Summary of the invention
The real time meteorological data that technical matters to be solved by this invention is to provide precision higher for industrial control field and energy industry and prediction weather data.
For solving the problems of the technologies described above, the invention provides a kind of intelligent weather station system of measurable weather data, comprise sensor assembly, it is characterized in that: sensor assembly is connected with the computer for controlling in intelligent weather station, in computer for controlling, be provided with SD card, computer for controlling is connected with GPRS communication module by SD card.
The intelligent weather station system of aforesaid measurable weather data, is characterized in that: also comprise power module, and described power module is connected with sensor assembly, computer for controlling, GPRS communication module respectively.
The intelligent weather station system of aforesaid measurable weather data, is characterized in that: sensor assembly comprises wind speed wind direction sensor, temperature sensor, humidity sensor, baroceptor, light intensity sensor and rain sensor.
The intelligent weather station system of aforesaid measurable weather data, is characterized in that: described computer for controlling comprises data processing module, for receiving and the real time meteorological data of processes sensor collection, and the data after process is sent to SD card.
The intelligent weather station system of aforesaid measurable weather data, it is characterized in that: described computer for controlling comprises weather data prediction module, according to history weather data, utilize adaptive logic network A LN method to train prediction algorithm model, and obtain prediction weather data according to actual measurement weather data and the model of having trained.
The intelligent weather station system of aforesaid measurable weather data, it is characterized in that: described computer for controlling comprises main control module, for controlling the operation of sensor assembly, weather data prediction module, GPRS communication module, comprising control data and flowing to, monitor each process ruuning situation.
The beneficial effect that the present invention reaches:
The present invention uses a kind of intelligent weather station system of measurable weather data, the real time meteorological data providing precision higher for industrial control field and energy industry and prediction weather data, these weather datas comprise: wind speed and direction, temperature, relative humidity, air pressure, light intensity, rainfall amount etc.The present invention predicts that weather data is realized by the method for the adaptive logic network (ALN) in intelligent algorithm.This method prediction accuracy is high, and modeling process is relatively simple, and model computing velocity is fast, is easy to realize in embedded systems.By the statistical study at least trimestral historical data, set up prediction algorithm model.Only needing when prediction weather data using surveying weather data as the initial conditions of prediction algorithm model, weather data can be obtained fast and predicting the outcome.
Embodiment
Below in conjunction with accompanying drawing, the present invention is further described with enforcement.
The present invention passes through the current weather information of sensor Real-time Obtaining, be stored in the SD card in intelligent weather station through data processing module, intelligence weather station sets up weather data prediction algorithm model by artificial intelligence prediction algorithm according to the geography information in configuration file and history weather data, in general after generation forecast algorithm model, from now on without the need to revising this prediction algorithm model again.When predicting weather data, using the initial conditions of the weather data of current collection as prediction algorithm model, prediction weather data result can be obtained fast.
The intelligent weather station system of a kind of measurable weather data of the present invention comprises: sensor assembly, data processing module, SD card, weather data prediction module, GPRS communication module and power module.
Sensor assembly comprises: wind speed wind direction sensor, temperature sensor, humidity sensor, baroceptor, light intensity sensor, rain sensor.The sampling period of each sensor is all 5 seconds.
Data processing module be used for processes sensor sampling after real time meteorological data.Maximal value, minimum value, the mean value of data processing module principal statistical every 5 minutes, 10 minutes, 15 minutes, 1 hour, 1 day all kinds of weather data.
The content that SD card stores comprises four parts: the weather data of history weather data, prediction, weather data prediction algorithm model file, intelligent weather station configuration information.The result that history weather data is through sensor Real-time Collection, data processing module statistics exports, comprise maximal value, minimum value, the mean value of every 5 minutes, 10 minutes, 15 minutes, 1 hour, 1 day, these data retain more than 2 years.The weather data of prediction comprises the weather data of following 4 hours, 24 hours, 48 hours, 72 hours.Intelligence weather station configuration information includes the geography information of this location, weather station, as longitude, latitude, height above sea level, geopotential unit; The configuration information of intelligence weather station collection, forecast, as gathered weather data interval, the forecast weather data time interval; Intelligence weather station dispatches from the factory information etc.
Weather data prediction module has two functions, is first according to history weather data training prediction algorithm model, and another function obtains prediction weather data according to actual measurement weather data and the model of having trained.Training prediction algorithm model needs at least trimestral history weather data.After prediction algorithm model training completes, generate dtr file, exist in SD card.During prediction weather data, first read the measured data of having trained model (dtr file) and just having collected, drawn by intelligent algorithm and predict the outcome and be stored in SD card.
GPRS communication module supports 4 frequency range: 850/900/1800/1900MHZ, supports GSM standard AT order, supports GSM/GPRS Phase2/2+ agreement, embedded ICP/IP protocol.GPRS module interface comprises: serial line interface (maximum serial ports speed can reach 115200bit/s) is for obtaining the weather data after acquisition process; The SIM card that standard SIM card interface (1.8V or 3V) provides for inserting operator.If use GPRS private-line mode to carry out data transmission, need the SIM card using operator's specific customization.
Main control module, for controlling the operation of other functional module, as the control data flow direction, monitors each process ruuning situation.
Power module provides power supply support to whole micro weather station.Whole system adopts solar powered.Power module comprises sun power electroplax, accumulator, DC/DC direct current cross-changing unit.
As shown in Figure 1, intelligent weather station obtains outside weather data by sensor.Sensor comprises: wind speed wind direction sensor, temperature sensor, humidity sensor, baroceptor, light intensity sensor, rain sensor.The data that sensor collects carry out statistical calculation by data processing module, count maximal value, minimum value, the mean value of every 5 minutes, 10 minutes, 15 minutes, 1 hour, 1 day respectively.The result that statistics exports is stored in SD card.Prediction module obtains historical data through training generation forecast algorithm model from SD card, and predicted data according to the prediction algorithm model of having trained and real time data generation forecast data, and can also be stored in SD card by prediction module.Main control module is responsible for controlling the collection of weather data, process, storage, forecasting process.Power module provides electric flux for whole intelligent weather station, ensures that intelligent meteorology is stood firm and determines work.
As shown in Figure 2, weather data forecast function mainly completes training pattern and predicted data two functions.Before predicted data, need a prediction algorithm model trained, be exactly the dtr file be stored on SD card specifically.In general a model can be predicted data service for a long time, does not need regeneration prediction algorithm model in forecasting process.In time finding that the root-mean-square error of predicted data and real data is larger, online retraining can be set, improve the precision of prediction of prediction algorithm model.
Training pattern carrys out generation model by reading history weather data and configuration file, and history weather data must be at least trimestral history weather data, and configuration file contains local geographic information data.Model training completes generation dtr file and is stored in SD card.During predicted data, need to read dtr file and real time data, can fast prediction go out to predict weather data.
Following explanation is carried out for adaptive logic network A LN method:
1.ALN basic structure
Any number of linear logic relational expression between any number of independently input variable and variable can be included in adaptive logic network.Linear equation form is as follows:
ALN is by changing the weight w in its system of linear equations
ijproduce the result of expectation.Usually, in order to the generality of model representation, regulation X
0≡ 1, namely represents the constant term of equation.
In neural network model, X is input, and Y is the output of network, to X
0constraint information can be understood to neuronic amount of bias.
Make L
j=0.Formula (1) defines straight line (n=1), a plane (n=2) or a lineoid (n>2), therefore obtains following system of equations expression formula:
ALN is by the linear relation (LTU, Linear Threshold Unit) passed judgment on the threshold value node as network.If make threshold value be 0, then each similar node will to L
jwhether>=0 set up and pass judgment on.The non-"True" of its result (1, do not reach threshold value) is i.e. "false" (0, reach threshold value).Therefore (2) are converted into Linear inequalities:
And the father node of these linear relation nodes is logical operator " AND " in ALN and " OR ".
Fig. 3 describes the ALN structure (only statistics contains the layer of LTU and logical operator) of three layers.OR(AND(2 can be write a Chinese character in simplified form to this ALN structure), AND(2)).
The size of ALN structure is bounded, because the LTU number determined in ALN for the finite data sample of parameter estimation is limited, if there be n independent variable, then needs the weight vector estimated to be n+1 dimension in LTU.In addition, the shape of ALN structure is also limited.Many ALN body structure surfaces look different, but actual be equivalent.Such as AND(LTU, AND(2)) and AND(3) be equivalent.So just greatly reduce the structure species needing to consider ALN.
2. based on the modeling method of ALN
ALN had both maintained the simplicity of traditional Corpus--based Method principle modeling, also had the superiority that it is incomparable.Because a continuous function can be approached with arbitrary accuracy by one group of straight-line segment, therefore ALN can construct logical relation between linear function and linear unit according to accuracy requirement, carrys out the arbitrary continuous function of matching.Fig. 4 describes the two-dimentional output region of ALN in Fig. 3.When in figure, the corresponding all sample points (X, Y) of dash area drop on this region, network exports as "True" (1).
And if only if when all sample points drop on two straight line L1 and below L2, and in Fig. 4, the AND of top exports as "True".L1 and L2 has taken out the shape of " pointed tent ", and its connection mode AND is also equivalent to the minimum value (MIN) of getting two straight lines.In like manner, second AND gives half " flat-top tent " shape.
And if only if when sample point drops in shadow region, and the output of OR is "True", is equivalent to and gets maximal value (MAX) to two " tents ".
Principle LTU being used to logical operator computing can be summed up: AND is equivalent to and gets minimum (MIN), and OR is equivalent to and gets maximum (MAX).
The final mask i.e. broken line of the handing-over in " 0 " space and " 1 " space of ALN matching, result is continuous function, but non-monotonic, convex function that neither be simple or concave function.At more higher dimensional space, it is the splicing of one group of lineoid that ALN exports.Its shape can be changed by the structure of amendment tree or LTU weights and threshold.Such as, concavity changes by logical operator.AND(4) convex surface is obtained, and with OR(4) obtain a concave surface.Monotonicity can positive and negatively by the weight coefficient revising LTU obtain.ALN allows to force as constraint condition these characteristics.
Within the scope of certain straight-line segment, the change of its output valve is directly proportional to input value, and the weight coefficient of LTU represents rate of change.And in the neural network structure of standard, to export and the relation of input cannot be learnt, unless every other input variable is endowed fixed value, otherwise the tiny change of other input variables all can cause the larger change of output quantity.
The prediction algorithm of 3.ALN
This process of ALN algorithm predicts is called as parameter estimation, and this method is with the test by there being planning, periodic observation or by schema extraction, ALN algorithm uses least square method to carry out the estimation of LTU weight coefficient.
At study initial phase, LTU all in ALN is assigned with random weight coefficient, if there is priori or constraint condition, can there is other distribution method.
According to the order of sample, training sample (X
t, Y
t) as the input of ALN, logical operation value, along tree network transmission, exports end value.
Below disclose the present invention with preferred embodiment, so it is not intended to limiting the invention, and all employings are equal to replacement or the technical scheme that obtains of equivalent transformation mode, all drop within protection scope of the present invention.