Mode data fusion method based on improved analytic hierarchy process
By improving the analytic hierarchy process and using Euclidean distance to calculate model data weights, the uncertainties and fluctuations in expert scores in multi-model climate predictions were resolved. This enabled rapid and efficient fusion of multi-model data and improved accuracy, making it suitable for meteorological and hydrological research.
Patent Information
- Application Number
- CN202510952729.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-10
- Publication Date
- 2025-11-25
AI Technical Summary
In existing multi-model climate prediction methods, the prediction results of a single climate model are uncertain, and the fusion quality of the traditional analytic hierarchy process is greatly affected by the fluctuation of expert scoring quality, lacking objectivity and efficiency.
An improved analytic hierarchy process is adopted, which uses the Euclidean distance between model data and measured data as an objective value to replace subjective expert scoring. A judgment matrix is constructed and the weight of each model is calculated to perform multi-model data fusion.
It enables rapid and efficient fusion of multi-model data, reduces uncertainty, and improves the objectivity and accuracy of fusion results. It is applicable to meteorological and hydrological research and enhances the efficiency and accuracy of hydrological model parameter calibration.
Smart Images
Figure CN121009486A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The application belongs to the technical field of multi-mode data fusion methods, and particularly relates to a mode data fusion method based on an improved analytic hierarchy process. BACKGROUND
[0002] When making future climate estimation, there is great uncertainty in the estimation results based on a single climate model due to different internal mechanisms and application ranges among modes. Research shows that the multi-mode ensemble method can effectively reduce the uncertainty of future estimation results caused by physical processes and parameterization schemes of climate modes to a certain extent.
[0003] The most commonly used scheme at present is the multi-mode arithmetic mean AME, but it should be noted that when using this method, the climate modes with poor simulation performance will reduce the simulation capability of the multi-mode ensemble as a whole. The most similar prior art to the patent application technical scheme is the analytic hierarchy process AHP, but this method needs subjective scoring by experts and is not an objective assignment method. The fusion quality of modes will change with the fluctuation of the scoring quality of experts.
[0004] Climate model data is basic data for studying meteorology and hydrology, and can provide important reference for studying climate change, basin runoff and confluence, and preventing flood disasters. Therefore, it is necessary to develop a method for reasonably and objectively calculating the fusion weight of each mode, condensing and improving subjective artificial experience into scientific laws, and automatically and batch processing by using computer language to quickly and efficiently fuse multi-mode data. Therefore, a mode data fusion method based on an improved analytic hierarchy process is needed to solve the above problems. SUMMARY
[0005] The technical problem to be solved by the application is to provide a mode data fusion method based on an improved analytic hierarchy process. Based on the basic principle of the analytic hierarchy process, starting from constructing the judgment matrix of original data, the rule that the smaller the Euclidean distance between each index mode data and measured data, the more accurate the mode data is skillfully used, the Euclidean distance is used as an objective assignment instead of subjective expert scoring, so as to calculate the weight of each mode.
[0006] To achieve the above technical effects, the technical scheme adopted by the application is: A mode data fusion method based on an improved analytic hierarchy process, comprising the following steps: S1, data collection: Obtain measured station data and mode data or reanalysis product data of a target basin; S2, data preprocessing: Resampling the measured station data into grid data, and downscaling and bias correction of the model data, so that the spatial scale is unified with the measured data; S3, index calculation: Calculate the evaluation index of the model data and the measured data, and zero-mean standardize the index, and calculate the Euclidean distance between the standardized measured index and the multi-mode index; S4, objective score: The Euclidean distance is used as an objective assignment to replace the expert score to construct the original judgment matrix; S5, weight calculation; Calculate the weight vector based on the judgment matrix, and get the weight of each mode after consistency check, and use the weight to fuse the multi-mode data or generalize the future mode.
[0007] Preferably, step S1 comprises the following steps: S11, obtaining the measured station data of the target basin from the existing data; S12, obtaining the model data or reanalysis product data from the relevant website; S13, checking the abnormal values in the data, and optimizing them, such as representing the missing data as-999, changing it to 0 or interpolating to complete.
[0008] Preferably, the step S2 comprises the following steps: S21, referring to the target basin, cutting the model data or reanalysis data product set; S22, resampling the measured station data, such as Kriging interpolation and inverse distance weighted interpolation; converting the station data into grid data; S23, downscaling the model data or reanalysis data product set, such as statistical downscaling and dynamic downscaling, and bias correction, such as cumulative frequency curve method and median method, so that the resampled measured data reaches the unified spatial scale.
[0009] Preferably, the downscaling adopts bilinear interpolation method, and the formula is: ; ; ; In the formula, represents the longitude, represents the latitude; and represent and on the axis and represent and exist Precipitation on the axis; the result of bilinear interpolation is obtained by cubic linear interpolation.
[0010] Preferably, the bias correction adopts the ECDF bias correction method. ECDF bias correction is mainly based on the cumulative probability distribution of measured, historical simulation, and predicted data. When correcting biases in historical simulation data of the climate model, the goal is to make the average temperature and precipitation simulated by the model within the same time period close to the corresponding observed values. When correcting biases in predicted data of the climate model, it is necessary to calculate the cumulative probability corresponding to a certain future value. It is assumed that the difference between the measured and historical simulation data corresponding to this cumulative probability remains unchanged, and the climate model data is corrected using this difference Δ. The specific formula is: ; = + ; In the formula, For climate elements; F is the cumulative probability distribution function (CDF); These are measured data from historical periods; Simulated data for historical periods; This is simulated data for the estimated period.
[0011] Preferably, step S3 includes: Evaluation metrics for computational model data and measured data, including the coefficient of determination. The formulas for Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Pearson Correlation Coefficient (CC), and Bias are as follows: ; ; ; ; ; in, Watershed grid points mode value, Watershed grid points The measured value, It is the average of the pattern values. It is the average of the measured values. It refers to the number of samples.
[0012] Preferably, step S3 includes zero-mean standardization of the index: After performing the above index calculations on each pattern, the original index matrix is obtained: ; In the formula, is the total number of evaluation indexes; is the number of modes; is the value of the i-th index of the j-th mode. is the value of the i-th index of the j-th mode. is the value of the i-th index of the j-th mode. is the row vector of the i-th index; represents the numerical value of a single index in each mode, and the column vector represents the value of each index of a single mode; represents the value of each index of a single mode; represents the value of each index of a single mode; Since the differences between the original index data are relatively small and the distribution is relatively concentrated, the effect of directly extracting data features may not be good. In order to eliminate the influence of the magnitude dimension and improve the accuracy of the subsequent judgment matrix, the original index data is processed by using the zero-mean standardization method, and the standardized matrix is: ; ; In the formula, is the average value of the multi-mode value of a single index; is the standard deviation of the multi-mode value of a single index; is the standardized matrix row vector, representing the standardized value of a single index in each mode. is the standardized matrix row vector, representing the standardized value of a single index in each mode. is the standardized matrix row vector, representing the standardized value of a single index in each mode. is the standardized matrix row vector, representing the standardized value of a single index in each mode.
[0013] Preferably, the step S3 comprises calculating the Euclidean distance between the measured index after eliminating the influence of the magnitude dimension and the multi-mode data index, and the calculation formula is as follows: = ; In the formula, , , , and are calculated by the evaluation index.
[0014] Preferably, the step S4 comprises the following steps: S41, the calculated Euclidean distance is used as the assignment instead of the subjective expert score; S42, constructing the original judgment data matrix.
[0015] Preferably, the step S5 comprises the following steps: S51, calculating the weight vector; S52. Perform a consistency check on the calculation results. If the consistency check is satisfied, the weights of each mode are determined; otherwise, the judgment matrix is adjusted and the calculation is repeated. S53. Obtain the weights corresponding to each mode, perform data fusion, or use the weights to generalize future modes without actual measured data.
[0016] The beneficial effects of this invention are as follows: This invention is based on strong statistical principles and cleverly and reasonably avoids the fluctuations in data fusion results caused by expert scoring in traditional analytic hierarchy process (AHP). It allows for flexible and varied selection of indicators to be considered and enables rapid and efficient optimization and fusion of multi-model data. The multi-model fusion results can be widely used in meteorological and hydrological research, such as in hydrological forecasting as meteorological input parameters for runoff generation and confluence calculations; and in hydrological model parameter calibration, improving the efficiency and accuracy of hydrological model parameter calibration to a certain extent. Attached Figure Description
[0017] Figure 1 This is a flowchart of a pattern data fusion method based on an improved analytic hierarchy process according to the present invention. Figure 2 This is a Taylor diagram of the traditional AHP and improved AHP set methods in the embodiments of the present invention; Figure 3 This is a scatter plot of traditional AHP and improved AHP in an embodiment of the present invention; Figure 4 This is a spatial distribution diagram of traditional AHP and improved AHP in an embodiment of the present invention. Detailed Implementation Example 1: like Figure 1 As shown, a pattern data fusion method based on an improved analytic hierarchy process includes the following steps: S1, Data Collection: Acquire measured station data for the target watershed, as well as model data or reanalysis product data; S2, Data Preprocessing: The measured station data is resampled and converted into grid data, and the model data is downscaled and biased to make it consistent with the spatial scale of the measured data. S3, Indicator Calculation: The evaluation indexes of the model data and the measured data are calculated, and the indexes are standardized to zero mean. The Euclidean distance between the standardized measured indexes and the multi-model indexes is then calculated. S4, Objective Rating: The Euclidean distance is used as an objective value to replace expert scoring, and the original judgment matrix is constructed. S5, Weight Calculation; The weight vector is calculated based on the judgment matrix, and each mode weight is obtained after consistency test, and the weight is used for multi-mode data fusion or future mode generalization.
[0018] Preferably, the step S1 comprises the following steps: S11, obtaining the measured station data of the target basin from the existing data; S12, obtaining the mode data or reanalysis product data from the related website; S13, checking the abnormal values in the data and optimizing them, such as representing the missing data as-999, changing it to 0 or interpolating to complete it.
[0019] Preferably, the step S2 comprises the following steps: S21, referring to the target basin, the mode data or reanalysis data product set is cropped; S22, resampling the measured station data, such as Kriging interpolation and inverse distance weighted interpolation; converting the station data into grid data; S23, performing downscaling on the mode data or reanalysis data product set, such as statistical downscaling and dynamic downscaling, and bias correction, such as cumulative frequency curve method and median method, so as to unify the spatial scale with the resampled measured data.
[0020] Preferably, the downscaling adopts bilinear interpolation method, and the formula is: ; ; ; In the formula, λ represents the longitude, represents the latitude; represents the precipitation on the axis; represents the precipitation on the axis; the result of bilinear interpolation is obtained by cubic linear interpolation. represents the precipitation on the axis; the result of bilinear interpolation is obtained by cubic linear interpolation. represents the precipitation on the axis; the result of bilinear interpolation is obtained by cubic linear interpolation.
[0021] Preferably, the bias correction adopts the ECDF bias correction method. ECDF bias correction is mainly based on the cumulative probability distribution of measured, historical simulation, and predicted data. When correcting biases in historical simulation data of the climate model, the goal is to make the average temperature and precipitation simulated by the model within the same time period close to the corresponding observed values. When correcting biases in predicted data of the climate model, it is necessary to calculate the cumulative probability corresponding to a certain future value. It is assumed that the difference between the measured and historical simulation data corresponding to this cumulative probability remains unchanged, and the climate model data is corrected using this difference Δ. The specific formula is: ; = + ; In the formula, For climate elements; F is the cumulative probability distribution function (CDF); These are measured data from historical periods; Simulated data for historical periods; This is simulated data for the estimated period.
[0022] Preferably, step S3 includes: Evaluation metrics for computational model data and measured data, including the coefficient of determination. The formulas for Root Mean Square Error (RMSE), Mean Absolute Error (MAE), Pearson Correlation Coefficient (CC), and Bias are as follows: ; ; ; ; ; in, Watershed grid points mode value, Watershed grid points The measured value, It is the average of the pattern values. It is the average of the measured values. It refers to the number of samples.
[0023] Preferably, step S3 includes zero-mean standardization of the index: After performing the above index calculations on each pattern, the original index matrix is obtained: ; In the formula, The total number of evaluation indicators; Number of patterns; the value of the first index of the first mode. the value of the first index of the first mode. the value of the first index of the first mode. the row vector of the value of the first index of the first mode. the value of the first index of the first mode. the column vector of the value of the first index of the first mode. the value of the first index of the first mode. the value of the first index of the first mode. Since the differences between the original index data are relatively small and the distribution is relatively concentrated, the effect of directly extracting data features may not be good. In order to eliminate the influence of the quantity dimension and improve the result accuracy of the subsequent judgment matrix, the zero mean standardization method is used to process the original index data, and the standardized matrix is: ; ; In the formula, the average value of the multi-mode value of the first index. the standard deviation of the multi-mode value of the first index. the standardized matrix row vector, representing the standardized value of the first index in each mode. the standardized matrix row vector, representing the standardized value of the first index in each mode. Preferably, the step S3 comprises calculating the Euclidean distance between the measured index after eliminating the influence of the quantity dimension and the multi-mode data index, and the calculation formula is as follows:
[0024] ; ; In the formula, , , , , and are calculated by evaluating the index.
[0025] Preferably, the step S4 comprises the following steps: S41, taking the calculated Euclidean distance as the evaluation instead of the subjective expert score; S42, constructing the original judgment data matrix.
[0026] Preferably, the step S5 comprises the following steps: S51, calculating the weight vector; S52, performing consistency check on the calculation result. If the consistency check is satisfied, the weight of each mode is obtained, and if the consistency check is not satisfied, the judgment matrix is adjusted and recalculated; S53, obtaining the weight corresponding to each mode, performing data fusion or using the weight to generalize the future mode without measured data.
[0027] Embodiment Two: This embodiment provides a case of implementing the method specifically, the process includes: Step S1 includes the following steps: S11, data collection from 1960 to 2014, length of 55 years. The precipitation data from 1960 to 2014 is from the National Meteorological Science Data Center, the website is https: / / data.cma.cn / ; S12, climate model data comes from the latest sixth international coupling model comparison plan (CMIP6), which has higher spatial resolution and improved parameterization scheme than previous versions. CMIP6 output results have been widely used in climate change and extreme climate related research.
[0028] This embodiment mainly uses the precipitation data of 15 CMIP6 global climate models, and the data is downloaded from the CMIP6 website (http: / / esgfnode.llnl.gov / search / cmip6 / ). The historical model time period is from 1961 to 2014. The fifteen models are ACCESS-CM2, ACCESS-ESM1-5, BCC-CSM2-MR, CanESM5, CESM2-WACCM, CIESM, CMCC-CM2-SR5, FIO-ESM-2-0, IITM-ESM, INM-CM4-8, INM-CM5-0, IPSL-CM6A-LR, MPI-ESM1-2-LR, MRI-ESM2-0, NESM3. S13, the measured data selects 733 sites with complete data and relatively uniform distribution in the Yangtze River Basin, and selects 603 sites with less than 200 missing days in the complete time series. This can not only reflect the systematicness and completeness of the precipitation time distribution calculation, but also help to reflect the regional characteristics of precipitation distribution.
[0029] Specifically, the step S2 includes the following steps: S21, using the shp file of the Yangtze River Basin, the fifteen global scale historical scenario model data are cut to obtain the model precipitation data of the Yangtze River Basin; S22, resample (such as Kriging interpolation, inverse distance weighted interpolation) the measured site data, and convert the 603 site data into grid data; S23, downscaling (such as statistical downscaling, dynamic downscaling) the model data or reanalysis data product set. The bilinear interpolation method (statistical downscaling) is used to interpolate the global climate model data with different resolutions to a unified 0.5°x0.5°, and the specific formula is: ; ; ; where x represents longitude and y represents latitude; and represent and in precipitation on the axis, and and represent and in precipitation on the axis. The results of the bilinear interpolation are obtained by cubic linear interpolation.
[0030] When bias correction is performed on historical simulation data of the climate model, the mean value of the precipitation simulated by the model in the same time period is close to the corresponding observation value. When bias correction is performed on prediction data of the climate model, the cumulative probability corresponding to a certain value in the future needs to be calculated, and it is assumed that the difference between the measured and historical simulation data at the cumulative probability remains unchanged. The climate model data is corrected by the difference Δ. The EDCDF formula is as follows:
[0031] = + ; wherein, is a climate element; F is a cumulative probability distribution function CDF; is the measured data in the historical period; is the simulation data in the historical period; is the simulation data in the prediction period.
[0032] Specifically, the step S3 includes the following steps: S31, according to the measured data, the model data or the reanalysis data product set is calculated. Five common evaluation indexes are selected as the reference indexes: the determination coefficient , the root mean square error (RMSE), the mean absolute error (MAE), the Pearson correlation coefficient (CC), and the bias (Bias) formula is as follows: ; ; ; ; ; wherein, is the grid point of the river basin mode value, Watershed grid points The measured value, It is the average of the pattern values. It is the average of the measured values. This refers to the sample size. Table 1 below shows the statistical table for the calculation of multi-mode indicators.
[0033] Table 1: Statistical Table of Multi-Mode Indicator Calculation;
[0034] S32. In order to eliminate the influence of unit dimensions, the indicators are standardized to zero mean. After performing the above index calculations on each pattern, the original index matrix is obtained. ; In the formula: The total number of evaluation indicators; Number of patterns; For the first The first mode The value of each indicator. row vectors Representing a single indicator Numerical values and column vectors in each mode Representing a single pattern The values of each indicator.
[0035] Because the differences between the original indicator data are relatively small and their distribution is relatively concentrated, direct data feature extraction may not be effective. To eliminate the influence of unit dimensions and improve the accuracy of the subsequent judgment matrix, a standardization method is used to process the original indicator data. Commonly used data standardization methods include decimal scaling standardization, maximum / minimum value standardization, and zero-mean standardization. This embodiment uses the zero-mean standardization method, which, by adjusting parameters, can amplify the differences between data while maintaining their distribution characteristics. The standardized matrix... ; ; In the formula: For a single indicator The average value of the multi-mode values; For a single indicator Standard deviation of multi-mode values; The standardized matrix row vectors represent individual indicators. The standardized values in each mode are shown in Table 2 below, which is a statistical table of zero-mean standardized multi-mode indicators.
[0036] Table 2: Statistical table for the calculation of zero-mean standardized multi-mode indicators;
[0037] S33. Calculate the Euclidean distance between the measured index and the multi-model data index after eliminating the influence of dimensions. The calculation formula is as follows. = ; Table 3 below shows the Euclidean distance statistics for multi-mode indicators.
[0038] Table 3: Statistics on Euclidean Distance of Multi-Mode Indicators;
[0039] Specifically, step S4 includes the following steps: S41. Use the calculated Euclidean distance as the assigned value instead of subjective expert scoring; S42. Construct the original judgment data matrix, as shown in Table 4 below.
[0040] Table 4: Judgment Matrix;
[0041] Specifically, step S5 includes the following steps: S51. Calculate the weight vector; S52. Perform a consistency check on the calculation results. If the consistency check is satisfied, the weights of each mode are determined; otherwise, the judgment matrix is adjusted and the calculation is repeated. S53. Obtain the weights corresponding to each mode. Data fusion can be performed, and the weights can also be used to generalize future modes without actual measured data, as shown in Table 5 below.
[0042] Table 5: Weight statistics for each mode;
[0043] The improved analytic hierarchy process (AHP) is demonstrated through examples to achieve better fusion results than the original AHP. Specific steps for each example are detailed in the implementation steps section.
[0044] pass Figure 2 As can be seen, the data fused using the improved analytic hierarchy process (AHP) IM-AHP is closer to the measured data reference; while the original AHP is further away from the measured data reference. This indicates that the improvement is beneficial and provides a new method for data fusion.
[0045] Scatter fitting Figure 3It can be seen that the slope changes from 0.17 to 0.33, which is closer to 1, and the R 2 The slope changes from 0.16 to 0.5, which shows that the improved AHP is better at simulating the observation data. The correlation coefficient CC increases from 0.41 to 0.71, which shows that the correlation is significantly improved. The improved AHP effectively improves the precision of the generalization data. It shows that the improvement is beneficial, and provides a new method for data fusion.
[0046] Through the spatial distribution Figure 4 It can be seen that the Pearson correlation coefficient CC of the im-AHP method is significantly improved compared with the AHP method in the upper reaches of the Yangtze River. The area of the high correlation region in the lower reaches of the Yangtze River also increases. The number of high correlation regions in the middle reaches of the Yangtze River also increases. The root mean square error RMSE of the im-AHP method and the AHP method does not change significantly. In some areas of the upper reaches of the Yangtze River, the RMSE decreases, which shows that the fusion data of the im-AHP method is closer to the actual value, and the data precision is better. The standard deviation STD of the im-AHP method is significantly reduced in some areas of the upper reaches of the Yangtze River and the northern part of the middle reaches of the Yangtze River. The area of the low standard deviation region in the lower reaches of the Yangtze River also increases, which shows that the representativeness of the sample mean is better.
[0047] The above-described embodiments are merely preferred embodiments of the present application and are not intended to limit the scope of the present application. Without departing from the principles and the essence of the present application, various modifications and improvements to the technical solutions of the present application made by those skilled in the art shall fall within the scope of protection of the present application as defined by the claims.
Claims
1. A pattern data fusion method based on improved analytic hierarchy process, characterized in that, It comprises the following steps: S1, data collection: Obtain the measured site data of the target basin, and mode data or reanalysis product data; S2, data preprocessing: Resample the measured site data to convert it into grid data, and downscale and bias correct the mode data to make it uniform with the measured data in spatial scale; S3, index calculation: Calculate the evaluation index of the mode data and the measured data, and zero-mean standardize the index, and calculate the Euclidean distance between the standardized measured index and the multi-mode index; S4, objective scoring: Use the Euclidean distance as the objective assignment to replace the expert score to construct the original judgment matrix; S5, weight calculation; Calculate the weight vector based on the judgment matrix, pass the consistency test to obtain the weight of each mode, and use the weight to fuse the multi-mode data or generalize the future mode.
2. The pattern data fusion method based on improved analytic hierarchy process according to claim 1, characterized in that, Step S1 comprises the following steps: S11, obtain the measured site data of the target basin from existing data; S12, obtain the mode data or reanalysis product data from related websites; S13, check the abnormal values in the data and optimize them.
3. The pattern data fusion method based on improved analytic hierarchy process according to claim 1, characterized in that, The step S2 comprises the following steps: S21, refer to the target basin to crop the mode data or reanalysis data product set; S22, resample the measured site data to convert the site data into grid data; S23, downscale and bias correct the mode data or reanalysis data product set to make it uniform with the resampled measured data in spatial scale.
4. The pattern data fusion method based on improved analytic hierarchy process according to claim 3, characterized in that, The downscaling adopts the bilinear interpolation method, and the formula is: ； ； ； wherein representing longitude, representing latitude; and representing and in precipitation on the axis, and representing and in precipitation on the axis; the result of the bilinear interpolation is obtained by a trilinear interpolation.
5. The pattern data fusion method based on improved analytic hierarchy process according to claim 4, characterized in that, The bias correction adopts the EDCDF bias correction method, and the formula is: ； = + ； wherein F is a climatic element; F is a cumulative probability distribution function CDF; is historical period measured data; is historical period simulated data; is forecast period simulated data.
6. The pattern data fusion method based on improved analytic hierarchy process according to claim 1, characterized in that, The step S3 comprises: Evaluation metrics of the calculated pattern data and the measured data, including the coefficient of determination , the root mean square error RMSE, the mean absolute error MAE, the Pearson correlation coefficient CC and the bias Bias formula as follows: ； ； ； ； ； wherein is the model value for a basin grid point is the observed value for a basin grid point is the model value for a basin grid point is the observed value for a basin grid point is the average of the model values, is the average of the observed values, is the number of samples.
7. The pattern data fusion method based on improved analytic hierarchy process according to claim 6, characterized in that, The step S3 comprises zero-mean standardization of the index: After the above index calculation for each mode, the original index matrix is obtained: ； wherein is the total number of indicators; is the number of patterns; is the value of the th indicator of the th pattern; is a row vector represents the value of a single indicator in each pattern, a column vector represents the values of each indicator of a single pattern ; The zero-mean standardization method is used to process the original index data, and the standardized matrix is: ； ； wherein is the average of the multi-modal values of the individual indicators ; is the standard deviation of the multi-modal values of the individual indicators ; is the normalized matrix row vector representing the normalized values of the individual indicators in each mode.
8. The pattern data fusion method based on improved analytic hierarchy process according to claim 1, characterized in that, The step S3 comprises calculating the Euclidean distance between the dimensionless measured index and the multi-mode data index, and the calculation formula is as follows: = ； wherein, , , , and are calculated by evaluating the indicators.
9. The pattern data fusion method based on improved analytic hierarchy process according to claim 1, characterized in that, The step S4 comprises the following steps: S41, use the calculated Euclidean distance as the assignment to replace the subjective expert score; S42, construct the original judgment data matrix.
10. The pattern data fusion method based on improved analytic hierarchy process according to claim 1, characterized in that, The step S5 comprises the following steps: S51, calculate the weight vector; S52, perform consistency test on the calculation result; if the consistency test is satisfied, the weight of each mode is obtained, if not, adjust the judgment matrix and recalculate; S53, obtain the weight corresponding to each mode, perform data fusion or use the weight to generalize the future mode without measured data.