CN113516766B - Neural network algorithm-based independent coordinate system parameter analysis method and system - Google Patents

Neural network algorithm-based independent coordinate system parameter analysis method and system Download PDF

Info

Publication number
CN113516766B
CN113516766B CN202110785374.9A CN202110785374A CN113516766B CN 113516766 B CN113516766 B CN 113516766B CN 202110785374 A CN202110785374 A CN 202110785374A CN 113516766 B CN113516766 B CN 113516766B
Authority
CN
China
Prior art keywords
classification
value
coordinate system
compensation surface
control point
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN202110785374.9A
Other languages
Chinese (zh)
Other versions
CN113516766A (en
Inventor
张禾
任成冕
曹扬
杨莎莎
韩莉
胡书
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Guizhou Zhengyuan Zhihua Technology Co ltd
Original Assignee
Guizhou Zhengyuan Zhihua Technology Co ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Guizhou Zhengyuan Zhihua Technology Co ltd filed Critical Guizhou Zhengyuan Zhihua Technology Co ltd
Priority to CN202110785374.9A priority Critical patent/CN113516766B/en
Publication of CN113516766A publication Critical patent/CN113516766A/en
Application granted granted Critical
Publication of CN113516766B publication Critical patent/CN113516766B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T17/00—Three dimensional [3D] modelling, e.g. data description of 3D objects
    • G06T17/05—Geographic models
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C15/00—Surveying instruments or accessories not provided for in groups G01C1/00 - G01C13/00
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
    • G06F17/10—Complex mathematical operations
    • G06F17/11—Complex mathematical operations for solving equations, e.g. nonlinear equations, general mathematical optimization problems
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
    • G06F18/24133—Distances to prototypes
    • G06F18/24137—Distances to cluster centroïds
    • G06F18/2414—Smoothing the distance, e.g. radial basis function networks [RBFN]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Abstract

The invention relates to an independent coordinate system parameter analysis method and system based on a neural network algorithm, wherein under the condition that a specific ellipsoid expansion algorithm is unknown when an independent coordinate system is converted into a national coordinate system, an optimal compensation surface is analyzed through a neural network single-layer perceptron algorithm in a repeated iteration mode, the defect that only a small amount of control point precision changes can be analyzed manually in the prior art is overcome, the error change conditions of control points in all project areas are analyzed at one time, an optimal conversion parameter value is obtained, high-precision conversion of measurement data is guaranteed, and efficiency and precision for obtaining specific parameters required by coordinate system conversion are improved.

Description

Neural network algorithm-based independent coordinate system parameter analysis method and system
Technical Field
The invention relates to the technical field of engineering construction, in particular to an independent coordinate system parameter analysis method and system based on a neural network algorithm.
Background
An independent coordinate system is required to be arranged in a large-scale engineering construction project, and the error from the design length to the implementation lofting length in the engineering project design and construction is ensured to be less than 2.5 centimeters per kilometer. Such errors are projection length errors, the main reason being that since the gaussian projection used is an equiangular projection, the angle element after projection remains constant before and after projection, while the length deformation increases the further away from the central meridian after projection. Meanwhile, a projection compensation elevation surface is required to be arranged in a high-altitude area during projection, and the compensation elevation surface is an elevation surface determined for mutually compensating the correction of the length of the Gaussian projection on the ground and the correction of the length of the Gaussian projection on the ground on the reference surface. General engineering applications use an ellipsoid expansion method to set a compensation elevation surface.
General large-scale engineering construction projects (such as highways, high-speed railways, large-scale city planning) and the like need to be provided with independent coordinate systems meeting the survey requirements of the projects. Independent coordinate system settings generally need to satisfy two requirements: firstly, the requirements of relevant design survey specifications are met, namely the deformation error of the projection length per kilometer is not more than 2.5 centimeters, and secondly, the set independent coordinate system can be converted with a national standard coordinate system.
The Gaussian-gram Luge projection algorithm is fixed in the setting process of the independent coordinate system, and no special requirement is changed; however, various methods exist on the setting of the projection compensation elevation surface, and the usable ellipsoid expansion algorithm is more than ten; the independent coordinate system setting unit generally does not describe the ellipsoid expansion algorithm in detail, so that a specific using unit does not know which ellipsoid expansion algorithm is used when the independent coordinate system is used, and a certain error exists after conversion calculation; therefore, the time consumption cannot meet the requirement of coordinate system conversion.
At present, three methods are generally adopted for converting an independent coordinate system into a national standard coordinate system in engineering application, firstly, an ellipsoidal expansion algorithm is adopted by a survey design and measurement unit for consulting engineering projects during the design of the independent coordinate system, and a related calculation formula and a setting method are provided, but the survey unit is difficult to provide related specific parameters under general conditions, and the main reason is that the expansion algorithm is not specified to provide data for projects; and the specific algorithms are more, and the design unit only needs to adopt parameters meeting the standard design requirements. Secondly, an ellipsoid expansion algorithm adopted by a manual one-by-one analysis survey measurement design unit is adopted; by adopting the method, the error conditions of all points in the area are difficult to analyze, and the manual calculation amount is large; generally, only a few control points in the engineering project range can be selected for analysis and comparison, and the whole area of the engineering project cannot be effectively analyzed. Thirdly, directly adopting a method of Boolean Sha Qi parameters to carry out coordinate system conversion, and using the method can avoid analyzing the specifically used independent coordinate system parameters; however, due to the fact that parameters are not clear, the error of the control point after conversion cannot be optimized; can only be used under the condition of low precision requirement.
Disclosure of Invention
The invention aims to provide an independent coordinate system parameter analysis method and system based on a neural network algorithm, which can improve the efficiency and the precision of obtaining specific parameters required by coordinate system conversion.
In order to achieve the purpose, the invention provides the following scheme:
a method for analyzing parameters of an independent coordinate system based on a neural network algorithm, the method comprising:
acquiring three-dimensional coordinates of a plurality of control points in an independent coordinate system and three-dimensional coordinates in a national coordinate system respectively; the control points are points which are distributed in a measurement area for measurement operation in an engineering construction project;
determining a plurality of reference compensation surface elevations at preset steps according to the compensation surface elevations of the independent coordinate system;
converting three-dimensional coordinates of the control points in the independent coordinate system into projected three-dimensional coordinates in a national coordinate system by using different height values of the reference compensation surface by using a Gaussian projection algorithm;
calculating error values of the projected three-dimensional coordinates of the control points under each reference compensation surface elevation and the three-dimensional coordinates of the control points in the national coordinate system to obtain the coordinate error value of each control point under each reference compensation surface elevation;
classifying and identifying each control point according to the coordinate error value of each control point, and combining the classification identification of each control point, the elevation of the reference compensation surface and the coordinate error value to form a classification matrix;
performing error classification calculation by using a neural network single-layer perceptron algorithm according to the classification matrix to obtain a classification linear equation of a classification value about the height of a compensation surface;
according to the classification matrix, obtaining an error value fitting linear equation of the classification value relative to the height of the compensation surface by using a linear fitting method;
acquiring an intersection point of the classification linear equation and the error value fitting linear equation, and judging whether a classification value in the intersection point is smaller than a classification value threshold value or not to obtain a first judgment result;
if the first judgment result shows that the reference surface height is not the same as the reference surface height, updating the preset step pitch, and returning to the step of determining a plurality of reference compensation surface heights by the preset step pitch according to the compensation surface heights of the independent coordinate system;
and if the first judgment result shows that the height of the compensation surface in the intersection point is positive, outputting the height of the compensation surface as the optimal height of the compensation surface.
Optionally, the converting, by using a gaussian projection algorithm, the three-dimensional coordinates of the plurality of control points in the independent coordinate system into the projected three-dimensional coordinates of each control point in the national coordinate system under the elevation of the plurality of reference compensation surfaces specifically includes:
initializing basic parameters of an independent coordinate system;
according to basic parameters of the independent coordinate system, performing Gaussian projection back calculation on three-dimensional coordinates of a plurality of control points in the independent coordinate system under the elevations of a plurality of reference compensation surfaces respectively to obtain longitude and latitude coordinates of the plurality of control points under each elevation of the reference compensation surface;
and calculating the longitude and latitude coordinates of the control points under each datum compensation surface elevation according to the national coordinate system parameters by utilizing Gaussian-Kluger projection forward calculation, and projecting the longitude and latitude coordinates of each control point under each datum compensation surface elevation to a three-dimensional coordinate in a national coordinate system to serve as the projected three-dimensional coordinate of each control point under each datum compensation surface elevation in the national coordinate system.
Optionally, the classifying and identifying each control point according to the coordinate error value of each control point specifically includes:
classifying and identifying the control points with the coordinate error value larger than 0 as 1, and classifying and identifying the control points with the coordinate error value smaller than 0 as-1; the coordinate error value is not equal to 0.
Optionally, according to the classification matrix, performing error classification calculation by using a neural network single-layer perceptron algorithm to obtain a classification linear equation of a classification value with respect to a compensation surface elevation, specifically including:
initializing the number n of iterations as 1 and parameters of a neuron network single-layer perceptron; the parameters of the neuron network single-layer perceptron comprise an initial weight matrix;
the neural network single-layer perceptron after the input parameters of the classification matrix are initialized obtains a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix;
judging whether the initial classification value of each control point under each reference compensation surface elevation calculated by the nth neural unit is the same as the classification identification of each control point or not, and obtaining a second judgment result;
if the second judgment result shows no, updating the initial weight matrix to the weight matrix calculated by the nth neural unit, increasing the value of n by 1, returning to the step of inputting the classification matrix into the neural network single-layer perceptron after the parameters are initialized, obtaining a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix;
if the second judgment result shows yes, the formula is used according to the weight matrix calculated by the nth neural unit
Figure GDA0003844833070000041
Determining a classification linear equation of the classification value about the compensation surface elevation; wherein Y is a dependent variable of the classification linear equation, X is an independent variable of the classification linear equation, K is a coefficient of the classification linear equation, B is a constant of the classification linear equation, W is a constant of the classification linear equation n(1) 、W n(2) 、W n(3) The weight values are respectively the first, the second and the third weight values in the weight matrix of the nth iteration nerve unit.
Optionally, the obtaining, by the neural network single-layer perceptron after the classification matrix input parameter is initialized, a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each reference compensation surface elevation based on the initial weight matrix specifically includes:
the neuron network single-layer perceptron after the input parameters of the classification matrix are initialized, and a formula is utilized based on the initial weight matrix
Figure GDA0003844833070000042
Calculating an initial classification value of each control point under each reference compensation surface elevation; wherein, y IJ For the initial classification value of the J-th reference compensation surface height Cheng Xiadi I control points, sgn () is a sign function, x i For the ith coordinate error value, ω, in the ith control point i The weight of the ith coordinate error value of the ith control point in the initial weight matrix is obtained, N is the number of the coordinate error values in the ith control point, and b is a bias value;
according to the initial classification value of each control point under each reference compensation surface elevation, using a formula F = lr X (X-T-dot (E))/X]Calculating the bias weight F of each control point under each reference compensation surface elevation; wherein lr is the learning rate, X.T is the matrix after the transformation of the classification matrix, dot () is the matrix point multiplication function, shape [, [ alpha ] ]]For a row in the row and column of the classification matrix, E = d I -y IJ ，d I The classification mark of the I-th control point, and E is the difference value between the classification mark of the I-th control point and the initial classification value;
and determining a matrix obtained by adding the bias weight of each control point under each reference compensation surface elevation and the initial weight matrix as the weight matrix calculated by the neural unit for the nth time.
A system for independent coordinate system parameter analysis based on neural network algorithms, the system comprising:
the three-dimensional coordinate acquisition module is used for acquiring three-dimensional coordinates of the control points in an independent coordinate system and three-dimensional coordinates in a national coordinate system respectively; the control points are points which are distributed in a measurement area for measurement operation in the engineering construction project;
the reference compensation surface elevation determination module is used for determining a plurality of reference compensation surface elevations at preset step distances according to the compensation surface elevations of the independent coordinate system;
the projection three-dimensional coordinate conversion module is used for converting three-dimensional coordinates of the control points in the independent coordinate system into projection three-dimensional coordinates in a national coordinate system by using different reference compensation surface height values by using a Gaussian projection algorithm;
the coordinate error value obtaining module is used for calculating the error value of the projection three-dimensional coordinates of the control points under each reference compensation surface elevation and the three-dimensional coordinates of the control points in the national coordinate system to obtain the coordinate error value of each control point under each reference compensation surface elevation;
the classification matrix forming module is used for performing classification identification on each control point according to the coordinate error value of each control point and forming a classification matrix by combining the classification identification, the reference compensation surface elevation and the coordinate error value of each control point;
the classification linear equation obtaining module is used for performing error classification calculation by utilizing a neural network single-layer perceptron algorithm according to the classification matrix to obtain a classification linear equation of a classification value about the elevation of the compensation surface;
an error value fitting linear equation obtaining module, configured to obtain an error value fitting linear equation of the classification value with respect to the height of the compensation surface by using a linear fitting method according to the classification matrix;
the first judgment result obtaining module is used for obtaining an intersection point of the classification linear equation and the error value fitting linear equation, judging whether a classification value in the intersection point is smaller than a classification value threshold value or not and obtaining a first judgment result;
a step returning module, configured to update a preset step pitch if the first determination result indicates no, and return to the step of "determining multiple reference compensation surface elevations at the preset step pitch according to the compensation surface elevations of the independent coordinate system";
and the optimal compensation surface elevation output module is used for outputting the compensation surface elevation in the intersection point as the optimal compensation surface elevation if the first judgment result shows that the elevation is positive.
Optionally, the projection three-dimensional coordinate conversion module specifically includes:
the basic parameter initialization submodule is used for initializing basic parameters of an independent coordinate system;
the longitude and latitude coordinate obtaining submodule is used for respectively carrying out Gaussian projection back calculation on the three-dimensional coordinates of the control points in the independent coordinate system under the elevations of the reference compensation surfaces according to the basic parameters of the independent coordinate system to obtain the longitude and latitude coordinates of the control points under the elevation of each reference compensation surface;
and the projection three-dimensional coordinate determination submodule is used for calculating the three-dimensional coordinates of the longitude and latitude coordinates of each control point under each reference compensation surface elevation projected into the national coordinate system by utilizing Gaussian-Kruger projection forward calculation according to the national coordinate system parameters, and using the three-dimensional coordinates as the projection three-dimensional coordinates of each control point under each reference compensation surface elevation in the national coordinate system.
Optionally, the classification matrix forming module specifically includes:
the classification identification submodule is used for classifying and identifying the control points with the coordinate error value larger than 0 as 1 and classifying and identifying the control points with the coordinate error value smaller than 0 as-1; the coordinate error value is not equal to 0.
Optionally, the classification linear equation obtaining module specifically includes:
the initialization submodule is used for initializing the parameters of the neuron network single-layer perceptron, wherein the iteration number n is 1; the parameters of the neuron network single-layer perceptron comprise an initial weight matrix;
the initial classification value obtaining submodule is used for inputting the classification matrix into the neural network single-layer perceptron after the parameters are initialized, and obtaining a weight matrix calculated by the neural unit for the nth time and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix;
the second judgment result obtaining submodule is used for judging whether the initial classification value of each control point under each reference compensation surface elevation calculated by the nth neural unit is the same as the classification identification of each control point or not so as to obtain a second judgment result;
an updating submodule, configured to, if the second determination result indicates no, the initial weight matrix is updated to the weight matrix calculated by the nth neural unit, and the value of n is increased by 1, returning to the step of obtaining a weight matrix calculated by the neural unit for the nth time and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix by the neuron network single-layer perceptron after the input parameters of the classification matrix are initialized;
a classification linear equation determining submodule for utilizing a formula according to the weight matrix calculated by the nth neural unit if the second judgment result represents yes
Figure GDA0003844833070000071
Determining a classification linear equation of the classification value relative to the height of the compensation surface; wherein Y is a dependent variable of the classification linear equation, X is an independent variable of the classification linear equation, K is a coefficient of the classification linear equation, B is a constant of the classification linear equation, W is a constant of the classification linear equation n(1) 、W n(2) 、W n(3) The first, second and third weight values in the weight matrix of the nth iteration neural unit are respectively.
Optionally, the initial classification value obtaining sub-module specifically includes:
an initial classification value calculation unit for initializing the input parameters of the classification matrix to the neuron network single-layer perceptron, and based on the initial weight matrix, using a formula
Figure GDA0003844833070000072
Calculating an initial classification value of each control point under each reference compensation surface elevation; wherein, y IJ For the initial classification value of the J-th reference compensation surface height Cheng Xiadi I control points, sgn () is a sign function, x i Is the I-th coordinate error value, omega, in the I-th control point i The weight of the ith coordinate error value of the ith control point in the initial weight matrix is obtained, N is the number of the coordinate error values in the ith control point, and b is a bias value;
an offset weight calculation unit, configured to use a formula F = lr × (x.t.dot (E))/x.shape [0] according to the initial classification value of each control point at each reference compensation surface elevation]Calculating the bias weight F of each control point under each reference compensation surface elevation; wherein lr is the learning rate, X.T is the matrix after the transformation of the classification matrix, dot () is the matrix point multiplication function, shape [, [ alpha ] ]]For a row in the row and column of the classification matrix, E = d I -y IJ ，d I The classification mark of the I-th control point, and E is the difference value between the classification mark of the I-th control point and the initial classification value;
and the weight matrix determining unit is used for determining a matrix obtained by adding the bias weight of each control point under each reference compensation surface elevation and the initial weight matrix as the weight matrix calculated by the neural unit for the nth time.
According to the specific embodiment provided by the invention, the invention discloses the following technical effects:
the invention provides an independent coordinate system parameter analysis method and system based on a neural network algorithm, wherein under the condition that a specific ellipsoid expansion algorithm is unknown when an independent coordinate system is converted into a national standard coordinate system, an optimal compensation surface is analyzed through a neural network single-layer perceptron algorithm in a repeated iteration mode, the defect that only a small number of control point precision changes can be analyzed manually in the prior art is overcome, the error change conditions of control points in all project areas are analyzed at one time, an optimal conversion parameter value is obtained, high-precision conversion of measurement data is guaranteed, and the efficiency and precision for obtaining specific parameters required by coordinate system conversion are improved.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings needed to be used in the embodiments will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without inventive exercise.
FIG. 1 is a flow chart of an independent coordinate system parameter analysis method based on neural network algorithm according to the present invention;
fig. 2 is a schematic diagram of an independent coordinate system parameter analysis method based on a neural network algorithm according to a second embodiment of the present invention;
FIG. 3 is a diagram of a single-layer sensor according to a second embodiment of the present invention;
fig. 4 is a schematic diagram of a classification line equation and an error value fitting line equation provided in the second embodiment of the present invention.
Detailed Description
The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
The invention aims to provide an independent coordinate system parameter analysis method and system based on a neural network algorithm, which can improve the efficiency and the precision of obtaining specific parameters required by coordinate system conversion.
In order to make the aforementioned objects, features and advantages of the present invention comprehensible, embodiments accompanied with figures are described in further detail below.
Example one
The invention discloses an independent coordinate system parameter analysis method based on a neural network algorithm, which comprises the following steps of:
s101, acquiring three-dimensional coordinates of a plurality of control points in an independent coordinate system and three-dimensional coordinates in a national coordinate system respectively; the control points are points which are distributed in the measuring area for measuring operation in the engineering construction project;
s102, determining a plurality of reference compensation surface elevations at preset step distances according to the compensation surface elevations of the independent coordinate system;
s103, converting three-dimensional coordinates of the control points in the independent coordinate system into projected three-dimensional coordinates in a national coordinate system by using different height values of the reference compensation surface by using a Gaussian projection algorithm, and specifically comprising the following steps:
initializing basic parameters of an independent coordinate system;
according to basic parameters of the independent coordinate system, performing Gaussian projection back calculation on three-dimensional coordinates of a plurality of control points in the independent coordinate system under the elevation of a plurality of reference compensation surfaces respectively to obtain longitude and latitude coordinates of the plurality of control points under the elevation of each reference compensation surface;
and calculating the three-dimensional coordinates of each control point under each reference compensation surface elevation in a national standard projection plane rectangular coordinate system by adopting Gaussian-Kluger projection forward calculation according to the national standard coordinate system parameters of the longitude and latitude coordinates of the control points under each reference compensation surface elevation, and taking the three-dimensional coordinates as the projection three-dimensional coordinates of each control point under each reference compensation surface elevation in the national coordinate system.
S104, calculating error values of the projected three-dimensional coordinates of the control points under each reference compensation surface elevation and the three-dimensional coordinates of the control points in the national coordinate system to obtain the coordinate error value of each control point under each reference compensation surface elevation;
s105, performing classification identification on each control point according to the coordinate error value of each control point, and combining the classification identification, the reference compensation surface elevation and the coordinate error value of each control point to form a classification matrix;
and (4) classification identification: classifying and marking the control points with the coordinate error values larger than 0 as 1, and classifying and marking the control points with the coordinate error values smaller than 0 as-1; the coordinate error value is not equal to 0.
S106, performing error classification calculation by using a neural network single-layer perceptron algorithm according to the classification matrix to obtain a classification linear equation of a classification value about the elevation of the compensation surface, and specifically comprising the following steps:
initializing the number n of iterations as 1 and parameters of a neuron network single-layer perceptron; the parameters of the neuron network single-layer perceptron comprise an initial weight matrix;
inputting the classification matrix into the neural network single-layer perceptron after the parameters are initialized, and obtaining a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix;
judging whether the initial classification value of each control point under each reference compensation surface elevation calculated by the nth neural unit is the same as the classification identification of each control point or not, and obtaining a second judgment result;
if the second judgment result shows no, updating the initial weight matrix to the weight matrix calculated by the nth neural unit, increasing the value of n by 1, returning to the step of inputting the classification matrix into the neural network single-layer perceptron after the parameters are initialized, obtaining a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix;
if the second judgment result shows yes, the formula is used according to the weight matrix calculated by the nth neural unit
Figure GDA0003844833070000101
Determining a classification linear equation of the classification value relative to the height of the compensation surface; wherein Y is a dependent variable of the equation of the classification straight line, X is an independent variable of the equation of the classification straight line, K is a coefficient of the equation of the classification straight line, B is a constant of the equation of the classification straight line, and W is n(1) 、W n(2) 、W n(3) The weight values are respectively the first, the second and the third weight values in the weight matrix of the nth iteration nerve unit.
The method for obtaining the weight matrix calculated by the neural unit for the nth time and the initial classification value of each control point under each datum compensation surface elevation includes the following steps:
the neuron network single-layer perceptron after the input parameters of the classification matrix are initialized, based on the initial weight matrix and by using a formula
Figure GDA0003844833070000102
Calculating an initial classification value of each control point under each datum compensation surface elevation; wherein, y IJ For the initial classification value of the J-th reference compensation surface height Cheng Xiadi I control points, sgn () is a sign function, x i Is the I-th coordinate error value, omega, in the I-th control point i The weight of the ith coordinate error value of the ith control point in the initial weight matrix is obtained, N is the number of the coordinate error values in the ith control point, and b is a bias value;
according to the initial classification value of each control point under each reference compensation surface elevation, using the formula F = lr x (X.T.dot (E))/X.shape [0]]Calculating the bias weight F of each control point under each reference compensation surface elevation; wherein lr is the learning rate, X.T is the matrix after the transformation of the classification matrix, dot () is the matrix point multiplication function, shape [, [ alpha ] ]]For a row in a row column of the classification matrix, E = d I -y IJ ，d I The classification mark of the I control point, and E is the difference value between the classification mark of the I control point and the initial classification value;
and determining a matrix obtained by adding the bias weight of each control point under each reference compensation surface elevation and the initial weight matrix as the weight matrix calculated by the neural unit for the nth time.
S107, obtaining an error value fitting linear equation of the classification value relative to the height of the compensation surface by using a linear fitting method according to the classification matrix;
s108, acquiring an intersection point of the classification linear equation and the error value fitting linear equation, and judging whether a classification value in the intersection point is smaller than a classification value threshold value or not to obtain a first judgment result;
s109, if the first judgment result shows no, updating the preset step pitch, and returning to the step of determining height values of a plurality of reference compensation surfaces by the preset step pitch according to the height of the compensation surfaces of the independent coordinate system;
and S110, if the first judgment result shows that the height of the compensation surface in the intersection point is positive, outputting the height of the compensation surface as the optimal height of the compensation surface.
The method solves the problem that when an independent coordinate system is converted into a national standard coordinate system, an artificial neural network algorithm is utilized to analyze an optimal compensation surface through multiple iterations under the condition that a specific ellipsoid expansion algorithm is unknown, and the conversion result precision is guaranteed. The invention utilizes the powerful algorithm of the artificial neural network, can effectively analyze the conversion precision of all control points in the whole engineering area, has great advantage in efficiency, and has the characteristics of high efficiency and intellectualization compared with the artificial analysis. The invention can provide rapid and accurate conversion basis and parameters for converting the independent coordinate system to the national standard coordinate system in engineering application.
Example two
Referring to fig. 2, the specific scheme of the present invention is as follows:
step 101: and reading the three-dimensional coordinates of the control point of the independent coordinate system in the organized csv format through a data reading module. The three-dimensional coordinate values are an eastern coordinate value, a northern coordinate value and an elevation value.
Step 102: and reading the three-dimensional coordinates of the control points of the organized csv format national standard coordinate system through a data reading module.
The 101 and 102 steps are required to read that the roll names of the control points of the two coordinate systems are consistent and are required to be in one-to-one correspondence, namely, one independent coordinate system control point corresponds to one national standard coordinate system control point.
Step 103: and setting basic parameters of an independent coordinate system, such as adopted ellipsoid information, adopted central meridian values adopted by Gaussian-gram projection, adopted compensation surface elevation values and the like.
Step 104: and performing Gaussian projection back calculation on the independent coordinate system coordinate read in the step 101 according to the acquired independent coordinate system parameter, wherein the Gaussian projection back calculation is to use a Gaussian projection calculation method to perform projection calculation on the coordinate value of the projection plane rectangular coordinate system in the unit of meter to the geographic coordinate system in the unit of degree.
The ellipsoid expansion calculation method adopted in the Gaussian projection reverse calculation process is a basic calculation method for turning an independent coordinate system to a national standard coordinate system, the ellipsoid expansion algorithms are more, no clear specification is provided for which expansion algorithm is adopted in the design process of the independent coordinate system, and the algorithm only needs to ensure that the deformation error of the projection length per kilometer in the control range of the control point is less than 2.5 centimeters according to the relevant specifications; therefore, a general ellipsoid expansion algorithm is adopted for calculation in the invention.
Step 105: and calculating according to the result calculated in the step 104 by adopting a Gaussian-Kruger projection forward algorithm according to the national standard coordinate system parameters to obtain banded data which are consistent with the control point result of the national standard coordinate system obtained in the step 102. The Gaussian projection forward calculation is to project and calculate the coordinate value of the geographic coordinate system with the degree as a unit onto a projection plane rectangular coordinate system with the meter as a unit by adopting a Gaussian projection calculation method.
Step 106: and calculating error values of the results obtained in the step 105 and the results of the control points read in the step 102, wherein the error values in the east direction and the north direction are mainly calculated.
Specifically, the result obtained after the gaussian projection calculation is performed according to the provided independent coordinate parameters should be consistent with the control point result of the provided national standard coordinate system, but because different ellipsoid expansion algorithms are adopted for compensation surface elevation correction, the result obtained by adopting a general ellipsoid expansion algorithm in the invention has an error with the standard result, and in order to find the minimum value of the error, the step is basic data of subsequent calculation.
Step 107: calculating new compensation surface elevation values by taking the independent coordinate system compensation surface elevation provided in the step 106 as a reference and taking +/-10 meters as a step pitch, and calculating 10 groups of new compensation surface elevation values; and respectively calculating the errors of the coordinate values of the independent coordinate system converted under the 10 groups of compensation surfaces and the national standard coordinate system according to the 10 groups of new compensation surfaces.
Specifically, the calculated error should be preliminarily determined, the error should include a positive error and a negative error, and the errors are classified according to different compensation surfaces to form an initial calculation matrix, which is imported into step 108 as initial data of neuron calculation.
The classification is based on whether the calculated error is greater than 0 or less than 0, and a label matrix is used to identify the classification, wherein the label with the error value greater than 0 is 1, and the label with the error value less than 1 is-1.
Step 108: and setting initial parameters for neuron calculation according to the error result calculated in the step 107, wherein the parameters comprise a learning rate, an initial weight matrix and the like.
Step 109: through the single-layer neuron calculation parameters provided in step 108 and the initial data provided in step 107, a classification linear equation with an error close to 0 is obtained through single-layer neuron classification calculation, as shown in fig. 4, the abscissa represents different compensation surfaces, and the ordinate represents error values under different compensation surfaces.
Specifically, the method comprises the following steps: the artificial neural network is an algorithmic mathematical model simulating animal neural network behavior characteristics and performing distributed parallel information processing. The network achieves the purpose of processing information by adjusting the mutual connection relationship among a large number of internal nodes depending on the complexity of the system, and has self-learning and self-adapting capabilities. The invention adopts a neural network single-layer perceptron algorithm to classify and calculate the error classification.
The calculation method is as follows: the single-layer perceptron is a feature vector with an input of an example and an output of an example, wherein a binary classification model is similar to a logistic regression model, and the value of +1 or-1 is taken. It can also be said that a straight line is drawn on the plane coordinate axis, and the error points are divided into two types, one type is larger than zero and the other type is smaller than zero.
The single layer perceptron consists of a linear combiner and a binary threshold element. Each component of the input vector is multiplied by the weight value, then the components are superposed in the linear combiner to obtain a scalar result, and the output of the scalar result is that the linear combination result passes through a binary threshold function. The binary threshold element is usually a rising function, and typical functions are non-negative numbers mapped to positive 1 and negative numbers mapped to negative 1.
The input is an N-dimensional vector x = [ x ] 1 ,x 2 ,...,x n ]Each of which is x i Corresponding to a weight value w i The hidden layer outputs are superimposed into a scalar value:
Figure GDA0003844833070000131
subsequently, the obtained values are compared in a binary threshold element
Figure GDA0003844833070000132
Judging to generate binary output:
Figure GDA0003844833070000133
data can be divided into two categories. In practical application, a bias is added, the value is always 1, and the weight is b.
At this time, the y output is:
Figure GDA0003844833070000134
taking the bias value of b as a special weight, the hyperplane of the single-layer perceptron for pattern recognition is determined by the following formula:
Figure GDA0003844833070000135
the structure of the single layer sensor is shown in fig. 3.
The input vector may be represented as a point in a planar rectangular coordinate system. The classification hyperplane is a straight line:
Figure GDA0003844833070000141
this allows the points to be divided into two categories along a straight line.
Calculating input data into 3 multiplied by 910 arrays, wherein the symbol values are 1 for convenience of calculation, and the compensation surface values and the errors of the coordinates and the standard values of each control point of different compensation surfaces are as follows:
[1,1090,-7.765867612],[1,1100,-3.011333824],[1,1110,1.743185056],[1,1120,6.49768903],[1,1130,11.2521781],[……,……,……]。
and each corresponding data error is greater than 0 and marked as 1, and each data error smaller than 0 and marked as-1 are classified by 1 multiplied by 910 groups of data and marked as T.
Input an initial weight W 1 Is a random matrix of 3 × 1, the initial value is a value between 0 and 1 randomly generated by a computer, and the specific data is as follows: [[0.12860615],[0.08277231],[0.01879966]]。
Training input data through a single-layer neuron, wherein in the training process, each component of an input vector is multiplied by an initial weight value firstly and then subjected to symbol calculation, and the symbol calculation is subtracted from identification data T to obtain a value E under the initial weight value; and calculating an offset value, wherein the calculation formula is as follows: lr × (X.T.dot (E))/X.shape [0]; wherein lr is the learning rate, and X.T is the matrix after the input matrix is inverted.
After calculating the offset value, adding the offset value and the initial weight to obtain a new weight W calculated by the next sub-neural unit 2 And participating in the next bias calculation.
And comparing the data obtained after the symbol calculation with the provided standard classification data T after each calculation, judging whether all the data obtained after the symbol calculation are consistent with the provided comparison data, and stopping the calculation if the data are consistent. Obtaining the final weight value W n Last weight W n W in n(2) ÷W n(3) For classification hyperplane straight line equations: y = K value, W in KX + B n(1) ÷W n(3) Is the B value of the straight line equation.
The classification value of each compensation surface on the classification straight line can be calculated, and a straight line function and a straight line data list are formed.
Step 110: according to the values calculated in step 107, a straight line function can be used for fitting, a straight line equation is fitted by using a straight line fitting method in machine learning to obtain error values under different compensation surfaces, as shown in fig. 4, and a straight line parameter and a straight line data list are obtained.
Step 111: solving an intersection point according to the linear functions obtained in the steps 109 and 110, wherein the solved intersection point is the minimum value of the error under the compensation surface of the neuron analysis; the intersection of the straight lines is shown in FIG. 4, wherein the red line is a straight line obtained by 109-step neuron analysis values, and the green line is a straight line generated by 110-step fitting.
Judging by solving the intersection point value of two intersecting straight lines, solving the intersection point abscissa as a new compensation surface elevation, the ordinate as a classification intermediate value of neuron analysis, wherein the intermediate value is theoretically infinitely close to 0, taking the intermediate value smaller than 0.001 as a judgment standard according to experience in actual use, repeating the steps of 103-110 for reanalysis when the intermediate value is larger than 0.001, wherein initial data needs to be recalculated, the calculation principle is that the calculated compensation surface elevation is taken as a reference, the step pitch is taken as a step pitch value according to one tenth of the previous calculated value, and 10 groups of data under the new compensation surface elevation value are calculated as input values for analysis; until the solving error is less than the decision value.
Specifically, the new compensation surface is calculated and then error calculation is carried out on the new compensation surface and a standard coordinate value, the error value is smaller than the basic requirement of engineering application and is used as a basis for ending the neuron analysis, and the analysis result is recorded and checked.
For the current using technology and method, the method provided by the invention can efficiently and accurately acquire the specific parameters required by conversion. The parameters calculated by the invention can be converted to the standard coordinate system result from the independent coordinate system in engineering application with high precision, and the standard coordinate system result can also be converted to the independent coordinate system with high precision. The method has an intelligent analysis method, and can complete the analysis and judgment of the required result through the self-learning and self-adaptive characteristics of the neurons under the condition of not knowing a body ellipsoid expansion algorithm by depending on the artificial neuron calculation of the step 109 of the part 3.
Because the artificial neuron algorithm is adopted to carry out intelligent analysis on the measured data, the problem that only a small amount of control point precision changes can be analyzed manually in the prior art is solved. The manual analysis result cannot effectively represent the error change condition of the whole project area, and the precision in the project area cannot be effectively guaranteed. The method provided by the invention can be used for analyzing the error change conditions of the control points in all the project areas at one time, so that an optimal conversion parameter value is obtained. High-precision conversion of the measurement data is ensured.
EXAMPLE III
A certain project is a long-distance linear project; in order to ensure normal design and construction of a project, an independent coordinate system used for design and construction is arranged at the project survey design stage; the independent coordinate system is provided with ellipsoid parameters adopting a 2000 national geodetic coordinate system, the projection central meridian is 105 degrees and 55 minutes, and the height of a compensation surface is 1100 meters. Meanwhile, the survey unit provides the achievement of a lower coordinate system of the national standard sub-band: a2000 national geodetic coordinate system ellipsoid is adopted, and the standard 3-degree zone banding central meridian is 105 degrees. The two sets of achievements respectively have 92 control points and correspond to one another.
Respectively reading the information of 92 control points of the two sets of coordinate systems, wherein the information comprises three-dimensional coordinate values, namely an east coordinate value, a north coordinate value and an elevation value.
Performing projection calculation on the control points of the independent coordinate system through the acquired parameters of the independent coordinate system; in the calculation process, the new ellipsoid parameters are calculated by using the compensation surface elevation, and the new ellipsoid parameters are calculated according to a general expansion ellipsoid method.
And calculating an error value by converting a set of 92 control points into a national standard coordinate system result through the provided 1100 m compensation surface elevation calculation and comparing the result with the provided national standard coordinate system result.
And calculating error values of 92 independent coordinate system control points which are calculated under the compensation elevation surface of 1050-1150 meters and converted into a national standard coordinate system and provide correct control points under the coordinate system respectively according to steps of +/-10 meters and with the height of the compensation surface of 1100 meters as a reference.
And (3) forming an initial calculation matrix by using the step pitch and the error value, setting parameters such as an initial learning rate and a random weight matrix, and analyzing and classifying initial data by using a single-layer neuron. And calculating data dividing straight line parameters.
The data obtained by analyzing and calculating the data generated by the compensation surface of 1050-1150 meters is 1106.228, -0.0583, the first data is the height value of the compensation surface with the minimum error obtained by utilizing neuron analysis, and the second data is the intersection value of straight lines and is the error value on the straight lines after the neuron analysis; the units are all meters.
The error value of the intersected straight line is found to be 0.058 m, which can not meet the engineering application requirement after conversion; then, 1100 meters is taken as the reference of the compensation surface, and error values of 92 independent coordinate system control points calculated under the 1101-1111 meter compensation surface and controlled under the correct coordinate system are provided after the control points are converted into a national standard coordinate system are calculated respectively according to +/-1 meter; repeating the above analysis steps to obtain analysis data under the compensation of 1101-1111 meters as follows: 1106.204, -0.0667, the error is not obviously reduced but increased compared with the last learning, which shows that the analysis compensation surface can not meet the use requirement.
And combining the second analysis situation, calculating 92 independent coordinate system control points under 1103.5-1108.5 m compensation surface respectively according to +/-0.1 m as a reference, converting the 92 independent coordinate system control points into a national standard coordinate system, and performing neuron analysis calculation again with error values for providing control under a correct coordinate system to obtain data 1106.3482,0.0013, and extracting an error of 0.0013 on the intersecting straight line calculated for the third time to be used as the basic judgment for determining the compensation surface. 5363 and 1106.348 m can be used as a new compensation face value to participate in the subsequent engineering application data processing.
And converting 92 independent coordinate system control points into a standard coordinate system by using 1106.348 meters obtained by calculation and analysis as compensation surface parameters of the independent coordinate system. By comparing the converted control point result with the provided standard coordinate system control point result, the medium error is: 0.011 meters, maximum error of 0.019 meters, and minimum error of-0.020 meters. The compensation surface calculated by using neuron analysis is used for comparing the result of converting the independent coordinate system into the standard coordinate system with the accurate result, the error meets the requirement of converting the data result of the engineering project, and 1106.348 can be used as a new compensation surface for data projection conversion.
Under the condition of an expansion ellipsoid algorithm of an unknown projection compensation surface, the invention adopts a neural network algorithm to perform machine learning analysis on the compensation surface elevation adopted by an independent coordinate system and find out the compensation surface elevation value with the precision meeting the related design requirement so as to meet the requirement of engineering application.
Example four
The invention provides an independent coordinate system parameter analysis system based on a neural network algorithm aiming at the principle shown in FIG. 2.
The setting module 201 and the control point reading module are used for reading the control point results of the independent coordinate system and the standard coordinate system, and ensuring the correctness of the initial data.
The Gaussian-gram-Luge projection forward and backward calculation module 202 is used for projecting independent coordinate system results to a standard coordinate system, basically, the independent coordinate system is subjected to Gaussian-gram projection backward calculation according to related parameters, a geographic coordinate system is calculated by projection of a projection plane rectangular coordinate system, and the coordinate value unit is calculated as a degree unit by projection of a meter unit; and the geographic coordinates are projected to calculate the plane rectangular coordinates according to the parameters of the national standard coordinate system by using a Gaussian-Criger projection forward calculation method.
And the ellipsoid expansion algorithm module 203 is used for calculating a new ellipsoid in which a compensation surface elevation parameter participates when the independent coordinate system is transformed into a standard coordinate system in a projection manner.
The neuron analysis data generation module 204 adopts neurons to analyze and calculate projection data under different compensation surfaces; before using the neuron calculation, basic data for the neuron calculation including error values of respective control points under different compensation planes, initial calculation random weight, learning rate, and the like need to be prepared.
The module is mainly used for organizing and calculating the error calculation and data organization of the accurate value after the independent coordinate system achievements under different compensation surfaces are projected to the standard coordinate system achievements.
The single-layer neuron analysis module 205 adopts a single-layer neuron to analyze and classify the conversion data, classifies the provided error values through parameters such as the learning rate and the initial weight set by the module 204, and calculates linear equation coefficients of linear segmentation so as to search a compensation surface height value with the error close to 0 under the current provided data.
The error straight line fitting and straight line intersection module 206 is used for performing straight line fitting calculation on the initial error data generated by the module 204 to calculate a straight line equation, wherein the fitted straight line equation accords with the distribution trend of the error data; and (4) performing line-line intersection solving by using the fitted straight line and the segmentation straight line analyzed by the neuron, wherein the lower value is the optimal compensation surface height value and the segmentation error value under the error data.
And the result data determination and analysis iteration module 207 is used for analyzing and determining the result calculated by the module 206, determining whether the segmentation error value is close to 0, calculating by using all the control points of the independent coordinate system of the new compensation surface height Cheng Zhidui after analysis, and performing error calculation with a standard result, wherein the calculated middle error is larger than the related technical requirement, and performing iterative calculation by adopting a one-tenth reduced step pitch repetition module 202-206 according to the newly acquired compensation surface height value as the standard until the error meets the related technical requirement.
EXAMPLE five
The invention provides an independent coordinate system parameter analysis system based on a neural network algorithm aiming at a flow chart shown in figure 1, and the system comprises:
the three-dimensional coordinate acquisition module is used for acquiring three-dimensional coordinates of the control points in an independent coordinate system and three-dimensional coordinates in a national coordinate system respectively; the control points are points which are distributed in a measurement area for measurement operation in the engineering construction project;
the reference compensation surface elevation determination module is used for determining a plurality of reference compensation surface elevations at preset step distances according to the compensation surface elevations of the independent coordinate system;
the projection three-dimensional coordinate conversion module is used for converting three-dimensional coordinates of the control points in the independent coordinate system into projection three-dimensional coordinates in a national coordinate system by using different reference compensation surface height values through a Gaussian projection algorithm;
the coordinate error value obtaining module is used for calculating the error value of the projection three-dimensional coordinates of the control points under each reference compensation surface elevation and the three-dimensional coordinates of the control points in the national coordinate system to obtain the coordinate error value of each control point under each reference compensation surface elevation;
the classification matrix forming module is used for performing classification identification on each control point according to the coordinate error value of each control point and forming a classification matrix by combining the classification identification, the reference compensation surface elevation and the coordinate error value of each control point;
the classification linear equation obtaining module is used for carrying out error classification calculation by utilizing a neural network single-layer perceptron algorithm according to the classification matrix to obtain a classification linear equation of a classification value about the height of the compensation surface;
the error value fitting linear equation obtaining module is used for obtaining an error value fitting linear equation of the classification value relative to the height of the compensation surface by utilizing a linear fitting method according to the classification matrix;
the first judgment result obtaining module is used for obtaining an intersection point of the classification linear equation and the error value fitting linear equation, judging whether a classification value in the intersection point is smaller than a classification value threshold value or not and obtaining a first judgment result;
a step returning module, configured to update the preset step pitch if the first determination result indicates no, and return to the step of determining height values of the multiple reference compensation surfaces at the preset step pitch according to the height of the compensation surfaces of the independent coordinate system;
and the optimal compensation surface elevation output module is used for outputting the compensation surface elevation in the intersection point as the optimal compensation surface elevation if the first judgment result shows that the intersection point is positive.
The projection three-dimensional coordinate conversion module specifically comprises:
the basic parameter initialization submodule is used for initializing basic parameters of an independent coordinate system;
the longitude and latitude coordinate obtaining submodule is used for respectively carrying out Gaussian projection back calculation on the three-dimensional coordinates of the control points in the independent coordinate system under the elevations of the reference compensation surfaces according to the basic parameters of the independent coordinate system to obtain the longitude and latitude coordinates of the control points under the elevation of each reference compensation surface;
and the projection three-dimensional coordinate determination submodule is used for calculating the longitude and latitude coordinates of each control point under each datum compensation surface elevation to be projected to the three-dimensional coordinates in the national coordinate system according to the national coordinate system parameters by utilizing Gaussian-Kruger projection forward calculation, and taking the longitude and latitude coordinates of each control point under each datum compensation surface elevation as the projection three-dimensional coordinates of each control point under each datum compensation surface elevation in the national coordinate system.
The classification matrix forming module specifically comprises:
the classification identification submodule is used for classifying and identifying the control points with the coordinate error value larger than 0 as 1 and classifying and identifying the control points with the coordinate error value smaller than 0 as-1; the coordinate error value is not equal to 0.
The classification linear equation obtaining module specifically comprises:
the initialization submodule is used for initializing the parameters of the neuron network single-layer perceptron, wherein the iteration number n is 1; the parameters of the neuron network single-layer perceptron comprise an initial weight matrix;
the initial classification value obtaining submodule is used for inputting the initialized parameters of the classification matrix into the neuron network single-layer perceptron, and obtaining a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix;
a second judgment result obtaining sub-module, configured to judge whether the initial classification value of each control point at each reference compensation surface elevation calculated by the nth neural unit is the same as the classification identifier of each control point, and obtain a second judgment result;
an updating submodule, configured to update the initial weight matrix to a weight matrix calculated by the nth neural unit if the second determination result indicates no, increase the value of n by 1, and return to the step "obtain the weight matrix calculated by the nth neural unit and the initial classification value of each control point at each reference compensation surface elevation based on the initial weight matrix by inputting the classification matrix into the neural network single-layer perceptron after the parameters are initialized";
a classification linear equation determination submodule for utilizing the formula according to the weight matrix calculated by the nth neural unit if the second judgment result indicates yes
Figure GDA0003844833070000201
Determining a classification linear equation of the classification value relative to the height of the compensation surface; wherein Y is a dependent variable of the classification linear equation and X is a dependent variable of the classification linear equationVariables, K being the coefficients of the equation of the classification straight line, B being the constants of the equation of the classification straight line, W n(1) 、W n(2) 、W n(3) The first, second and third weight values in the weight matrix of the nth iteration neural unit are respectively.
The initial classification value obtaining submodule specifically includes:
an initial classification value calculation unit for initializing the input parameters of the classification matrix of the neuron network single-layer perceptron, based on the initial weight matrix, by using a formula
Figure GDA0003844833070000202
Calculating an initial classification value of each control point under each datum compensation surface elevation; wherein, y IJ For the initial classification value of the J-th reference compensation surface height Cheng Xiadi I control points, sgn () is a sign function, x i Is the I-th coordinate error value, omega, in the I-th control point i The weight of the ith coordinate error value of the ith control point in the initial weight matrix is obtained, N is the number of the coordinate error values in the ith control point, and b is a bias value;
an offset weight calculation unit, configured to use a formula F = lr × (x.t.dot (E))/x.shape [0] according to the initial classification value of each control point at each reference compensation surface elevation]Calculating the bias weight F of each control point under each reference compensation surface elevation; wherein lr is the learning rate, X.T is the matrix after the transformation of the classification matrix, dot () is the matrix point multiplication function, shape [, [ alpha ] ]]For a row in the row and column of the classification matrix, E = d I -y IJ ，d I The classification mark of the I-th control point, and E is the difference value between the classification mark of the I-th control point and the initial classification value;
and the weight matrix determining unit is used for determining a matrix obtained by adding the bias weight of each control point under each reference compensation surface elevation and the initial weight matrix as the weight matrix calculated by the neural unit for the nth time.
The embodiments in the present description are described in a progressive manner, each embodiment focuses on differences from other embodiments, and the same and similar parts among the embodiments are referred to each other. For the system disclosed by the embodiment, the description is relatively simple because the system corresponds to the method disclosed by the embodiment, and the relevant points can be referred to the method part for description.
The principles and embodiments of the present invention have been described herein using specific examples, which are provided only to help understand the method and the core concept of the present invention; meanwhile, for a person skilled in the art, according to the idea of the present invention, the specific embodiments and the application range may be changed. In view of the above, the present disclosure should not be construed as limiting the invention.

Claims (6)

1. An independent coordinate system parameter analysis method based on a neural network algorithm is characterized by comprising the following steps:
acquiring three-dimensional coordinates of a plurality of control points in an independent coordinate system and three-dimensional coordinates in a national coordinate system respectively; the control points are points which are distributed in a measurement area for measurement operation in the engineering construction project;
determining a plurality of reference compensation surface elevations at preset steps according to the compensation surface elevations of the independent coordinate system;
converting three-dimensional coordinates of the control points in the independent coordinate system into projected three-dimensional coordinates in a national coordinate system by using different height values of the reference compensation surface by using a Gaussian projection algorithm;
calculating error values of the projected three-dimensional coordinates of the control points under each reference compensation surface elevation and the three-dimensional coordinates of the control points in the national coordinate system to obtain the coordinate error value of each control point under each reference compensation surface elevation;
classifying and identifying each control point according to the coordinate error value of each control point, and combining the classification identification of each control point, the elevation of the reference compensation surface and the coordinate error value to form a classification matrix;
according to the classification matrix, error classification calculation is carried out by utilizing a neural network single-layer perceptron algorithm, and a classification linear equation of a classification value about the elevation of the compensation surface is obtained; the method specifically comprises the following steps: number of initialization iterations n is 1, and neuralParameters of a meta-network single-layer perceptron; the parameters of the neuron network single-layer perceptron comprise an initial weight matrix; the neural network single-layer perceptron after the input parameters of the classification matrix are initialized obtains a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix; judging whether the initial classification value of each control point under each reference compensation surface elevation calculated by the nth neural unit is the same as the classification identification of each control point or not, and obtaining a second judgment result; if the second judgment result shows no, updating the initial weight matrix to the weight matrix calculated by the nth neural unit, increasing the value of n by 1, returning to the step of inputting the classification matrix into the neural network single-layer perceptron after the parameters are initialized, obtaining a weight matrix calculated by the nth neural unit and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix; if the second judgment result shows yes, the formula is used according to the weight matrix calculated by the nth neural unit
Figure FDA0003844833060000011
Determining a classification linear equation of the classification value about the compensation surface elevation; wherein Y is a dependent variable of the classification linear equation, X is an independent variable of the classification linear equation, K is a coefficient of the classification linear equation, B is a constant of the classification linear equation, W is a constant of the classification linear equation n(1) 、W n(2) 、W n(3) The weight values are respectively the first weight value, the second weight value and the third weight value in the weight matrix of the nth iteration neural unit; the method for obtaining the weight matrix calculated by the neural unit for the nth time and the initial classification value of each control point under each datum compensation surface elevation includes the following steps: the neuron network single-layer perceptron after the input parameters of the classification matrix are initialized, and a formula is utilized based on the initial weight matrix
Figure FDA0003844833060000021
Each calculationThe initial classification value of each control point under each datum compensation surface elevation; wherein, y IJ For the initial classification value of Cheng Xiadi I control points of the J-th reference compensation level height, sgn () is a sign function, x i For the ith coordinate error value, ω, in the ith control point i The weight of the ith coordinate error value of the ith control point in the initial weight matrix is obtained, N is the number of the coordinate error values in the ith control point, and b is a bias value; according to the initial classification value of each control point under each reference compensation surface elevation, using the formula F = lr x (X.T.dot (E))/X.shape [0]]Calculating the bias weight F of each control point under each reference compensation surface elevation; wherein lr is the learning rate, X.T is the matrix after the transformation of the classification matrix, dot () is the matrix point multiplication function, shape [, [ alpha ] ]]For a row in the row and column of the classification matrix, E = d I -y IJ ，d I The classification mark of the I control point, and E is the difference value between the classification mark of the I control point and the initial classification value; determining a matrix obtained by adding the bias weight of each control point under each datum compensation surface elevation and the initial weight matrix as a weight matrix calculated by the neural unit for the nth time;
according to the classification matrix, obtaining an error value fitting linear equation of the classification value relative to the height of the compensation surface by using a linear fitting method;
acquiring an intersection point of the classification linear equation and the error value fitting linear equation, and judging whether a classification value in the intersection point is smaller than a classification value threshold value or not to obtain a first judgment result;
if the first judgment result shows that the reference coordinate system is not the reference coordinate system, updating a preset step pitch, and returning to the step of determining a plurality of reference compensation surface elevations by the preset step pitch according to the compensation surface elevations of the independent coordinate system;
and if the first judgment result shows that the height of the compensation surface in the intersection point is positive, outputting the height of the compensation surface as the optimal height of the compensation surface.
2. The method for analyzing parameters of independent coordinate system based on neural network algorithm according to claim 1, wherein the transforming the three-dimensional coordinates of the plurality of control points in the independent coordinate system into projected three-dimensional coordinates in the national coordinate system by using different height values of the reference compensation surface by using the gaussian projection algorithm specifically comprises:
initializing basic parameters of an independent coordinate system;
according to basic parameters of the independent coordinate system, performing Gaussian projection back calculation on three-dimensional coordinates of a plurality of control points in the independent coordinate system under the elevation of a plurality of reference compensation surfaces respectively to obtain longitude and latitude coordinates of the plurality of control points under the elevation of each reference compensation surface;
and calculating the longitude and latitude coordinates of the control points under each datum compensation surface elevation according to the national coordinate system parameters by utilizing Gaussian-Kluger projection forward calculation, and projecting the longitude and latitude coordinates of each control point under each datum compensation surface elevation to a three-dimensional coordinate in a national coordinate system to serve as the projected three-dimensional coordinate of each control point under each datum compensation surface elevation in the national coordinate system.
3. The method for analyzing the parameters of the independent coordinate system based on the neural network algorithm of claim 1, wherein the classifying and identifying each control point according to the coordinate error value of each control point specifically comprises:
classifying and identifying the control points with the coordinate error value larger than 0 as 1, and classifying and identifying the control points with the coordinate error value smaller than 0 as-1; the coordinate error value is not equal to 0.
4. An independent coordinate system parameter analysis system based on neural network algorithm, the system comprising:
the three-dimensional coordinate acquisition module is used for acquiring three-dimensional coordinates of the control points in an independent coordinate system and three-dimensional coordinates in a national coordinate system respectively; the control points are points which are distributed in a measurement area for measurement operation in the engineering construction project;
the reference compensation surface elevation determination module is used for determining a plurality of reference compensation surface elevations at preset step distances according to the compensation surface elevations of the independent coordinate system;
the projection three-dimensional coordinate conversion module is used for converting three-dimensional coordinates of the control points in the independent coordinate system into projection three-dimensional coordinates in a national coordinate system by using different reference compensation surface height values through a Gaussian projection algorithm;
the coordinate error value obtaining module is used for calculating the error value of the projection three-dimensional coordinates of the control points under each reference compensation surface elevation and the three-dimensional coordinates of the control points in the national coordinate system to obtain the coordinate error value of each control point under each reference compensation surface elevation;
the classification matrix forming module is used for performing classification identification on each control point according to the coordinate error value of each control point and forming a classification matrix by combining the classification identification, the reference compensation surface elevation and the coordinate error value of each control point;
the classification linear equation obtaining module is used for performing error classification calculation by utilizing a neural network single-layer perceptron algorithm according to the classification matrix to obtain a classification linear equation of a classification value about the elevation of the compensation surface; the method specifically comprises the following steps: the initialization submodule is used for initializing the parameters of the neuron network single-layer perceptron, wherein the iteration number n is 1; the parameters of the neuron network single-layer perceptron comprise an initial weight matrix; the initial classification value obtaining submodule is used for inputting the classification matrix into the neural network single-layer perceptron after the parameters are initialized, and obtaining a weight matrix calculated by the neural unit for the nth time and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix; the second judgment result obtaining submodule is used for judging whether the initial classification value of each control point under each reference compensation surface elevation calculated by the nth neural unit is the same as the classification identification of each control point or not so as to obtain a second judgment result; an updating submodule, configured to, if the second determination result indicates no, the initial weight matrix is updated to the weight matrix calculated by the nth neural unit, and the value of n is increased by 1, returning to the step of obtaining a weight matrix calculated by the neural unit for the nth time and an initial classification value of each control point under each datum compensation surface elevation based on the initial weight matrix by the neuron network single-layer perceptron after the input parameters of the classification matrix are initialized; a classification linear equation determination submodule for determining the nth order of the nerves if the second judgment result indicates yesWeight matrix of cell calculation using formula
Figure FDA0003844833060000041
Determining a classification linear equation of the classification value relative to the height of the compensation surface; wherein Y is a dependent variable of the classification linear equation, X is an independent variable of the classification linear equation, K is a coefficient of the classification linear equation, B is a constant of the classification linear equation, W is a constant of the classification linear equation n(1) 、W n(2) 、W n(3) The weight values are respectively the first weight value, the second weight value and the third weight value in the weight matrix of the nth iteration neural unit; wherein, the initial classification value obtaining submodule specifically includes: an initial classification value calculation unit for initializing the input parameters of the classification matrix to the neuron network single-layer perceptron, and based on the initial weight matrix, using a formula
Figure FDA0003844833060000042
Calculating an initial classification value of each control point under each datum compensation surface elevation; wherein, y IJ For the initial classification value of the J-th reference compensation surface height Cheng Xiadi I control points, sgn () is a sign function, x i Is the I-th coordinate error value, omega, in the I-th control point i The weight of the ith coordinate error value of the ith control point in the initial weight matrix is obtained, N is the number of the coordinate error values in the ith control point, and b is a bias value; an offset weight calculation unit, configured to use a formula F = lr × (x.t.dot (E))/x.shape [0] according to the initial classification value of each control point at each reference compensation surface elevation]Calculating the bias weight F of each control point under each reference compensation surface elevation; wherein lr is the learning rate, X.T is the matrix after the transformation of the classification matrix, dot () is the matrix point multiplication function, shape [ 2 ]]For a row in the row and column of the classification matrix, E = d I -y IJ ，d I The classification mark of the I control point, and E is the difference value between the classification mark of the I control point and the initial classification value; the weight matrix determining unit is used for determining a matrix obtained by adding the bias weight of each control point under each datum compensation surface elevation and the initial weight matrix as a weight matrix calculated by the neural unit for the nth time;
an error value fitting linear equation obtaining module, configured to obtain an error value fitting linear equation of the classification value with respect to the height of the compensation surface by using a linear fitting method according to the classification matrix;
the first judgment result obtaining module is used for obtaining an intersection point of the classification linear equation and the error value fitting linear equation, judging whether a classification value in the intersection point is smaller than a classification value threshold value or not and obtaining a first judgment result;
a step returning module, configured to update a preset step pitch if the first determination result indicates no, and return to the step of "determining multiple reference compensation surface elevations at the preset step pitch according to the compensation surface elevations of the independent coordinate system";
and the optimal compensation surface elevation output module is used for outputting the compensation surface elevation in the intersection point as the optimal compensation surface elevation if the first judgment result shows that the elevation is positive.
5. The system for analyzing independent coordinate system parameters based on neural network algorithm according to claim 4, wherein the projected three-dimensional coordinate transformation module specifically comprises:
the basic parameter initialization submodule is used for initializing basic parameters of an independent coordinate system;
the longitude and latitude coordinate obtaining submodule is used for respectively carrying out Gaussian projection back calculation on the three-dimensional coordinates of the control points in the independent coordinate system under the elevations of the reference compensation surfaces according to the basic parameters of the independent coordinate system to obtain the longitude and latitude coordinates of the control points under the elevation of each reference compensation surface;
and the projection three-dimensional coordinate determination submodule is used for calculating the longitude and latitude coordinates of each control point under each datum compensation surface elevation to be projected to the three-dimensional coordinates in the national coordinate system according to the national coordinate system parameters by utilizing Gaussian-Kruger projection forward calculation, and taking the longitude and latitude coordinates of each control point under each datum compensation surface elevation as the projection three-dimensional coordinates of each control point under each datum compensation surface elevation in the national coordinate system.
6. The system of claim 4, wherein the classification matrix constitutes a module, specifically comprising:
the classification identification submodule is used for classifying and identifying the control points with the coordinate error value larger than 0 as 1 and classifying and identifying the control points with the coordinate error value smaller than 0 as-1; the coordinate error value is not equal to 0.
CN202110785374.9A 2021-07-12 2021-07-12 Neural network algorithm-based independent coordinate system parameter analysis method and system Active CN113516766B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110785374.9A CN113516766B (en) 2021-07-12 2021-07-12 Neural network algorithm-based independent coordinate system parameter analysis method and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110785374.9A CN113516766B (en) 2021-07-12 2021-07-12 Neural network algorithm-based independent coordinate system parameter analysis method and system

Publications (2)

Publication Number Publication Date
CN113516766A CN113516766A (en) 2021-10-19
CN113516766B true CN113516766B (en) 2022-10-18

Family

ID=78067215

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110785374.9A Active CN113516766B (en) 2021-07-12 2021-07-12 Neural network algorithm-based independent coordinate system parameter analysis method and system

Country Status (1)

Country Link
CN (1) CN113516766B (en)

Families Citing this family (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN114054302B (en) * 2022-01-17 2022-03-25 南通西屋智能科技有限公司 Accurate positioning control method and system of dispenser

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107167119A (en) * 2017-03-29 2017-09-15 中交烟台环保疏浚有限公司 The data processing method of distortion of projection
CN107908808A (en) * 2017-08-11 2018-04-13 山东交通学院 3 ° of points of band coordinate transformation systems based on Compensating level surface or Mean height plane

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110220505A (en) * 2019-06-11 2019-09-10 江苏海事职业技术学院 Control suitable for multiple sensors measures target and space coordinate method for registering

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN107167119A (en) * 2017-03-29 2017-09-15 中交烟台环保疏浚有限公司 The data processing method of distortion of projection
CN107908808A (en) * 2017-08-11 2018-04-13 山东交通学院 3 ° of points of band coordinate transformation systems based on Compensating level surface or Mean height plane

Non-Patent Citations (3)

* Cited by examiner, † Cited by third party
Title
Joint Alignment of Multiple Generalized Point Sets with Anisotropic Positional Uncertainty Based on Expectation Maximization;Zhe Min等;《2018 International Conference on 3D Vision (3DV)》;20181014;170-179 *
农村地籍测量中抵偿高程面独立坐标系建立和转换;胡著翱等;《科技创新与应用》;20200408(第10期);79-83 *
线路工程测量坐标系的建立方法及精度分析;王文彬;《中国优秀硕士学位论文全文数据库 (基础科学辑)》;20120115(第1期);A008-73 *

Also Published As

Publication number Publication date
CN113516766A (en) 2021-10-19

Similar Documents

Publication Publication Date Title
Zhang et al. Short-term rainfall forecasting using multi-layer perceptron
CN110132263B (en) Star map identification method based on representation learning
CN109829399A (en) A kind of vehicle mounted road scene point cloud automatic classification method based on deep learning
CN107679556A (en) The zero sample image sorting technique based on variation autocoder
CN112633350A (en) Multi-scale point cloud classification implementation method based on graph convolution
CN111310965A (en) Aircraft track prediction method based on LSTM network
CN110619432A (en) Feature extraction hydrological forecasting method based on deep learning
CN114332578A (en) Image anomaly detection model training method, image anomaly detection method and device
CN110866934A (en) Normative coding-based complex point cloud segmentation method and system
CN104504709A (en) Feature ball based classifying method of three-dimensional point-cloud data of outdoor scene
CN115510174A (en) Road network pixelation-based Wasserstein generation countermeasure flow data interpolation method
CN114444378A (en) Short-term power prediction method for regional wind power cluster
CN109034232B (en) Automatic output system and control method for urban planning condition verification result report
CN113516766B (en) Neural network algorithm-based independent coordinate system parameter analysis method and system
CN116518960B (en) Road network updating method, device, electronic equipment and storage medium
CN110826244A (en) Conjugate gradient cellular automata method for simulating influence of rail transit on urban growth
CN113591997B (en) Assembly feature graph connection relation classification method based on graph learning convolutional neural network
CN114169502A (en) Rainfall prediction method and device based on neural network and computer equipment
CN109977131A (en) A kind of house type matching system
CN109116300A (en) A kind of limit learning position method based on non-abundant finger print information
CN111353525A (en) Modeling and missing value filling method for unbalanced incomplete data set
CN115601745A (en) Multi-view three-dimensional object identification method facing application end
CN113903016B (en) Bifurcation point detection method, bifurcation point detection device, computer equipment and storage medium
CN115359197A (en) Geological curved surface reconstruction method based on spatial autocorrelation neural network
CN114221334A (en) Rapid state estimation method based on graph neural network

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant