A method for hourly downscaling of monthly meteorological forecast data based on distribution-adjusted time mapping

Abstract

This invention belongs to the field of building performance simulation and provides an hourly downscaling method for monthly weather forecast data based on distribution-adjusted time mapping. The method includes the following steps: identifying representative cities, collecting data from a typical meteorological year, determining a global climate model and a monthly average resolution dataset, acquiring a primary variable dataset, determining candidate statistical distributions, performing maximum likelihood estimation, screening optimal distributions, constructing an empirical cumulative distribution function, modifying distributions, calculating quantiles, mapping quantiles, adjusting distributions, and finally model prediction. By using specific distribution adjustment and quantile mapping methods, the method achieves hourly downscaling of monthly forecast data for different weather variables in an atmospheric circulation model (GCM). This method not only maintains the physical properties of the data and the statistical characteristics of historical data, but also reflects long-term trends and extreme weather conditions predicted by GCMs, providing reliable future weather data for building performance simulation.

Description

Technical Field

[0001] The present invention relates to the field of building performance simulation, and in particular to an hourly downscaling method for monthly meteorological forecast data based on distribution-adjusted time mapping. Background Art

[0002] Buildings contribute approximately 28% of global greenhouse gas emissions and 30% of global energy consumption during their operational phase. As climate conditions change, buildings are likely to experience increased cooling demand, altered heating demand, and increased stress on building materials and systems, necessitating adaptive design and retrofit strategies. The necessity and importance of downscaling future weather data for building performance simulations cannot be underestimated when assessing the impacts of climate change on building performance. Building performance simulations, or BPSs, are essential tools for predicting building thermal behavior, energy consumption, and indoor comfort under future climate conditions. Accurate future weather data are crucial for these simulations, providing the granular information necessary for reliable BPS results. Downscaling global climate model outputs to high-resolution local weather data enables architects and engineers to design and retrofit buildings that are resilient, energy-efficient, and comfortable under future climate conditions, thereby promoting sustainable development and climate adaptation. Global climate models (GCMs) typically have a coarse spatial resolution of 100 to 250 kilometers. This resolution is far from sufficient to capture local climate changes and the specific weather patterns that affect individual buildings or urban areas, necessitating downscaling. Downscaling methods are used to bridge this gap, transforming coarse-grid GCM outputs into the high-resolution local weather information required for BPS under future climate conditions. Existing downscaling methods can be divided into two main categories: statistical downscaling and dynamical downscaling. Dynamical downscaling methods utilize high-resolution regional climate models to simulate local climate processes based on GCM boundary conditions. They provide a detailed physical representation of the climate system and are able to capture the complex interactions between atmospheric, oceanic, and land surface processes. However, they are computationally intensive and require significant resources to run long-term climate simulations. Statistical downscaling methods involve developing empirical relationships between large-scale climate variables and local-scale climate characteristics. They are computationally efficient and can be applied to a wide range of GCM outputs, including regression analysis, weather typing, and simulation day methods. Most of these techniques assume that the historical relationship between large-scale and local-scale variables will remain unchanged in the future. Due to its computational efficiency and flexibility, most studies related to building performance employ statistical downscaling rather than dynamical downscaling. Compared to dynamical downscaling methods, statistical downscaling methods require significantly fewer resources and time, allowing statistical techniques to be applied to multiple GCM outputs and a variety of scenarios, enhancing simulation robustness. Furthermore, statistical methods can leverage historical weather data, such as typical meteorological year data, and adapt them to specific local conditions, thereby providing more convenient and context-sensitive climate predictions. These advantages make statistical downscaling a more practical option for extensive and detailed building performance simulations. In recent years, numerous studies have developed and employed various statistical downscaling methods to improve the accuracy of future local hourly weather data. These methods aim to transform coarse-grid GCM outputs into high-resolution local weather information.However, most statistical downscaling methods are not suitable for application in BPS due to their complexity and computational requirements. The morphing method proposed by the researchers aims to adjust historical weather data to account for predicted future climate changes by applying shifts, stretching and interpolation. It modifies temperature, humidity and solar radiation based on GCM projections and typical meteorological year data to create future weather files. The method is simple, efficient, easy to implement, and has low computational requirements, making it suitable for BPS-related applications. Its flexibility allows it to adapt to various climates and scenarios, and it integrates well with typical meteorological year data using a reliable historical baseline. In addition, the practicality of the method is obvious because it can work without the need for long-term historical data.

[0003] However, the morphing method still has some limitations. It relies on linear assumptions, cannot fully capture local climate patterns, and cannot accurately predict extreme meteorological conditions. Summary of the Invention

[0004] To overcome the shortcomings of the prior art, the present invention aims to provide an hourly downscaling method for monthly meteorological forecast data based on distribution-adjusted time mapping. By adjusting specific distributions and using quantile mapping methods, hourly downscaling of monthly forecast data for different weather variables in the atmospheric circulation model (GCM) is achieved. This method not only maintains the physical properties of the data and the statistical characteristics of historical data, but also reflects the long-term trends and extreme weather conditions predicted by the GCM, providing reliable future meteorological data for building performance simulation.

[0005] To achieve the above object, the present invention provides the following solutions:

[0006] A method for hourly downscaling of monthly meteorological forecast data based on distribution-adjusted time mapping includes:

[0007] Identify representative cities for pre-defined climate zones;

[0008] Compile historical weather data based on the representative cities to obtain data for a typical meteorological year;

[0009] Determine global climate models and monthly mean resolution datasets;

[0010] Extracting hourly typical meteorological year data of preset weather variables according to the typical meteorological year data to obtain a main variable data set; the main variable data set includes: hourly temperature data, hourly relative humidity data, hourly wind speed data and hourly solar radiation data;

[0011] setting a candidate statistical distribution of each of the preset weather variables according to the primary variable data set;

[0012] Verifying the candidate statistical distribution of each of the preset weather variables using maximum likelihood estimation to obtain a verification data set;

[0013] Screening the candidate statistical distributions of the preset weather variables according to the validation data set to obtain an optimal distribution of each of the preset weather variables;

[0014] constructing an empirical cumulative distribution function based on the monthly average resolution data set and the main variable data set;

[0015] The optimal distribution is corrected using a preset correction function to obtain a corrected distribution; the correction function is: Y adjusted is the parameter correction result; is the parameter correction value; μ fut is the average value of the prediction results of the global climate model; μ historical is the mean value of the main variable data set;

[0016] Determining the quantile corresponding to each value in the prediction result of the global climate model using the empirical cumulative distribution function;

[0017] Mapping the quantile to the calculation result of the empirical cumulative distribution function of the main variable data set to obtain an original prediction model;

[0018] The prediction model is adjusted using a preset distribution adjustment formula to obtain a target prediction model; the distribution adjustment formula includes:

[0019]

[0020] Among them, loc and scale are distribution parameters; a and b are shape parameters; σ obs is the standard deviation of the main variable data set; Γ(·) is the gamma function;

[0021] The target prediction model is used to predict future hourly weather data to obtain a target prediction result.

[0022] Preferably, the screening criteria for the representative cities include: weather patterns, temperature ranges and humidity levels.

[0023] Preferably, the climate scenarios of the global climate model include: SSP126, SSP245, SSP370, SSP434 and SSP585.

[0024] Preferably, historical weather data is compiled based on the representative cities to obtain typical meteorological year data, including:

[0025] Collect raw meteorological data through public weather databases;

[0026] Performing integrity testing and data cleaning on the raw meteorological data using a preset program to obtain preprocessed data;

[0027] Formatting the preprocessed data according to a preset data structure to obtain formatted data;

[0028] The missing values ​​obtained by the integrity detection of the preset program are supplemented by using the difference technology, and the supplemented data of the missing values ​​are matched to the formatted data to obtain the typical meteorological year data.

[0029] Preferably, the candidate statistical distributions include: normal distribution, lognormal distribution and skewed normal distribution.

[0030] Preferably, the judgment data of the maximum likelihood estimation is the residual sum of squares; the calculation formula of the residual sum of squares is: Wherein, SSE is the calculation result of the residual sum of squares; is the i-th observation data point; x i is the i-th fitted distribution value with parameter θ; n is the number of data points.

[0031] Preferably, the empirical cumulative distribution function is: Among them, F X (x) is the empirical cumulative distribution function; II() is the index function; N is the number of observations; X i The i-th set of meteorological parameters; x is the meteorological parameter value.

[0032] The present invention discloses the following technical effects:

[0033] The present invention provides an hourly downscaling method for monthly meteorological forecast data based on distribution-adjusted time mapping. Through the adjustment of specific distributions and the quantile mapping method, it solves the problem that conventional technologies cannot reflect the long-term trends and extreme weather conditions predicted by GCMs while maintaining the physical properties of the data and the statistical characteristics of historical data. It realizes the hourly downscaling of monthly forecast data of different weather variables in the atmospheric circulation model and the reliable simulation of future meteorological data. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0035] Figure 1A schematic diagram of a downscaling process for meteorological forecast data based on distribution-adjusted time mapping provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] The purpose of this invention is to provide a meteorological forecast data downscaling method based on distribution-adjusted time mapping. By adjusting specific distributions and using quantile mapping methods, the downscaling of monthly forecast data of different weather variables in the atmospheric circulation model (GCM) is achieved. This method not only maintains the physical properties of the data and the statistical characteristics of historical data, but also reflects the long-term trends and extreme weather conditions predicted by the GCM, providing reliable future meteorological data for building performance simulation.

[0038] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0039] Figure 1 A schematic diagram of a downscaling process for meteorological forecast data based on distribution adjustment time mapping provided by an embodiment of the present invention is shown in FIG. Figure 1 As shown, the present invention provides an hourly downscaling method for monthly meteorological forecast data based on distribution-adjusted time mapping, comprising:

[0040] Identify representative cities for pre-defined climate zones;

[0041] Compile historical weather data based on the representative cities to obtain data for a typical meteorological year;

[0042] Determine global climate models and monthly mean resolution datasets;

[0043] Extracting hourly typical meteorological year data of preset weather variables according to the typical meteorological year data to obtain a main variable data set; the main variable data set includes: hourly temperature data, hourly relative humidity data, hourly wind speed data and hourly solar radiation data;

[0044] setting a candidate statistical distribution of each of the preset weather variables according to the primary variable data set;

[0045] Verifying the candidate statistical distribution of each of the preset weather variables using maximum likelihood estimation to obtain a verification data set;

[0046] Screening the candidate statistical distributions of the preset weather variables according to the validation data set to obtain an optimal distribution of each of the preset weather variables;

[0047] constructing an empirical cumulative distribution function based on the monthly average resolution data set and the main variable data set;

[0048] Correcting the optimal distribution using a preset correction function to obtain a corrected distribution;

[0049] Determining the quantile corresponding to each value in the prediction result of the global climate model using the empirical cumulative distribution function;

[0050] Mapping the quantile to the calculation result of the empirical cumulative distribution function of the main variable data set to obtain an original prediction model;

[0051] Adjusting the prediction model using a preset distribution adjustment formula to obtain a target prediction model;

[0052] The target prediction model is used to predict future hourly weather data to obtain a target prediction result.

[0053] Optionally, the screening criteria for the representative cities include: weather patterns, temperature ranges and humidity levels.

[0054] Preferably, the climate scenarios of the global climate model include: SSP126, SSP245, SSP370, SSP434 and SSP585.

[0055] Specifically, historical weather data of the representative cities are compiled to obtain data for a typical meteorological year, including:

[0056] Collect raw meteorological data through public weather databases;

[0057] Performing integrity testing and data cleaning on the raw meteorological data using a preset program to obtain preprocessed data;

[0058] Formatting the preprocessed data according to a preset data structure to obtain formatted data;

[0059] The missing values ​​obtained by the integrity detection of the preset program are supplemented by using the difference technology, and the supplemented data of the missing values ​​are matched to the formatted data to obtain the typical meteorological year data.

[0060] Preferably, the candidate statistical distributions include: normal distribution, lognormal distribution and skewed normal distribution.

[0061] Optionally, the judgment data of the maximum likelihood estimation is the residual sum of squares; the calculation formula of the residual sum of squares is: Wherein, SSE is the calculation result of the residual sum of squares; is the i-th observation data point; x i is the i-th fitted distribution value with parameter θ; n is the number of data points.

[0062] Preferably, the empirical cumulative distribution function is: F X (x); where F X (x) is the empirical cumulative distribution function; II() is the index function; N is the number of observations; X i The i-th set of meteorological parameters; x is the meteorological parameter value.

[0063] Furthermore, the correction function is: Among them, Y adjusted is the parameter correction result; is the parameter correction value; μ fut is the average value of the prediction results of the global climate model; μ historical is the mean value of the main variable data set.

[0064] Specifically, the distribution adjustment formula includes:

[0065]

[0066] Among them, loc and scale are distribution parameters; a and b are shape parameters; σ obs is the standard deviation of the main variable data set; Γ(·) is the gamma function.

[0067] Specifically, this embodiment collects typical meteorological year (TMY) data for representative cities in different ASHRAE climate zones. TMY data provides a reliable historical baseline of weather conditions, which is crucial for developing accurate downscaling methods. Each city selected in this embodiment represents a specific climate zone, characterized by unique weather patterns, temperature ranges, and humidity levels. This diverse selection ensures that the downscaling method can be tested and validated under a wide range of climatic conditions, thereby enhancing the robustness and applicability of the distribution adjusted time mapping (DATM) method. The TMY data for these cities come from a reputable platform, providing high-quality and consistent historical weather data for developing and validating the downscaling method. Table 1 lists the selected representative cities and their climate zones, respectively.

[0068] Table 1

[0069]

[0070]

[0071] The list of cities was referenced from a typical building model database to ensure comprehensive representation of different ASHRAE climate zones. By selecting these representative cities to cover a wide range of climate conditions, it aims to provide a solid foundation for the application and validation of the proposed DATM method under different climate conditions.

[0072] Furthermore, the global climate model (GCM) used in this embodiment is MRI-ESM2-0 developed by a meteorological research institute. This model is part of the Coupled Model Intercomparison Project Phase 6 (CMIP6) project and is widely recognized for its comprehensive characterization of the climate system. MRI-ESM2-0 integrates various atmospheric, oceanic and land surface processes to simulate the Earth's climate and predict future climate scenarios based on different greenhouse gas emission trajectories. This embodiment uses the GCM output under the latest Shared Socioeconomic Pathway (SSP). SSP is a set of standardized scenarios that describe different trajectories of future social development, including greenhouse gas emissions, land use and economic growth. The SSPs selected in this embodiment include:

[0073] SSP126: This scenario represents a low greenhouse gas emissions pathway that emphasizes sustainable development and limiting global warming to below 2°C;

[0074] SSP245: Medium emissions scenario, where global development continues along historical trends with some mitigation efforts, leading to medium levels of greenhouse gas concentrations;

[0075] SSP370: a high-emissions scenario characterized by dispersed development and minimal mitigation efforts, leading to large increases in greenhouse gas concentrations and significant global warming;

[0076] SSP434: This scenario represents a pathway with moderate levels of mitigation and adaptation measures, leading to moderate greenhouse gas emissions and associated climate impacts;

[0077] SSP585: An extreme scenario where emissions are very high due to rapid economic growth and continued reliance on fossil fuels, leading to severe climate impacts;

[0078] These SSP monthly outputs from MRI-ESM2-0 are obtained for selected representative cities. These outputs include key meteorological variables such as temperature, relative humidity, wind speed, and global horizontal irradiance (GHI). To integrate these GCM projections with local historical weather data, the monthly GCM outputs are downscaled using the proposed Distribution Adjusted Time Mapping (DATM) method. Using MRI-ESM2-0 and SSP scenarios ensures that the generated future weather data encompass a wide range of potential future climate scenarios, providing a basis for comparing and evaluating the performance of downscaling methods.

[0079] Preferably, this embodiment compiles historical weather data from 2015 to 2023 based on selected representative cities. This period was selected to reflect the real situation of climate conditions affected by ongoing climate change. Historical data can be used as a benchmark for evaluating the accuracy of downscaling methods in capturing real-world climate change and their ability to accurately predict future weather scenarios. Historical weather data comes from publicly accessible weather databases such as the National Oceanic and Atmospheric Administration (NOAA) and other local meteorological agencies, ensuring high quality and consistent records. The collected data include hourly observations of key meteorological variables such as temperature, relative humidity, and wind speed. However, global horizontal irradiance (GHI) data is missing in most stations and is therefore not considered in this embodiment. Data processing is performed using Python 3.9 and involves several key steps:

[0080] 1) Check the raw data for missing values, outliers, and inconsistencies; automated scripts are used to identify and correct errors to ensure the integrity of the dataset;

[0081] 2) formatting the cleaned data to be consistent with the structure of TMY, thus achieving seamless integration with the DATM method;

[0082] 3) For the period of missing data, statistical methods and interpolation techniques are used to accurately estimate the missing values ​​and maintain the continuity and reliability of the data set.

[0083] The Distribution Adjusted Time Mapping (DATM) method, proposed in this embodiment, provides an alternative approach for forecasting future weather data by combining monthly GCM outputs with historical TMY data. This method involves fitting the statistical distribution of historical weather data and adjusting these distributions based on GCM projections to generate high-resolution hourly future weather data. DATM can address the downscaling problem of multiple variables. The three weather variables primarily downscaled and analyzed in this embodiment are temperature, wind speed, and relative humidity.

[0084] Furthermore, the DATM method first identifies and parameterizes the statistical distribution of weather data from the TMY dataset. This process includes the following key steps to produce appropriate and correct distributions of weather variables. First, hourly TMY data for each variable, temperature, relative humidity, wind speed, and solar radiation are obtained. The data are then analyzed to find the distribution that best fits the data. Commonly used distributions include normal distribution for temperature, lognormal distribution for wind speed and relative humidity, and skewed or lognormal distribution for solar radiation. In order to determine the most appropriate distribution, the method requires comparing several potential distributions and estimating them on historical data through maximum likelihood estimation (MLE). The goodness of fit of each distribution is evaluated based on statistical measures, such as the sum of squared errors (SSE) between the actual data and the fitted distribution. SSE is calculated using the following formula:

[0085]

[0086] Where x i is the observed data point; is x i The parameter is the fitted distribution value of θ; n is the number of data points.

[0087] Furthermore, the distribution with the smallest SSE is considered the best fit. After selecting the optimal distribution, the distribution's parameters are estimated so that the mean of the fitted distribution is as close as possible to the mean of the historical data. Some modifications are made to the parameters, particularly for the Weibull, Rayleigh, skew-normal, and lognormal distributions, to eliminate any significant mean shift. This is crucial for adjusting historical data to future climate conditions estimated from GCM outputs. The DATM method uses a certain probability distribution for each climate variable, chosen to be appropriate for the climate variable in question. Table 2 summarizes the assumed distribution for each variable.

[0088] Table 2

[0089]

[0090] These distributions were chosen for their ability to represent the physical characteristics and statistical behavior of each variable. For temperature, the normal distribution and the β distribution were used to accommodate the nature of the temperature data. Relative humidity, the ratio of the amount of water vapor in the air to the maximum amount of water vapor the air can hold at a given temperature, is expressed as a percentage, ranging from 0 to 100%, and is best described by a lognormal or beta distribution. Wind speed is non-negative and often right-skewed; therefore, it is modeled using a lognormal, Weibull, or Rayleigh distribution. For solar radiation, since only positive values ​​are considered, skewed normal, normal, and Rayleigh distributions are used to model the variability of this variable as affected by atmospheric conditions. This helps maintain the statistical properties of each climate variable during the downscaling process to the greatest extent possible.

[0091] The DATM method preferably uses so-called statistical mapping to match the frequency distribution of historical TMY data with projected future climate conditions obtained from GCMs. This process makes it possible to obtain downscaled weather data that accurately describe the expected changes in climate variables. The method consists of several steps:

[0092] 1) In order to make historical observations consistent with future forecasts, the first step is to construct empirical cumulative distribution functions (CDFs). For the hourly resolution TMY data X, construct empirical CDFs F X (x)(i.e. F X (x) or CDF), CDF represents the probability distribution of historical TMY data:

[0093]

[0094] 2) A set of candidate distributions is used to determine the best fitting distribution for each climate variable. Each candidate distribution is fitted to the observed data using maximum likelihood estimation (MLE). The best fitting distribution and its parameters are then selected based on the minimum sum of squared errors (SSE). To ensure that the scaled-down future data reflects the mean changes projected by the climate model, the revised values ​​are adjusted to match the projected mean values ​​of the future models. For a given variable Y (i.e., a meteorological variable such as temperature, wind speed, etc.), the revised value is and the average output μ of future GCMs fut , the adjustment can be described as:

[0095]

[0096] Y adjusted represents the final adjusted variable value. This is the final result after the correction is completed, reflecting the projected changes in future climate conditions; represents the corrected value before mean adjustment. This step ensures that the long-term trend of the model forecast is preserved while correcting the distribution. After the best fitting distribution is determined and corrected, the quantiles of the future model forecasts need to be mapped to the historical observations. For the future model M with a monthly mean resolution future Each value m in the estimate i , CDFF of typical meteorological year data in TMY m (m i ) to determine the corresponding quantile:

[0097] q i =F m (m i )F m Represents the future model prediction value M futureThe cumulative distribution function (CDF) of , which has a monthly mean resolution. This is similar to the CDF representing the historical TMY (Typical Meteorological Year) data, but F m Specifically refers to the distribution of the model's future predicted values. m (m i ), it describes the use of CDF to find the specific value m in the future model forecast. i The corresponding probability (or quantile) process. This probability q i It is then used to map the Find the corresponding value in the historical distribution. The specific breakdown is as follows:

[0098] F m is the cumulative distribution function of future model data;

[0099] m i It is a specific value from future predictions;

[0100] F m (m i ) gives the probability / quantile q of this value appearing in the future model distribution i ;

[0101] This quantile is then used to map back to the historical distribution.

[0102] This process is a key part of the statistical downscaling method, which aims to adjust future climate projections while maintaining the statistical properties of historical observations.

[0103] Using quantiles, we can map this quantile to the observed history (F X (inverse function of x) to obtain the scaled-down hourly value (The function value corresponding to the quantile in the probability distribution after mean adjustment):

[0104]

[0105] This mapping ensures that the future model data distribution is as close as possible to the historical data distribution. For the DATM method, 95% of the mapped values ​​are contained within the calculated margin. While preserving the characteristics of the historical data and the predicted future mean, the 5% mapped quantile is clipped to prevent the weather data from being shrunk beyond realistic limits due to distribution mapping.

[0106] Furthermore, the DATM method employs distribution-specific adjustment techniques to map historical data into future forecasts based on monthly averaged GCM outputs. In this context, the mathematical expressions for modifying the Rayleigh, Weibull, and Beta distributions are discussed separately. For the commonly used Rayleigh distribution for wind speed, the adjustment function minimizes the mean and standard deviation errors:

[0107]

[0108] where μ fut is the predicted future mean, σ obs is the observed standard deviation, loc and scale are the distribution parameters.

[0109] The Weibull distribution is also suitable for wind speed, and the scale parameter is optimized to adjust while maintaining the shape:

[0110]

[0111] Where Γ represents the gamma function.

[0112] For the Beta distribution used for relative humidity, the shape parameters a and b are optimized:

[0113]

[0114] These optimization problems are solved numerically using Sequential Least Squares Programming (SLSQP) for the Rayleigh distribution, Brent's method for the Weibull distribution, and the Limited Memory Broyden-Fletcher-Goldfarb-Shanno (L-BFGS-B) algorithm for the Beta distribution. This approach ensures that the adjusted distributions are as close as possible to the expected mean and the variability of the historical data. Depending on the distribution, different algorithms are used to numerically solve these optimization problems. For the Rayleigh distribution, Sequential Least Squares Programming (SLSQP) is used. SLSQP is an iterative method for solving constrained nonlinear optimization problems. It works by solving a series of quadratic programming subproblems. At each iteration k, the objective function f(u) and the constraints c(u) are approximated using a quadratic model and a linear model, respectively:

[0115]

[0116] Where p is the search direction, H k is an approximation of the Hessian matrix, E and I are the sets of equality constraints and inequality constraints respectively; u k represents the solution vector of the kth iteration; when i belongs to E, it means it is under the equality constraint, and when i belongs to I, it means it is under the inequality constraint. Then the algorithm updates the solution:

[0117] uk+1 =u k +α k ·p k

[0118] where α k Determined by the row search routine.

[0119] For the Weibull distribution, Brent's method is used. This is a root-finding algorithm that combines bisection, the secant method, and inverse quadratic interpolation. Brent's method starts with an interval [a, b] known to contain the root. At each iteration, it chooses between three methods:

[0120] Inverse quadratic interpolation:

[0121]

[0122] Secant method

[0123]

[0124] dichotomy

[0125]

[0126] This method selects the most efficient method among the three methods in each iteration, ensuring the reliability of the bisection method and the fast convergence of the inverse quadratic interpolation as much as possible.

[0127] For the Beta distribution, the L-BFGS-B algorithm is used. It is a quasi-Newton method for solving large-scale optimization problems with simple bounds on the variables. This algorithm approximates the inverse Hessian matrix using limited computer memory. The core idea is to approximate the inverse Hessian using the following update formula:

[0128]

[0129] but:

[0130]

[0131] s k =u k+1 -u k

[0132]

[0133] The algorithm only stores and uses the last memory size parameter m (a preset value used to approximate the previous iteration information of the Hessian matrix) k ,y k) to approximate the Hessian, where m is typically between 3 and 20. This allows it to handle large-scale problems efficiently. The L-BFGS-B variant also handles box constraints (i.e., simple bounds on the variables) by using a projected gradient method. At each iteration, it identifies the set of active bounds and performs the optimization in a subspace of the free variables. The algorithm uses this approximation to determine the search direction and performs a line search, similar to other quasi-Newton methods, but with the added ability to efficiently handle bounds constraints. This approach is particularly well-suited for optimizing the shape parameter of the Beta distribution while being mindful of the presence of bounds, allowing the distribution to be precisely adjusted to match the predicted future mean while preserving the statistical properties of the historical data. m effectively represents the memory length, or history length, that the algorithm uses to construct its approximation of the inverse Hessian matrix. Larger values ​​of m generally provide better approximations but require more memory and computational effort per iteration, while smaller values ​​of m use less memory but may require more iterations to converge.

[0134] The beneficial effects of the present invention are as follows:

[0135] The present invention achieves hourly downscaling of monthly forecast data of different weather variables in the General Circulation Model (GCM) through specific distribution adjustment and quantile mapping methods. This not only maintains the physical properties of the data and the statistical characteristics of historical data, but also reflects the long-term trends and extreme weather conditions predicted by the GCM, providing reliable future meteorological data for building performance simulation.

[0136] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0137] This document uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only intended to help understand the method and core concept of the present invention. At the same time, those skilled in the art will find that the specific implementation methods and application scopes may vary based on the concept of the present invention. In summary, the contents of this specification should not be construed as limiting the present invention.

Claims

1. A method for hourly downscaling of monthly meteorological forecast data based on distribution-adjusted time mapping, characterized in that: include: Identify representative cities for pre-defined climate zones; Compile historical weather data based on the representative cities to obtain data for a typical meteorological year; Determine global climate models and monthly mean resolution datasets; Extracting hourly typical meteorological year data of preset weather variables according to the typical meteorological year data to obtain a main variable data set; the main variable data set includes: hourly temperature data, hourly relative humidity data, hourly wind speed data and hourly solar radiation data; setting a candidate statistical distribution of each of the preset weather variables according to the primary variable data set; Verifying the candidate statistical distribution of each of the preset weather variables using maximum likelihood estimation to obtain a verification data set; Screening the candidate statistical distributions of the preset weather variables according to the validation data set to obtain an optimal distribution of each of the preset weather variables; constructing an empirical cumulative distribution function based on the monthly average resolution data set and the main variable data set; The optimal distribution is corrected using a preset correction function to obtain a corrected distribution; the correction function is: Among them, Y adjusted is the parameter correction result; is the parameter correction value; μ fut is the average value of the prediction results of the global climate model; μ historical is the mean value of the main variable data set; Determining the quantile corresponding to each value in the prediction result of the global climate model using the empirical cumulative distribution function; Mapping the quantile to the calculation result of the empirical cumulative distribution function of the main variable data set to obtain an original prediction model; The prediction model is adjusted using a preset distribution adjustment formula to obtain a target prediction model; the distribution adjustment formula includes: Among them, loc and scale are distribution parameters; a and b are shape parameters; σ obs is the standard deviation of the main variable data set; Γ(·) is the gamma function; The target prediction model is used to predict future hourly weather data to obtain a target prediction result.

2. The method for hourly downscaling of monthly weather forecast data based on distribution-adjusted time mapping according to claim 1, characterized in that: The representative cities were selected based on criteria including weather patterns, temperature ranges and humidity levels.

3. The method for hourly downscaling of monthly weather forecast data based on distribution-adjusted time mapping according to claim 1, characterized in that: The climate scenarios of the global climate model include: SSP126, SSP245, SSP370, SSP434 and SSP585.

4. The method for hourly downscaling of monthly weather forecast data based on distribution-adjusted time mapping according to claim 1, characterized in that: Based on the historical weather data of the representative cities, typical meteorological year data were obtained, including: Collect raw meteorological data through public weather databases; Performing integrity testing and data cleaning on the raw meteorological data using a preset program to obtain preprocessed data; Formatting the preprocessed data according to a preset data structure to obtain formatted data; The missing values ​​obtained by the integrity detection of the preset program are supplemented by using the difference technology, and the supplemented data of the missing values ​​are matched to the formatted data to obtain the typical meteorological year data.

5. The method for hourly downscaling of monthly weather forecast data based on distribution-adjusted time mapping according to claim 1, characterized in that: The candidate statistical distributions include: normal distribution, lognormal distribution and skewed normal distribution.

6. The method for hourly downscaling of monthly weather forecast data based on distribution-adjusted time mapping according to claim 1, characterized in that: The judgment data of the maximum likelihood estimation is the residual sum of squares; the calculation formula of the residual sum of squares is: Wherein, SSE is the calculation result of the residual sum of squares; is the i-th observation data point; x i is the i-th fitted distribution value with parameter θ; n is the number of data points.

7. The method for hourly downscaling of monthly weather forecast data based on distribution-adjusted time mapping according to claim 1, characterized in that: The empirical cumulative distribution function is: Among them, F X (x) is the empirical cumulative distribution function; II() is the index function; N is the number of observations; X i The i-th set of meteorological parameters; x is the meteorological parameter value.

Images