing a comprehensive prosperity index by measuring and screening features based on feature priority of the preceding feature pool;

[0032] The characterization module is used to characterize the warning situation of the supply and demand of refined oil by combining the supply and demand warning data and the comprehensive prosperity index.

[0033] Based on the same inventive concept, the present invention also provides an electronic device, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0034] Memory for storing computer programs;

[0035] The processor is used to implement the supply and demand early warning method of refined oil when executing the program stored in the memory.

[0036] Based on the same inventive concept, the present invention also provides a computer-readable storage medium on which a computer program is stored, and when the computer program is executed by a processor, a method for early warning of supply and demand of refined oil is implemented.

[0037] Beneficial effects of the present invention: The present invention constructs supply and demand early warning data and a comprehensive prosperity index, wherein the supply and demand early warning data qualitatively indicates the supply and demand status and predicts the direction of the price availability rate, and the comprehensive prosperity index quantitatively predicts the prosperity interval of the price availability rate. The combined application of the two can characterize the current status and predict the direction of future supply and demand changes, providing an important reference for making production and operation plans, formulating sales strategies, and optimizing inventory operations, eliminating the problem of error transmission, and also preventing the sudden change in demand for refined oil from causing a waste of resources due to a failure to supply or oversupply.

[0038] Other features and advantages of the present invention will be described in the following description, and partly become obvious from the description, or be understood by implementing the present invention. The purpose and other advantages of the present invention can be realized and obtained by the structures pointed out in the description and the drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A schematic diagram of the method for early warning of supply and demand and price prediction of refined oil of the present invention is shown;

[0041] Figure 2 A schematic diagram of a construction path for the supply and demand early warning data of refined oil products of the present invention is shown;

[0042] Figure 3 The basic process diagram of seasonal adjustment of refined oil products of the present invention is shown;

[0043] Figure 4 The data graphs of the gasoline price of the present invention showing the leading, consistent and lagging arrival;

[0044] Figure 5 The data graphs of the diesel price of the present invention showing the first, the same and the lagging;

[0045] Figure 6 The data graphs of the leading, consistent and lagging growth rates of gasoline price arrival rates in the same period of the present invention are shown;

[0046] Figure 7 The data graphs of the leading, consistent and lagging growth rates of the diesel price availability rate in the same period of the present invention are shown;

[0047] Figure 8 The relationship between the gasoline price availability rate and the advance warning index of the present invention is shown;

[0048] Fig. 9 The preceding data graph of the gasoline price arrival rate growth rate during the same period of the present invention is shown;

[0049] Fig.10 The consistent data graph of the diesel price availability rate during the same period of the present invention is shown;

[0050] Fig.11 The present invention shows the lagged data graph of gasoline price arrival rate growth rate over the same period;

[0051] Fig.12 The relationship diagram between the gasoline price availability rate and the supply-demand ratio of the present invention is shown;

[0052] Fig.13 The relationship diagram between the diesel price availability rate and the supply-demand ratio of the present invention is shown;

[0053] Fig.14 A schematic diagram of the process of selecting the economic characteristics of the present invention is shown;

[0054] Fig.15 A schematic diagram of the supply and demand early warning system for refined oil of the present invention is shown;

[0055] Fig.16 A schematic diagram of an electronic device in the present invention is shown. DETAILED DESCRIPTION

[0056] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0057] It should be noted that the terms "first", "second" etc. in the present application are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequential order. It should be understood that the data used in this way can be interchanged in appropriate circumstances, so that the embodiments of the present application described here. In the present application, the orientation or positional relationship indicated by the terms "upper", "lower", "left", "right", "front", "back", "top", "bottom", "inside", "outside", "middle", "vertical", "horizontal", "lateral", "longitudinal" etc. are based on the orientation or positional relationship shown in the accompanying drawings.

[0058] The embodiment of the present invention provides a method for early warning of supply and demand of refined oil, see Figure 1 ,include:

[0059] S101: Obtaining refined oil supply and demand data, and establishing a supply and demand early warning database based on the characteristics of the oil-using industry;

[0060] S102: Dividing the warning feature pool into a leading feature pool, a consistent feature pool and a lagging feature pool by time difference analysis, BB peak-valley method and KL information method;

[0061] S103: Filtering the leading features according to the leading feature pool to construct supply and demand warning data; wherein the leading data, the consistent data and the lagging data are filtered from the leading feature pool, the consistent feature pool and the lagging feature pool by a time difference inspection method;

[0062] S104: constructing a comprehensive prosperity index by measuring the feature priority of the preceding feature pool and screening the features;

[0063] S105: Characterizing the supply and demand situation of refined oil by combining the supply and demand warning data and the comprehensive prosperity index.

[0064] Among them, the screening of feature data in the leading feature pool, the consistent feature pool and the lagging feature pool by the time difference test method includes: screening leading features, consistent features and lagging features that have a significant time difference relationship with the availability rate of gasoline and diesel prices.

[0065] In some optional embodiments, after the warning feature pool is divided into a leading feature pool, a consistent feature pool and a lagging feature pool by time difference analysis, BB peak-valley method and KL information method, each feature standard change rate Sij is obtained by the American composite index method.

[0066] The method includes: obtaining the symmetric change rate of the feature and standardizing the symmetric change rate, and obtaining the average change rate of each group of features according to the standardized symmetric change rate.

[0067] The method of obtaining each characteristic standard change rate Sij by the American composite index method further includes: obtaining an initial composite index, obtaining an index trend according to the composite index and adjusting the trend, and calculating a composite characteristic.

[0068] Specifically, there are five steps in the calculation of synthetic features: S11: finding the symmetrical change rate of the feature and standardizing it; S12: finding the standardized average change rate of each feature group; S13: finding the initial synthetic index; S14: trend adjustment; S15: calculating the synthetic feature. Since the present invention constructs the synthetic index based on the periodic fluctuation of the growth rate, trend adjustment is not required, so the trend adjustment is not performed in the calculation process of the present invention, and 3-5 are for calculating the synthetic features.

[0069] Specifically, find the symmetric change rate of the feature and standardize it. Let feature Y ij(t) is the i-th feature of the j-th feature group, j=1, 2, 3 are the leading, consistent, and lagging feature groups respectively, i=1, 2, …, k j ,k j is the number of features in the jth feature group, then feature Y ij The symmetric rate of change of (t) C ij (t) is:

[0070]

[0071] When the feature Y ij When (t) has zero or negative value or the feature is a ratio sequence, take the first-order difference

[0072] C ij (t) = Y ij (t)-Y ij (t-1)(t=2,3,…,T).

[0073] In order to prevent the features with large fluctuation range from dominating the composite index, the symmetrical change rate C of each feature is ij (t) are all standardized so that their average absolute value is equal to 1. First, find the normalization factor A ij :

[0074]

[0075] Use A ij C ij (t) is standardized to obtain the standardized rate of change S ij (t):

[0076]

[0077] Find the standardized average change rate V of each feature group j (t)

[0078] First, find the average rate of change R within different feature groups j (t)

[0079]

[0080] Among them, ω ij is the weight of the i-th feature of the j-th group. Considering that the KL information reflects the shape similarity of each feature, the smaller the value, the higher the similarity. Therefore, here we take the reciprocal of the KL information value of each feature in each feature group and then calculate the weight ω of each feature. ij (t), the reason why weights change over time is that different features may have missing data in different periods, so the sum of weights is the sum of features with data.

[0081] Calculate the exponential normalization factor F j

[0082]

[0083] Among them, F2 = 1,

[0084] Finally, calculate the standardized average rate of change V j (t)

[0085] V j (t) = R j (t) / F j (t=2,3,...,T).

[0086] Find the composite index I j (t),

[0087] Order I j (1) = 100, and the initial synthesis index I is obtained. j (t):

[0088]

[0089] Create a composite index CI with the base period (base year) as 100 j (t):

[0090]

[0091] in, For I j (t) Average value in the base year.

[0092] In some optional embodiments, the importance of each feature is obtained by principal component analysis to determine the weight, including: taking the weight based on the inverse of KL information to obtain the standardized average change rate Rj of each feature group.

[0093] It should be noted that the present invention uses the inverse of the KL information amount to determine the weight of each feature, and the principal component analysis method is applied here to determine the weight of each feature obtained through the PCA process. The main idea of ​​the principal component analysis method is to combine several features that have a certain internal connection in the fluctuation trend, through a series of operations, into a new set of comprehensive features, and then use these new features to describe the overall change trend of all the original features. From a popular understanding, the principal component analysis method can be regarded as a "dimensionality reduction" process, that is, the original multi-dimensional features are converted into features of a few dimensions after changes, and the main factors behind the changes of all features are extracted, thereby reducing the difficulty of analyzing the overall change trend.

[0094] Among them, according to the features screened by the preceding feature pool, the consistent feature pool and the lagging feature pool, the principal component scores Pj of the feature combinations are obtained by the rolling principal component analysis method.

[0095] The standardized average change rate Vj is calculated based on the standardized average change Rj of each feature group and the principal component score Pj of the feature group, and the synthesis index lj is calculated based on the standardized average change rate Vj.

[0096] Specifically, the specific steps of the composite index method based on principal component analysis are as follows:

[0097] S11: Preprocess the initial data of the selected features. Defined as:

[0098]

[0099] in, Represents the original data of the jth feature at the i-th moment; the matrix row vector It represents the data of all features corresponding to the same time node, that is, cross-sectional data; the matrix column vector It represents the data of the same feature at each time point, that is, time series data. The initial data matrix is ​​preprocessed by Z-score standardization to obtain the standardized matrix X = (x ij ) (n×m) , the calculation formula for the standardized transformation is:

[0100]

[0101] Where: x ij Represents the standardized data of the jth feature at the i-th moment; and The original data matrix The sample mean and standard deviation of the j-th column of data in .

[0102] S12: Find the relationship matrix R between features. Wherein, the relationship matrix and R = (r gk ) m×m The element r in gk It represents the characteristic variable after Z-score standardization and The specific calculation formula of the correlation coefficient is:

[0103]

[0104] S13: Calculate the variance contribution rate and cumulative variance contribution rate of each feature. By solving the characteristic equation |(λE-|R|)|=0, the eigenvalue λ of the relationship matrix R can be obtained.g (g=1,2,…,m) and all its corresponding eigenvectors u g =(u1g,u2g,…,u m g) T According to the characteristics of the relationship matrix R, its eigenvalue λ g They must all be greater than zero, and the obtained λ g Rearrange according to the rule from large to small, that is, λ1≥λ2≥…≥λ g ≥ 0. Furthermore, the contribution rate of each component is the percentage of the corresponding eigenvalue to the sum of all eigenvalues, that is, the contribution rate Z of the g-th eigenvariable g And the cumulative contribution rate W g They are:

[0105]

[0106]

[0107] S14: Determine the principal component. According to the principle of determining the number of principal components in the principal component analysis theory, it is necessary to select the principal component with an eigenvalue greater than 1, or directly take the eigenvalues ​​corresponding to the first t eigenvalues ​​with a cumulative contribution rate greater than 70% to 100% as the principal component, denoted as y l (l=1,2,…,t):

[0108]

[0109] Where: is the row vector of the normalized matrix X.

[0110] S15: Calculate the score coefficients of all principal components. The component score coefficients F of p principal components can be obtained by dividing the eigenvector of each principal component by the square root of the corresponding eigenvalue: l , the specific calculation formula is:

[0111]

[0112] S16: Calculate the scores of all principal components. Component score coefficient Q l It is the elements x in the original matrix after Z-score standardization. ij The component score coefficient F of each principal component l Multiplying together gives:

[0113] Q l =x ij XF l , l=1,2,...,t.

[0114] S17: Calculate the comprehensive score. Linearly weight the scores of the selected p principal components according to their corresponding contribution rates to obtain the final comprehensive score of all principal components:

[0115]

[0116] Where: Z l W represents the variance contribution rate of the lth principal component calculated by the formula. t It represents the cumulative variance contribution rate of the tth principal component calculated by the formula, that is, the sum of the variance contributions of the t selected principal components.

[0117] S18: Calculate the index normalization factor F j

[0118]

[0119] Among them, F2 = 1

[0120] Finally, calculate the standardized average rate of change V j (t)

[0121] V j (t) = R j (t) / F j (t=2, 3, ..., T).

[0122] S19: Find the composite index I j (t)

[0123] Order I j (1) = 100, and the initial synthesis index I is obtained. j (t):

[0124]

[0125] Create a composite index CI with the base period (base year) as 100 j (t):

[0126]

[0127] in, For I j (t) Average value in the base year.

[0128] In some optional embodiments, according to the composite index lj, the divided leading data, consistent data and lagging data are selected by using the BB peak-valley method, the KL information method and the time difference correlation coefficient method;

[0129] The selected leading data, consistent data and lagging data are used as supply and demand warning data;

[0130] The supply and demand warning situation of refined oil is characterized according to the supply and demand warning data.

[0131] It should be noted that the steps to construct the refined oil supply and demand early warning data are: first, collect independent variables that can objectively reflect changes and construct an independent variable feature pool; second, use corresponding methods such as time difference analysis to divide the independent variables into leading, consistent, and lagging characteristics; third, determine the weight of each feature, and use the composite index method to calculate the composite index.

[0132] See also Figure 2 , the method for determining key features such as feature weights is as follows:

[0133] S21: When dividing the leading, consistent, and lagging feature pools, the BB peak-valley method, KL information method, and time difference correlation coefficient method are used. In addition, after the composite index is calculated, the composite index is tested using three methods.

[0134] S22: In the calculation method of the US composite index, the inverse of the KL information value of each feature is used as the weight of each feature in each feature group. In the composite index of the principal component analysis method, due to the long time span, the influencing factors of the independent variables in different time periods are different. Therefore, a rolling method is used to obtain the comprehensive principal component score of each feature group at each time point. The rolling period is set to 18 months, and the rolling filling method uses the data at the last time point of each rolling period.

[0135] S23: When testing the composite index, select the composite index that meets the attribute characteristics according to the principle recommended by the majority method. Among them, the leading data meets the recommendation of the majority method and its lead month is greater than or equal to 3 months, the consistent data meets the recommendation of the majority method and its delayed month is between [-2,2], and the lagged data meets the recommendation of the majority method and its lagged month is greater than or equal to 3 months. The critical value will be adjusted appropriately according to feasibility. If there are multiple composite indexes that meet the conditions, the final composite index will be selected according to the principle of minimum KL information value.

[0136] It should be noted that in order to ensure that the model can be put into practical application, when building the feature pool, the independent variable features to be selected need to meet the following conditions:

[0137] The first is relevance. The independent variable features must be related to the supply and demand of refined oil. When selecting such features, we can consider the impact on the supply and demand of refined oil. For example, tractor sales are selected as an important factor affecting the demand side; the average exchange rate will affect the cost of imported crude oil for local refineries to a certain extent. However, when selecting features, we must consider the transmission effect. Although some features are also located in the industrial chain, their effects have obvious time lags and cannot be included in the candidate interval. For example, the number of new crude oil wells may affect the future supply of refined oil to a certain extent, but since crude oil will go through a series of complex processes from drilling to mining to refining, it takes a long time and has no obvious correlation, so it is not considered as a candidate feature.

[0138] The second is availability. The selected features must be easy to obtain. If data is often missing, the effectiveness of the model will be greatly reduced. Therefore, try to select independent variables that can be obtained in a timely manner through official channels.

[0139] The third is high frequency. In order to better guide production and operation, the model prediction dimension is refined to the monthly level. Therefore, the independent variables must be features that are counted and released monthly.

[0140] Fourth, timeliness. Features must be released in a timely manner without too long a lag. If there is a lag of more than 2 months, the model may not be able to run or its effect may be greatly reduced due to missing data.

[0141] Before dividing the features, it is necessary to remove the seasonal trend of the features, and then make the leading, consistent and lagging measurements and divisions. The seasonal adjustment method of the data and the related algorithms of time difference analysis are described in detail below.

[0142] There are two main types of classic time series models: additive models and multiplicative models:

[0143] Y t =T t +C t +S t +I t .

[0144] Y t =T t ×C t ×S t ×I t .

[0145] Among them: the first formula is the additive model, the second formula is the multiplicative model, Y t is the original sequence; T t is the trend factor; C t is the circulation factor; S t I is the seasonal factor; t The irregular factor.

[0146] Generally speaking, if the four factors are independent, the additive model is used, and if they are interrelated, the multiplicative model is used. For general economic problems, the multiplicative model is more commonly used. Based on the information criterion, the multiplicative or additive model is automatically selected.

[0147] X-12 The basic process of seasonal adjustment is shown in Figure 3 shown.

[0148] The basic idea of ​​the BB method is to smooth the original sequence appropriately, infer the time of occurrence of its peaks and valleys on the smooth curve, and then gradually approach the time of occurrence of the peaks and valleys of the original sequence. To use the BB method to determine the time of occurrence of the peaks and valleys of the economic time series, two constraints must be added:

[0149] 1. The duration between peaks and valleys, and valleys and peaks is required to be more than 6 months. Considering the lack of data in the present invention and the relatively short time period, the short-term cyclic fluctuations are mainly measured, so the duration is determined to be more than 3 months here.

[0150] 2. The duration of a cycle, i.e. the interval between two turning points (peak-peak, valley-valley) is greater than 15 months, which is determined to be more than 6 months in the present invention.

[0151] The basic steps of the peak-to-valley correspondence method are as follows:

[0152] S31: Eliminate outliers and determine turning points on the 12-item moving average curve

[0153] First, we need to determine the upper and lower bounds of the outliers, and to do this, we need to find the irregular elements of the sequence. v is the sequence obtained by Spencer moving average of the x sequence, so the irregular elements have been eliminated in v, so the irregular elements can be separated from the sequence: c

[0154] I t =x t / v t (Multiplicative model).

[0155] I t =x t -v t (Additive model).

[0156] Where i = 1, 2, ..., n. Find the mean of the irregular elements and standard deviation σ, that is

[0157]

[0158]

[0159] According to the actual situation, take parameter A (1, 2 or 3) to determine whether the data is a unique value. The upper bound of A is

[0160] Suppose the sequence after singular value correction is ω, and the correction method of singular value is

[0161]

[0162] Where, the v sequence is the Spencer curve.

[0163] Secondly, after completing the above revisions, perform a 12-term moving average on the ω sequence to further eliminate the residual seasonal factors. The sequence after the 12-term moving average is recorded as ω1, and then the peak and valley occurrence time is preliminarily determined on the ω1 curve through a cyclic comparison method and considering the previous two constraints. The peaks and valleys preliminarily determined are numbered as p1, p2, …, p m , where m is the tentative number of turning points.

[0164] S32: Further identify turning points on the Spencer curve

[0165] First, perform a Spencer moving average on the ω sequence to further eliminate irregular elements.

[0166]

[0167] but

[0168]

[0169] In the formula, is the extended sequence, its length is n+14, The sequence is subjected to Spencer moving average, denoted as ω2.

[0170] Based on the turning point determined in the first step, determine the turning point on the ω2 curve. Let the variable r be

[0171]

[0172] MCD measures whether the sequence is smooth. The MCD value in the above formula is calculated by the x sequence. If p i is a tentative peak, in its neighborhood [p i -r,p i +r] select the maximum value, that is Then this maximum value ω 2j' is the tentative peak on the W curve, and the corresponding serial number If P tis a tentative valley, in its neighborhood [p t -r,p t +r] select the minimum value, that is Then this minimum value ω 2j' is the valley on the W curve, and the corresponding serial number

[0173] S33: Perform moving average of MCD items and determine turning points on the MCD curve

[0174] Perform MCD moving average on the original sequence x to preliminarily eliminate irregular elements. The moving average sequence ω3. Determine the corresponding turning point on the ω3 curve based on the turning point on the ω2 curve determined in the second step. The method of selecting the turning point is similar to the second step. Finally, record the turning point sequence number p i (i=1,2,…,m) is updated to the serial number of the turning point on the ω3 curve.

[0175] S34: Determine the turning point of the original sequence x

[0176] Based on the turning point of the ω3 curve determined in the third step, the corresponding turning point of the original sequence x is determined in a similar way. The turning points obtained in the fourth step gradually approach the peaks and valleys of the smooth sequence to the peaks and valleys of the original sequence, and become the final turning points. Finally, check whether each pair of final turning points meets the two constraints mentioned above. If not, discard the lower peak or higher valley until the constraints are met.

[0177] In the actual application of the classification of the prosperity index, the value of the KL information is to calculate the goodness of fit of the selected features relative to the benchmark features in turn. When the value of the KL information is closer to zero, it can be said that the assumed probability distribution model matches the true distribution model of the benchmark features more closely, that is, it can be classified as a consistency feature; when the value of the KL information is negative, it can be said that the selected feature is a leading feature; when the value of the KL information is positive, it can be said that the selected feature is a lagging feature. According to the judgment basis of the KL information method, it can be seen that the KL information method is very suitable for the classification of prosperity features when there are many features.

[0178] The basic idea is to select the reference feature and set it as y = {y1, y2, ..., y T}, T is the number of samples, select the selected feature and set it to x = {x1, x2, ..., x T}, T is the number of samples. By changing the delay number l between the selected feature and the benchmark feature, and calculating the KL information corresponding to the time difference, when the KL information of a feature and the benchmark feature is adjusted to the minimum, the corresponding delay number is used as the delay number of the selected feature. l takes a positive number to indicate lag, and the corresponding selected feature is classified as a lagging feature. l takes a negative number to indicate advance, and the corresponding selected feature is classified as a leading feature. When l is 0, the selected feature is classified as a consistent feature. The core steps of KL information calculation are as follows.

[0179] S41: Standardize the reference feature y, and the processed sequence is recorded as p, then:

[0180]

[0181] S42: Standardize the selected feature x, and the processed sequence is recorded as q, then:

[0182]

[0183] S43: Calculate KL information amount k l

[0184]

[0185] Time difference analysis is also a very important and commonly used method for classifying economic characteristics. The main theoretical basis of time difference analysis is the correlation analysis in the statistical principle. By calculating the correlation coefficients between the selected characteristics and the benchmark characteristics, the cross-correlation relationship is obtained. Then, according to the lag number in the cross-correlation relationship, the selected characteristics can be divided into three types: leading characteristics, consistency characteristics and lagging characteristics. Time difference analysis is to calculate the coefficients of the cross-correlation relationship between the selected characteristics, including the cross-correlation coefficients when the time series are aligned and the cross-correlation coefficients when the characteristics are staggered by several time nodes. According to the series of cross-correlation coefficients calculated, the one with the largest absolute value is selected, and the lag number of the feature is determined according to the number of time nodes staggered between the features. If the lag number is less than zero, it means that the selected feature is a leading feature; if the lag number is equal to zero, it means that the selected feature is a consistency feature; if the lag number is greater than zero, it means that the selected feature is a lagging feature.

[0186] Let y = {y1, y2, …, y n} is the data sequence of the referenced feature, x={x1,x2,…,x n} is the data sequence of other characteristics for which the lag period is to be calculated, then:

[0187]

[0188] In the formula, rl It represents the correlation coefficient between the selected reference feature and the feature to be calculated; l represents the number of time nodes that the feature leads or lags behind the reference feature. If l is less than zero, it means that the feature is a lagging feature, and if l is greater than zero, it means that the feature is a leading feature. L represents the maximum number of delay periods. This value is generally given in advance. This article selects 7, that is, the maximum moving period is 7 quarters. l It is the number of data that will be aligned after moving.

[0189] Select the largest time difference correlation coefficient as:

[0190]

[0191] r l It is calculated based on the Pearson correlation coefficient. In addition to the Pearson correlation coefficient, there are two non-parametric correlation coefficient calculation methods, Spearmon and Kendall. The present invention calculates these three correlation coefficients when using the time difference correlation coefficient method.

[0192] According to the BB peak-valley method, KL information method, and time-difference correlation coefficient method, the delayed month relationship between gasoline price availability rate, diesel price availability rate, gasoline price availability rate growth rate over the same period, diesel price availability rate growth rate over the same period and the alternative features was obtained. According to the principles recommended by more than half of the methods, the leading, consistent, and lagging features of the original sequence and the growth rate sequence were finally determined.

[0193] The smaller the KL information, the more similar the shape of the feature is to the gasoline price availability rate, and the larger the absolute value of the Pearson correlation coefficient, the higher the degree of linear correlation.

[0194] After the feature classification is completed, the leading, consistent and lagging warning indexes are synthesized directly through the American synthetic index method and the principal component analysis method. Since this process is directly produced by the algorithm, the specific process is no longer shown. After the early warning index is generated, in order to screen the index with better early warning effect, this study again uses the BB peak and valley method, KL information method, time difference correlation coefficient method and other synthetic indexes to select the synthetic index with the best leading, consistent and lagging effects as the refined oil supply and demand early warning data.

[0195] It should be noted that the present invention defines the red line as the price availability rate or the growth rate of the price availability rate, the blue line as the leading warning index, the yellow line as the consistent warning index, and the green line as the lagging warning index. Figure 4 ,According to the selection rules of the candidate early warning index, the best gasoline price obtained from the candidate early warning index is finally in place of the leading, consistent, and lagging early warning index.

[0196] See also Figure 5, according to the selection rules of the candidate early warning index, the best diesel price obtained from the candidate early warning index is finally in place of the leading, consistent and lagging early warning index.

[0197] See also Figure 6 According to the selection rules of the candidate early warning index, the best gasoline price arrival rate growth rate leading, consistent, and lagging early warning indexes are finally obtained from the candidate early warning indexes.

[0198] See also Figure 7 , according to the selection rules of the candidate early warning index, the best diesel price arrival rate growth rate in the same period is finally obtained from the candidate early warning index, including the leading, consistent and lagging early warning index.

[0199] The present invention obtains supply and demand warning data based on the original series of gasoline price availability rate, diesel price availability rate and their corresponding growth rate series. Figure 8 , showing the relationship between the gasoline price availability rate and the leading warning index, but note that the leading data only reflects the trend of the future price availability rate, not its actual value. For example: from September 2014 to March 2015, the price availability rate will decline. In November 2017, the leading data began to turn downward, while the price availability rate did not start to decline until April 2018. Similarly, the leading data in January 2019 warned of an increase in the price availability rate in July 2019. Overall, the lead time between the leading data and the price availability rate is about 6 months.

[0200] See also Fig. 9 , showing the relationship between the use of leading data and the growth rate of the price-in-place rate over the same period, it can be roughly seen that the leading data has certain lead-time characteristics. For example, six months before the price-in-place rate hit bottom in July 2016, the leading data had already hit bottom and quickly rebounded; by April 2017, the leading data peaked and fell back, and six months later the growth rate of the price-in-place rate reached a plateau. However, around June 2021, the growth rate reached a high level, but the price-in-place rate was not captured. Finally, the increase in the leading data value does not have much reference value for the time being, because due to the reasons after the data was updated, some features are missing, and the calculation results of the leading data have limited reference significance.

[0201] See also Fig.10, showing the relationship between the growth rate of diesel price availability and consistent data. For example: from July 2013 to September 2017, the two trends were very close; but from September 2017 to January 2019, there was a certain divergence between the two trends, the growth rate of price availability continued to rise, but the consistent data continued to fluctuate; after January 2019, the two trends were basically consistent. Especially from July to September 2020 and November-December 2021, the consistent data and the growth rate of price availability were very similar. This shows that consistent data is effective in tracking the growth rate of diesel price availability.

[0202] See also Fig.11 , showing the relationship between the gasoline price availability rate and the lagged data of the growth rate over the same period. The lagged data clearly grasps its trend. In the peak and valley states in October 2016, October 2018, July 2019 and April 2021, the trend change of the price availability rate was accurately grasped. The average lag period is 3.5 months.

[0203] Overall, the supply and demand warning data using the growth rate sequence has a better grasp of the future state, but its interpretation is slightly more complicated than the original sequence. Because the growth rate sequence is obtained by differentiating the original sequence, it is equivalent to the first-order derivative in mathematical principle. Therefore, when the growth rate is greater than 0, it means that the price fulfillment rate continues to rise; when the growth rate is equal to 0, it means that the price fulfillment rate has reached the critical point; when the growth rate is less than 0, it means that the price fulfillment rate will decline.

[0204] In some optional embodiments, a comprehensive prosperity index is constructed by performing feature priority measurement and screening on the leading feature pool.

[0205] It should be noted that the main function of the supply and demand warning data is to judge the changes in the fluctuation cycle of refined oil supply and demand, and it cannot tell the extent to which the supply and demand will rise or fall. In order to more accurately grasp the changes in the price availability rate, it is necessary to extract the most important features from the feature pool and construct a comprehensive prosperity index to quantitatively predict the future range of the price availability rate. The role of the comprehensive prosperity index is to quantitatively judge the status of the price availability rate in advance and prompt the decision-making department to take corresponding regulatory measures.

[0206] Specifically, the present invention selects the price availability rate as the benchmark feature for measuring the supply and demand situation of refined oil products. In the early warning model, signal lights are generally used to warn of the situation. The colors of the early warning signal lights are divided into five types, as shown in Table 1:

[0207] :Table 1

[0208] Overcooling Cool normal Hot overheat Price availability ≤85％ (85％，88％) (88％，94％) (94％，98％) ≥98％

[0209] They correspond to the five areas of "overheating", "slightly hot", "normal", "slightly cold" and "overcooling" when each characteristic develops. Therefore, when using the economic signal light, four thresholds should be calculated accordingly, and then the values ​​of the benchmark characteristics should be divided into five intervals according to these four thresholds, and each interval corresponds to a signal light of a different color. Each signal light is given a different score, generally 5 points for "overheating", 4 points for "slightly hot", 3 points for "normal", 2 points for "slightly cold" and 1 point for "overcooling".

[0210] See also Fig.12 and Fig.13 The relationship curves are respectively between gasoline price availability and supply-demand ratio, and diesel price availability and supply-demand ratio. It can be seen that in extreme cases, gasoline and diesel price availability is below 83% and above 98%, and most cases are between the two. Combined with the relationship between supply-demand ratio and price availability, the price availability is divided into five prosperity intervals.

[0211] Specifically, see Fig.14 To construct a comprehensive prosperity index and predict the range of future price availability, it is necessary to compare all independent variable features with the price availability to determine which features contribute to the predicted prosperity range of the price availability. Therefore, the refined oil supply and demand industry feature pool and the constructed supply and demand early warning data are used as alternatives. According to the timeliness and lag of feature data acquisition, the price availability range for the next month is predicted. The present invention lags the above-mentioned alternative features by one period and two periods, and uses the lag data as an alternative object for early warning features.

[0212] Based on the lagged data of the above characteristics, we first use one-way ANOVA, multiple comparison test, variable importance measure, categorized out-of-bag error measure and other methods to obtain the measurement value of each feature. Then, we use the TOPSIS method and entropy weight method to obtain the comprehensive score of the measurement value of each feature under multiple methods. Finally, we calculate the contribution of each feature, that is, the comprehensive score of each feature / the comprehensive score of all features. Then, we select the features with prosperity effect according to the cumulative contribution threshold.

[0213] The first step in constructing a comprehensive prosperity index feature system based on the price availability rate is to select features. The principle of selection is that the feature has a significant identification effect on the interval where the price availability rate is predicted. Therefore, the present invention selects methods such as one-way variance analysis, multiple comparison test, variable importance measurement, and classification out-of-bag error measurement to determine the identification effect of the features in the candidate feature pool.

[0214] One-way ANOVA is used to study whether different levels of a control variable have a significant impact on the observed variable. Here, the observed variable is the warning interval score of the price availability rate, and the control variable is the selected feature.

[0215] One-way ANOVA believes that the change of the observed variable value will be affected by both the control variable and the random variable. Based on this, one-way ANOVA decomposes the total deviation sum of squares of the observed variable into two parts: the sum of squares of the deviations between groups and the sum of squares of the deviations within groups, which can be expressed in mathematical form as: SST = SSA + SSE. By comparing the proportion of each part of the total deviation sum of squares of the observed variable, it is inferred whether the control variable has a significant impact on the observed variable.

[0216] In the total sum of squares of observed variables, if the proportion of the sum of squares of between-group deviations is large, it means that the changes in the observed variables are mainly caused by the control variables, can be mainly explained by the control variables, and the control variables have a significant impact on the observed variables; conversely, if the proportion of the sum of squares of between-group deviations is small, it means that the changes in the observed variables are not mainly caused by the control variables, cannot be mainly explained by the control variables, different levels of the control variables do not have a significant impact on the observed variables, and the changes in the values ​​of the observed variables are caused by random variable factors.

[0217] The specific analysis steps are as follows:

[0218] S51 proposes the null hypothesis: H0 - no difference; H1 - there is a significant difference.

[0219] S52 Select the test statistic: The test statistic used in variance analysis is the F statistic, that is, the F value test.

[0220] S53 calculates the observed value and probability P value of the test statistic: The purpose of this step is to calculate the observed value of the test statistic and the corresponding probability P value.

[0221] S54 gives a significance level and makes a decision.

[0222] The present invention is based on the one-way variance analysis method to obtain the test statistic P value of the warning interval score variable of each feature on the price availability rate, and the smaller the value, the better.

[0223] It should be noted that the basic analysis of one-way ANOVA can only determine whether the control variable has a significant impact on the observed variable. If the control variable does have a significant impact on the observed variable, it is necessary to further determine the degree of influence of different levels of the control variable on the observed variable, which level has a significantly different effect from other levels, which level has an insignificant effect, etc. The multiple comparison test uses all the observed variable values ​​to achieve a pairwise comparison of the overall mean of the observed variables at each level.

[0224] The present invention is based on the idea of ​​pairwise comparison test and obtains the significant logarithms and the average value of pairwise comparison P values ​​in pairwise comparison.

[0225] The variable importance measure is a variable importance measure feature obtained by the decision tree machine learning method when splitting each node based on the extent to which the Gini index decreases due to the split variable. The larger the feature value, the greater the role of the feature in distinguishing the prosperity range of the price availability rate.

[0226] In order to better integrate the results of the above methods, the TOPSIS method and entropy weight method are used to weight the feature recognition effect, so as to judge the recognition effect of the alternative features.

[0227] TOPSIS method is a sequential optimization technique for the similarity of ideal targets, and is a very effective method in multi-target decision analysis. Through the normalized data normalization matrix, the best and worst targets among multiple targets are found, and the distance between each evaluation target and the ideal solution and the anti-ideal solution is calculated respectively. The closeness of each target to the ideal solution is obtained, and the closeness is sorted according to the size of the closeness to the ideal solution, which is used as the basis for evaluating the quality of the target. The closeness value is between 0 and 1. The closer the value is to 1, the closer the corresponding evaluation target is to the optimal level. Conversely, the closer it is to 0, the closer the evaluation target is to the worst level.

[0228] Assume there are m evaluation targets D1, D2, ..., Dm, and each target has n evaluation features X1, X2, ...Xn. The rating feature scores of each target can be expressed in the following mathematical matrix form:

[0229]

[0230] Normalize the feature matrix to obtain a normalized vector, and establish a normalized matrix about the normalized vector:

[0231]

[0232] By calculating the weight normalization value, a weight normalization matrix about the weight normalization value is established:

[0233] v ij =w j r ij ,i=1,2,…,m,j=1,2,…,n. .

[0234] Determine the ideal solution and anti-ideal solution based on the weight normalization values:

[0235]

[0236]

[0237] Calculate the distance scale, that is, calculate the distance from each target to the ideal solution and the anti-ideal solution. The distance scale can be calculated by the n-dimensional Euclidean distance. The distance from the target to the ideal solution is S *, the distance to the anti-ideal solution is S - .

[0238]

[0239] Compute the closeness of the ideal solution:

[0240]

[0241] The solutions are sorted according to their closeness to the ideal solution. The larger the closeness value of the sorting result, the better the target.

[0242] Entropy is a physical concept in thermodynamics, and is a measure of the degree of chaos or disorder in a system. The larger the entropy, the more chaotic the system is, which means it carries less information. The smaller the entropy, the more ordered the system is, which means it carries more information. Information entropy borrows the concept of entropy in thermodynamics to describe the amount of information in an event on average. Therefore, mathematically, information entropy is the expected mean, or mean, or expectation, of the amount of information contained in an event. It is the sum of the probability of each possible result in the experiment multiplied by its result. Based on the definition of expectation, it can be imagined that the formula for information entropy is roughly:

[0243] Information entropy = ∑ the probability of each possible event * the amount of information contained in each possible event. The amount of information contained in each possible event is related to the uncertainty of this event. In other words, it is related to the probability of the event occurring. The greater the probability, the smaller the amount of information.

[0244] According to the definition of information entropy, for a certain feature, the entropy value can be used to judge the discreteness of a certain feature. The smaller the entropy value, the greater the discreteness of the feature, and the greater the influence of the feature on the comprehensive evaluation, that is, the weight. If the values ​​of a certain feature are all equal, then the feature has no effect in the comprehensive evaluation.

[0245] The calculation steps of the entropy weight method are as follows:

[0246] S61: Normalization

[0247] Normalization is to make features of different dimensions the same dimension. For example, the jth feature of the i-th user is x ij , after normalization, it is x' ij , there are two formulas as follows:

[0248] or

[0249] If the feature is a positive feature, the first formula is used; if the feature is a negative feature, the second formula is used. min(x j ) is the minimum value of the jth feature, max(x j ) is the maximum value of the j-th feature.

[0250] S62: Calculate the entropy and weight of features

[0251] a. First calculate the weight of the jth feature of the i-th user:

[0252] b. Calculate the information entropy of the jth feature Where k is a constant,

[0253] Calculate the weight of the jth feature

[0254] In general, this study has constructed 12 leading, consistent, and lagging early warning indices from the original sequence of refined oil price availability and the growth rate sequence. Among them, the original sequence can directly indicate the price availability, but due to the distinct characteristics of past economic growth, the effect is slightly worse than the growth rate sequence; the growth rate sequence better grasps the direction of change of the price availability, but cannot be directly used as a direct reference for the price availability, but needs to be further interpreted and analyzed in the form of its first-order derivative, which represents the direction and rate of increase and decrease of the price availability, and has a better tracking effect but relatively complex interpretation.

[0255] The construction of leading, consistent and lagging data is convenient for detailed tracking and description of the current situation. It is impossible to clearly state the status by relying on a single index. As shown in Table 2, if the leading data rises, it means that the supply and demand will enter a tense state in the future, but the time of entry is unclear. If the consistent data also shows an upward trend, it means that the supply has entered a state of insufficient demand. When the leading data falls and the consistent data rises, it means that the tense state of supply and demand cannot continue, and the price availability rate will decline in the future. When the consistent data declines, it cannot represent the impending exit from the tense state of supply and demand. At this time, it can be judged in combination with the changes in the lagging data. For example, if the lagging data also declines, it clearly indicates that the tense state of supply and demand has ended, and the downward trend of the price availability rate will continue. The changes in the early warning index during supply and demand tension are shown in the table below, which shows the combined application of the early warning index during supply and demand tension.

[0256] Table 2

[0257]

[0258] Compared with the supply and demand early warning data, the prosperity comprehensive index in the present invention is mainly used to quantitatively determine the prosperity range of the price availability rate in the T+1 period and the T+2 period.

[0259] For example: after calculation, the model outputs the comprehensive gasoline prosperity index in the third interval, which means that the T+1 month price availability rate will be between 88% and 94%, and the overall supply and demand are balanced, with little possibility of upward or downward movement; for finished oil sales companies, they can stabilize the current sales strategy and maintain normal inventory operations.

[0260] If the comprehensive gasoline prosperity index is in the fifth interval, it means that the price availability rate in T+1 month will reach 98%, and it may face a short-term shortage of supply and demand, and the price availability rate may soar. For sales companies, they should seize the opportunity as soon as possible, prepare stocks as soon as possible, and purchase gasoline to take the opportunity to realize the inventory value when the price availability rate is high.

[0261] In general, when the prosperity index is in different prosperity ranges, the supply and demand conditions it indicates, as well as the strategies that sales companies should adopt, are different. Table 3 shows its specific meaning and recommended measures.

[0262] Table 3

[0263]

[0264] The present invention constructs supply and demand early warning data and a comprehensive prosperity index, wherein the supply and demand early warning data qualitatively indicates the supply and demand status and predicts the direction of the price availability rate, and the comprehensive prosperity index quantitatively predicts the prosperity interval of the price availability rate. The combined application of the two can characterize the current status and predict the direction of future supply and demand changes, providing an important reference for making production and operation plans, formulating sales strategies, and optimizing inventory operations, eliminating the problem of error transmission, and also preventing the sudden change in demand for refined oil from causing a waste of resources due to a lack of supply or oversupply.

[0265] Based on the same inventive concept, the embodiment of the present invention also provides a supply and demand early warning system for refined oil, see Fig.15 ,include:

[0266] The acquisition module 151 is used to obtain the supply and demand data of refined oil products and establish a supply and demand early warning database according to the characteristics of the oil-using industry;

[0267] A division module 152 is used to divide the warning feature pool into a leading feature pool, a consistent feature pool and a lagging feature pool by using time difference analysis, BB peak-valley method and KL information method;

[0268] A screening module 153 is used to screen the leading features according to the leading feature pool to construct supply and demand warning data; wherein the leading data, the consistent data and the lagging data are screened from the leading feature pool, the consistent feature pool and the lagging feature pool by a time difference inspection method;

[0269] Prediction module 154, used for constructing a comprehensive prosperity index by measuring and screening features based on feature priority of the preceding feature pool;

[0270] The characterization module 155 is used to characterize the supply and demand of refined oil by combining the supply and demand warning data and the comprehensive prosperity index.

[0271] Based on the same inventive concept, the present invention also provides an electronic device 161, see Fig.16 , including a processor 164, a communication interface 165, a memory 162 and a communication bus, wherein the processor 164, the communication interface 165 and the memory 162 communicate with each other through the communication bus;

[0272] A memory 162 for storing a computer program 163;

[0273] The processor 164 is used to implement the supply and demand early warning method of refined oil when executing the program stored in the memory 162.

[0274] The above communication bus may be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus may be divided into an address bus, a data bus, a control bus, etc.

[0275] The communication interface 165 is used for communication between the electronic device 161 and other devices.

[0276] The memory 162 may include a random access memory 162 (RAM), or may include a non-volatile memory 162 (non-volatile memory), such as at least one disk memory 162. Optionally, the memory 162 may also be at least one storage device located away from the processor 164.

[0277] The above-mentioned processor 164 can be a general-purpose processor 164, including a central processing unit 164 (CPU), a network processor 164 (NP), etc.; it can also be a digital signal processor 164 (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.

[0278] Based on the same inventive concept, the present invention also provides a computer-readable storage medium on which a computer program 163 is stored. When the computer program 163 is executed by the processor 164, the supply and demand early warning method of refined oil is implemented.

[0279] The computer-readable storage medium may be included in the device / apparatus described in the above embodiment; or it may exist independently without being assembled into the device / apparatus. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the supply and demand early warning method of refined oil according to the embodiment of the present disclosure is implemented.

[0280] Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent substitutions for some of the technical features therein; and these modifications or substitutions do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for early warning of supply and demand of refined oil, characterized in that: include: Obtain refined oil supply and demand data and establish a supply and demand early warning database based on the characteristics of the oil-using industry; By using time difference analysis, BB peak-valley method and KL information method, the supply and demand early warning database is divided into a leading feature pool, a consistent feature pool and a lagging feature pool; The supply and demand warning data are screened according to the preceding feature pool, the consistent feature pool and the lagging feature pool; wherein the preceding data, the consistent data and the lagging data are screened from the preceding feature pool, the consistent feature pool and the lagging feature pool by a time difference inspection method; By measuring and screening the features by feature priority on the preceding feature pool, a comprehensive prosperity index is constructed; The supply and demand warning data and the comprehensive prosperity index are combined to represent the warning situation of refined oil supply and demand.

2. The method according to claim 1, characterized in that The method of screening the leading data, consistent data and lagging data from the leading feature pool, the consistent feature pool and the lagging feature pool by the time difference inspection method includes: screening the leading feature data, consistent feature data and lagging feature data that have a significant time difference relationship with the gasoline and diesel price arrival rate.

3. The method according to claim 1, characterized in that After dividing the supply and demand early warning database into a leading feature pool, a consistent feature pool and a lagging feature pool by using time difference analysis, BB peak-valley method and KL information method, the method further includes: The standard change rate Sij of each feature is obtained by the American composite index method, including: obtaining the symmetrical change rate of the feature ； The symmetric change rate is standardized, and the average change rate of each group of features is calculated based on the standardized symmetric change rate.

4. The method according to claim 3, characterized in that The method of obtaining each characteristic standard change rate Sij by the American composite index method further includes: obtaining an initial composite index, obtaining an index trend according to the composite index and adjusting the trend, and calculating a composite characteristic.

5. The method according to claim 1 or 4, characterized in that: The method further comprises: The importance of each feature is determined by principal component analysis, including: The weight is taken based on the inverse of the KL information to obtain the standardized average change rate Rj of each feature group.

6. The method according to claim 1, characterized in that The method further comprises: According to the features screened by the preceding feature pool, the consistent feature pool and the lagging feature pool, a principal component score Pj of each feature combination is obtained by a rolling principal component analysis method.

7. The method according to claim 6, characterized in that The method further comprises: The standardized average change rate Vj is calculated based on the standardized average change Rj of each feature group and the principal component score Pj of the feature group, and the synthesis index lj is calculated based on the standardized average change rate Vj.

8. The method according to claim 1 or 7, characterized in that: The method further comprises: According to the composite index lj, the divided leading data, consistent data and lagged data are selected by BB peak-valley method, KL information method and time difference correlation coefficient method; The selected leading data, consistent data and lagging data are used as supply and demand warning data; The supply and demand warning situation of refined oil is characterized according to the supply and demand warning data.

9. A supply and demand early warning system for refined oil, characterized in that: include: The acquisition module is used to obtain the supply and demand data of refined oil products and establish a supply and demand early warning database based on the characteristics of the oil-using industry; A division module is used to divide the warning feature pool into a leading feature pool, a consistent feature pool and a lagging feature pool through time difference analysis, BB peak-valley method and KL information method; A screening module, used to screen the leading features according to the leading feature pool to construct supply and demand warning data; wherein the leading data, consistent data and lagging data are screened from the leading feature pool, the consistent feature pool and the lagging feature pool by a time difference inspection method; A prediction module, used for constructing a comprehensive prosperity index by measuring and screening features based on feature priority of the preceding feature pool; The characterization module is used to characterize the warning situation of the supply and demand of refined oil by combining the supply and demand warning data and the comprehensive prosperity index.

10. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus; Memory for storing computer programs; A processor is used to implement the supply and demand early warning method for refined oil according to any one of claims 1 to 8 when executing the program stored in the memory.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for early warning of supply and demand of refined oil according to any one of claims 1 to 8 is implemented.

Images