Construction progress management system based on BIM data and application method thereof

Abstract

The invention discloses a BIM data-based construction progress management system and an application method thereof, and the system comprises a BIM data collection module which is used for collecting corresponding noise data in a BIM construction process and transmitting the corresponding noise data to a plurality of networks for training; the construction data management module is used for carrying out data separation iteration management operation in the trained network; and the management optimization module is used for performing algorithm optimization on the separated and iterated data and outputting the data to the BIM cloud network.

Description

Technical Field

[0001] This invention relates to the field of vehicle energy balance control, and in particular to a construction progress management system based on BIM data and its application method. Background Technology

[0002] In BIM construction management, blind source signal management can be used to detect and analyze various noise and interference signals at the construction site to ensure safety and quality during construction. For example, blind source signal management technology can be used in BIM construction management to detect and analyze factors such as equipment noise, electromagnetic interference, and vibration at the construction site to ensure the safe and stable operation of construction equipment, thereby improving construction efficiency and optimizing project schedule. This necessitates that those skilled in the art solve the corresponding technical problems. Summary of the Invention

[0003] This invention aims to improve the construction efficiency and quality of building projects, and innovatively proposes a construction progress management system based on BIM data and its application method.

[0004] To achieve the above-mentioned objectives of the present invention, the present invention provides a construction progress management system based on BIM data, comprising:

[0005] The BIM data acquisition module is used to collect relevant noise data during BIM construction and transmit the relevant noise data to a complex network for training.

[0006] The construction data management module is used to perform data separation and iterative management operations in the trained network.

[0007] The management optimization module is used to perform algorithmic optimization on the separated iterative data and output it to the BIM cloud network.

[0008] This invention also discloses an application method for a construction progress management system based on BIM data, comprising the following steps:

[0009] S1: During BIM construction, each frequency point is estimated using the input time-frequency data in a complex network;

[0010] S2: Integrate the trained network framework as a module into the Fast-ICA algorithm;

[0011] S3: A two-channel complex neural network based on the time-frequency domain, where the weights are parameterized by a full-dimensional unitary matrix:

[0012] S4: Separate iterative calculations are performed by the Fast-ICA algorithm module;

[0013] S5: Match network output and target label to perform mapping between real and complex numbers;

[0014] S6: The Fast-ICA algorithm is optimized using artificial bee colony optimization and then transmitted to the BIM construction terminal.

[0015] In a preferred embodiment of the above technical solution, S1 includes:

[0016] During the training phase, the complex network directly estimates each frequency point using the input time-frequency data, extracts the complex form of the TF bin of the input data block, obtains the initial separation matrix W0(f) through the CgRNN separation module, initializes the Fast-ICA separation matrix using W0(f), and obtains the final separation matrix W(k). A minimum distortion principle smoothing module is added, and the smoothed output is transformed by inverse Fourier transform to obtain the numerical filter form of the separation matrix. The input signal X(t) is passed through an FIR filter to obtain the output signal Y(t).

[0017] In a preferred embodiment of the above technical solution, step S2 includes:

[0018] Fast-ICA processes data as a block of data consisting of the frequency harmonics of a mixed signal. The data block x(k,t) represents the complex value at frequency k and time t of the mixed signal. The data block is composed of the complex spectrum expansion of b frames of the mixed signal. Each frequency point corresponds to a separation matrix W(k). Each data block contains k separation matrices. W(k) is the learned label of the complex network. The trained network framework is then incorporated into the Fast-ICA algorithm as a module.

[0019] In a preferred embodiment of the above technical solution, step S3 includes:

[0020] Based on the time-frequency domain, a dual-channel complex neural network (CgRNN) is introduced. CgRNN introduces a novel complex RNN unit for blind source separation of complex signals. The weights are parameterized by a full-dimensional unitary matrix. This complex RNN network outperforms real-valued networks with the same number of parameters in blind separation tasks. The CgRNN network estimates each complex element in the separation matrix based on the input complex spectrum information, and uses the trained model to calculate the separation matrix of the output signal.

[0021] In a preferred embodiment of the above technical solution, step S4 includes:

[0022] The Fast-ICA algorithm module performs separation and iterative calculations. During the training phase, the mixed signal is subjected to a short-time Fourier transform to obtain the complex form of the TF bin features, which are then used as X input to the network for training. The estimated value Wo(f) of the network output is calculated, and the separation matrix W(k) of the true source is calculated. The mean squared error loss function is optimized using the PIT method, and finally the separation matrix W(k) is obtained. During the testing phase, the TF bin of the test signal is input into the trained CgRNN model to obtain the estimated value Wo(f).

[0023] In a preferred embodiment of the above technical solution, step S5 includes:

[0024] Match the network output and target label, perform a mapping between real and complex numbers, using the real part of the original complex number as the real part, and setting its imaginary part to 0, i.e.:

[0025] z = x + i·0

[0026] In the complex state, the complex number is mapped to the real state by using a linear combination of the real and imaginary parts of the complex number, with Wo and bo as the weight and bias factor, respectively, where R(z) is the real part of the complex number and S(z) is the imaginary part of the complex number.

[0027] o r =Wo[R(z)·S(z)]+bo.

[0028] In a preferred embodiment of the above technical solution, step S6 includes:

[0029] The mixed signal x(k,t) is input into the complex network using a method based on PIT and CgRNN to initialize the blind separation matrix. The MSE loss function is optimized using the PIT principle to reduce the error between the estimated value and the label W(k). Finally, the estimated complex matrix is ​​used to initialize the separation matrix in the algorithm.

[0030] In a preferred embodiment of the above technical solution, step S6 further includes:

[0031] The Fast-ICA algorithm is optimized using artificial bee colony optimization. The maximum number of iterations kma is initialized, and the number of iterations that do not yield a better solution is allowed k_limit. The EB optimization stage is simulated, and the best possible solution θbest so far is selected from θ. The OB optimization stage and the SB optimization stage are simulated, and an estimation method based on cross-validation is applied.

[0032] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0033] Improved construction efficiency and quality: The collection and management of BIM data enables effective planning and monitoring of construction progress, thereby improving construction efficiency and quality. Reduced costs: Effective management of construction progress reduces losses caused by project changes and delays, thus lowering costs. Improved safety: Effective management of construction progress avoids safety accidents caused by construction delays, thereby improving safety. Enhanced decision-making capabilities: Data analysis and processing of construction progress improves the decision-making capabilities of managers, thereby better guiding project construction.

[0034] In BIM construction management, blind source signal management is an effective technical means that can be used to detect and analyze various noise and interference signals at the construction site to ensure safety and quality during the construction process.

[0035] In BIM-based construction, the Fast-ICA algorithm is used. Based on higher-order statistics of the signal, it uses statistical information above the second order to correctly separate and recover the source signal. It decomposes these independent components from the observation signal obtained by multi-channel measurement, which is a linear combination of several independent signal sources. It has good stability, fast convergence speed, and can batch process data.

[0036] 2. Bee colony optimization algorithm is used for management optimization in construction progress monitoring. The artificial bee colony algorithm is improved and can quickly converge to the region where the optimal solution is located in the initial stage. The absolute value of the signal kurtosis is used as the objective function. The initial separation matrix is ​​optimized by the improved artificial bee colony algorithm. Finally, the NGA algorithm is used for signal separation. It uses fewer control parameters and has high convergence accuracy.

[0037] This invention employs the Fast-ICA algorithm in BIM construction progress management. Based on higher-order statistics of the signal, it performs independent component analysis, exhibiting good stability and fast convergence speed. An improved version of the Fast-ICA algorithm is implemented using an artificial bee colony optimization algorithm to search for global and local optimal solutions, increasing the probability of finding the optimal solution. It uses fewer control parameters, demonstrating strong robustness and high convergence accuracy, significantly improving signal separation efficiency and speed, and resulting in a marked improvement in separation performance. Furthermore, this scheme provides stable information output during BIM construction progress management.

[0038] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0039] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the description of the embodiments taken in conjunction with the following drawings, in which:

[0040] Figure 1 This is a schematic diagram of the overall invention.

[0041] Figure 2 This is a schematic diagram of the process of this invention. Detailed Implementation

[0042] Embodiments of the present invention are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as limiting the present invention.

[0043] like Figure 1 and 2 As shown, Example 1

[0044] Reference Figure 1 This invention discloses a construction progress management system based on BIM data and its application method, including:

[0045] The BIM data acquisition module is used to collect relevant noise data during BIM construction and transmit the relevant noise data to a complex network for training.

[0046] The construction data management module is used to perform data separation and iterative management operations in the trained network.

[0047] The management optimization module is used to perform algorithmic optimization on the separated iterative data and output it to the BIM cloud network.

[0048] Since the stability and accuracy of network data are crucial in BIM construction, appropriate optimization algorithms need to be adopted during the data separation and iteration process. Ensuring the stability of transmitted data has a significant effect on reducing signal noise in construction projects.

[0049] The application method of this system includes the following steps:

[0050] S1: During BIM construction, each frequency point is estimated using the input time-frequency data in a complex network;

[0051] S2: Integrate the trained network framework as a module into the Fast-ICA algorithm;

[0052] S3: A two-channel complex neural network based on the time-frequency domain, where the weights are parameterized by a full-dimensional unitary matrix:

[0053] S4: Separate iterative calculations are performed by the Fast-ICA algorithm module;

[0054] S5: Match network output and target label to perform mapping between real and complex numbers;

[0055] S6: The Fast-ICA algorithm is optimized using artificial bee colony optimization and then transmitted to the BIM construction terminal.

[0056] In this embodiment, during the training phase, the complex network directly estimates each frequency point using the input time-frequency data, extracts the complex form TF bin of the input data block, obtains the initial separation matrix W0(f) through the CgRNN separation module, initializes the Fast-ICA separation matrix using W0(f), obtains the final separation matrix W(k), adds a minimum distortion principle smoothing module, and the smoothed output is transformed by inverse Fourier transform to obtain the numerical filter form of the separation matrix. The input signal X(t) is passed through an FIR filter to obtain the output signal Y(t).

[0057] In this embodiment, the data processed by Fast-ICA is a data block composed of the frequency harmonics of the mixed signal. The data block x(k,t) represents the complex value at frequency k and time t of the mixed signal. The data block is composed of the complex spectrum expansion of b frames of the mixed signal. Each frequency point corresponds to a separation matrix W(k). Each data block contains k separation matrices. W(k) is the learning label of the complex network. Then, the trained network framework is combined into the Fast-ICA algorithm as a module of the algorithm.

[0058] In this embodiment, a dual-channel complex neural network CgRNN based on the time-frequency domain is used. CgRNN introduces a new complex RNN unit for blind source separation of complex signals. The weights are parameterized by a full-dimensional unitary matrix. This complex RNN network outperforms real-valued networks with the same number of parameters in blind separation tasks. The CgRNN network is used to estimate each complex element in the separation matrix based on the input complex spectrum information, and the separation matrix of the output signal is calculated using the trained model.

[0059] In this embodiment, the Fast-ICA algorithm module performs separation iterative calculations. During the training phase, the complex form of the TF bin features is obtained after performing a short-time Fourier transform on the mixed signal. These features are then used as X input to the network for training. The estimated value Wo(f) output by the network and the separation matrix W(k) of the true source are calculated. The mean squared error loss function is optimized using the PIT method, and finally the separation matrix W(k) is obtained. During the testing phase, the TF bin of the test signal is input into the trained CgRNN model to obtain the estimated value Wo(f).

[0060] In this embodiment, the network output and target label are matched, and a mapping between real and complex numbers is performed. The real part of the original complex number is retained as the real part, and its imaginary part is set to 0, that is:

[0061] z = x + i·0

[0062] In the complex state, the complex number is mapped to the real state by using a linear combination of the real and imaginary parts of the complex number, with Wo and bo as the weight and bias factor, respectively, where R(z) is the real part of the complex number and S(z) is the imaginary part of the complex number.

[0063] o r =Wo[R(z)·S(z)]+bo.

[0064] In this embodiment, the mixed signal x(k,t) is input into the complex network using the method of initializing the blind separation matrix based on PIT and CgRNN. The MSE loss function is optimized using the PIT principle to reduce the error between the estimated value and the label W(k). Finally, the estimated complex matrix is ​​used to initialize the separation matrix in the algorithm.

[0065] In this embodiment, the artificial bee colony optimization algorithm is used to optimize Fast-ICA. The maximum number of iterations kmax is initialized, and the number of iterations without obtaining a better solution is allowed k_limit. The EB optimization stage is simulated, and the best possible solution θbest so far is selected from θ. The OB optimization stage and the SB optimization stage are simulated. An estimation method based on cross-validation is applied, and the data is divided into several parts. One part is used to extract the features of the data, and the other part is used to verify these features. Using the idea of ​​cross-validation, an estimation criterion for the number of complex source signals is given.

[0066] Example 2

[0067] Reference Figure 1 This invention discloses an application method for a construction progress management system based on BIM data, comprising the following steps:

[0068] S1: During BIM construction, each frequency point is estimated using the input time-frequency data in a complex network;

[0069] S2: Integrate the trained network framework as a module into the Fast-ICA algorithm;

[0070] S3: A two-channel complex neural network based on the time-frequency domain, where the weights are parameterized by a full-dimensional unitary matrix:

[0071] S4: Separate iterative calculations are performed by the Fast-ICA algorithm module;

[0072] S5: Match network output and target label to perform mapping between real and complex numbers;

[0073] S6: The Fast-ICA algorithm is optimized using an artificial bee colony optimization algorithm.

[0074] In this embodiment, during the training phase, the complex network directly estimates each frequency point using the input time-frequency data, extracts the complex form of the TF bin of the input data block, obtains the final separation matrix W(k), adds a minimum distortion principle smoothing module, and the smoothed output is transformed by inverse Fourier transform to obtain the numerical filter form of the separation matrix. The input signal X(t) is passed through an FIR filter to obtain the output signal Y(t).

[0075] In this embodiment, the data processed by Fast-ICA is a data block composed of the frequency harmonics of the mixed signal. Each frequency point corresponds to a separation matrix W(k). Each data block contains k separation matrices. W(k) is the learning label of the complex network. Then, the trained network framework is combined into the Fast-ICA algorithm as a module of the algorithm.

[0076] In this embodiment, a dual-channel complex neural network CgRNN based on the time-frequency domain is used. CgRNN introduces a new complex RNN unit for blind source separation of complex signals. The weights are parameterized by a full-dimensional unitary matrix. This complex RNN network outperforms real-valued networks with the same number of parameters in blind separation tasks. The CgRNN network is used to estimate each complex element in the separation matrix based on the input complex spectrum information, and the separation matrix of the output signal is calculated using the trained model.

[0077] In this embodiment, the Fast-ICA algorithm module performs separation iterative calculations. During the training phase, the complex form of the TF bin features is obtained after performing a short-time Fourier transform on the mixed signal. These features are then used as X input to the network for training. The estimated value Wo(f) output by the network and the separation matrix W(k) of the true source are calculated. The mean squared error loss function is optimized using the PIT method, and finally the separation matrix W(k) is obtained. During the testing phase, the TF bin of the test signal is input into the trained CgRNN model to obtain the estimated value Wo(f).

[0078] In this embodiment, the network output and target label are matched, and a mapping between real and complex numbers is performed. The real part of the original complex number is retained as the real part, and its imaginary part is set to 0, that is:

[0079] z = x + i·0

[0080] In the complex state, the complex number is mapped to the real state by using a linear combination of the real and imaginary parts of the complex number, with Wo and bo as the weight and bias factor, respectively, where R(z) is the real part of the complex number and S(z) is the imaginary part of the complex number.

[0081] o r =Wo[R(z)·S(z)]+bo.

[0082] In this embodiment, the mixed signal x(k,t) is input into the complex network using the method of initializing the blind separation matrix based on PIT and CgRNN. The MSE loss function is optimized using the PIT principle to reduce the error between the estimated value and the label W(k). Finally, the estimated complex matrix is ​​used to initialize the separation matrix in the algorithm.

[0083] In this embodiment, the artificial bee colony optimization algorithm is used to optimize Fast-ICA. The maximum number of iterations kmax is initialized, and the number of iterations without obtaining a better solution is allowed k_limit. The EB optimization stage is simulated, and the best possible solution θbest so far is selected from θ. The OB optimization stage and the SB optimization stage are simulated. An estimation method based on cross-validation is applied, and the data is divided into several parts. One part is used to extract the features of the data, and the other part is used to verify these features. Using the idea of ​​cross-validation, an estimation criterion for the number of complex source signals is given.

[0084] Example 3

[0085] Reference Figure 1 This invention discloses an application method for a construction progress management system based on BIM data, comprising the following steps:

[0086] S1: During BIM construction, each frequency point is estimated using the input time-frequency data in a complex network;

[0087] S2: Integrate the trained network framework as a module into the Fast-ICA algorithm;

[0088] S3: A two-channel complex neural network based on the time-frequency domain, where the weights are parameterized by a full-dimensional unitary matrix:

[0089] S4: Separate iterative calculations are performed by the Fast-ICA algorithm module;

[0090] S5: Match network output and target label to perform mapping between real and complex numbers;

[0091] S6: The Fast-ICA algorithm is optimized using an artificial bee colony optimization algorithm.

[0092] In this embodiment, during the training phase, the complex network directly estimates each frequency point using the input time-frequency data, extracts the complex form TF bin of the input data block, obtains the initial separation matrix W0(f) through the CgRNN separation module, initializes the Fast-ICA separation matrix using W0(f), obtains the final separation matrix W(k), adds a minimum distortion principle smoothing module, and the smoothed output is transformed by inverse Fourier transform to obtain the numerical filter form of the separation matrix. The input signal X(t) is passed through an FIR filter to obtain the output signal Y(t).

[0093] In this embodiment, the data processed by Fast-ICA is a data block composed of the frequency harmonics of the mixed signal. The data block x(k,t) represents the complex value at frequency k and time t of the mixed signal. The data block is composed of the complex spectrum expansion of b frames of the mixed signal. Each frequency point corresponds to a separation matrix W(k). Each data block contains k separation matrices. W(k) is the learning label of the complex network. Then, the trained network framework is combined into the Fast-ICA algorithm as a module of the algorithm.

[0094] In this embodiment, a dual-channel complex neural network CgRNN based on the time-frequency domain is used. CgRNN introduces a new complex RNN unit for blind source separation of complex signals. The weights are parameterized by a full-dimensional unitary matrix. This complex RNN network outperforms real-valued networks with the same number of parameters in blind separation tasks. The CgRNN network is used to estimate each complex element in the separation matrix based on the input complex spectrum information, and the separation matrix of the output signal is calculated using the trained model.

[0095] In this embodiment, the Fast-ICA algorithm module performs separation iterative calculations. During the training phase, the complex form of the TF bin features is obtained after performing a short-time Fourier transform on the mixed signal. These features are then used as X input to the network for training. The estimated value Wo(f) output by the network and the separation matrix W(k) of the true source are calculated. The mean squared error loss function is optimized using the PIT method, and finally the separation matrix W(k) is obtained. During the testing phase, the TF bin of the test signal is input into the trained CgRNN model to obtain the estimated value Wo(f).

[0096] In this embodiment, the network output and target label are matched, and a mapping between real and complex numbers is performed. The real part of the original complex number is retained as the real part, and its imaginary part is set to 0, that is:

[0097] z = x + i·0

[0098] In the complex state, the complex number is mapped to the real state by using a linear combination of the real and imaginary parts of the complex number, with Wo and bo as the weight and bias factor, respectively, where R(z) is the real part of the complex number and S(z) is the imaginary part of the complex number.

[0099] o r =Wo[R(z)·S(z)]+bo.

[0100] In this embodiment, the mixed signal x(k,t) is input into the complex network using the method of initializing the blind separation matrix based on PIT and CgRNN. The MSE loss function is optimized using the PIT principle to reduce the error between the estimated value and the label W(k). Finally, the estimated complex matrix is ​​used to initialize the separation matrix in the algorithm.

[0101] In this embodiment, the artificial bee colony optimization algorithm is used to optimize Fast-ICA. The maximum number of iterations kmax is initialized, and the OB optimization stage and SB optimization stage are simulated. An estimation method based on cross-validation technology is applied to divide the data into several parts. One part is used to extract the features of the data, and the other part is used to verify these features. Using the idea of ​​cross-validation technology, an estimation criterion for the number of complex source signals is given.

[0102] Comparative Example 1

[0103] The difference from Example 1 is that, in S1, the complex network estimates each frequency point using the input time-frequency data. During the training phase, the complex network directly estimates each frequency point using the input time-frequency data, extracts the complex form of the TF bin of the input data block, obtains the initial separation matrix W0(f) through the CgRNN separation module, initializes the Fast-ICA separation matrix using W0(f), obtains the final separation matrix W(k), adds a minimum distortion principle smoothing module, and obtains the numerical filter form of the separation matrix after the smoothed output undergoes inverse Fourier transform. The input signal X(t) is passed through an FIR filter to obtain the output signal Y(t).

[0104] Comparative Example 2

[0105] The difference from Example 2 is that, in S2, the trained network framework is incorporated into the Fast-ICA algorithm as a module. The data processed by Fast-ICA is a data block composed of the frequency harmonics of the mixed signal. Each frequency point corresponds to a separation matrix W(k). Each data block contains k separation matrices. W(k) is the learning label of the complex network. Then, the trained network framework is incorporated into the Fast-ICA algorithm as a module.

[0106] Comparative Example 3

[0107] The difference from Example 3 is that S5: a dual-channel complex neural network based on the time-frequency domain, with weights parameterized by a full-dimensional unitary matrix, is used. The dual-channel complex neural network CgRNN based on the time-frequency domain introduces a new complex RNN unit for blind source separation of complex signals. The weights are parameterized by a full-dimensional unitary matrix. This complex RNN network outperforms real-valued networks with the same number of parameters in blind separation tasks. The CgRNN network is used to estimate each complex element in the separation matrix based on the input complex spectrum information, and the separation matrix of the output signal is calculated using the trained model.

[0108] Experimental Example

[0109] The blind source separation methods for complex signals described in Examples 1, 2, and 3 were tested, and the results are as follows:

[0110] Example 1 Example 2 Example 3 Existing methods Separation accuracy 97％ 66％ 58％ 66％ Separation efficiency 91％ 47％ 63％ 56％

[0111] Compared with existing methods for separating blind sources of complex signals, the blind source separation methods in Examples 1, 2, and 3 significantly improve separation efficiency and accuracy, with Example 1 being the best example.

[0112] Test report

[0113] The purpose of this invention is to address the problems of insufficient stability, poor signal separation effect, slow convergence speed, low signal separation efficiency, and inability to perform independent component analysis in existing algorithms. This invention proposes a blind source separation method for complex signals. By employing the Fast-ICA algorithm, based on higher-order statistics of the signal, it performs independent component analysis, exhibiting good stability and fast convergence speed. The Fast-ICA algorithm is improved by using an artificial bee colony optimization algorithm to search for global and local optimal solutions, increasing the probability of finding the optimal solution. It uses fewer control parameters, achieves high convergence accuracy, and significantly improves signal separation efficiency and speed, resulting in a marked improvement in separation accuracy.

[0114] Although embodiments of the invention have been shown and described, those skilled in the art will understand that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.

Claims

1. A construction progress management system based on BIM data, characterized in that, include: The BIM data acquisition module is used to collect relevant noise data during BIM construction and transmit the relevant noise data to a complex network for training. The construction data management module is used to perform data separation and iterative management operations in the trained network. The management optimization module is used to perform algorithmic optimization on the separated iterative data and output it to the BIM cloud network.

2. An application method for a construction progress management system based on BIM data, characterized in that, Includes the following steps: S1: During BIM construction, each frequency point is estimated using the input time-frequency data in a complex network; S2: Integrate the trained network framework as a module into the Fast-ICA algorithm; S3: A two-channel complex neural network based on the time-frequency domain, where the weights are parameterized by a full-dimensional unitary matrix: S4: Separate iterative calculations are performed by the Fast-ICA algorithm module; S5: Match network output and target label to perform mapping between real and complex numbers; S6: The Fast-ICA algorithm is optimized using artificial bee colony optimization and then transmitted to the BIM construction terminal.

3. The application method of the construction progress management system based on BIM data according to claim 2, characterized in that, S1 includes: During the training phase, the complex network directly estimates each frequency point using the input time-frequency data, extracts the complex form of the TF bin of the input data block, obtains the initial separation matrix W0(f) through the CgRNN separation module, initializes the Fast-ICA separation matrix using W0(f), and obtains the final separation matrix W(k). A minimum distortion principle smoothing module is added, and the smoothed output is transformed by inverse Fourier transform to obtain the numerical filter form of the separation matrix. The input signal X(t) is passed through an FIR filter to obtain the output signal Y(t).

4. The application method of the construction progress management system based on BIM data according to claim 2, characterized in that, S2 includes: Fast-ICA processes data as a block of data consisting of the frequency harmonics of a mixed signal. The data block x(k,t) represents the complex value at frequency k and time t of the mixed signal. The data block is composed of the complex spectrum expansion of b frames of the mixed signal. Each frequency point corresponds to a separation matrix W(k). Each data block contains k separation matrices. W(k) is the learned label of the complex network. The trained network framework is then incorporated into the Fast-ICA algorithm as a module.

5. The application method of the construction progress management system based on BIM data according to claim 2, characterized in that, S3 includes: Based on the time-frequency domain, a dual-channel complex neural network (CgRNN) is introduced. CgRNN introduces a novel complex RNN unit for blind source separation of complex signals. The weights are parameterized by a full-dimensional unitary matrix. This complex RNN network outperforms real-valued networks with the same number of parameters in blind separation tasks. The CgRNN network estimates each complex element in the separation matrix based on the input complex spectrum information, and uses the trained model to calculate the separation matrix of the output signal.

6. The application method of the construction progress management system based on BIM data according to claim 2, characterized in that, S4 includes: The Fast-ICA algorithm module performs separation and iterative calculations. During the training phase, the mixed signal is subjected to a short-time Fourier transform to obtain the complex form of the TF bin features, which are then used as X input to the network for training. The estimated value Wo(f) of the network output is calculated, and the separation matrix W(k) of the true source is calculated. The mean squared error loss function is optimized using the PIT method, and finally the separation matrix W(k) is obtained. During the testing phase, the TF bin of the test signal is input into the trained CgRNN model to obtain the estimated value Wo(f).

7. The application method of the construction progress management system based on BIM data according to claim 2, characterized in that, S5 includes: Match the network output and target label, perform a mapping between real and complex numbers, using the real part of the original complex number as the real part, and setting its imaginary part to 0, i.e.: z = x + i·0 In the complex state, the complex number is mapped to the real state by using a linear combination of the real and imaginary parts of the complex number, with Wo and bo as the weight and bias factor, respectively, where R(z) is the real part of the complex number and S(z) is the imaginary part of the complex number. or=Wo[R(z)·S(z)]+bo.

8. The application method of the construction progress management system based on BIM data according to claim 2, characterized in that, S6 includes: The mixed signal x(k,t) is input into the complex network using a method based on PIT and CgRNN to initialize the blind separation matrix. The MSE loss function is optimized using the PIT principle to reduce the error between the estimated value and the label W(k). Finally, the estimated complex matrix is ​​used to initialize the separation matrix in the algorithm.

9. The application method of the construction progress management system based on BIM data according to claim 2, characterized in that, S6 further includes: The Fast-ICA algorithm is optimized using artificial bee colony optimization. The maximum number of iterations kma is initialized, and the number of iterations that do not yield a better solution is allowed k_limit. The EB optimization stage is simulated, and the best possible solution θbest so far is selected from θ. The OB optimization stage and the SB optimization stage are simulated, and an estimation method based on cross-validation is applied.

Images