Disclosure of Invention
Aiming at the problems of the background art, the invention provides a short-time passenger flow prediction system for urban rail transit stations, which is innovative in that: the urban rail transit station short-time passenger flow prediction system consists of an AFC device, a video passenger flow statistic device, a data screening module, a data preprocessing module, a Federal Kalman filtering module, a neural network prediction module and a database module;
the AFC equipment and the video passenger flow statistics equipment are both connected with the data screening module; the data screening module is respectively connected with the data preprocessing module, the Federal Kalman filtering module and the database module; the data preprocessing module is respectively connected with the Federal Kalman filtering module and the database module; the federal Kalman filtering module is respectively connected with the neural network prediction module and the database module; the database module is connected with the neural network prediction module;
the AFC equipment can output the counted passenger flow volume data to the data screening module in real time, and the passenger flow volume data output by the AFC equipment is recorded as AFC data;
the video passenger flow statistics equipment can output the counted passenger flow data to the data screening module in real time, and the passenger flow data output by the video passenger flow statistics equipment is recorded as video data;
the data screening module can identify the current date type, and after the date type is identified, the data screening module transmits the received AFC data and video data to the database module for classified storage; recording AFC data stored in the database module as AFC historical data, and recording video data stored in the database module as video historical data; then, the data screening module controls the data preprocessing module and the Federal Kalman filtering module according to the identified date type: the date types comprise normal days and special days, wherein working days and double holidays belong to the normal days, and holidays and activity days belong to the special days; if the date type of the current date is identified as the normal date, the data screening module outputs a first control signal to the data preprocessing module and outputs a second control signal to the Federal Kalman filtering module; if the date type of the current date is identified as the special date, the data screening module outputs a control signal III to the data preprocessing module and outputs a control signal IV to the Federal Kalman filtering module;
when the data preprocessing module receives the control signal III, the data preprocessing module does not work; when the data preprocessing module receives a control signal, the data preprocessing module calls a plurality of AFC historical data and a plurality of video historical data from the database module, and then the data preprocessing module respectively processes the AFC historical data and the video historical data;
when the data preprocessing module processes AFC historical data, the average value of a plurality of AFC historical data is calculated firstly
And standard deviation s
AObtaining a first threshold range
Then judging whether each AFC historical data is in a first threshold range: recording the AFC historical data in the first threshold range as good values, recording the AFC historical data not in the first threshold range as bad values, and if all the AFC historical data are good values, outputting the good values to the Federal Kalman filtering module by the data preprocessing module
If all AFC historical data are bad values, the data preprocessing module compares x with the multiple AFC historical data
AThe one with the minimum difference value is output to a Federal Kalman filtering module; if the AFC historical data have good values and bad values, the bad values are removed, and the average value of the remaining good values is calculated again
Then will be
Outputting to a Federal Kalman filtering module;
when the data preprocessing module processes the historical video data, the average value of a plurality of historical video data is calculated firstly
And standard deviation s
BObtaining a second threshold range
Then judging whether each video history data is in a second threshold value range: recording the video history data within the second threshold value range as effective values, recording the video history data not within the second threshold value range as invalid values, and if all the video history data are effective values, outputting the effective values to the Federal Kalman filtering module by the data preprocessing module
If all the video history data are invalid values,the data preprocessing module neutralizes the plurality of video history data
The one with the minimum difference value is output to a Federal Kalman filtering module; if the plurality of video historical data have both effective values and invalid values, rejecting the invalid values and recalculating the average value of the remaining effective values
Then will be
Outputting to a Federal Kalman filtering module;
in the output signals of the data preprocessing module, the output signal corresponding to AFC historical data is marked as a signal I, and the output signal corresponding to video historical data is marked as a signal II;
when the Federal Kalman filtering module receives the control signal II, performing information fusion processing on the signal I and the signal II output by the data preprocessing module, and then outputting a processing result to the neural network prediction module; when the Federal Kalman filtering module receives the control signal, the Federal Kalman filtering module calls the latest AFC historical data and video historical data from the database module, then performs information fusion processing on the called AFC historical data and video historical data, and then outputs a processing result to the neural network prediction module;
after the neural network prediction module receives the processing result, the neural network prediction module calls a plurality of current data from the database module, then the processing result and the plurality of current data are used as input vectors to carry out passenger flow prediction processing to obtain passenger flow prediction data, and then the passenger flow prediction data are output outwards; and when the passenger flow volume prediction data are output outwards, the neural network prediction module also sends the passenger flow volume prediction data to the database module for storage.
The principle of the invention is as follows: in the invention, passenger flow information comes from AFC equipment (automatic fare collection system) and video passenger flow statistical equipment, AFC data and video data obtained by the two equipment are used for a neural network prediction module after being subjected to fusion processing by a Federal Kalman filtering module, and the two data are subjected to fusion processing, so that data errors caused by factors such as missing reading, misreading and ticket mixing of the AFC equipment can be reduced, the problem that the video passenger flow statistical equipment is greatly influenced by light and environment, particularly, when passenger flows are dense and overlapped with each other, the detection precision is reduced is solved, the actual passenger flow is reflected more accurately, and compared with the method that data is obtained by adopting single equipment, the detection coverage range is wider when the two equipment is used for obtaining the data, and the data source is more comprehensive; aiming at the problems of normality and abnormality in the background technology, the inventor sets a data screening module and a data preprocessing module in the scheme, wherein the data screening module is used for sorting and classifying and storing data; meanwhile, the data screening module can also output the data belonging to the normal day to the data preprocessing module and directly output the data belonging to the special day to the Federal Kalman filtering module; the passenger flow data on the normal day has a change rule taking a week as a cycle, the passenger flow distribution rules on the same day in different weeks are basically consistent in time, and the existing conventional method is to take the average passenger flow of the previous week or history. When the last week passenger flow is taken, the fluctuation influence of the last week due to some accidental factors is also input into the neural network predictor. Thus, the last week's passenger flow volume will no longer be appropriate as a predictive input; the historical average passenger flow is a compromise method adopting averaging. Therefore, the invention is provided with a data preprocessing module which can remove abnormal values in the historical data so as to eliminate the influence of abnormal data generated by accidental factors such as interference on the system, finally, the data used for the neural network prediction module is more reasonable, the prediction stability is enhanced, and the prediction precision is improved.
Based on the scheme, the invention also provides the following preferable scheme aiming at the Federal Kalman filtering module: the Federal Kalman filtering module comprises a local filtering module I, a local filtering module II and an information fusion module; the information fusion module is provided with two input ends and three output ends, and the three output ends of the information fusion module are respectively marked as a first feedback signal output end, a second feedback signal output end and a main output end; the output end of the first local filtering module and the output end of the second local filtering module are connected with the two input ends of the information fusion module in a one-to-one correspondence manner, the first feedback signal output end is connected with the feedback signal receiving end of the first local filtering module, the second feedback signal output end is connected with the feedback signal receiving end of the second local filtering module, and the main output end is connected with the neural network prediction module; when the data preprocessing module outputs a signal to the Federal Kalman filtering module, the data preprocessing module outputs the signal I to the local filtering module I, and the data preprocessing module outputs the signal II to the local filtering module II; when the Federal Kalman filtering module calls AFC historical data and video historical data from the database module, the corresponding AFC historical data is input into the local filtering module I, and the corresponding video historical data is input into the local filtering module II.
Based on the system, the invention also provides a method for predicting the short-term passenger flow of the urban rail transit station, and the method is based on the system for predicting the short-term passenger flow of the urban rail transit station, and the specific method is as follows: the short-time passenger flow prediction method for the urban rail transit station comprises the following steps:
in the running process of the short-time passenger flow prediction system of the urban rail transit station, AFC equipment and video passenger flow statistical equipment periodically output AFC data and video data to a data screening module; after AFC data and video data are received each time, the urban rail transit station short-time passenger flow prediction system carries out passenger flow prediction operation according to the following steps:
1) the data screening module identifies the current date type, and then transmits the received AFC data and video data to the database module for classified storage; then: if the date type of the current date is identified as the normal date, the data screening module outputs a first control signal to the data preprocessing module and outputs a second control signal to the Federal Kalman filtering module, and the step 2) is carried out; if the date type of the current date is identified as the special date, the data screening module outputs a control signal three to the data preprocessing module and outputs a control signal four to the Federal Kalman filtering module, and the step 3) is carried out;
2) the data preprocessing module calls a plurality of AFC historical data and a plurality of video historical data from the database module, then the data preprocessing module respectively processes the AFC historical data and the video historical data to generate a corresponding signal I and a signal II, and the step 4 is carried out;
3) the Federal Kalman filtering module calls the latest AFC historical data and video historical data from the database module, then performs information fusion processing on the called AFC historical data and video historical data, then outputs the processing result to the neural network prediction module, and enters step 5);
4) the Federal Kalman filtering module performs information fusion processing on the signal I and the signal II output by the data preprocessing module, then outputs a processing result to the neural network prediction module, and the step 5 is carried out;
5) the neural network prediction module calls a plurality of current data from the database module, then the processing result and the plurality of current data are used as input vectors to carry out passenger flow prediction processing to obtain passenger flow prediction data, then the passenger flow prediction data are output outwards, and the neural network prediction module sends the passenger flow prediction data to the database module for storage while the passenger flow prediction data are output outwards.
Based on the method, the invention also provides the following preferable scheme aiming at the Federal Kalman filtering module and the data processing mode thereof, the specific structure of the Federal Kalman filtering module is as described above, and the Federal Kalman filtering module carries out Federal Kalman filtering processing according to the following mode:
(1) the first local filtering module processes the first signal by adopting a linear Kalman filtering method to obtain a first filtering result, the first filtering result is output to the information fusion module, meanwhile, the second local filtering module processes the second signal by adopting a linear Kalman filtering method to obtain a second filtering result, and the second filtering result is output to the information fusion module;
(2) the information fusion module carries out information fusion processing on the first filtering result and the second filtering result and then outputs the processing results to the neural network prediction module;
the information fusion module adopts a fusion algorithm for information fusion processing, which comprises the following steps:
wherein k represents the number of steps,
for a global optimum estimate, P
g(k) For the variance of the global optimum estimate value,
is an estimate, P, of the filtering result
Z(k) For the result of the filtering-a corresponding variance,
represents a pair P
Z(k) The inversion is carried out on the basis of the obtained data,
is an estimated value, P, corresponding to the filtering result two
V(k) For the variance corresponding to the filtering result two,
represents a pair P
V(k) Inversion is carried out;
the information fusion module is to
Feeding back to the local filtering module I, and the information fusion module will
And
feedback to local filtering modeBlock two, β
Z(k) Distribution coefficient of the corresponding local filter block one in the k step, β
V(k) For the partition coefficient of the corresponding local filter block two in step k, β
Z(k)+β
V(k)=1,
Representation pair β
Z(k) The inversion is carried out on the basis of the obtained data,
representation pair β
V(k) And (6) inversion.
The beneficial technical effects of the invention are as follows: the invention provides a short-time passenger flow prediction system and method for urban rail transit stations, which can improve the diversity, comprehensiveness and accuracy of data sources, improve the accuracy of data and finally enable prediction results to be more accurate.
Detailed Description
A short-time passenger flow prediction system for urban rail transit stations is innovative in that: the urban rail transit station short-time passenger flow prediction system is composed of an AFC device 1, a video passenger flow statistic device 2, a data screening module 3, a data preprocessing module 4, a Federal Kalman filtering module 5, a neural network prediction module 6 and a database module 7;
the AFC device 1 and the video passenger flow statistical device 2 are both connected with a data screening module 3; the data screening module 3 is respectively connected with the data preprocessing module 4, the Federal Kalman filtering module 5 and the database module 7; the data preprocessing module 4 is respectively connected with the Federal Kalman filtering module 5 and the database module 7; the federal Kalman filtering module 5 is respectively connected with the neural network prediction module 6 and the database module 7; the database module 7 is connected with the neural network prediction module 6;
the AFC device 1 can output the counted passenger flow volume data to the data screening module 3 in real time, and the passenger flow volume data output by the AFC device 1 is recorded as AFC data;
the video passenger flow statistics device 2 can output the counted passenger flow data to the data screening module 3 in real time, and the passenger flow data output by the video passenger flow statistics device 2 is recorded as video data;
the data screening module 3 can identify the current date type, and after the date type is identified, the data screening module 3 transmits the received AFC data and video data to the database module 7 for classified storage; the AFC data stored in the database module 7 is recorded as AFC historical data, and the video data stored in the database module 7 is recorded as video historical data; then, the data screening module 3 controls the data preprocessing module 4 and the federal kalman filter module 5 according to the identified date type: the date types comprise normal days and special days, wherein working days and double holidays belong to the normal days, and holidays and activity days belong to the special days; if the date type of the current date is identified as the normal date, the data screening module 3 outputs a first control signal to the data preprocessing module 4 and outputs a second control signal to the federal Kalman filtering module 5; if the date type of the current date is identified as the special date, the data screening module 3 outputs a control signal III to the data preprocessing module 4 and outputs a control signal IV to the Federal Kalman filtering module 5;
when the data preprocessing module 4 receives the control signal III, the data preprocessing module 4 does not work; when the data preprocessing module 4 receives the control signal one, the data preprocessing module 4 calls a plurality of AFC historical data and a plurality of video historical data from the database module 7, and then the data preprocessing module 4 respectively processes the AFC historical data and the video historical data;
when the
data preprocessing module 4 processes the AFC historical data, the average value of a plurality of AFC historical data is calculated firstly
And standard deviation s
AObtaining a first threshold range
Then judging whether each AFC historical data is in a first threshold range: recording the AFC historical data in the first threshold range as good values, recording the AFC historical data not in the first threshold range as bad values, and if all the AFC historical data are good values, outputting the good values to the Federal Kalman
filtering module 5 by the
data preprocessing module 4
If all AFC historical data are bad values, the
data preprocessing module 4 neutralizes the multiple AFC historical data
The one with the minimum difference value is output to a Federal Kalman
filtering module 5; if the AFC historical data have good values and bad values, the bad values are removed, and the average value of the remaining good values is calculated again
Then will be
Outputting to a Federal Kalman
filtering module 5;
when the
data preprocessing module 4 processes the video history data, the average value of a plurality of video history data is calculated first
And standard deviation s
BObtaining a second threshold range
Then judging whether each video history data is in a second threshold value range: recording the video history data in the second threshold range as effective values, and recording the videos not in the second threshold rangeRecording the historical data as invalid values, and if all the video historical data are valid values, outputting the data to the Federal Kalman
filtering module 5 by the
data preprocessing module 4
If all the video history data are invalid values, the
data preprocessing module 4 neutralizes the plurality of video history data
The one with the minimum difference value is output to a Federal Kalman
filtering module 5; if the plurality of video historical data have both effective values and invalid values, rejecting the invalid values and recalculating the average value of the remaining effective values
Then will be
Outputting to a Federal Kalman
filtering module 5;
in the output signals of the data preprocessing module 4, the output signal corresponding to the AFC historical data is recorded as a signal I, and the output signal corresponding to the video historical data is recorded as a signal II;
when receiving the control signal II, the Federal Kalman filtering module 5 performs information fusion processing on the signal I and the signal II output by the data preprocessing module 4, and then outputs a processing result to the neural network prediction module 6; when the federal kalman filter module 5 receives the control signal, the federal kalman filter module 5 calls the latest AFC historical data and video historical data from the database module 7, then performs information fusion processing on the called AFC historical data and video historical data, and then outputs the processing result to the neural network prediction module 6;
after the neural network prediction module 6 receives the processing result, the neural network prediction module 6 calls a plurality of current data from the database module 7, then the processing result and the plurality of current data are used as input vectors to perform passenger flow prediction processing to obtain passenger flow prediction data, then the passenger flow prediction data are output outwards, and the neural network prediction module 6 sends the passenger flow prediction data to the database module 7 for storage while the passenger flow prediction data are output outwards.
Further, the federal Kalman filtering module 5 comprises a local filtering module I5-1, a local filtering module II 5-2 and an information fusion module 5-3; the information fusion module 5-3 is provided with two input ends and three output ends, and the three output ends of the information fusion module 5-3 are respectively marked as a first feedback signal output end, a second feedback signal output end and a main output end; the output end of the local filtering module I5-1 and the output end of the local filtering module II 5-2 are correspondingly connected with the two input ends of the information fusion module 5-3, the first feedback signal output end is connected with the feedback signal receiving end of the local filtering module I5-1, the second feedback signal output end is connected with the feedback signal receiving end of the local filtering module II 5-2, and the main output end is connected with the neural network prediction module 6; when the data preprocessing module 4 outputs a signal to the Federal Kalman filtering module 5, the data preprocessing module 4 outputs the signal I to the local filtering module I5-1, and the data preprocessing module 4 outputs the signal II to the local filtering module II 5-2; when the Federal Kalman filtering module 5 calls AFC historical data and video historical data from the database module 7, the corresponding AFC historical data is input into the local filtering module I5-1, and the corresponding video historical data is input into the local filtering module II 5-2.
A short-time passenger flow prediction method for an urban rail transit station comprises a short-time passenger flow prediction system for the urban rail transit station, wherein the short-time passenger flow prediction system for the urban rail transit station consists of an AFC (automatic frequency control) device 1, a video passenger flow statistic device 2, a data screening module 3, a data preprocessing module 4, a Federal Kalman filtering module 5, a neural network prediction module 6 and a database module 7;
the AFC device 1 and the video passenger flow statistical device 2 are both connected with a data screening module 3; the data screening module 3 is respectively connected with the data preprocessing module 4, the Federal Kalman filtering module 5 and the database module 7; the data preprocessing module 4 is respectively connected with the Federal Kalman filtering module 5 and the database module 7; the federal Kalman filtering module 5 is respectively connected with the neural network prediction module 6 and the database module 7; the database module 7 is connected with the neural network prediction module 6;
the AFC device 1 can output the counted passenger flow volume data to the data screening module 3 in real time, and the passenger flow volume data output by the AFC device 1 is recorded as AFC data;
the video passenger flow statistics device 2 can output the counted passenger flow data to the data screening module 3 in real time, and the passenger flow data output by the video passenger flow statistics device 2 is recorded as video data;
the data screening module 3 can identify the current date type, and after the date type is identified, the data screening module 3 transmits the received AFC data and video data to the database module 7 for classified storage; the AFC data stored in the database module 7 is recorded as AFC historical data, and the video data stored in the database module 7 is recorded as video historical data; then, the data screening module 3 controls the data preprocessing module 4 and the federal kalman filter module 5 according to the identified date type: the date types comprise normal days and special days, wherein working days and double holidays belong to the normal days, and holidays and activity days belong to the special days; if the date type of the current date is identified as the normal date, the data screening module 3 outputs a first control signal to the data preprocessing module 4 and outputs a second control signal to the federal Kalman filtering module 5; if the date type of the current date is identified as the special date, the data screening module 3 outputs a control signal III to the data preprocessing module 4 and outputs a control signal IV to the Federal Kalman filtering module 5;
when the data preprocessing module 4 receives the control signal III, the data preprocessing module 4 does not work; when the data preprocessing module 4 receives the control signal one, the data preprocessing module 4 calls a plurality of AFC historical data and a plurality of video historical data from the database module 7, and then the data preprocessing module 4 respectively processes the AFC historical data and the video historical data;
when the
data preprocessing module 4 processes the AFC historical data, the average value of a plurality of AFC historical data is calculated firstly
And standard deviation s
AObtaining a first threshold range
Then judging whether each AFC historical data is in a first threshold range: recording the AFC historical data in the first threshold range as good values, recording the AFC historical data not in the first threshold range as bad values, and if all the AFC historical data are good values, outputting the good values to the Federal Kalman
filtering module 5 by the
data preprocessing module 4
If all AFC historical data are bad values, the
data preprocessing module 4 sums x in the multiple AFC historical data
AThe one with the minimum difference value is output to a Federal Kalman
filtering module 5; if the AFC historical data have good values and bad values, the bad values are removed, and the average value of the remaining good values is calculated again
Then will be
Outputting to a Federal Kalman
filtering module 5;
when the
data preprocessing module 4 processes the video history data, the average value of a plurality of video history data is calculated first
And standard deviation s
BObtaining a second threshold range
Then judging whether each video history data is in a second threshold value range: recording the video history data within the second threshold range as effective values, recording the video history data not within the second threshold range as invalid values, and if all the video history data are effective values, outputting the video history data to the Federal
Kalman filtering module 5 by the
data preprocessing module 4
If all the video history data are invalid values, the
data preprocessing module 4 neutralizes the plurality of video history data
The one with the minimum difference value is output to a Federal
Kalman filtering module 5; if the plurality of video historical data have both effective values and invalid values, rejecting the invalid values and recalculating the average value of the remaining effective values
Then will be
Outputting to a Federal
Kalman filtering module 5;
in the output signals of the data preprocessing module 4, the output signal corresponding to the AFC historical data is recorded as a signal I, and the output signal corresponding to the video historical data is recorded as a signal II;
when receiving the control signal II, the Federal Kalman filtering module 5 performs information fusion processing on the signal I and the signal II output by the data preprocessing module 4, and then outputs a processing result to the neural network prediction module 6; when the federal kalman filter module 5 receives the control signal, the federal kalman filter module 5 calls the latest AFC historical data and video historical data from the database module 7, then performs information fusion processing on the called AFC historical data and video historical data, and then outputs the processing result to the neural network prediction module 6;
after the neural network prediction module 6 receives the processing result, the neural network prediction module 6 calls a plurality of current data (the current data is actual passenger flow volume data obtained by adopting a technical means) from the database module 7, then the processing result and the plurality of current data are used as input vectors to carry out passenger flow volume prediction processing to obtain passenger flow volume prediction data, then the passenger flow volume prediction data are output outwards, and the neural network prediction module 6 also sends the passenger flow volume prediction data to the database module 7 for storage while the passenger flow volume prediction data are output outwards;
the innovation lies in that: the short-time passenger flow prediction method for the urban rail transit station comprises the following steps:
in the running process of the short-time passenger flow prediction system of the urban rail transit station, AFC equipment 1 and video passenger flow statistical equipment 2 periodically output AFC data and video data to a data screening module 3; after AFC data and video data are received each time, the urban rail transit station short-time passenger flow prediction system carries out passenger flow prediction operation according to the following steps:
1) the data screening module 3 identifies the current date type, and then transmits the received AFC data and video data to the database module 7 for classified storage; then: if the date type of the current date is identified as the normal date, the data screening module 3 outputs a control signal I to the data preprocessing module 4 and outputs a control signal II to the Federal Kalman filtering module 5, and the step 2) is carried out; if the date type of the current date is identified as the special date, the data screening module 3 outputs a control signal three to the data preprocessing module 4 and outputs a control signal four to the federal Kalman filtering module 5, and the step 3) is carried out;
2) the data preprocessing module 4 calls a plurality of AFC historical data and a plurality of video historical data from the database module 7, then the data preprocessing module 4 respectively processes the AFC historical data and the video historical data to generate a corresponding signal I and a signal II, and the step 4 is carried out;
3) the Federal Kalman filtering module 5 calls the latest AFC historical data and video historical data from the database module 7, then performs information fusion processing on the called AFC historical data and video historical data, then outputs the processing result to the neural network prediction module 6, and enters step 5);
4) the Federal Kalman filtering module 5 performs information fusion processing on the signal I and the signal II output by the data preprocessing module 4, then outputs a processing result to the neural network prediction module 6, and enters the step 5);
5) the neural network prediction module 6 calls a plurality of current data from the database module 7, then the processing result and the plurality of current data are used as input vectors to carry out passenger flow volume prediction processing to obtain passenger flow volume prediction data, then the passenger flow volume prediction data are output outwards, and the neural network prediction module 6 also sends the passenger flow volume prediction data to the database module 7 for storage while the passenger flow volume prediction data are output outwards.
Further, the federal Kalman filtering module 5 comprises a local filtering module I5-1, a local filtering module II 5-2 and an information fusion module 5-3; the information fusion module 5-3 is provided with two input ends and three output ends, and the three output ends of the information fusion module 5-3 are respectively marked as a first feedback signal output end, a second feedback signal output end and a main output end; the output end of the local filtering module I5-1 and the output end of the local filtering module II 5-2 are correspondingly connected with the two input ends of the information fusion module 5-3, the first feedback signal output end is connected with the feedback signal receiving end of the local filtering module I5-1, the second feedback signal output end is connected with the feedback signal receiving end of the local filtering module II 5-2, and the main output end is connected with the neural network prediction module 6; when the data preprocessing module 4 outputs a signal to the Federal Kalman filtering module 5, the data preprocessing module 4 outputs the signal I to the local filtering module I5-1, and the data preprocessing module 4 outputs the signal II to the local filtering module II 5-2; when the Federal Kalman filtering module 5 calls AFC historical data and video historical data from the database module 7, the corresponding AFC historical data is input into the local filtering module I5-1, and the corresponding video historical data is input into the local filtering module II 5-2;
the federal kalman filter module 5 performs the federal kalman filter processing in the following manner:
(1) the local filtering module I5-1 processes the signal I by adopting a linear Kalman filtering method to obtain a filtering result I, the filtering result I is output to the information fusion module 5-3, meanwhile, the local filtering module II 5-2 processes the signal II by adopting a linear Kalman filtering method to obtain a filtering result II, and the filtering result II is output to the information fusion module 5-3;
(2) the information fusion module 5-3 carries out information fusion processing on the first filtering result and the second filtering result and then outputs the processing results to the neural network prediction module 6;
the information fusion module 5-3 adopts a fusion algorithm for information fusion processing, which is as follows:
wherein k represents the number of steps,
for a global optimum estimate, P
g(k) For the variance of the global optimum estimate value,
is an estimate, P, of the filtering result
Z(k) For the result of the filtering-a corresponding variance,
represents a pair P
Z(k) The inversion is carried out on the basis of the obtained data,
is an estimated value, P, corresponding to the filtering result two
V(k) For the variance corresponding to the filtering result two,
represents a pair P
V(k) Inversion is carried out;
the information fusion module 5-3 will
And
feeding back to the local filtering module I5-1, and the information fusion module 5-3 will
And
feeding back to local filtering module two 5-2 and β
Z(k) For the partition coefficient corresponding to local filter block one 5-1 at step k, β
V(k) For the partition coefficient corresponding to local filter block two 5-2 at step k, β
Z(k)+β
V(k)=1,
Representation pair β
Z(k) The inversion is carried out on the basis of the obtained data,
representation pair β
V(k) And (6) inversion.