A multi-algorithm fusion multi-mode integrated air temperature prediction method and related device

Abstract

The application provides a kind of multi-algorithm fusion multi-mode integrated air temperature prediction method and related device, the method includes according to the mode forecast value of observation site at the forecast time, the observation live value of observation site at the same time of previous day and forecast time, the mode forecast value of observation site at the same time of previous day and forecast time, the forecast deviation value of observation site at the same time of previous day and forecast time, adopt multiple air temperature prediction algorithm, construct first air temperature prediction model, obtain the first air temperature prediction result of observation site at the forecast time;According to the first air temperature prediction result of observation site at the forecast time, utilize dynamic weighted set average algorithm, construct second air temperature prediction model, obtain the second air temperature prediction result of observation site at the forecast time;The result is determined as the air temperature prediction product of observation site at the forecast time.The application can adapt to the demand of complex terrain area air temperature accurate prediction, improve the accuracy and stability of air temperature prediction product.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of air temperature prediction, in particular to a multi-algorithm fusion multi-mode integrated air temperature prediction method and related device. BACKGROUND

[0002] Air temperature prediction is an important part of weather service, which is of great significance to agricultural production, energy supply, forest fire prevention, traffic management, public health and daily life. Influenced by global warming, accelerated urbanization and human activities, high temperature drought, low temperature rain and snow freezing and other meteorological disasters directly related to air temperature occur frequently. Society has put forward higher requirements for the accuracy and stability of air temperature prediction. Especially in complex terrain areas such as mountains, the terrain has a significant impact on air temperature, and the weather system is complex, so it is difficult to accurately predict the air temperature. Therefore, it is an urgent problem to improve the accuracy and stability of air temperature prediction in complex terrain areas. SUMMARY

[0003] In view of the defects in the prior art, the present application provides a multi-algorithm fusion multi-mode integrated air temperature prediction method and related device to improve the accuracy and stability of air temperature prediction products in complex terrain areas.

[0004] A multi-algorithm fusion multi-mode integrated air temperature prediction method, comprising:

[0005] According to the mode prediction value of the observation site at the prediction time, the observation live value of the observation site at the same time as the prediction time on the previous day, the mode prediction value of the observation site at the same time as the prediction time on the previous day, and the prediction deviation value of the observation site at the same time as the prediction time on the previous day, a plurality of air temperature prediction algorithms are used to construct a first air temperature prediction model to obtain a first air temperature prediction result of the observation site at the prediction time;

[0006] According to the first air temperature prediction result of the observation site at the prediction time, a dynamic weighted ensemble average algorithm is used to construct a second air temperature prediction model to obtain a second air temperature prediction result of the observation site at the prediction time;

[0007] The second air temperature prediction result is determined as the air temperature prediction product of the observation site at the prediction time.

[0008] Further, the method further comprises:

[0009] A bilinear interpolation method is used to interpolate the air temperature prediction product of the numerical prediction mode at the prediction time to the observation site to obtain the mode prediction value of the observation site at the prediction time.

[0010] Further, the numerical prediction model includes a global numerical prediction model of the European Centre for Medium-Range Weather Forecasts, a global numerical prediction model of the China Meteorological Administration, a regional numerical prediction model of the China Meteorological Administration, and a mesoscale regional numerical prediction model of Chongqing.

[0011] Further, the plurality of temperature prediction algorithms includes a linear regression equation of one variable, a linear regression equation of two variables, a linear regression equation of three variables, a bias linear regression equation, a model prediction, and a bias linear regression equation.

[0012] Further, according to the model prediction value of the observation site at the prediction time, a plurality of temperature prediction algorithms are used to construct a first temperature prediction model to obtain a first temperature prediction result of the observation site at the prediction time, including:

[0013] According to the model prediction value of the observation site at the prediction time, a linear regression equation of one variable, a linear regression equation of two variables, a linear regression equation of three variables, a bias linear regression equation, a model prediction, and a bias linear regression equation are used to construct a first temperature prediction model to obtain a first temperature prediction result of the observation site at the prediction time, wherein:

[0014] The linear regression equation of one variable is: F'(t+n) = a1F(t+n) + a0

[0015] The linear regression equation of two variables is: F'(t+n) = a1F(t+n) + a2O(t0+n0) + a0

[0016] The linear regression equation of three variables is: F'(t+n) = a1F(t+n) + a2O(t0+n0) + a3F(t0+n0) + a0

[0017] The bias linear regression equation is: F'(t+n) = F(t+n) - a1B(t0+n0) + a0

[0018] The model prediction and bias linear regression equation is: F'(t+n) = a1F(t+n) + a2B(t0+n0) + a0

[0019] In the formula, F'(t+n) represents the first temperature prediction result of the observation site at the prediction time t+n, F(t+n) represents the model prediction value of the observation site at the prediction time t+n, a1, a2, a3, and a0 represent regression coefficients calculated based on the least square method, O(t0+n0) represents the observation value of the observation site at the same time as the prediction time t+n on the previous day, F(t0+n0) represents the model prediction value of the observation site at the same time as the prediction time t+n on the previous day, and B(t0+n0) represents the prediction bias value of the observation site at the same time as the prediction time t+n on the previous day.

[0020] Further, according to the first air temperature prediction result of the observation station at the prediction time, a second air temperature prediction model is constructed by using a dynamic weighted ensemble average algorithm, including:

[0021] According to the first air temperature prediction result of the observation station at the prediction time, a second air temperature prediction model is constructed by using a dynamic weighted ensemble average algorithm, including:

[0022]

[0023] In the formula, MAE ij represents the average absolute error of the first air temperature prediction result of the i th numerical prediction mode under the j th air temperature prediction algorithm of the first air temperature prediction model and the observation value of all samples in the sliding period; M represents the number of samples in the sliding period, and k is the sample serial number in the sliding period; F / ij,k is the first air temperature prediction result of the i th numerical prediction mode under the j th air temperature prediction algorithm of the first air temperature prediction model for the k th sample; O k is the observation value of the k th sample, w ij is the weight coefficient of the i th numerical prediction mode for the j th air temperature prediction algorithm of the first air temperature prediction model, F / ij is the first air temperature prediction result of the i th numerical prediction mode under the j th air temperature prediction algorithm of the first air temperature prediction model, F WEM is the second air temperature prediction result of the observation station at the prediction time output by the second air temperature prediction model, and X is the product of the number of numerical prediction modes participating in the ensemble average and the number of air temperature prediction algorithms in the first air temperature prediction model.

[0024] A kind of air temperature prediction device of multi-algorithm fusion multi-mode integration, including:

[0025] The first prediction unit is used for the mode prediction value of observation station at the prediction time, the observation value of observation station at the same time of the previous day and the prediction time, the mode prediction value of observation station at the same time of the previous day and the prediction time, the prediction deviation value of observation station at the same time of the previous day and the prediction time, adopts multiple air temperature prediction algorithms, constructs the first air temperature prediction model, obtains the first air temperature prediction result of observation station at the prediction time;

[0026] The second prediction unit is used for according to the first air temperature prediction result of the observation station at the prediction time, using dynamic weighted ensemble average algorithm, constructs the second air temperature prediction model, obtains the second air temperature prediction result of observation station at the prediction time;

[0027] The determination unit is used for determining the second air temperature prediction result as the air temperature prediction product of the observation station at the prediction time.

[0028] An apparatus comprising a memory, a processor, and a computer program stored on the memory and executable on the processor, wherein the processor implements the steps of the method for predicting air temperature by multi-algorithm fusion and multi-mode integration according to any one of the preceding embodiments when executing the computer program.

[0029] A storage medium having stored thereon a computer program, wherein the computer program, when executed by a processor, implements the steps of the method for predicting air temperature by multi-algorithm fusion and multi-mode integration according to any one of the preceding embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0030] In order to more clearly illustrate the specific embodiments of the present application, the drawings required to be used in the specific embodiments or prior art description will be briefly introduced below. In all the drawings, the elements or parts are not necessarily drawn according to the actual scale.

[0031] Figure 1 A flow chart of the method for predicting air temperature by multi-algorithm fusion and multi-mode integration according to an embodiment of the present application is provided.

[0032] Figure 2 A schematic diagram of the observation station mode air temperature prediction of the bilinear interpolation process according to an embodiment of the present application is provided.

[0033] Figure 3 A structural schematic diagram of the device for predicting air temperature by multi-algorithm fusion and multi-mode integration according to an embodiment of the present application is provided.

[0034] In the drawings, (x, y) is the coordinate of the corresponding position of the observation station, (x1, y1), (x2, y1), (x1, y2), (x2, y2) are respectively the coordinates of the corresponding positions of the left lower, right lower, left upper and right upper four grid points near the observation station at the prediction time, and (x, y1), (x, y2) are respectively the coordinates of the corresponding positions of the two middle grid points with the same horizontal coordinate as the observation station. DETAILED DESCRIPTION

[0035] The embodiments of the technical solutions of the present application will be described in detail below with reference to the drawings. The following embodiments are only used to more clearly illustrate the technical solutions of the present application, and therefore only serve as examples, but cannot limit the protection scope of the present application.

[0036] It should be noted that, unless otherwise specified, the technical terms or scientific terms used in the present application should be understood as the usual meanings understood by the skilled in the art to which the present application belongs.

[0037] In one of the embodiments, as shown in Figure 1 a method for predicting air temperature by multi-algorithm fusion and multi-mode integration is provided, comprising:

[0038] 1. Constructing a first temperature forecast model by using a plurality of temperature forecast algorithms according to a mode forecast value of an observation site at a forecast time, an observation real-time value of the observation site at a same time as the forecast time on a previous day, a mode forecast value of the observation site at the same time as the forecast time on the previous day, and a forecast bias value of the observation site at the same time as the forecast time on the previous day, to obtain a first temperature forecast result of the observation site at the forecast time;

[0039] Preferably, a bilinear interpolation method is used to interpolate the temperature forecast product of the numerical prediction model to the observation site to obtain the mode forecast value of the observation site at the forecast time.

[0040] Preferably, the numerical prediction model includes a global numerical prediction model of the European Centre for Medium-Range Weather Forecasts (EC), a global numerical prediction model of the China Meteorological Administration (CMA-GFS), a regional numerical prediction model of the China Meteorological Administration (CMA-MESO), and a mesoscale regional numerical prediction model of Chongqing (CQ-MESO).

[0041] Actual business applications show that both forecasters and forecast users usually refer to multiple numerical prediction models in their work, because each model has its own advantages, and different numerical prediction models have different prediction performances for different regions and different prediction times. For example, the global numerical prediction model of the European Centre for Medium-Range Weather Forecasts performs stably in predicting large-scale weather systems, but often lacks the ability to predict sudden or local weather events; while the regional high-resolution numerical model performs well in predicting small-scale convective weather, but is less stable than the global numerical prediction model in predicting large-scale weather systems.

[0042] Currently, in weather forecast business, site observation data with high credibility are used to develop mode statistical interpretation (MOS) at the site to improve the forecast skill. The MOS method can effectively reduce the systematic bias in temperature forecast by establishing a statistical relationship between the numerical model output and the historical observation data. Based on the site MOS forecast, a grid temperature MOS temperature guidance product can also be generated by the CRESSMAN step-by-step correction method. Currently, the difficulty of the MOS forecast method lies in the optimization of the model and the selection of the forecast factors, and many studies use numerical prediction products and site real-time data at the same time to establish a statistical relationship for site correction forecast, without considering the relationship between historical site real-time data, mode forecast, mode bias and current forecast. If the optimal forecast factor can be extracted, different MOS forecast algorithms can be used to integrate their advantages, and the temperature forecast performance will be significantly improved. Therefore, the present application uses a multi-model integrated multi-algorithm fusion method to integrate the correction results of multiple numerical prediction models under different forecast algorithms, reduce the uncertainty in single numerical prediction, and fuse the forecast advantages of different algorithms to obtain more accurate and stable forecast results.

[0043] In numerical weather prediction, the temperature forecast product is usually generated on a two-dimensional or three-dimensional grid system, which covers the forecast area (i.e. the target area). Each grid point represents a specific geographical location and contains the model forecast value of the meteorological element (such as temperature, precipitation, wind speed, etc.) at that location.

[0044] The bilinear interpolation method is a commonly used numerical analysis method for interpolating data on grid points to non-grid points (such as observation sites). When the numerical prediction model forecasts the temperature of the target area, it generates a series of temperature forecast values on the grid system. Then, in order to obtain the model temperature forecast at the observation site (i.e. the model forecast value), the bilinear interpolation method is used to interpolate the temperature forecast values on the grid points to the location of the observation site.

[0045] Specifically, as shown in FIG. 1, the bilinear interpolation method is used to interpolate the temperature forecast product of the target area by the numerical prediction model to the observation site to obtain the model temperature forecast at the observation site in the target area, which is referred to as the model forecast value (DMO). Figure 2

[0046] In the formula, F(x, y) is the model forecast value of the observation site at the forecast time, (x, y) is the coordinate of the corresponding position of the observation site; F(x1, y1), F(x2, y1), F(x1, y2), F(x2, y2) are respectively the model forecast values of the left lower, right lower, left upper and right upper four grid points near the observation site at the forecast time, (x1, y1), (x2, y1), (x1, y2), (x2, y2) are respectively the coordinates of the four grid points; F(x, y1), F(x, y2) are respectively the model forecast values of the two middle grid points with the same horizontal coordinate as the horizontal coordinate of the observation site.

[0047] The first temperature forecast model inputs are the model forecast value of the observation site at the forecast time, the observed value of the observation site at the same time as the forecast time on the previous day, the model forecast value of the observation site at the same time as the forecast time on the previous day, and the forecast bias value of the observation site at the same time as the forecast time on the previous day, and outputs the first temperature forecast result of the observation site at the forecast time.

[0048]

[0049] ​The date in which the forecasting time is located is a forecasting date, and the forecasting date includes multiple forecasting times. The forecasting date refers to a specific date of publication or prediction in weather forecasting, and the forecasting time refers to a specific time point of publication or prediction in weather forecasting. In weather forecasting, in order to provide more detailed and timely weather information, multiple forecasting times are usually set for a forecasting date to publish weather conditions in different time periods.

[0050] Preferably, a quasi-symmetry mixed sliding training period method is adopted to construct a sample data set by sliding the training period with the forecasting start date, and the first temperature forecasting model is pre-trained.

[0051] The first day of the forecasting date is referred to as the forecasting start date.

[0052] Preferably, the training period includes 3n days in total.

[0053] Specifically, the training period includes:

[0054] n days before the forecasting start date of the current year;

[0055] n days before the same day as the forecasting start date of the previous year;

[0056] 1 day of the same day as the forecasting start date of the previous year plus n-1 days after the same day as the forecasting start date of the previous year.

[0057] The sample data set specifically includes:

[0058] observed live values, model predicted values, and prediction bias values of the observation site at the same time as the forecasting time in the n days before the forecasting start date;

[0059] observed live values, model predicted values, and prediction bias values of the observation site at the same time as the forecasting time in the n days before the same day as the forecasting start date of the previous year;

[0060] observed live values, model predicted values, and prediction bias values of the observation site at the same time as the forecasting time on the same day as the forecasting start date of the previous year and observed live values, model predicted values, and prediction bias values at the same time as the forecasting time in the n-1 days after the same day as the forecasting start date of the previous year.

[0061] Preferably, the multiple temperature forecasting algorithms include a linear regression equation, a binary linear regression equation, a ternary linear regression equation, a bias linear regression equation, a model prediction and bias linear regression equation.

[0062] Preferably, according to the model predicted value of the observation site at the forecasting time, a first temperature forecasting model is constructed by using multiple temperature forecasting algorithms to obtain a first temperature forecasting result of the observation site at the forecasting time, including:

[0063] According to the mode prediction value of the observation station at the prediction time, a unary linear regression equation, a binary linear regression equation, a ternary linear regression equation, a bias linear regression equation, and a mode prediction and bias linear regression equation are used to construct a first air temperature prediction model, so as to obtain a first air temperature prediction result of the observation station at the prediction time, wherein:

[0064] The unary linear regression equation is F'(t+n) = a1F(t+n) + a0.

[0065] The binary linear regression equation is F'(t+n) = a1F(t+n) + a2O(t0+n0) + a0.

[0066] The ternary linear regression equation is F'(t+n) = a1F(t+n) + a2O(t0+n0) + a3F(t0+n0) + a0.

[0067] The bias linear regression equation is F'(t+n) = F(t+n) - a1B(t0+n0) + a0.

[0068] The mode prediction and bias linear regression equation is F'(t+n) = a1F(t+n) + a2B(t0+n0) + a0.

[0069] In the formula, F'(t+n) represents the first air temperature prediction result of the observation station at the prediction time t+n, F(t+n) represents the mode prediction value of the observation station at the prediction time t+n, a1, a2, a3, and a0 represent regression coefficients calculated based on the least square method, O(t0+n0) represents an observation actual value of the observation station at the same time as the prediction time t+n on the previous day, F(t0+n0) represents a mode prediction value of the observation station at the same time as the prediction time t+n on the previous day, and B(t0+n0) represents a prediction bias value of the observation station at the same time as the prediction time t+n on the previous day.

[0070] Preferably, B(t0+n0) is calculated by the following formula:

[0071] B(t0+n0) = F(t0+n0) - O(t0+n0).

[0072] In the formula, t represents the reporting time, t+n represents the prediction time, t0 = t - 24h is the reporting time of the previous day, and t0+n0 represents the time at the same time as the prediction time t+n on the previous day.

[0073] Preferably, the method further comprises constructing a historical sample data set according to an optimal training period n, solving linear regression equations in various air temperature prediction algorithms, and obtaining an optimal first air temperature prediction model (containing five linear regression algorithms).

[0074] The selection steps of the optimal training period n are as follows:

[0075] The value range of n is set to 5-60 days, the interval is 5 days, 12 types of training period historical sample data sets are constructed by sliding with the forecast starting day, and the regression coefficients in the first air temperature forecast model are obtained by training and solving the five linear regression equations (unary linear regression equation, binary linear regression equation, ternary linear regression equation, bias linear regression equation, pattern forecast and bias linear regression equation) in the first air temperature forecast model respectively.

[0076] Then, the first air temperature forecast result of the previous year is obtained by using the first air temperature forecast model to correct the pattern air temperature forecast value of the previous year, and then the first air temperature forecast result of the previous year is tested and evaluated. The first air temperature forecast result of the previous year is closest to the actual air temperature, that is, the optimal training period. The specific test and evaluation index is the mean absolute error, and the calculation method is as follows:

[0077]

[0078] In the formula, F / i,j , O i,j are the first air temperature forecast value and the observed actual value at the same time of the i-th day in the j-th air temperature forecast algorithm of the previous year, N is the number of days in the previous year, MAE j is the mean absolute error of the first air temperature forecast value of the j-th air temperature forecast algorithm in the previous year. When MAE i is the minimum, the training period at this time is the optimal training period. Thus, five optimal training periods can be obtained, and five optimal linear regression equations, that is, five optimal air temperature forecast algorithms, can be obtained, thereby constructing the optimal first air temperature forecast model.

[0079] Then, the mode forecast value of the observation station at the forecast time, the observed actual value at the same time as the forecast time of the previous day, the mode forecast value at the same time as the forecast time of the previous day, and the forecast bias value at the same time as the forecast time of the previous day are input into the first air temperature forecast model, and the first air temperature forecast result of the observation station at the forecast time is output.

[0080] 2. According to the first air temperature forecast result of the observation station at the forecast time, a second air temperature forecast model is constructed by using a dynamic weighted ensemble average algorithm, and a second air temperature forecast result of the observation station at the forecast time is obtained.

[0081] According to the first air temperature forecast result of the observation station at the forecast time, a second air temperature forecast model is constructed by using a dynamic weighted ensemble average algorithm, and a second air temperature forecast result of the observation station at the forecast time is obtained.

[0082]

[0083] In the formula, MAE ijMAEi,j(k) represents the average absolute error of the first temperature prediction result of the i-th numerical prediction model under the j-th temperature prediction algorithm of the first temperature prediction model for all samples in the sliding period; M represents the number of samples in the sliding period, and k is the sample number in the sliding period; F / ij,k Fik represents the first temperature prediction result of the i-th numerical prediction model under the j-th temperature prediction algorithm of the first temperature prediction model for the k-th sample (each numerical prediction model has five first temperature prediction results, and each linear regression equation algorithm corresponds to one result); O k O represents the observed value of the k-th sample, w ij wi,j represents the weight coefficient of the i-th numerical prediction model for the j-th temperature prediction algorithm of the first temperature prediction model, F / ij Fik represents the first temperature prediction result of the i-th numerical prediction model under the j-th temperature prediction algorithm of the first temperature prediction model, F WEM X represents the second temperature prediction result of the observation station at the prediction time output by the second temperature prediction model, and X is the product of the number of numerical prediction models participating in the ensemble average and the number of temperature prediction algorithms in the first temperature prediction model (in fact, i=4 because there are four numerical prediction models; j=5 because there are five algorithms in the first temperature prediction model, and the last second temperature prediction result is the fusion integration of 20=4*5 first temperature prediction results).

[0084] Preferably, the sliding period of the second temperature prediction model is 15 days, and MAE ij MAEi,j(k) represents the average absolute error of the first temperature prediction result of the i-th numerical prediction model under the j-th linear regression equation of the first temperature prediction model for all samples at the same time as the prediction time within the previous 15 days of the prediction time of the observation station.

[0085] The input of the second temperature prediction model is the first temperature prediction result of the observation station at the prediction time, and the output is the second temperature prediction result of the observation station at the prediction time.

[0086] 3. The second temperature prediction result is determined as the temperature prediction product of the observation station at the prediction time.

[0087] The present application effectively integrates temperature prediction results from different models and different algorithms, and adopts dynamic weight allocation to obtain the best temperature prediction effect, and also considers the relationship between the temperature prediction product of the numerical prediction model, the observed value and the prediction bias value, thereby improving the practicability and accuracy of the temperature prediction product.

[0088] In one embodiment, as shown in Figure 3 a multi-algorithm fusion multi-model integrated temperature prediction device is provided, comprising:

[0089] The first prediction unit is configured to construct a first temperature prediction model by using a plurality of temperature prediction algorithms according to the mode prediction value of the observation site at the prediction time, the observation live value of the observation site at the same time as the prediction time on the previous day, the mode prediction value of the observation site at the same time as the prediction time on the previous day, and the prediction deviation value of the observation site at the same time as the prediction time on the previous day, to obtain a first temperature prediction result of the observation site at the prediction time.

[0090] The second prediction unit is configured to construct a second temperature prediction model by using a dynamic weighted ensemble algorithm according to the first temperature prediction result of the observation site at the prediction time, to obtain a second temperature prediction result of the observation site at the prediction time.

[0091] The determination unit is configured to determine the second temperature prediction result as the temperature prediction product of the observation site at the prediction time.

[0092] An apparatus includes a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the steps of the multi-algorithm fusion multi-mode integrated temperature prediction method of any one of the above embodiments when executing the computer program.

[0093] A storage medium has a computer program stored thereon, and the computer program implements the steps of the multi-algorithm fusion multi-mode integrated temperature prediction method of any one of the above embodiments when executed by a processor.

[0094] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in any one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples without contradiction.

[0095] The above examples are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing examples, those skilled in the art should understand that the technical solutions recorded in the foregoing examples can be modified, or some or all of the technical features can be replaced equivalently; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application, and they should be covered in the scope of the claims and the specification of the present application.

Claims

1. A temperature prediction method integrating multiple algorithms and multiple models, characterized by: include: Based on the model forecast value of the observation station at the forecast time, the actual observation value of the observation station at the same time as the forecast time the previous day, the model forecast value of the observation station at the same time as the forecast time the previous day, and the forecast deviation value of the observation station at the same time as the forecast time the previous day, a plurality of temperature forecast algorithms are used to construct a first temperature forecast model and obtain the first temperature forecast result of the observation station at the forecast time; According to the first temperature forecast result of the observation station at the forecast time, a second temperature forecast model is constructed using the dynamic weighted ensemble average algorithm and the following formula to obtain the second temperature forecast result of the observation station at the forecast time: ， ， ， Where, MAE ij represents the average absolute error between the first temperature forecast results of all samples and the observed actual values ​​under the jth temperature forecast algorithm of the first temperature forecast model in the i-th numerical forecast model within the sliding period; M represents the number of samples in the sliding period, k is the sample sequence number in the sliding period; F / ij,k is the first temperature forecast result of the kth sample under the jth temperature forecast algorithm of the first temperature forecast model by the i-th numerical forecast model; k is the observed value of the kth sample, w ij is the weight coefficient of the i-th numerical forecast model to the j-th temperature forecast algorithm of the first temperature forecast model, F / ij is the first temperature forecast result of the i-th numerical forecast model under the j-th temperature forecast algorithm in the first temperature forecast model, F WEM is the second temperature forecast result of the observation station at the forecast time output by the second temperature forecast model, and X is the product of the type of numerical forecast model participating in the ensemble average and the types of multiple temperature forecast algorithms in the first temperature forecast model; The second temperature forecast result is determined as the temperature forecast product of the observation site at the forecast time.

2. The temperature prediction method of multi-algorithm fusion and multi-mode integration according to claim 1, characterized in that: The method further comprises: The bilinear interpolation method is used to interpolate the temperature forecast product of the numerical forecast model for the target area at the forecast time to the observation station to obtain the model forecast value of the observation station at the forecast time.

3. The temperature prediction method of multi-algorithm fusion and multi-mode integration according to claim 2 is characterized in that: The numerical forecast models include the European Centre for Medium-Range Weather Forecasts global numerical forecast model, the China Meteorological Administration global numerical forecast model, the China Meteorological Administration regional numerical forecast model and the Chongqing mesoscale regional numerical forecast model.

4. The temperature prediction method of multi-algorithm fusion and multi-mode integration according to claim 1, characterized in that: The multiple temperature forecasting algorithms include a univariate linear regression equation, a bivariate linear regression equation, a ternary linear regression equation, a deviation linear regression equation, a model forecast and a deviation linear regression equation.

5. The temperature prediction method of multi-algorithm fusion and multi-mode integration according to claim 4 is characterized in that: The method comprises: constructing a first temperature forecast model based on a model forecast value of the observation station at the forecast time, an actual observation value of the observation station at the same time as the forecast time on the previous day, a model forecast value of the observation station at the same time as the forecast time on the previous day, and a forecast deviation value of the observation station at the same time as the forecast time on the previous day, and obtaining a first temperature forecast result of the observation station at the forecast time by using multiple temperature forecast algorithms. According to the model forecast value of the observation station at the forecast time, the actual observation value of the observation station at the same time as the forecast time the day before, the model forecast value of the observation station at the same time as the forecast time the day before, and the forecast deviation value of the observation station at the same time as the forecast time the day before, a single-variable linear regression equation, a two-variable linear regression equation, a three-variable linear regression equation, a deviation linear regression equation, a model forecast, and a deviation linear regression equation are used to construct the first temperature forecast model and obtain the first temperature forecast result of the observation station at the forecast time, where: The univariate linear regression equation is: The binary linear regression equation is: The three-variable linear regression equation is: The deviated linear regression equation is: The linear regression equation of model forecast and bias is: Where, It represents the first temperature forecast result of the observation station at the forecast time t+n, represents the model forecast value of the observation station at the forecast time t+n, 、 、 、 represents the regression coefficient calculated based on the least squares method, It represents the actual observation value of the observation station at the same time as the forecast time t+n on the previous day. It represents the model forecast value of the observation station at the same time as the forecast time t+n on the previous day, It represents the forecast deviation value of the observation station at the same time as the forecast time t+n on the previous day.

6. A temperature prediction device integrating multiple algorithms and multiple modes, characterized by: It includes a first forecasting unit, a second forecasting unit and a determining unit, wherein: The first forecasting unit is configured to construct a first temperature forecast model using multiple temperature forecast algorithms based on the model forecast value of the observation station at the forecast time, the observed actual value of the observation station at the same time as the forecast time on the previous day, the model forecast value of the observation station at the same time as the forecast time on the previous day, and the forecast deviation value of the observation station at the same time as the forecast time on the previous day, to obtain a first temperature forecast result for the observation station at the forecast time; The second forecasting unit is configured to construct a second temperature forecast model based on the first temperature forecast result of the observation station at the forecast time using a dynamic weighted ensemble average algorithm and the following formula to obtain the second temperature forecast result of the observation station at the forecast time: ， ， ， Where, MAE ij represents the average absolute error between the first temperature forecast results of all samples and the observed actual values ​​under the jth temperature forecast algorithm of the first temperature forecast model in the i-th numerical forecast model within the sliding period; M represents the number of samples in the sliding period, k is the sample sequence number in the sliding period; F / ij,k is the first temperature forecast result of the kth sample under the jth temperature forecast algorithm of the first temperature forecast model by the i-th numerical forecast model; k is the observed value of the kth sample, w ij is the weight coefficient of the i-th numerical forecast model to the j-th temperature forecast algorithm of the first temperature forecast model, F / ij is the first temperature forecast result of the i-th numerical forecast model under the j-th temperature forecast algorithm in the first temperature forecast model, F WEM is the second temperature forecast result of the observation station at the forecast time output by the second temperature forecast model, and X is the product of the type of numerical forecast model participating in the ensemble average and the types of multiple temperature forecast algorithms in the first temperature forecast model. The determining unit is used to determine the second temperature forecast result as the temperature forecast product of the observation site at the forecast time.

7. A device, characterized in that The method comprises a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the steps of the method according to any one of claims 1 to 5 when executing the computer program.

8. A storage medium, characterized in that: A computer program is stored thereon, and when the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images