CN105718756A - Real-time computing method and system for nonrandom informational probability in continuous game data streams - Google Patents

Abstract

The invention discloses a real-time computing method and system for nonrandom informational probability in continuous game data streams. The real-time computing method comprises the steps of maintaining a current trading volume data bucket by virtue of a series of input time series data, simultaneously maintaining a queue with a length equal to the number n of sampled data buckets, adding VtauB and VtauS computed in the current trading volume data bucket into the head of the current queue whenever the current trading volume data bucket is full, and computing the nonrandom informational probability according to n latest VtauB and VtauS by virtue of a sliding window analogue method. According to the real-time computing method, the online rolling real-time computation can be realized, furthermore, the stability and the adaptivity are relatively good, obstacles for estimating unobservable parameters are eliminated, and defect of inaccuracy of a time dimension algorithm during real-time trading and particularly high-frequency trading is overcome.

Description

The real-time computing technique of nonrandom informational probability and system in Continuous Game data stream

Technical field

The invention belongs to field of computer technology, particularly relate to a kind of data analysing method and system.

Background technology

Continuous Game, such as auction, Ask-Bid System, real-time deal etc., the data stream formed, there is the factor such as information asymmetry, irrational decision-making in the main body owing to participating in game, is mingled with the sticgastuc deicision information of a large amount of main body in data stream.Stochastic Decision-making probability in this characteristic data flow is calculated, or informational probability nonrandom in this characteristic data flow is calculated, the effective information in betting data time series can be better profited from, thus helping to solve many practical problems, such as the undulatory property of prediction markets, prevention market risk etc..

But, current conventional nonrandom informational probability evaluation method is based on what the statistical method that some unobservable parameters carry out maximum likelihood method carried out, inevitably need parameter estimation is carried out artificial hypothesis, and the result of parameter estimation is inconvenient to upgrade in time, thus the instant situation of nonrandom informational probability in this characteristic data flow can not be reflected.Therefore, find a kind of on-line data analysis method and system, the real-time data stream formed for Continuous Game implements a series of objective technical finesses, thus obtaining the evaluation of nonrandom informational probability in this characteristic data flow, is technical problem currently in the urgent need to address.

Summary of the invention

In view of this, it is an object of the invention to provide the real-time computing technique of nonrandom informational probability in a kind of Continuous Game data stream and system, nonrandom informational probability in the Continuous Game time series data calculating different time granularity in real time, the undulatory property of timely prediction markets can be rolled online.

The real-time computing technique of nonrandom informational probability in a kind of Continuous Game data stream, comprises the following steps:

(1) data stream Continuous Game formed is sampled by constant duration, and obtaining sampling time interval is T₀Time series data stream, be designated as: { T₁,P₁,V₁}、{T₂,P₂,V₂}…{T_i,P_i,V_i..., wherein, T_iFor ith sample moment, P_iFor the End-price in ith sample moment, V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；

(2) from described time series data stream, extract activity data stream, be designated as: { V₁,V₂,…,V_i... }, wherein, the T between ith sample moment and the i-th-1 sampling instant₀Total volume V in interval_iIt is regarded as the V that can split_iPart unit exchange hand；

(3) by described activity data stream according to etc. exchange hand interval carry out being polymerized and splitting, and be sequentially filled in n data bucket, the maximum exchange hand summation that each data bucket holds is V；

(4) whether full described data bucket is detected in real time, i.e. in described data bucket, whether the summation of activity data has reached V, if it is, carry out step (5)；If it is not, then return to step (1)；

(5) (I) and (II) calculates the exchange hand V bought in each described data bucket according to the following formula respectively_τ ^BWith the exchange hand V sold_τ ^SValue, and by (V_τ ^B,V_τ ^S) it is filled into sampling buffer area as a paired sampled value；

$V_{\mathrm{\τ}}^B=\underset{i=t(\mathrm{\τ}-1)+1}{\overset{t\left(\mathrm{\τ}\right)}{\Σ}}{V}_i\·S\left(\frac{P_i-P_{i-1}}{ATR}\right)---\left(I\right)$

$V_{\mathrm{\τ}}^S=\underset{i=t(\mathrm{\τ}-1)+1}{\overset{t\left(\mathrm{\τ}\right)}{\Σ}}{V}_i\·\[1-S\left(\frac{P_i-P_{i-1}}{ATR}\right)\]=V-V_{\mathrm{\τ}}^B---\left(II\right)$

Wherein, t (τ) is last sampling instant, V in τ described data bucket_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；P_i、P_i-1The respectively End-price of ith sample moment and the i-th-1 sampling instant, ATR is the average true wave amplitude (AverageTrueRange) of described time series data stream, and V is the maximum exchange hand summation held in each described data bucket；S is Sigmoid function, and its definition is as follows:B is the parameter of control function smoothness；

(6) whether full detect in real time described sampling buffer area, namely detect the value of the number m of paired sampled value in described sampling buffer area, if m is < n, return to step (1) or (5)；If m=n, then carry out step (7)；

(7) according to the following formula (III) calculates nonrandom informational probability ρ output in real time in the data stream that presently described Continuous Game is formed:

$\mathrm{\&rho;}\&ap;\frac{{\mathrm{\&Sigma;}}_{\mathrm{\&tau;}=1}^n|V_{\mathrm{\&tau;}}^S-V_{\mathrm{\&tau;}}^B|}{nV}---\left(III\right).$

In the present invention, the data stream that described Continuous Game is formed is obtained from metadata provider by application programming interface (ApplicationProgrammingInterface, API).

In the present invention, described interval T₀Value be preferably 1 minute, 10 seconds or 30 seconds.

In the present invention, the value of the number n of described data bucket is preferably 50.

In the present invention, ATR is true wave amplitude (TrueRange, TR) rolling average, and TR=Max{ (H-L), (H-PC), (PC-L) }, wherein, the ceiling price of H and L respectively present sample time period and lowest price, PC is the End-price of previous sampling time section.When calculating, the number n of described data bucket can be considered sliding window number, then, in n described data bucket, there is Y value T in time series data corresponding to activity data₀Sampling time interval, then ATR can value be the rolling average of the true wave amplitude (TrueRange, TR) of Y.

Meanwhile, present invention also offers the real time computation system of nonrandom informational probability in a kind of Continuous Game data stream, including:

(1) sampling module: the data stream for Continuous Game is formed is sampled by constant duration, and obtaining sampling time interval is T₀Time series data stream, be designated as: { T₁,P₁,V₁}、{T₂,P₂,V₂}…{T_i,P_i,V_i..., wherein, T_iFor ith sample moment, P_iFor the End-price in ith sample moment, V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；

(2) data extraction module: for extracting activity data stream from the described time series data stream exported by described sampling module, be designated as: { V₁,V₂,…,V_i... }, wherein, the T between ith sample moment and the i-th-1 sampling instant₀Total volume V in interval_iIt is regarded as the V that can split_iPart unit exchange hand；

(3) data concentrating module: for by the described activity data stream exported by described data extraction module according to etc. exchange hand interval carry out being polymerized and splitting, and it being sequentially filled in n data bucket, the maximum exchange hand summation that each data bucket holds is V；Whether full it is simultaneously used for detecting described data bucket in real time, i.e. in described data bucket, whether the summation of activity data has reached V, if it is, proceed to exchange hand classification of type computing module；If it is not, then return to described sampling module；

(4) exchange hand classified counting module: for calculating by the exchange hand V bought in each described data bucket of described data concentrating module output_τ ^BWith the exchange hand V sold_τ ^SValue, wherein, the exchange hand V bought_τ ^BWith the exchange hand V sold_τ ^SValue respectively according to the following formula (I) and (II) be calculated:

$V_{\mathrm{\&tau;}}^B=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;S\left(\frac{P_i-P_{i-1}}{ATR}\right)---\left(I\right)$

$V_{\mathrm{\&tau;}}^S=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;\&lsqb;1-S\left(\frac{P_i-P_{i-1}}{ATR}\right)\&rsqb;=V-V_{\mathrm{\&tau;}}^B---\left(II\right)$

Wherein, t (τ) is last sampling instant, V in τ described data bucket_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；P_i、P_i-1The respectively End-price of ith sample moment and the i-th-1 sampling instant, ATR is the average true wave amplitude (AverageTrueRange) of described time series data stream, and V is the maximum exchange hand summation held in each described data bucket；S is Sigmoid function, and its definition is as follows:B is the parameter of control function smoothness；

(5) sampling cache module: for the exchange hand V that will buy in each described data bucket exported by described exchange hand classified counting module_τ ^BWith the exchange hand V sold_τ ^SValue be filled into sampling buffer area as a paired sampled value, and it is whether full for detecting described sampling buffer area in real time, namely the value of the number m of paired sampled value in described sampling buffer area is detected, if m is < n, return to described sampling module or described exchange hand classified counting module；If m=n, then proceed to nonrandom informational probability estimation module；

(6) nonrandom informational probability estimation module: for calculating nonrandom informational probability ρ in the data stream that presently described Continuous Game is formed according to the data of described sampling cache module output, described nonrandom informational probability ρ (III) according to the following formula is calculated:

$\mathrm{\&rho;}\&ap;\frac{{\mathrm{\&Sigma;}}_{\mathrm{\&tau;}=1}^n|V_{\mathrm{\&tau;}}^S-V_{\mathrm{\&tau;}}^B|}{nV}---\left(III\right);$

(7) data outputting module: for the value of calculated for described nonrandom informational probability estimation module described nonrandom informational probability is exported.

In the present invention, the data stream that described Continuous Game is formed is obtained from metadata provider by application programming interface (ApplicationProgrammingInterface, API).

In the present invention, described interval T₀Value be preferably 1 minute, 10 seconds or 30 seconds.

In the present invention, the value of the number n of described data bucket is preferably 50.

In the present invention, ATR is true wave amplitude (TrueRange, TR) rolling average, and TR=Max{ (H-L), (H-PC), (PC-L) }, wherein, the ceiling price of H and L respectively present sample time period and lowest price, PC is the End-price of previous sampling time section.When calculating, the number n of described data bucket can be considered sliding window number, then, in n described data bucket, there is Y value T in time series data corresponding to activity data₀Sampling time interval, then ATR can value be the rolling average of the true wave amplitude (TrueRange, TR) of Y.

In the present invention, described data outputting module is digital output module, exports result with the form of numerical value；Can also be the combination of digital output module and images outputting module, export result in a graphical form.

In the present invention, described data outputting module can be docked with other system, so that other system can call the data processed result of described data outputting module output.

In the present invention, for the deficiency in being constituted with the time for dimension research exchange hand in prior art, adopt the statistical method being dimension with exchange hand, calculate nonrandom informational probability in the data stream that Continuous Game is formed based on the exchange hand of different time granularity and knock-down price, can be used for the prediction of price wave motion to the financial market such as stock, futures.

Specifically, the present invention is that a series of time series data first passing through input safeguards a current activity data bucket, safeguard a long queue for sampled data bucket number n simultaneously, whenever current activity data bucket has been expired time, can by calculated V in current activity data bucket_τ ^BAnd V_τ ^SJoin in current queue heads, adopt the method that sliding window is similar here, take n nearest V_τ ^BAnd V_τ ^SCalculate nonrandom informational probability.In the present invention, for V_τ ^BAnd V_τ ^SCalculating introduces ATR index absolute drift is normalized, there is better stability and adaptivity；Further, in the present invention, being calculated as of nonrandom informational probability ρ calculates online in real time, it is not limited to certain time period of analysis of history.Adopt object oriented designing pattern to be encapsulated in the real time computation system of nonrandom informational probability by the calculating of ρ when realizing, by constantly to the incoming online time series data of system, system can according to the nonrandom informational probability ρ value of the situation real-time update of current exchange hand.

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

(1) eliminate the obstacle that unobservable parameter is estimated, overcome the inaccurate defect that time dimension algorithm is concluded the business at present deal especially high frequency.

(2)V_τ ^BAnd V_τ ^SPartitioning algorithm in creativeness introduce ATR index absolute drift be normalized, there is better stability and adaptivity.

(3) achieve the calculating of nonrandom informational probability in the Continuous Game time series data for different grain size (day, time, grade), and real-time calculating can be rolled online.

Accompanying drawing explanation

Fig. 1 is microcosmic Trading Model figure.

Fig. 2 is that the present invention carries out the flow chart of calculating in real time for nonrandom informational probability in Continuous Game data stream.

Fig. 3 be the present invention a specific embodiment in sampling time interval be the time series data stream of 1 minute.

Fig. 4 be the present invention a specific embodiment in informational probability nonrandom in the time series data stream of Fig. 3 carried out to the result of in real time calculating.

Fig. 5 be the present invention a specific embodiment in for the real-time result of calculation of nonrandom informational probability of Fig. 4 in the time interval of subsequently 2 hours with the dependency of price wave motion.

Detailed description of the invention

Hereinafter, in conjunction with specific embodiments, technical scheme is described in further detail.Following example are carried out under premised on technical solution of the present invention, give detailed embodiment and process, but protection scope of the present invention is not limited to following embodiment.

For nonrandom informational probability in Continuous Game data stream, illustrated by the microcosmic Trading Model built as shown in Figure 1, wherein, α is that message event produces probability, μ be in the know (namely, nonrandom information) the trading instruction speed of dealer, ε is the trading instruction speed of non-(that is, the random information) dealer that knows the inside story.Thus, the probability P IN (that is, nonrandom Transaction Information probability) of Informed Trade information can be expressed as follows:

$PIN=\frac{\mathrm{\&alpha;}\mathrm{\&mu;}}{\mathrm{\&alpha;}\mathrm{\&mu;}+2\mathrm{\&epsiv;}}$

As in figure 2 it is shown, in a specific embodiment of the present invention, the real-time computing technique of nonrandom informational probability in a kind of Continuous Game data stream, comprise the following steps:

Step S1: the data stream (can be obtained from metadata provider by application programming interface API) that Continuous Game is formed is sampled by constant duration, and obtaining sampling time interval is T₀Time series data stream, be designated as: { T₁,P₁,V₁}、{T₂,P₂,V₂}…{T_i,P_i,V_i..., wherein, T_iFor ith sample moment, P_iFor the End-price in ith sample moment, V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval, T₂-T₁=T_i-T_i-1=T₀；

Step S2: extract activity data stream from described time series data stream, be designated as: { V₁,V₂,…,V_i... }, wherein, the T between ith sample moment and the i-th-1 sampling instant₀Total volume V in interval_iIt is regarded as the V that can split_iPart unit exchange hand；

Step S3: by described activity data stream according to etc. exchange hand interval carry out being polymerized and splitting, and be sequentially filled in n data bucket, the maximum exchange hand summation that each data bucket holds is V；

Step S4: whether detection current data bucket is full in real time, i.e. in current data bucket, whether the summation of activity data has reached V, if it is, carry out step S5；If it is not, then return to step S1；

Step S5: (I) and (II) calculates the exchange hand V bought in current data bucket according to the following formula respectively_τ ^BWith the exchange hand V sold_τ ^SValue；

$V_{\mathrm{\&tau;}}^B=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;S\left(\frac{P_i-P_{i-1}}{ATR}\right)---\left(I\right)$

$V_{\mathrm{\&tau;}}^S=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;\&lsqb;1-S\left(\frac{P_i-P_{i-1}}{ATR}\right)\&rsqb;=V-V_{\mathrm{\&tau;}}^B---\left(II\right)$

Wherein,

T (τ) is last sampling instant in τ described data bucket,

V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；

P_i、P_i-1The respectively End-price of ith sample moment and the i-th-1 sampling instant；

ATR is the average true wave amplitude (AverageTrueRange) of described time series data stream, value is true wave amplitude (TrueRange, TR) rolling average, and TR=Max{ (H-L), (H-PC), (PC-L) }, wherein, the ceiling price of H and L respectively present sample time period and lowest price, PC is the End-price of previous sampling time section；

V is the maximum exchange hand summation held in each described data bucket；

S is Sigmoid function, and its definition is as follows:Wherein, b is the parameter of control function smoothness, and b can be set to 1；

Step S6: by (V_τ ^B,V_τ ^S) it is filled into sampling buffer area as a paired sampled value；

Step S7: whether full detect in real time described sampling buffer area, namely detects the value of the number m of paired sampled value in described sampling buffer area, if m is < n, returns to step S1 or S5；If m=n, then carry out step S8；

Step S8: (III) calculates nonrandom informational probability ρ output in real time in the data stream that presently described Continuous Game is formed according to the following formula:

$\mathrm{\&rho;}\&ap;\frac{{\mathrm{\&Sigma;}}_{\mathrm{\&tau;}=1}^n|V_{\mathrm{\&tau;}}^S-V_{\mathrm{\&tau;}}^B|}{nV}---\left(III\right).$

For the ease of understanding and illustrating, give one example the exchange hand V illustrating to buy at this_τ ^BWith the exchange hand V sold_τ ^SThe calculating process of value:

Step S1 obtains the data stream that Continuous Game is formed, interval T by application programming interface (ApplicationProgrammingInterface, API) from metadata provider₀Value 1 minute；Activity data stream { V in step S2₁,V₂,…,V_i... }, it is assumed that for { 5,2,4,1,7,13,5 ... }；In step S3 etc. exchange hand interval V be 10, owing to each exchange hand is V_iData regard as and split into V_iIndividual exchange hand is the data of 1, then when certain activity data is filled in current data bucket to make this data bucket exchange hand reach 10, it is necessary to according to exchange hand, these data are split into two parts, a part fills up current data bucket, and another part the next data bucket to be added such as then, therefore, activity data stream after polymerization and fractionation is { (5,2,3), (1,1,7,1), (10), (2,5 ...) ... }；By (5,2,3), (1,1,7,1), (10), (2,5 ...) ... it is sequentially filled in n data bucket, step S5 calculates each data bucket such as (5,2,3) respectively, for another example (1,1,7,1) accounting of the exchange hand bought in the data of totally 10 unit exchange hands and the exchange hand sold.For time series data, the granularity of data can split, and such as daily the data at interval can be split as Hour, or be polymerized to and monthly count.According to the security bargain data minute counted, (5,2,3) are the data of 3 minutes, totally 10 unit exchange hands for modal, for each data and per minute, exist out, high and low, receive four prices, P_i、P_i-1Respectively " receipts " price of ith sample moment and the i-th-1 sampling instant, ATR is the rolling average of true wave amplitude (TrueRange, TR), for instance, value is the average of 250 minutes true wave amplitude TR,Meet Sigmoid function

Correspondingly, in a specific embodiment of the present invention, the real time computation system of nonrandom informational probability in a kind of Continuous Game data stream, including:

(1) sampling module: the data stream (can be obtained from metadata provider by application programming interface API) for Continuous Game is formed is sampled by constant duration, and obtaining sampling time interval is T₀Time series data stream, be designated as: { T₁,P₁,V₁}、{T₂,P₂,V₂}…{T_i,P_i,V_i..., wherein, T_iFor ith sample moment, P_iFor the End-price in ith sample moment, V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；

(2) data extraction module: for extracting activity data stream from the described time series data stream exported by described sampling module, be designated as: { V₁,V₂,…,V_i... }, wherein, the T between ith sample moment and the i-th-1 sampling instant₀Total volume V in interval_iIt is regarded as the V that can split_iPart unit exchange hand；

(3) data concentrating module: for by the described activity data stream exported by described data extraction module according to etc. exchange hand interval carry out being polymerized and splitting, and it being sequentially filled in n data bucket, the maximum exchange hand summation that each data bucket holds is V；Whether full it is simultaneously used for detecting described data bucket in real time, i.e. in described data bucket, whether the summation of activity data has reached V, if it is, proceed to exchange hand classification of type computing module；If it is not, then return to described sampling module；

(4) exchange hand classified counting module: for calculating by the exchange hand V bought in each described data bucket of described data concentrating module output_τ ^BWith the exchange hand V sold_τ ^SValue, wherein, the exchange hand V bought_τ ^BWith the exchange hand V sold_τ ^SValue respectively according to the following formula (I) and (II) be calculated:

$V_{\mathrm{\&tau;}}^B=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;S\left(\frac{P_i-P_{i-1}}{ATR}\right)---\left(I\right)$

$V_{\mathrm{\&tau;}}^S=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;\&lsqb;1-S\left(\frac{P_i-P_{i-1}}{ATR}\right)\&rsqb;=V-V_{\mathrm{\&tau;}}^B---\left(II\right)$

Wherein,

T (τ) is last sampling instant in τ described data bucket,

V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；

P_i、P_i-1The respectively End-price of ith sample moment and the i-th-1 sampling instant；

ATR is the average true wave amplitude (AverageTrueRange) of described time series data stream, value is true wave amplitude (TrueRange, TR) rolling average, and TR=Max{ (H-L), (H-PC), (PC-L) }, wherein, the ceiling price of H and L respectively present sample time period and lowest price, PC is the End-price of previous sampling time section；

V is the maximum exchange hand summation held in each described data bucket；

S is Sigmoid function, and its definition is as follows:Wherein, b is the parameter of control function smoothness, and b can be set to 1；

(5) sampling cache module: for the exchange hand V that will buy in each described data bucket exported by described exchange hand classified counting module_τ ^BWith the exchange hand V sold_τ ^SValue be filled into sampling buffer area as a paired sampled value, and it is whether full for detecting described sampling buffer area in real time, namely the value of the number m of paired sampled value in described sampling buffer area is detected, if m is < n, return to described sampling module or described exchange hand classified counting module；If m=n, then proceed to nonrandom informational probability estimation module；

(6) nonrandom informational probability estimation module: for calculating nonrandom informational probability ρ in the data stream that presently described Continuous Game is formed according to the data of described sampling cache module output, described nonrandom informational probability ρ (III) according to the following formula is calculated:

$\mathrm{\&rho;}\&ap;\frac{{\mathrm{\&Sigma;}}_{\mathrm{\&tau;}=1}^n|V_{\mathrm{\&tau;}}^S-V_{\mathrm{\&tau;}}^B|}{nV}---\left(III\right);$

(7) data outputting module: for the value of calculated for described nonrandom informational probability estimation module described nonrandom informational probability is exported.Data outputting module can be digital output module, exports result with the form of numerical value；Can also be the combination of digital output module and images outputting module, export result in a graphical form.Data outputting module can be docked with other system, so that other system can call the data processed result of this system data output module output.

In the process implemented, by the real time computation system of nonrandom informational probability, time series data in the on-line continuous betting data stream of input is processed in real time, when current activity data bucket is full time, calculate V according to all of exchange hand in current activity data bucket and conclusion of the business price differential_τ ^BAnd V_τ ^S, then update the activity data bucket queue calculating ρ；When current activity data bucket queue fills up sampling buffer, calculate nonrandom informational probability ρ value according to current queue immediately.The time produced due to ρ value is unfixed, so needing joining day stamp conveniently to display as index in the process of Data Integration.

Utilize above-mentioned real time computation system, the value of nonrandom informational probability under current time can be obtained by above-mentioned real-time computing technique, and take following methods to calculate the dependency of itself and price fluctuation further:

(1) calculate the extreme difference from the initial certain time concluded price in interval of current time or standard deviation, be designated as price wave motion VarPrice；

(2) calculate the covariance Cov (ρ, VarPrice) of the value of nonrandom informational probability and price wave motion in this time interval, be in this time interval nonrandom informational probability and the dependency of price wave motion.

Above-mentioned time interval is set in 1.5 to 2.5 hours, and the value of nonrandom informational probability is the most notable with the dependency of the price wave motion of corresponding time interval.

In one specific embodiment of Fig. 3, Fig. 4, Fig. 5 respectively present invention sampling time interval be 1 minute time series data stream, nonrandom informational probability in this time series data stream is carried out to the real-time result of calculating, the dependency of this nonrandom informational probability and price wave motion in the time interval of 120 minutes.Be can be seen that by Fig. 4 and Fig. 5, nonrandom informational probability and price wave motion significant correlation.As can be seen here, the method that the value of the nonrandom informational probability that said method calculates may be used for judging Continuous Game market fluctuation.

Additionally, set the number n of activity data bucket filling up sampling buffer and sliding window size value is 50, set the maximum V of the summation of activity data in each data bucket as V_D/ 200, and set the sampling time interval T obtaining time series data stream respectively₀It it is 1 minute, 10 seconds, 30 seconds, utilize above-mentioned real time computation system by the value of the nonrandom informational probability of above-mentioned real-time computing technique acquisition same day, result shows: the value performance of nonrandom informational probability is similar, and all shows the strong correlation with price wave motion.

Additionally, set the number n of activity data bucket filling up sampling buffer and sliding window size value is 50, set the sampling time interval T obtaining time series data stream₀It is 1 minute, and sets the maximum V of the summation of activity data in each data bucket respectively as V_D/ 200, V_D/ 20, V_D/ 5 (is wherein V_DThe per day exchange hand on past 50 days), utilize above-mentioned real time computation system by the value of the nonrandom informational probability of above-mentioned real-time computing technique acquisition same day, result shows: the value performance of nonrandom informational probability is similar, and all shows the strong correlation with price wave motion.

As can be seen here, the present invention not only can be implemented in line and rolls calculating in real time, but also there is better stability and adaptivity, and eliminate the obstacle that unobservable parameter is estimated, overcome the inaccurate defect that time dimension algorithm is concluded the business at present deal especially high frequency.

It should be understood by those skilled in the art that the embodiments of the invention shown in foregoing description are only used as citing for the present invention is described, and should not be taken as limiting the scope of the invention.

As can be seen here, the purpose of the present invention completely and is effectively achieved.The function of the present invention and structural principle are shown in an embodiment and are illustrated, when without departing substantially from described principle, embodiment can do any amendment.So, present invention comprises all variant embodiment based on claim spirit and right.

Claims (8)

1. the real-time computing technique of nonrandom informational probability in a Continuous Game data stream, it is characterised in that comprise the following steps:

(1) data stream Continuous Game formed is sampled by constant duration, and obtaining sampling time interval is T₀Time series data stream, be designated as: { T₁,P₁,V₁}、{T₂,P₂,V₂}…{T_i,P_i,V_i..., wherein, T_iFor ith sample moment, P_iFor the End-price in ith sample moment, V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；

(2) from described time series data stream, extract activity data stream, be designated as: { V₁,V₂,…,V_i... }, wherein, the T between ith sample moment and the i-th-1 sampling instant₀Total volume V in interval_iIt is regarded as the V that can split_iPart unit exchange hand；

(3) by described activity data stream according to etc. exchange hand interval carry out being polymerized and splitting, and be sequentially filled in n data bucket, the maximum exchange hand summation that each data bucket holds is V；

(4) whether full described data bucket is detected in real time, i.e. in described data bucket, whether the summation of activity data has reached V, if it is, carry out step (5)；If it is not, then return to step (1)；

(5) (I) and (II) calculates the exchange hand V bought in each described data bucket according to the following formula respectively_τ ^BWith the exchange hand V sold_τ ^SValue, and by (V_τ ^B,V_τ ^S) it is filled into sampling buffer area as a paired sampled value；

$V_{\mathrm{\&tau;}}^B=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;S\left(\frac{P_i-P_{i-1}}{ATR}\right)---\left(I\right)$

$V_{\mathrm{\&tau;}}^S=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;\&lsqb;1-S\left(\frac{P_i-P_{i-1}}{ATR}\right)\&rsqb;=V-V_{\mathrm{\&tau;}}^B---\left(II\right)$

Wherein, t (τ) is last sampling instant, V in τ described data bucket_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；P_i、P_i-1The respectively End-price of ith sample moment and the i-th-1 sampling instant, ATR is the average true wave amplitude of described time series data stream, and V is the maximum exchange hand summation held in each described data bucket；S is Sigmoid function, and its definition is as follows:B is the parameter of control function smoothness；

(6) whether full detect in real time described sampling buffer area, namely detect the value of the number m of paired sampled value in described sampling buffer area, if m is < n, return to step (1) or (5)；If m=n, then carry out step (7)；

(7) according to the following formula (III) calculates nonrandom informational probability ρ output in real time in the data stream that presently described Continuous Game is formed:

$\mathrm{\&rho;}\&ap;\frac{{\mathrm{\&Sigma;}}_{\mathrm{\&tau;}=1}^n|V_{\mathrm{\&tau;}}^S-V_{\mathrm{\&tau;}}^B|}{nV}---\left(III\right).$

2. the real-time computing technique of nonrandom informational probability in Continuous Game data stream as claimed in claim 1, it is characterised in that described interval T₀Value be 1 minute, 10 seconds or 30 seconds.

3. the real-time computing technique of nonrandom informational probability in Continuous Game data stream as claimed in claim 1, it is characterised in that the value of the number n of described data bucket is 50.

4. the real time computation system of nonrandom informational probability in a Continuous Game data stream, it is characterised in that including:

(1) sampling module: the data stream for Continuous Game is formed is sampled by constant duration, and obtaining sampling time interval is T₀Time series data stream, be designated as: { T₁,P₁,V₁}、{T₂,P₂,V₂}…{T_i,P_i,V_i..., wherein, T_iFor ith sample moment, P_iFor the End-price in ith sample moment, V_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；

(2) data extraction module: for extracting activity data stream from the described time series data stream exported by described sampling module, be designated as: { V₁,V₂,…,V_i... }, wherein, the T between ith sample moment and the i-th-1 sampling instant₀Total volume V in interval_iIt is regarded as the V that can split_iPart unit exchange hand；

(3) data concentrating module: for by the described activity data stream exported by described data extraction module according to etc. exchange hand interval carry out being polymerized and splitting, and it being sequentially filled in n data bucket, the maximum exchange hand summation that each data bucket holds is V；Whether full it is simultaneously used for detecting described data bucket in real time, i.e. in described data bucket, whether the summation of activity data has reached V, if it is, proceed to exchange hand classification of type computing module；If it is not, then return to described sampling module；

(4) exchange hand classified counting module: for calculating by the exchange hand V bought in each described data bucket of described data concentrating module output_τ ^BWith the exchange hand V sold_τ ^SValue, wherein, the exchange hand V bought_τ ^BWith the exchange hand V sold_τ ^SValue respectively according to the following formula (I) and (II) be calculated:

$V_{\mathrm{\&tau;}}^B=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;S\left(\frac{P_i-P_{i-1}}{ATR}\right)---\left(I\right)$

$V_{\mathrm{\&tau;}}^S=\underset{i=t(\mathrm{\&tau;}-1)+1}{\overset{t\left(\mathrm{\&tau;}\right)}{\&Sigma;}}{V}_i\&CenterDot;\&lsqb;1-S\left(\frac{P_i-P_{i-1}}{ATR}\right)\&rsqb;=V-V_{\mathrm{\&tau;}}^B---\left(II\right)$

Wherein, t (τ) is last sampling instant, V in τ described data bucket_iFor the T between ith sample moment and the i-th-1 sampling instant₀Total volume in interval；P_i、P_i-1The respectively End-price of ith sample moment and the i-th-1 sampling instant, ATR is the average true wave amplitude of described time series data stream, and V is the maximum exchange hand summation held in each described data bucket；S is Sigmoid function, and its definition is as follows:B is the parameter of control function smoothness；

(5) sampling cache module: for the exchange hand V that will buy in each described data bucket exported by described exchange hand classified counting module_τ ^BWith the exchange hand V sold_τ ^SValue be filled into sampling buffer area as a paired sampled value, and it is whether full for detecting described sampling buffer area in real time, namely the value of the number m of paired sampled value in described sampling buffer area is detected, if m is < n, return to described sampling module or described exchange hand classified counting module；If m=n, then proceed to nonrandom informational probability estimation module；

(6) nonrandom informational probability estimation module: for calculating nonrandom informational probability ρ in the data stream that presently described Continuous Game is formed according to the data of described sampling cache module output, described nonrandom informational probability ρ (III) according to the following formula is calculated:

$\mathrm{\&rho;}\&ap;\frac{{\mathrm{\&Sigma;}}_{\mathrm{\&tau;}=1}^n|V_{\mathrm{\&tau;}}^S-V_{\mathrm{\&tau;}}^B|}{nV}---\left(III\right);$

(7) data outputting module: for the value of calculated for described nonrandom informational probability estimation module described nonrandom informational probability is exported.

5. the real time computation system of nonrandom informational probability in Continuous Game data stream as claimed in claim 4, it is characterised in that described interval T₀Value be 1 minute, 10 seconds or 30 seconds.

6. the real time computation system of nonrandom informational probability in Continuous Game data stream as claimed in claim 4, it is characterised in that the value of the number n of described data bucket is 50.

7. the real time computation system of nonrandom informational probability in Continuous Game data stream as claimed in claim 4, it is characterised in that described data outputting module is digital output module.

8. the real time computation system of nonrandom informational probability in Continuous Game data stream as claimed in claim 4, it is characterised in that described data outputting module is the combination of digital output module and images outputting module.