CN117709530B - Case quantity prediction method for eliminating new year influence of lunar calendar - Google Patents
Case quantity prediction method for eliminating new year influence of lunar calendar Download PDFInfo
- Publication number
- CN117709530B CN117709530B CN202311714164.6A CN202311714164A CN117709530B CN 117709530 B CN117709530 B CN 117709530B CN 202311714164 A CN202311714164 A CN 202311714164A CN 117709530 B CN117709530 B CN 117709530B
- Authority
- CN
- China
- Prior art keywords
- month
- lunar calendar
- year
- prediction result
- new year
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 28
- 241000728173 Sarima Species 0.000 claims abstract description 48
- 238000012360 testing method Methods 0.000 claims abstract description 28
- 238000013112 stability test Methods 0.000 claims abstract description 18
- 238000005457 optimization Methods 0.000 claims abstract description 14
- 238000012545 processing Methods 0.000 claims abstract description 8
- 230000001932 seasonal effect Effects 0.000 claims description 22
- 230000009466 transformation Effects 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 8
- 230000000694 effects Effects 0.000 claims description 7
- 238000007689 inspection Methods 0.000 claims description 5
- 238000013507 mapping Methods 0.000 claims description 2
- 238000005259 measurement Methods 0.000 claims 2
- 230000004927 fusion Effects 0.000 claims 1
- PEDCQBHIVMGVHV-UHFFFAOYSA-N Glycerine Chemical compound OCC(O)CO PEDCQBHIVMGVHV-UHFFFAOYSA-N 0.000 abstract 2
- 238000013499 data model Methods 0.000 description 3
- 238000004364 calculation method Methods 0.000 description 2
- 230000008859 change Effects 0.000 description 2
- 238000011156 evaluation Methods 0.000 description 2
- 102100037651 AP-2 complex subunit sigma Human genes 0.000 description 1
- 238000012935 Averaging Methods 0.000 description 1
- 101000806914 Homo sapiens AP-2 complex subunit sigma Proteins 0.000 description 1
- 101001095088 Homo sapiens Melanoma antigen preferentially expressed in tumors Proteins 0.000 description 1
- 102100037020 Melanoma antigen preferentially expressed in tumors Human genes 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 230000009545 invasion Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000007781 pre-processing Methods 0.000 description 1
- 230000008569 process Effects 0.000 description 1
- 238000001744 unit root test Methods 0.000 description 1
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a case quantity prediction method for eliminating the influence of new years of lunar calendar, which comprises the following steps: 1) Acquiring and marking the month case quantity of the same case in the next calendar year; 2) Processing the marked month case quantity, and taking the average value of the case quantities of 1 month and 2 months in the same year as the case quantities of 1 month and 2 months in the same year; 3) Calculating a lunar calendar new year influence coefficient according to the marked month case; 4) Carrying out stability test on the moon case quantity data, and carrying out step 5) if the moon case quantity data passes the test; 5) Predicting the data subjected to the stability test through an SARIMA model to obtain a preliminary prediction result of the future month; 6) Residual error checking is carried out on residual errors obtained by constructing the model, and if the residual error checking is passed, the obtained preliminary prediction result is used as a preliminary optimization prediction result; 7) And if the future month is the month of the new lunar calendar year, multiplying the preliminary optimization prediction result by a coefficient to obtain a final prediction result of the future month.
Description
Technical Field
The invention belongs to the technical field of time sequence data prediction, and relates to a case quantity prediction method for eliminating the influence of new years of lunar calendar.
Background
The prediction of future trends based on historical data has been widely used, for example, patent literature of grant bulletin CN113919160B discloses a method and system for predicting urban crimes with fine granularity, which improves prediction performance by fusing time, space and category correlations, wherein fine granularity refers to predicting the number of cases a given area belongs to each crime category in a certain period of time in the future. The patent CN110503267B discloses a city financial invasion case prediction system and a prediction method based on a space-time scale self-adaptive model, which are used for revealing the geographical differentiation characteristics of crimes, can meet the requirements of the prediction precision of no accomplice crimes and realize the optimal arrangement of police cruising and defending work. The patent of the grant bulletin number CN113065347B discloses a criminal case judgment and prediction method and a criminal case judgment and prediction system based on multitask learning, which are used for predicting legal judgment by using original data of a case text. The patent of the grant bulletin number CN106845723B discloses a prediction method of occurrence of criminal cases, which is used for predicting a plurality of dimensions of time, space and occurrence probability by inputting the obtained influence factor data into a prediction system after preprocessing. Patent document CN109543909B discloses a method, apparatus and computer device for predicting the number of vehicle cases, for improving accuracy of predicting the number of vehicle cases occurring in a target geographical area.
Currently, the national court has a large amount of judicial statistics, and the case quantity is one of the important indexes. The scientific utilization of judicial statistics to predict the case quantity can help the court scientific judgment situation, predict the case quantity trend, play the role of case-by-early warning and assist decision-making.
At present, a learner adopts a regression model method for predicting the case quantity, regression fitting and prediction are required to be carried out on the case quantity through various influence factors, and the method has high dependence on the selected influence factors. Through observation, the case quantity of the court receipts shows seasonal fluctuation, so that the SARIMA model fused with seasonal factors can be adopted to perform initial prediction on the case quantity, information in residual errors is further extracted to further predict residual errors of the model, prediction errors of the SARIMA model are reduced, meanwhile, the case quantity is obviously influenced by the arrival of new lunar calendar, the traditional SARIMA model cannot judge the occurrence time of special events, and the prediction errors of the traditional SARIMA model on the case quantity of 1 month and 2 month in the past year are large.
Disclosure of Invention
Aiming at the problems existing in the prior art, the invention aims to provide a case quantity prediction method for eliminating the influence of the new year of lunar calendar, which considers that the case quantity data belongs to data with a one-dimensional structure and has obvious seasonal characteristics, so that the SARIMA model is adopted to perform preliminary prediction on the data for eliminating the influence of the new year of lunar calendar, meanwhile, the useful information for further extracting the residual items is further predicted by adopting the Bi-LSTM model and added with the preliminary prediction result, and then the month for eliminating the influence of the new year of lunar calendar is reduced to obtain the predicted data with better effect.
Firstly, acquiring month case quantity registered by the calendar year collection of a court, marking the month where the lunar calendar of the calendar is located, and establishing a time sequence of the month case quantity; then calculating to obtain an influence coefficient alpha of the new lunar calendar year through an original time sequence; meanwhile, the influence of the new year of the calendar in the original data is eliminated; predicting data for eliminating the influence of the new year of the lunar calendar through an SARIMA model to obtain a preliminary prediction result; meanwhile, the useful information of the residual error item is further extracted, further predicted by adopting a Bi-LSTM model and added with the primary prediction result; and finally multiplying the preliminary prediction result with the lunar calendar new year influence coefficient to obtain a final prediction result, so that a prediction model of the lunar calendar new year influence is considered, and the prediction effect is better.
The specific scheme comprises the following steps:
1) Acquiring month case quantity of the same case from the next calendar year, and marking data corresponding to the new calendar year of the calendar year; for the month where the lunar calendar is in the new year, the marked month case quantity is x (y,m′,l′), otherwise, is x (y,m,l); wherein x is the case quantity of month, y is year, m and m ′ are months, l is the month of the new year of the non-lunar calendar, and l' is the month of the new year of the lunar calendar; m ′ =1 or 2, m+.m ′, l+.l';
2) Processing the month case quantity marked in the step 1), taking the average value of the case quantities of 1 month and 2 months in the same year as the case quantities of 1 month and 2 months in the same year, and eliminating the influence of the new lunar calendar year on data;
3) Calculating the lunar calendar new year influence coefficient according to the lunar calendar case marked in the step 1);
4) Carrying out stability test on the data which is eliminated from the influence of the lunar calendar in the new year, if the test is passed, carrying out step 5), otherwise, carrying out linear transformation of multi-model combination on the data until the test is passed;
5) Constructing an SARIMA model based on the data subjected to the stability test, and predicting the data subjected to the stability test by the SARIMA model to obtain a preliminary prediction result of the future month;
6) Carrying out residual error test on residual error obtained by constructing the SARIMA model, if the residual error test is passed, the preliminary predicted result is the predicted result after preliminary optimization, if the residual error test is not passed, carrying out differential treatment on the residual error, and then predicting residual error item information by utilizing the Bi-LSTM model, wherein the predicted result after preliminary optimization is obtained by adding the predicted result of the residual error item and the preliminary predicted result;
7) And if the future month is not the month of the lunar calendar new year, taking the preliminary optimization prediction result obtained in the step 6) as the final prediction result of the future month, otherwise multiplying the preliminary optimization prediction result obtained in the step 6) by the lunar calendar new year influence coefficient obtained in the step 3) to obtain the final prediction result of the future month.
Further, in the step 2), the original data is processed to eliminate the influence of the lunar calendar new year on the data, and the lunar calendar new year has the most obvious influence on the 1 month and 2 month case amount of the calendar, so that the data of 1 month and 2 months are mainly processed when the lunar calendar new year and month are eliminated, and the specific operation method is as follows: the case amounts of 1 month and 2 months of calendar year are added respectively, and then the average value is given to 1 month and 2 months of calendar year, so that the data of eliminating the influence of lunar calendar new year is obtained, and the data expression of 1 month and 2 months obtained after the influence of lunar calendar new year is eliminated is as follows:
In formula (1), y represents year, l is a label for lunar New year, when l=1, represents month as lunar New year of the year, when l=0, represents non-lunar New year of the year, and l' are different labels intended to distinguish 0 from 1.
Further, in the step 3), the lunar calendar new year influence coefficient is calculated according to the months of the lunar calendar new year marked and the lunar calendar new year not marked in the original data, and the lunar calendar new year influence coefficient is mainly applied to the lunar calendar new year 1 month and 2 month case amount because the lunar calendar new year 1 month and 2 month case amount change is most sensitive to the lunar calendar new year, corresponding to the step 2), and the specific calculation steps are as follows: the lunar calendar month and lunar calendar month in 1 month and 2 months are added to the lunar calendar month and lunar calendar month, the respective duty ratio of lunar calendar month and lunar calendar month is calculated, and the obtained result is the lunar calendar month influence coefficient of lunar calendar month and lunar calendar month of lunar calendar month in 1 month and 2 months.
The expression of the lunar calendar new year influence coefficient is as follows:
In the formula (2), x (y,m,l) represents original time series data, y represents year, m represents month, m and m ′ represent different values, the value ranges are 1,2, l is a mark for new lunar calendar year, l and l ′ represent different values, when l=1, the month is new lunar calendar year of the current year, and when l=0, the month is non-new lunar calendar year of the current year; the formula (2) represents the lunar calendar new year influence coefficient of 1 month and 2 months.
Further, in the step 4), the stability test is performed on the data for eliminating the influence of the new year of lunar calendar, if the test passes, the next step is directly performed, otherwise, the data is subjected to linear transformation such as x=ln (x), x=cosx, x=x 2, x=ln (x) + cosx, and the like until the test passes.
Further, in the step 5), data which is subjected to the stability test and eliminates the influence of the new year of the lunar calendar is predicted by using a SARIMA model to obtain a preliminary prediction result, wherein the SARIMA model has the expression:
SARIMA ＝ (p,d,q)×(D,P,Q,s) (3)
In formula 3), P represents a non-seasonal autoregressive order, D represents a differential order, Q represents an order of a non-seasonal moving average, P represents a seasonal autoregressive order, D represents a seasonal differential order, Q represents an order of a seasonal moving average, and s represents a seasonal length, also called a period size.
Further, modeling is performed according to the SARIMA model described in the step 5), and the obtained preliminary prediction result expression should be:
Y(y,m,l)＝ SARIMA(p,d,q)(D,P,Q,s) (4)
Further, in the step 6), residual error checking is performed on the residual error obtained by modeling in the step 5), and if the residual error checking is passed, the preliminary prediction result is the result after preliminary optimization, namely Y' (y,m,l)＝Y(y,m,l); if the residual error test is not passed, carrying out differential processing on the residual error, then inputting the differential residual error and the original residual error into a Bi-LSTM model for predicting the residual error to obtain a residual error prediction result epsilon (y,m,l), wherein Y represents year, m represents month, l is a mark for lunar calendar, l=1 represents that the month is the lunar calendar of the current year, l=0 represents that the month is the non-lunar calendar of the current year, and the optimized result is the sum of the preliminary prediction result and the residual error prediction result, namely Y' (y,m,l)＝Y(y,m,l)+ε(y,m,l).
Further, in the step 7), the prediction result obtained by the SARIMA model is multiplied by the lunar calendar new year influence coefficient obtained according to the original time sequence to obtain a final prediction result, and the expression of the prediction result is as follows:
Wherein, alpha represents the new year influence coefficient of lunar calendar, and Y' (y,m,l) represents the prediction result of the step 6).
The invention also provides a server comprising a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program comprising instructions for performing the steps of the above method.
The invention also provides a computer readable storage medium having stored thereon a computer program, characterized in that the computer program when executed by a processor realizes the steps of the above method.
The invention has the following advantages:
The method comprises the steps of firstly eliminating the influence of the new year of the lunar calendar on the case quantity in the original data, and simultaneously obtaining the influence coefficient of the new year of the lunar calendar from the original data so as to calculate the data added with the influence of the new year of the lunar calendar subsequently; the method comprises the steps of firstly, carrying out a linear transformation on a data model, wherein the data model is used for solving the problem that the data model has a good prediction effect on data with obvious seasonal fluctuation, but the traditional SARIMA model has a certain requirement on the stability of a sequence, the stability test is required to be passed, and the prediction result is an effective result; and finally, multiplying the predicted value obtained after preliminary optimization by the lunar calendar new year influence coefficient to obtain a final predicted value. The model considers the influence of the new lunar calendar on the case quantity, introduces a method for improving the utilization rate of the SARIMA model, has higher effectiveness and reliability on the prediction result of the case quantity, and provides a new method and a new thought for the case quantity prediction.
Drawings
FIG. 1 is a flow chart of the present invention.
Detailed Description
The invention will now be described in further detail with reference to the accompanying drawings, which are given by way of illustration only and are not intended to limit the scope of the invention.
According to the figure 1, the case quantity prediction method based on SARIMA-lunar calendar new year comprises the following steps:
1) Acquiring month case quantity of the same case from the next calendar year, and marking data corresponding to the new calendar year of the calendar year;
2) Processing the original data to eliminate the influence of the lunar calendar on the data in the new year;
the method mainly processes the data of 1 month and 2 months when the new year and month of the lunar calendar are eliminated, and specifically comprises the steps of adding the data of 1 month and 2 months of the lunar calendar respectively, and then taking the average value to endow the data with the 1 month and 2 months of the lunar calendar respectively, so that the data of the new year of the lunar calendar are eliminated, and the time sequence expression obtained after the new year of the lunar calendar is eliminated is as follows:
In formula (1), y represents year, l is a label for lunar New year, when l=1, represents month as lunar New year of the year, when l=0, represents non-lunar New year of the year, and l' are different labels intended to distinguish 0 from 1.
3) Calculating the lunar calendar new year influence coefficient according to the month of the lunar calendar new year marked and the lunar calendar new year not marked in the original data;
The lunar calendar new year influence coefficient is calculated according to the months marked with lunar calendar new year and the lunar calendar new year not marked in the original data, and the lunar calendar new year influence coefficient is mainly acted on lunar calendar new year 1 month and 2 month case quantity because the change of lunar calendar new year 1 month and 2 month case quantity is most sensitive, the specific calculation steps are as follows: the method comprises the steps of adding the case quantities of the lunar New year in 1 month and the lunar New year in 2 months and the case quantity of the lunar New year in non-lunar New year, calculating the respective duty ratios of the lunar New year month in 1 month and the lunar New year and the case quantity of the lunar New year in 2 months, and obtaining the lunar New year influence coefficient of the lunar New year in 1 month and the lunar New year in 2 months and the lunar New year influence coefficient of the lunar New year in non-lunar New year, wherein the lunar New year influence coefficient of the other months is 1.
The expression of the lunar calendar new year influence coefficient is as follows:
In the formula (2), x (y,m,l) represents original time series data, y represents year, m is month, m and m ′ represent different values, the value ranges are 1,2, l is a mark for new year and month of lunar calendar, l and l ′ represent different values, when l=1, the month is new year and month of lunar calendar of the current year, and when l=0, the month is new year and month of non-lunar calendar of the current year; the formula (2) represents the lunar calendar new year influence coefficient of 1 month and 2 months.
4) Carrying out stability test on the data which is eliminated from the influence of the lunar calendar in the new year, if the test is passed, carrying out step 5), otherwise, carrying out linear transformation of multi-model combination on the data until the test is passed;
The linear transformation principle of the multi-model combination is from simple to complex, for example, x=ln (x) can be made to eliminate the unstable influence caused by extreme values, x= cosx is made to perform triangular mapping, sequences are made to be more stable in a certain interval, and the like, and x=ln (x) + cosx can be made to be combined and added.
5) Predicting the data subjected to the stability test through a SARIMA model to obtain a preliminary prediction result;
the expression of the SARIMA model is:
SARIMA ＝ (p,d,q)×(D,P,Q,s) (3)
In formula 3), P represents a non-seasonal autoregressive order, D represents a differential order, Q represents an order of a non-seasonal moving average, P represents a seasonal autoregressive order, D represents a seasonal differential order, Q represents an order of a seasonal moving average, and s represents a seasonal length, also called a period size.
The SARIMA model constructed as described above is based on data passing the stationarity check in step 4), namely: if the data eliminating the influence of the lunar calendar in the new year passes the stability test, directly using the data to construct an SARIMA model; otherwise, the SARIMA model is built based on the linearly transformed data passing the stationarity test.
When constructing the SARIMA model based on the data passing the test in step 4), the values of the parameters of the model need to be determined, bayesian information metrics (BICs) of different parameter values are calculated, and the SARIMA model with the minimum BIC value is selected from the parameter combinations to be used as the final SARIMA model.
The preliminary predicted outcome expression obtained should be:
Y(y,m,l)＝ SARIMA(p,d,q)(D,P,Q,s) (4)
6) Carrying out residual error test on residual error obtained by constructing the SARIMA model, and if the residual error test is passed, obtaining a preliminary prediction result which is a result after preliminary optimization, namely Y' (y,m,l)＝Y(y,m,l); if the residual error test is not passed, carrying out differential processing on the residual error, then inputting a differential sequence of the residual error and an original sequence of the residual error into a Bi-LSTM model for predicting the residual error to obtain a residual error prediction result epsilon (y,m,l), wherein Y represents year, m represents month, l is a mark for lunar calendar, when l=1, the month is lunar calendar of the current year, when l=0, the month is non-lunar calendar of the current year, and the optimized result is the sum of the preliminary prediction result and the residual error prediction result, namely Y' (y,m,l)＝Y(y,m,l)+ε(y,m,l).
7) Multiplying the predicted result obtained in the step 6) with the lunar calendar new year influence coefficient obtained in the step 3) to obtain a final predicted result, wherein the expression of the predicted result is as follows:
Wherein, alpha represents the new year influence coefficient of lunar calendar, and Y' (y,m,l) represents the prediction result of the step 6).
Examples: a prediction method for case quantity based on SARIMA-lunar calendar new year comprises the following steps:
1) Acquiring the month case quantity of a certain case from the next calendar year, and marking data corresponding to the new calendar year of the calendar year;
In this example, taking the month collection amount of the tending disputes accepted by national court as an example, the collection amount data from 1 month in 2018 to 12 months in 2022 is taken as a history fitting sample, and the data from 1 month in 2023 to 3 months in 2023 is taken as a prediction comparison sample.
2) Processing the original data to eliminate the influence of the lunar calendar on the data in the new year;
adding the values of the month 2018 1 and the month 2018 2 in the original data, averaging, and endowing the data with the month 2018 1 and the month 2018 2; data from 2019 to 2022 are also processed as described above, thereby yielding new data that eliminates the effects of new lunar calendar years.
3) Calculating the lunar calendar new year influence coefficient according to the month of the lunar calendar new year marked and the lunar calendar new year not marked in the original data;
Only the new lunar calendar of 2020 occurs in 1 month and the rest of the original data occurs in 2 months, so that the case quantity of 1 month in each year from 2018 to 2022 (except 2020) is added with the case quantity of 2 months in 2020 to obtain the sum of the case quantities of the months of non-lunar calendar; similarly, adding the case quantity of 2 months in each year from 2018 to 2022 (except 2020) to the case quantity of 1 month in 2020 to obtain the sum of the case quantities of the month in the new lunar calendar; and then calculating the respective duty ratio of the sum of the case amounts of the months of the non-lunar New year and the sum of the case amounts of the months of the lunar New year, wherein the obtained result is the lunar New year influence coefficient of the month of the lunar New year in 1 month and 2 months and the lunar New year influence coefficient of the month of the non-lunar New year, the lunar New year influence coefficients of the rest months are 1, and the lunar New year influence coefficients are obtained according to the method.
4) And (3) carrying out stability test on the data which is eliminated from the influence of the lunar calendar in the new year, if the test passes, carrying out step 5), otherwise, carrying out linear transformation of multi-model combination on the data until the test passes.
And 2) carrying out stability (ADF) inspection on the data which is processed in the step 2) and is eliminated from the new year of lunar calendar, so that the p value is 0.892 and is more than 0.05, the data is unstable, and the unit root inspection is not passed. Then, the data which is influenced by the new year of lunar calendar is subjected to difference, then the data subjected to difference is subjected to stability test, the data subjected to difference twice still fails to pass the stability test, so that the data is subjected to linear transformation, x=ln (x), the data is subjected to the stability test, the P value is 0.046 and is smaller than 0.05, the data is stable, the unit root test passes, and therefore d=0 is determined, and a SARIMA (P, 0, Q) (P, D, Q, s) model is preliminarily constructed.
5) Predicting the data subjected to the stability test through a SARIMA model to obtain a preliminary prediction result;
When the model is constructed, white noise test is carried out on the data subjected to linear transformation, the hysteresis order is set to be 10, p values of all hysteresis orders are smaller than 0.05, the white noise test passes, and the next analysis can be carried out. Next, bayesian information metrics (BIC) at different parameter values are calculated and a model is selected in which the BIC value is the smallest, so the final selected SARIMA model is SARIMA (0, 2) x (0,1,1,12). Model parameter effects are shown in table 1:
Table 1 shows the effect of model parameters
| coef | std err | z | P>|z| | [0.025 | 0.975] | |
| ma.L1 | 0.2292 | 0.151 | 1.517 | 0.129 | -0.067 | 0.525 |
| ma.L2 | 0.3504 | 0.161 | 2.172 | 0.03 | 0.034 | 0.667 |
| ma.S.L12 | -0.1739 | 0.101 | -1.716 | 0.086 | -0.372 | 0.025 |
| sigma2 | 6.42E+08 | 5.29E-11 | 1.21E+19 | 0 | 6.42E+08 | 6.42E+08 |
And then, carrying out preliminary prediction on data from 1 month to 3 months in 2023 according to the constructed SARIMA model to obtain a preliminary prediction result.
6) Carrying out residual error test on residual error obtained by constructing the SARIMA model, if the residual error test is passed, the preliminary prediction result is a result after preliminary optimization, if the residual error test is not passed, extracting to carry out differential treatment on the residual error, and then predicting residual error item information by utilizing the Bi-LSTM model, wherein the residual error item prediction result and the preliminary prediction result are added to obtain a prediction result after preliminary optimization;
Through inspection, the residual error of SARIMA (0, 2) x (0,1,1,12) fails to pass inspection, and information in the residual error can not be extracted, so that the residual error is predicted by using a Bi-LSTM model, and the predicted result of the residual error is added with the predicted result of the last step to obtain a predicted result after preliminary improvement.
7) Multiplying the prediction result obtained in the step 6) with the lunar calendar new year influence coefficient obtained in the step 3) to obtain a final prediction result.
Since the lunar calendar new year of 2023 occurs in 2 months, 2 months are marked as the month in which the lunar calendar new year is located, and the predicted values of 1 month and 2 months of 2023 are added and multiplied by the respective lunar calendar new year influence coefficients respectively to obtain the final prediction result.
Finally, evaluating results obtained by the traditional SARIMA model and the SARIMA-lunar calendar new year model, wherein the evaluation indexes are as follows: average absolute percent error (MAPE), average absolute error (MAE), and Root Mean Square Error (RMSE), the evaluation results are shown in Table 2:
Table 2 is a model prediction error table
As can be seen from Table 2, the indexes of the NY-SARIMA model are better improved compared with the SARIMA model, and the error value is greatly reduced compared with the traditional SARIMA model.
The foregoing describes a specific embodiment of the present invention and is merely exemplary of one embodiment of the present invention.
Although specific embodiments of the invention have been disclosed for illustrative purposes, it will be appreciated by those skilled in the art that the invention may be implemented with the help of a variety of examples: various alternatives, variations and modifications are possible without departing from the spirit and scope of the invention and the appended claims. Therefore, it is intended that the invention not be limited to the particular embodiment disclosed as the best mode contemplated for carrying out this invention, but that the invention will have the scope indicated by the scope of the appended claims.
Claims (3)
1. A case quantity prediction method for eliminating the influence of lunar calendar new year comprises the following steps:
1) Acquiring and marking the month case quantity of the same case in the next calendar year; for the month where the lunar calendar is in the new year, the marked month case quantity is x (y,m′,l′), otherwise, is x (y,m,l); wherein x is the case quantity of month, y is year, m and m 'are months, l is the month of the new year of the non-lunar calendar, and l' is the month of the new year of the lunar calendar; m ' =1 or 2, m+.m ', l+.l ';
2) Processing the month case quantity marked in the step 1), and taking the average value of the case quantities of 1 month and 2 months in the same year as the case quantities of 1 month and 2 months in the same year;
3) Calculating an lunar calendar new year influence coefficient alpha according to the lunar calendar case marked in the step 1); lunar calendar new year influence coefficient
4) Performing stationarity test on the month case quantity data processed in the step 2), if the month case quantity data passes the test, performing the step 5), otherwise performing linear transformation of multi-model combination on the month case quantity data processed in the step 2) until the stationarity test passes; the linear transformation of the multimodal fusion includes: the transformation x=ln (x) for eliminating the unstable effect caused by the extreme value, so that the sequence is more stable in a certain interval, the triangular mapping x= cosx, or x=ln (x) + cosx;
5) Constructing an SARIMA model based on the data subjected to the stability test, and predicting the data subjected to the stability test by using the SARIMA model to obtain a preliminary prediction result of the future month; when the SARIMA model is built, calculating the Bayesian information measurement of each parameter when different parameter values are taken according to the value range of each parameter of the SARIMA model, and taking a parameter value corresponding to the minimum value of the Bayesian information measurement as a parameter value corresponding to each parameter of the SARIMA model to obtain the SARIMA model; the SARIMA model is SARIMA= (P, D, Q) x (D, P, Q, s); wherein P represents a non-seasonal autoregressive order, D represents a differential order, Q represents an order of a non-seasonal moving average, P represents a seasonal autoregressive order, D represents a seasonal differential order, Q represents an order of a seasonal moving average, and s represents a seasonal length, also called a period size;
6) Carrying out residual error test on residual error obtained by constructing the SARIMA model, and taking the obtained preliminary prediction result in the step 5) as a preliminary optimization prediction result if the residual error test is passed; if the residual error is not passed through residual error inspection, carrying out differential processing on the residual error, then inputting a differential sequence of the residual error and the residual error into a Bi-LSTM model for predicting the residual error to obtain a residual error prediction result, and adding a residual error item prediction result and a preliminary prediction result to obtain a preliminary optimized prediction result;
7) And if the future month is not the month of the lunar calendar new year, taking the preliminary optimization prediction result obtained in the step 6) as the final prediction result of the future month, otherwise multiplying the preliminary optimization prediction result obtained in the step 6) by the lunar calendar new year influence coefficient obtained in the step 3) to obtain the final prediction result of the future month.
2. A server comprising a memory and a processor, the memory storing a computer program configured to be executed by the processor, the computer program comprising instructions for performing the steps of the method of claim 1.
3. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of claim 1.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311714164.6A CN117709530B (en) | 2023-12-13 | 2023-12-13 | Case quantity prediction method for eliminating new year influence of lunar calendar |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311714164.6A CN117709530B (en) | 2023-12-13 | 2023-12-13 | Case quantity prediction method for eliminating new year influence of lunar calendar |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117709530A CN117709530A (en) | 2024-03-15 |
| CN117709530B true CN117709530B (en) | 2024-06-04 |
Family
ID=90154794
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311714164.6A Active CN117709530B (en) | 2023-12-13 | 2023-12-13 | Case quantity prediction method for eliminating new year influence of lunar calendar |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117709530B (en) |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2428235A1 (en) * | 2000-11-09 | 2002-05-16 | Spss, Inc. | System and method for building a time series model |
| CN113516845A (en) * | 2021-04-27 | 2021-10-19 | 中南大学 | Road monthly traffic volume prediction method based on SARIMA-NAR combined model |
| CN114595861A (en) * | 2021-12-22 | 2022-06-07 | 国网宁夏电力有限公司 | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| SG126001A1 (en) * | 2005-03-30 | 2006-10-30 | Nanyang Polytechnic | Chinese lunar calendar-based tidal prediction system and method thereof |
-
2023
- 2023-12-13 CN CN202311714164.6A patent/CN117709530B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CA2428235A1 (en) * | 2000-11-09 | 2002-05-16 | Spss, Inc. | System and method for building a time series model |
| CN113516845A (en) * | 2021-04-27 | 2021-10-19 | 中南大学 | Road monthly traffic volume prediction method based on SARIMA-NAR combined model |
| CN114595861A (en) * | 2021-12-22 | 2022-06-07 | 国网宁夏电力有限公司 | MSTL (modeling, transformation, simulation and maintenance) and LSTM (least Square TM) model-based medium-and-long-term power load prediction method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117709530A (en) | 2024-03-15 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Binner* et al. | A comparison of linear forecasting models and neural networks: an application to Euro inflation and Euro Divisia | |
| Castle et al. | Modelling our changing world | |
| CN109739844B (en) | Data classification method based on attenuation weight | |
| Laitinen et al. | International applicability of corporate failure risk models based on financial statement information: Comparisons across European countries | |
| Bollen et al. | Automating the selection of model-implied instrumental variables | |
| CN112990386B (en) | User value clustering method and device, computer equipment and storage medium | |
| Fedorova et al. | Models for bankruptcy forecasting: Case study of Russian enterprises | |
| Svabova et al. | Being an outlier: a company non-prosperity sign? | |
| Poot | Meta-analysis of previous empirical research findings | |
| Lohmann et al. | Using accounting‐based information on young firms to predict bankruptcy | |
| CN115760380A (en) | Enterprise credit assessment method and system integrating electricity utilization information | |
| CN117436653A (en) | Prediction model construction method and prediction method for travel demands of network about vehicles | |
| CN117709530B (en) | Case quantity prediction method for eliminating new year influence of lunar calendar | |
| Abdullah et al. | Short Guides to Static Panel Data Regression Model Estimator | |
| Woo et al. | Modern definitions in reliability engineering | |
| Gouriéroux | The econometrics of risk classification in insurance | |
| CN115147091A (en) | Intelligent salary query method and system | |
| Horrace et al. | Nonparametric tests of tail behavior in stochastic frontier models | |
| CN115860505A (en) | Object evaluation method and device, terminal equipment and storage medium | |
| CN113064883A (en) | Method for constructing logistics wind control model, computer equipment and storage medium | |
| Vehmanen | Measurement of assets and the classical measurement theory | |
| Vasco et al. | Segmentation of potential fraud taxpayers and characterization in personal income tax using data mining techniques | |
| CN112598082A (en) | Method and system for predicting generalized error of image identification model based on non-check set | |
| Sprenger et al. | Corporate governance and investment: evidence from Russian unlisted firms | |
| Iqbal et al. | Multivariate tests of asset pricing: simulation evidence from an emerging market |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |