CN108196317A - A kind of Predictive meteorological methods for micro-grid system - Google Patents

Abstract

The invention discloses a weather prediction method for a microgrid system, specifically as follows: step 1, regularly collect and store local weather information R _t at time t; step 2, obtain the address from the National Meteorological Administration and the The following data with the same address and the same time point as in step 1: obtain and store the historical real weather information of the Meteorological Bureau; obtain the historical forecast weather information of the Meteorological Bureau corresponding to the historical real weather information of the Meteorological Bureau. The historical forecast weather information is called F1 _t ; the future weather forecast information of the Meteorological Bureau is obtained, and the future weather forecast information of the Meteorological Bureau is represented by F1'_t; Step 3, using the smoothing index method to predict the weather information at the next moment: Step 4, using the linear regression method , taking F1 _t and F2 _t as input, and R _t as the target column, fits an optimized meteorological information regression model, and predicts the meteorological information for a period of time in the future according to the model; the present invention obtains more accurate meteorological forecast data .

Description

A Meteorological Forecasting Method for Microgrid System

technical field

The invention relates to the field of micro-grid technology, in particular to a weather prediction method for a micro-grid system.

Background technique

With the continuous expansion of the scale of the power grid, the disadvantages of the ultra-large-scale power system are becoming more and more prominent, such as high cost, difficult operation, and difficulty in adapting to the increasing safety and reliability requirements of users and the diversified power supply needs. Especially after several large-scale blackouts occurred in the world in recent years, the vulnerability of the power grid was fully exposed, so distributed generation has been put on the agenda. Renewable energy power generation has become an important driving force for the development of the power system, is an important part of the smart grid, and will play an increasingly important role in the future power system. As a result, countries around the world have begun to pay attention to an environmentally friendly, efficient and flexible power generation method - distributed power generation. In order to eliminate various problems of distributed power generation, to coordinate the contradiction between large power grid and distributed power generation, and to fully tap the value and benefits brought by distributed power generation to power grid and users, a microgrid is proposed.

The microgrid is a relatively independent self-contained power grid system. Since the capacity of the power supply side and the power consumption side of the microgrid is relatively small compared with the large power grid, its own controllability is limited, and sometimes there will be excess power or power consumption. In the case of shortage, we need to predict the power generation and power consumption in advance. The more accurate the forecast of power generation and power consumption, the more beneficial it is for the decision-making of the microgrid. In the microgrid, the energy source of power generation is mainly new energy (such as photovoltaic power generation, wind power generation, etc.), and the power generation of new energy is closely related to weather factors. To a certain extent, the accuracy of weather information determines the upper limit of power generation forecast , the same user power consumption is also highly correlated with weather factors. The scale of the microgrid is relatively small, and the area is not very large, so weather information with relatively precise geographical location is required. However, the spatial accuracy of real-time weather information and forecast weather information provided by the National Weather Service is not very high, so some methods are needed to solve this problem. At present, there are problems such as low prediction accuracy of new energy power generation and weather prediction without network conditions.

Contents of the invention

The technical problem to be solved by the present invention is to overcome the deficiencies of the prior art and provide a weather prediction method for the microgrid system. The present invention obtains more accurate weather forecasts with higher location accuracy through the fusion of online and local weather data. forecast data.

The present invention adopts the following technical solutions for solving the problems of the technologies described above:

A kind of meteorological prediction method for microgrid system proposed according to the present invention comprises the following steps:

Step 1, regularly collect the local meteorological information R _t at time t and store it;

Step 2. Obtain the following data from the National Meteorological Administration with the same address as in step 1 and the same time point as in step 1:

Obtain and store historical real weather information from the Meteorological Bureau;

Obtain the historical forecast weather information of the Meteorological Bureau corresponding to the historical real weather information of the Meteorological Bureau, and the historical forecast weather information of the Meteorological Bureau is called F1 _t ;

Obtain the future weather forecast information of the Meteorological Bureau, and the future weather forecast information of the Meteorological Bureau is represented by F1'_t;

Step 3. Using the smoothing index method to predict the weather information at the next moment:

The specific process of the smoothing index method is as follows:

(1) The local weather information R _t at time t and the future weather forecast information of the Meteorological Bureau at time t-1 are known

F1't _-1 ;

(2) Training smoothing coefficient Require The size of the smoothing coefficient is based on F1' _t-1 and R _t , trained by formula (1)

Among them, R _t+1 represents the local weather information at time t+1;

(3) A certain one is trained by step (2) value;

(4) by formula (1) and It is concluded that the predicted meteorological value at time t obtained by the smoothing index method is F2 _t ,

Step 4: Using the linear regression method, taking F1 _t and F2 _t as input and R _t as the target column, fit an optimized meteorological information regression model, and predict the meteorological information for a period of time in the future according to the model.

As a further optimization scheme for a weather prediction method for a microgrid system according to the present invention, step 4 is specifically as follows:

Using the linear regression method, taking F1 _t and F2 _t as input and R _t as the target column, an optimized meteorological information regression model is fitted;

Among them, R _t ={r _t1 ,r _t2 ,...,r _ti ,...,r _tn }, r _ti represents the local weather information R _t of the i-th weather type at time t, and n represents the type of weather type , R _t has a total of n meteorological types, F1 _t = {f1 _t1 ,f1 _t2 ,...,f1 _ti ,...,f1 _tn }, f1 _ti represents the i-th meteorological type of F1 _t , and F1 _t has a total of There are n kinds of weather types, and the local forecast weather information F2 _t = {f2 _t1 , f2 _t2 ,..., f2 _ti ,..., f2 _tn }, f2 _ti represents the i-th weather type of F2 _t , and F2 _t has a total of There are n weather types;

R _t , F1 _t , and F2 _t are used to train the meteorological information regression model at different times according to t as follows:

Among them, the superscript T represents matrix transposition, c ^T =[c ₁ ,c ₂ ,..., _ci ,...,c _n ] ^T is the parameter of the linear regression model, c _i represents the i-th parameter, c has n parameters in total; θ ^T =[θ ₁ ,θ ₂ ,...,θ _i ,...,θ _n ] ^T is the parameter of the linear regression model, θ _i represents the i-th parameter, and θ has a total of n parameters; γ ^T ＝[γ ₁ ,γ ₂ ,...,γ _i ,...,γ _n ] ^T is the parameter of the linear regression model, γ _i represents the i-th parameter, and γ has n parameters in total ;

According to the trained meteorological information regression model, the predicted value R _t ' _{+ 1} of the meteorological information at the next moment relative to R _t can be obtained by formula (4)

As a further optimization scheme for a meteorological prediction method for a microgrid system according to the present invention, after step 4, the following steps are also included:

Train the K-nearest neighbor model according to the real-time collected R _t and the weather type corresponding to time t, and calculate the real-time weather type by calculating the predicted weather information through the trained K-nearest neighbor model, which is used to display the weather information in the offline state .

As a further optimization scheme of the meteorological prediction method for the microgrid system described in the present invention, in step 1, the local weather information R _t at time t is regularly collected by the local weather information collector and stored.

As a further optimization scheme of the meteorological forecasting method for the microgrid system described in the present invention, the second step is completed by using a python-related crawler library.

Compared with the prior art, the present invention adopts the above technical scheme and has the following technical effects:

The present invention obtains more accurate weather prediction data with higher location accuracy through the fusion of online and local weather data, and at the same time uses the K-nearest neighbor algorithm to realize the display of weather types in an offline state.

Description of drawings

Fig. 1 is a flow chart of the present invention.

Detailed ways

In order to make the object, technical solution and advantages of the present invention clearer, the present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

Based on the problems in the current practical research, the present invention proposes a weather prediction method for microgrid systems, which includes a local weather information collector, a weather information crawler, a local weather information predictor, an optimized weather information model output device and Real-time weather type display, in which the local weather information collector uses locally installed sensors to collect relevant weather information (such as temperature, humidity, etc.); the weather information crawler mainly includes two parts, the first part is to crawl the historical real weather information of the weather website , the second part is to crawl the historical forecast weather information of the weather station; the local weather information predictor is to predict the weather information in the future based on the weather information collected by the local collector; the optimized weather information model output device crawls according to the weather information crawler The historical forecast weather information of the weather station and the weather information predicted by the weather information predictor are compared with the information collected by the local weather information collector to train an optimized weather information model; the real-time weather type display is output according to the optimized weather information model Meteorological information value, forecast real-time weather type (sunny, cloudy, etc.).

The present invention provides a weather prediction method for a microgrid system, including a local weather information collector, a weather information crawler, a local weather information predictor, an optimized weather information model output device, and a real-time weather type display, which solves the problem of When forecasting power generation and power consumption in the microgrid, the forecast weather information used is inaccurate.

The present invention designs a weather prediction method for the microgrid system. The local weather information collector uses locally installed sensors to collect relevant weather information (such as temperature, humidity, etc.); the weather information crawler mainly includes two parts, the first One part is to crawl the historical real weather information of the weather website, and the second part is to crawl the forecast weather information of the weather station; the local weather information predictor uses the smoothing index algorithm to predict the weather information for a period of time in the future based on the weather information collected by the local collector; The training data of the optimized weather information model exporter is the predicted weather information of the weather station history crawled by the weather information crawler and the weather information predicted by the local weather information predictor. The training target is the real weather information collected by the local weather information collector. The linear regression algorithm trains an optimized weather information model, and predicts the weather information for a period of time in the future according to the model; the predicted weather information is used as the input of the real-time weather type display, and then the K-nearest neighbor algorithm is used to give the real-time weather type.

1. Architecture

Fig. 1 has provided the block diagram of the present invention, mainly is made up of five parts: local weather information collector, weather information crawler, local weather information predictor, optimization weather information model exporter and real-time weather type display, wherein local weather information The collector is to use locally installed sensors to collect relevant weather information (such as temperature, humidity, etc.); the weather information crawler mainly includes two parts, the first part is to crawl the historical real weather information of the weather website, and the second part is to crawl the weather station Historical forecast weather information; the local weather information predictor is to predict the weather information for a period of time in the future based on the weather information collected by the local collector; the optimized weather information model output device is based on the forecast weather information and the weather station history crawled by the weather information crawler The weather information predicted by the weather information predictor is compared with the information collected by the local weather information collector to train an optimized weather information model; the real-time weather type display predicts the real-time weather type according to the weather information value output by the optimized weather information model (sunny, cloudy, etc.).

The following specific introduction:

Local weather information collector: store local historical weather information data and collect local weather data in real time. Write the data transmission collected from the terminal sensor into the database to facilitate subsequent data reading and processing.

Meteorological Information Crawler: The so-called crawler is to use Python language tools to automatically download data from the Internet. There are two main modules. The first module is to crawl historical real meteorological data from the website of the National Meteorological Administration and make corresponding local meteorological data. In contrast, it will be found that the data are not completely consistent, because the location accuracy of the meteorological data collected by the Meteorological Bureau is not high, and it cannot accurately describe the meteorological information of the location of the microgrid. The second module crawls the forecasted weather information from the Meteorological Bureau, which includes historical forecasted weather information and future forecasted weather information, and combines the historical forecasted weather information with the weather information predicted by the local weather information predictor to train and optimize the weather information model.

Local weather information predictor: This module is set to use the smoothing index algorithm to predict the weather information for a period of time in the future at regular intervals. The forecast period corresponds to the forecast weather data crawled by the weather information crawler.

Optimize the weather information model output device: the function of this module is mainly to use two sets of predicted weather data from the weather information crawler and the local weather information predictor to approximate the local weather information collected at the corresponding time as much as possible. Specific method: use the linear regression algorithm, take two forecasted meteorological data as input, and local meteorological data as the target column (output), fit an optimized meteorological information regression model, and predict the meteorological information for a period of time in the future according to the model, The weather information has higher location accuracy and is more in line with the environment where the microgrid is located.

Real-time weather type display: The function of this module is mainly based on the data collected by the local collector and the weather type in the weather crawler at the corresponding time, and the weather information predicted by the outputter of the optimized weather information model is used as input, and then the K-nearest neighbor algorithm is used to give Display the real-time weather type.

2. Method flow

1. Local weather information collector:

Table 1 Data storage form (example)

The collected meteorological data are stored in the format of Table 1, and each meteorological data is collected from the sensor with time as the index and stored for use in other steps. The real data is recorded as R.

2. Meteorological information crawler:

The first module crawls and stores the historical real weather information, the storage format is the same as Table 1, and the time granularity is also consistent with Table 1; the function of the second module is mainly to crawl the forecast weather data corresponding to the real weather information, and also It is called the historical forecast weather information F ₁ , and the future weather forecast information is also crawled, which is also called the future forecast weather information F' ₁ . The above steps are completed using Python-related crawler libraries.

3. Local weather information predictor: This module is set to use the smoothing index algorithm to predict the weather information for a period of time in the future. The predicted period corresponds to the time granularity of the forecast weather data crawled by the weather information crawler .

The specific process of the smoothing index algorithm is as follows:

(5) The local weather information at this time and the local weather forecast information at this time are known;

(6) Training smoothing coefficient (Require ), the size of the smoothing coefficient is trained according to the formula (1) based on the comparison between the past forecast number F ₂ and the actual number R

Among them, R _next represents the actual meteorological data at the next moment.

(3) From step (2), a certain value, so that the weather data of the next moment can be predicted based on the weather information of the previous moment.

4. Optimize the weather information model output device:

The main function of this module is to use two sets of predicted weather data from the weather information crawler and the local weather information predictor to approximate the collected local weather information at the corresponding time as much as possible. Specific method: use the linear regression algorithm, take the two predicted weather data F ₁ and F ₂ as input, and the local weather data R as the target column (output), fit an optimized weather information regression model, and predict according to the model Provide weather information for a period of time in the future. Local meteorological information set R={r ₁ ,r ₂ ,...,r _i ,...,r _n }, where r _i represents the i-th weather type, historical forecast weather information F ₁ ={f ₁₁ ,f ₁₂ ,...,f _1i ,...,f _1n }, where f _1i represents the i-th weather type, local forecast weather information F ₂ ={f ₂₁ ,f ₂₂ ,...,f _2i ,.. ., f _2n }.

According to the regression equation as follows:

Among them, f _{predicts 1} , f _{predicts 2} , ..., f _{predicts i} ..., f _{predicts n} is the weather type to be predicted, and the real local weather information is used when training the model, c _i , θ _i , γ _i respectively Indicates the parameters in the linear regression training process. Finally, taking R _next and F' ₁ as input, a set of optimized forecast weather information can be obtained F _forecast = {f _{forecast 1} , f _{forecast 2} ,..., f _{forecast i} , ..., f _{forecast n} } . F _prediction can better reflect the meteorological information, and the location accuracy is higher. It can provide more accurate data for other modules in the microgrid, optimize the operation of the microgrid, and improve the efficiency of the microgrid.

5. Real-time weather type display:

The function of this module is mainly based on the data collected by the local collector and the weather type in the weather crawler at the corresponding time. For example, the local weather information at time A is R _A , and the corresponding weather type is sunny, and the model K is trained according to the real weather value- The nearest neighbor model takes the meteorological information predicted by the optimized weather information model output device as input, and then uses the K-nearest neighbor model to give the real-time weather type, so that the offline weather information can be displayed more accurately.

The present invention proposes a method for optimizing a microgrid based on meteorological information, which is mainly composed of five parts: a local weather information collector, a weather information crawler, a local weather information predictor, an optimized weather information model output device, and a real-time weather type display. , Through the fusion of online and local meteorological data, more accurate and location-accurate weather forecast data can be obtained, and at the same time, the K-nearest neighbor algorithm can be used to display the weather type in the offline state by using the locally collected meteorological data.

For the convenience of description, we assume the following application examples:

The micro-grid system of a school subject building includes photovoltaic power generation, fan power generation, classroom power consumption and energy management system. The following will take this system as an example to introduce it in detail.

(1) Local weather information collector. The system collects temperature, humidity, wind direction, wind speed, atmospheric pressure and irradiance through sensors and stores them in the database.

(2) Meteorological information crawler. The weather information of the weather station is downloaded through the Python crawler, including historical real weather information and forecast weather information, and the forecast weather information includes historical forecast weather information and future forecast weather information. The meteorological types to be crawled are temperature, humidity, wind direction, wind speed, atmospheric pressure, irradiance, and weather type (sunny, rainy, etc.), and the time granularity is consistent with that of the local weather information collector.

(3) Local weather information predictor. According to the existing collected historical meteorological information, the smoothing index algorithm is used to predict the value of meteorological information in the future.

(4) Optimize the weather information model exporter. The main function of this module is to use two sets of predicted weather data from the weather information crawler and the local weather information predictor to approximate the collected local weather information at the corresponding time as much as possible. Specific method: use the linear regression method, take F1 _t and F2 _t as input, and R _t as the target column, and fit an optimized meteorological information regression model;

Among them, R _t ={r _t1 ,r _t2 ,...,r _ti ,...,r _tn }, r _ti represents the local weather information R _t of the i-th weather type at time t, and n represents the type of weather type , R _t has a total of n meteorological types, F1 _t = {f1 _t1 ,f1 _t2 ,...,f1 _ti ,...,f1 _tn }, f1 _ti represents the i-th meteorological type of F1 _t , and F1 _t has a total of There are n kinds of weather types, and the local forecast weather information F2 _t = {f2 _t1 , f2 _t2 ,..., f2 _ti ,..., f2 _tn }, f2 _ti represents the i-th weather type of F2 _t , and F2 _t has a total of There are n weather types;

R _t , F1 _t , and F2 _t are used to train the meteorological information regression model at different times according to t as follows:

Among them, the superscript T represents matrix transposition, c ^T =[c ₁ ,c ₂ ,..., _ci ,...,c _n ] ^T is the parameter of the linear regression model, c _i represents the i-th parameter, c has n parameters in total; θ ^T =[θ ₁ ,θ ₂ ,...,θ _i ,...,θ _n ] ^T is the parameter of the linear regression model, θ _i represents the i-th parameter, and θ has a total of n parameters; γ ^T ＝[γ ₁ ,γ ₂ ,...,γ _i ,...,γ _n ] ^T is the parameter of the linear regression model, γ _i represents the i-th parameter, and γ has n parameters in total ;

According to the trained meteorological information regression model, the predicted value R _t ' _{+ 1} of the meteorological information at the next moment relative to R _t can be obtained by formula (4)

(5) Real-time information display. According to the collected meteorological information, use the K-nearest neighbor algorithm to get the real-time weather type and display it on the microgrid control panel.

The above is only a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto, any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention, All should be covered within the protection scope of the present invention.

Claims (5)

1. a kind of Predictive meteorological methods for micro-grid system, which is characterized in that include the following steps：

Step 1: the local weather information R of timing acquiring t moment_tAnd it stores；

Step 2: being obtained from National Meteorological Bureau, address is identical with the address in step 1, time point phase in time point and step 1 Same data below：

It obtains the true weather information of weather bureau's history and stores；

It obtains and the corresponding weather bureau's historical forecast weather information of the true weather information of weather bureau's history, weather bureau's history Prediction weather information is known as F1_t；

Obtain weather bureau's long-range weather forecast information, weather bureau long-range weather forecast information F1'_tIt represents；

Step 3: subsequent time weather information is predicted using Smoothness Index method：

The detailed process of Smoothness Index method is as follows：

(1) the local weather information R of t moment known to_tAnd the weather bureau long-range weather forecast information F1 ' at t-1 moment_t-1；

(2) training smoothing factorIt is required thatThe size of smoothing factor is according to F1'_t-1And R_t, trained by formula (1) It arrives

Wherein, R_t+1Represent the local weather information at t+1 moment；

(3) one is trained by step (2) to determineValue；

(4) by formula (1) andIt obtains, the meteorological value of t moment prediction obtained by Smoothness Index method is F2_t,

Step 4: using linear regression method, by F1_t、F2_tAs input, R_tAs target column, it is meteorological to fit an optimization Information regression model, and go out according to the model prediction weather information of following a period of time.

2. a kind of Predictive meteorological methods for micro-grid system according to claim 1, which is characterized in that step 4 has Body is as follows：

Using linear regression method, by F1_t、F2_tAs input, R_tAs target column, fit an optimization weather information and return Model；

Wherein, R_t={ r_t1,r_t2,...,r_ti,...,r_tn, r_tiRepresent the local weather information R of t moment_tI-th kind of weather category, N represents the type of weather category, R_tOne shared n kinds weather category, F1_t={ f1_t1,f1_t2,...,f1_ti..., f1_tn, f1_tiTable Show F1_tI-th kind of weather category, F1_tOne shared n kinds weather category, it is local to predict weather information F2_t={ f2_t1,f2_t2,..., f2_ti..., f2_tn, f2_tiRepresent F2_tI-th kind of weather category, F2_tOne shared n kinds weather category；

R when taking different moments according to t_t,F1_t,F2_tTraining weather information regression model is as follows：

Wherein, subscript T representing matrixes transposition, c^T=[c₁,c₂,...,c_i,...,c_n]^TIt is the parameter of linear regression model (LRM), c_iIt represents I-th of parameter, c mono- share n parameter；θ^T=[θ₁,θ₂,...,θ_i,...,θ_n]^TIt is the parameter of linear regression model (LRM), θ_iIt represents I-th of parameter, θ mono- share n parameter；γ^T=[γ₁,γ₂,...,γ_i,...,γ_n]^TIt is the parameter of linear regression model (LRM), γ_iRepresent i-th of parameter, γ mono- shares n parameter；

Weather information regression model according to training is obtained relative to R_tThe weather information predicted value R ' at lower a moment_t+1By formula (4) it obtains

3. a kind of Predictive meteorological methods for micro-grid system according to claim 1, which is characterized in that after step 4 It is further comprising the steps of：

According to the R acquired in real time_tAnd the weather pattern of t moment is corresponded to train K- Neighborhood Models, the weather information of prediction is passed through Real-time weather type is calculated in trained K- Neighborhood Models, is used to implement the display of the Weather information of off-line state.

A kind of 4. Predictive meteorological methods for micro-grid system according to claim 1, which is characterized in that step 1 In, using the local weather information R of local meteorological information acquisition device timing acquiring t moment_tAnd it stores.

5. a kind of Predictive meteorological methods for micro-grid system according to claim 1, which is characterized in that step 2 is It is completed with reptile library related Python.