Disclosure of Invention
The invention provides an intelligent detection system for building settlement, which effectively solves the problems that the existing building settlement does not have the influence on the settlement of the whole building according to the nonlinearity, large hysteresis, complex settlement change and the like of the settlement change of each detection point of the building, and the building settlement is not accurately detected, predicted and early warned, so that the early warning and management of the settlement of the building are greatly influenced.
The invention is realized by the following technical scheme:
the utility model provides a building settlement intelligent detection system which characterized in that: the system comprises a building settlement parameter acquisition platform based on a wireless sensor network and a building settlement intelligent early warning system, wherein the building settlement parameter acquisition platform based on the wireless sensor network realizes the detection and management of the building settlement parameters, the building settlement intelligent early warning system comprises a plurality of building settlement interval number neural network models, a plurality of building settlement prediction models and an interval number probability neural network classifier, the outputs of the building settlement sensors are respectively used as the inputs of the corresponding building settlement interval number neural network models, the outputs of the building settlement interval number neural network models are respectively used as the inputs of the corresponding building settlement prediction models, the outputs of the building settlement prediction models are used as the inputs of the interval number probability neural network classifier, the output of the interval number probability neural network classifier is the interval number representing the settlement risk degree of the building, the intelligent early warning system for the settlement of the building realizes detection, prediction and early warning of the settlement of the detected building.
The invention further adopts the technical improvement scheme that:
the building settlement interval number neural network model consists of an RR time recurrent neural network model, the system comprises an interval number DRNN neural network model and 3 beat-to-beat delay lines TDL, wherein the output of a building settlement sensor is the input of an RR time recurrent neural network model, the output of the RR time recurrent neural network model is the input of 1 corresponding beat-to-beat delay line TDL, the output of 3 beat-to-beat delay lines TDL is the input of an interval number DRNN neural network model, the output of the interval number DRNN neural network model is the output of the interval number and the building settlement interval number neural network model, the upper limit value and the lower limit value of the interval number DRNN neural network model are respectively used as the input of 2 corresponding beat-to-beat delay lines TDL, and the building settlement interval number neural network model converts the measured building settlement sensed by the building settlement sensor in a period into a dynamic interval value of building settlement.
The invention further adopts the technical improvement scheme that:
the building settlement prediction model comprises 2 small wave decomposition models, 2 groups of a plurality of Elman neural network prediction models, 2 LSTM neural networks, 2 time delay line TDL and interval number ridge neural networks; the upper limit value and the lower limit value of the number of output intervals of the building settlement interval number neural network model are respectively used as the input of 2 small wavelength decomposition models, the 2 small wavelength decomposition models respectively decompose the upper limit value and the lower limit value of the number of output intervals of the building settlement interval number neural network model into 2 groups of low-frequency trend parts and a plurality of high-frequency fluctuation part signals, the 2 groups of low-frequency trend parts and the plurality of high-frequency fluctuation part signals are respectively used as the input of corresponding 2 groups of a plurality of Elman neural network prediction models, the output of the 2 groups of a plurality of Elman neural network prediction models is respectively used as the input of corresponding 2 LSTM neural networks, the output of the 2 LSTM neural networks and the output of 2 beat delay lines TDL are respectively used as the input of the interval number ridge wave neural network, the upper limit value and the lower limit value of the interval number ridge wave neural network output are respectively used as the input of corresponding 2 beat delay lines TDL, and the interval number ridge wave neural network output is used as the predicted value of the detected point interval number of the building and the settlement value of the building settlement interval number And reducing the output of the prediction model.
The invention further adopts the technical improvement scheme that:
the interval number probability neural network classifier is a probability neural network with the input of a plurality of groups of interval numbers and the output of a group of interval numbers, the input of the interval number probability neural network classifier is a plurality of groups of interval values output by a plurality of building settlement prediction models, and the output of the interval number probability neural network classifier is an interval number representing the degree of danger of settlement and collapse of a detected building; according to engineering practice of building settlement detection and building deformation measurement specifications (JGJ8-2007), the interval number probability neural network classifier constructs a corresponding relation table of the interval number probability neural network classifier and 5 settlement and collapse risk degrees of the building, the 5 settlement and collapse risk degrees of the building are respectively in a normal state, relatively dangerous, very dangerous and very dangerous, the similarity between the interval number output by the interval number probability neural network classifier and 5 interval numbers representing the settlement and collapse risk degrees of the building is calculated, and the settlement and collapse risk degree of the building corresponding to the interval number with the maximum similarity is determined as the settlement and collapse risk degree of the building.
The invention further adopts the technical improvement scheme that:
the building settlement parameter acquisition platform based on the wireless sensor network consists of a detection node and an on-site monitoring end, the detection node and the on-site monitoring end are constructed into a building settlement parameter acquisition and intelligent prediction platform through a wireless communication module NRF2401 in a self-organizing manner, the detection node consists of a sensor group module, a single chip microcomputer MSP430 and a wireless communication module NRF2401 respectively, the sensor group module is responsible for detecting building temperature, humidity, settlement and water level building parameters, and the single chip microcomputer controls sampling intervals and sends the building settlement parameters to the on-site monitoring end through the wireless communication module NRF 2401; the on-site monitoring end consists of an industrial control computer, and realizes the management of detecting the building settlement parameters of the detection nodes and the prediction and early warning of the building settlement.
Compared with the prior art, the invention has the following obvious advantages:
the invention relates to a method for measuring the settlement of a building, which aims at the uncertainty and randomness of the problems of sensor precision error, interference, measured settlement abnormity and the like in the measurement process of the settlement parameters of the building.
The RNN time recursive neural network is used for processing time series data of building settlement. In the network, the state value of the hidden neuron at the current moment is reserved by a cycle structure, and the state value is input into the hidden layer neuron at the next moment as a part of a building settlement input signal input in the next cycle. The input signal of RNN is the building settlement time sequence input, each layer shares the network weight and bias when inputting one step, which greatly reduces the parameters needed to be learned in the network and reduces the complexity of the network.
The RNN time recursive neural network fully utilizes the correlation among the building settlement data based on the time sequence, is a neural network with a directional circulation structure added in a hidden layer, has a special structure, can better process the problem of the building settlement data based on the time sequence, shows stronger capability of learning essential characteristics of a building settlement data set by representing distributed representation of input building settlement data, realizes approximation of complex functions, better delineates rich intrinsic information of the building settlement data, has stronger generalization capability, and improves the accuracy and reliability of calculating the size of the building settlement.
The RNN time recursive neural network is a neural network introducing a time sequence concept, has a feedback mechanism, and is widely applied to time sequence data modeling. The RNN may store the learned information in the network, enabling the model to learn the dependency of the current time on past information. Given an input building settlement sequence, the RNN time recurrent neural network hides the layer state h at any time ttAre all based on the building settlement input x at the current momenttAnd hidden layers at past timesState ht-1The state of the hidden layer at each moment can be transmitted to the next moment by the RNN time recursive neural network; and finally, mapping the building settlement for a period of time by the RNN time recursive neural network through an output layer to obtain the output quantity of the building settlement.
And fifthly, 2 groups of a plurality of Elman neural networks are adopted to realize the prediction of the building settlement parameters within a period of time of the upper and lower limit values of the building settlement interval number, the Elman neural networks are generally divided into 4 layers, namely an input layer, a middle layer (hidden layer), a carrying layer and an output layer, the connection of the input layer, the hidden layer and the output layer is similar to a feedforward network, the units of the input layer only play a role in signal transmission, and the units of the output layer play a role in linear weighting. The transfer function of the hidden layer unit can adopt a linear or non-linear function, and the accepting layer is also called a context layer or a state layer, and is used for memorizing the output value of the hidden layer unit at the previous moment, which can be regarded as a time delay operator. The Elman neural network is characterized in that the output of the hidden layer is self-connected to the input of the hidden layer through the delay and storage of the receiving layer, the self-connection mode enables the output to have sensitivity to the data of the historical state, and the addition of the internal feedback network increases the capability of the network for processing dynamic information, so that the purpose of dynamically predicting the number of the settlement intervals of the building is achieved. The Elman neural network regression neuron network is characterized in that the output of the hidden layer is self-connected to the input of the hidden layer through the delay and storage of a structural unit, the self-connection mode enables the hidden layer to have sensitivity to data of a historical state, and the addition of an internal feedback network increases the capability of the network for processing dynamic information, thereby being beneficial to modeling of a dynamic process; the neural network fuses information of a future prediction network and information of a past prediction network by using feedback connection of dynamic neurons of a related layer, so that the memory of the network to time series characteristic information is enhanced, and the prediction accuracy of the number of the building settlement intervals is improved.
The building settlement prediction model comprises 2 small wave decomposition models, 2 groups of a plurality of Elman neural network prediction models, 2 LSTM neural networks, 2 beat-to-beat delay lines TDL and interval number ridge wave neural networks, and the building settlement prediction model is decomposed and then respectively predicted and fused, so that the accuracy and robustness of prediction of building settlement are improved. The interval number ridge wave neural network simulates visual cortex of human brain, neurons in the area can receive specific direction information of building settlement, namely, the neurons have optimal response to targets in the specific direction, the hidden layer excitation function of the ridge wave neural network is a ridge wave function, the neurons have directivity which is consistent with the direction of the building settlement, the ridge wave neural network has more dimension information and can process higher-dimension data, the approximation effect on the nonlinear high-dimension function is good, the input of the interval number ridge wave neural network is the upper limit value and the lower limit value of the interval number of the building settlement in a period of time, and the output represents the predicted value of the building settlement.
The interval number probability neural network classifier quantifies the dynamic degree of the influence of the predicted value of the settlement interval number of the detected building on the safety of the building into the risk degree of the collapse and the operation of the building according to engineering practice experience of the collapse of the building, building deformation measurement specifications (JGJ8-2007) and related maintenance control standards of the country of the subsidence of the building, and the interval number output by the interval number probability neural network classifier represents the risk degree of the collapse of the building; and constructing a corresponding relation table of 5 interval numbers and 5 settlement collapse danger degrees of the building according to the 5 settlement collapse danger degrees of the building, wherein the 5 settlement collapse danger degrees are respectively in a normal state, relatively dangerous, very dangerous and correspond to the different 5 different interval numbers. And calculating the similarity between the output interval number of each detection point interval number probability neural network classifier and 5 interval numbers representing 5 types of settlement and collapse of the building at different degrees, wherein the building settlement and collapse risk corresponding to the interval number with the maximum similarity is determined as the settlement and collapse risk of the building at the detection point, and the dynamic performance and the scientific classification of the building settlement risk grade classification are realized.
Detailed Description
The technical scheme of the invention is further described by combining the attached drawings 1-5:
1. design of overall system function
The system is used for detecting and predicting the building settlement parameters and early warning the building settlement, and consists of a building settlement parameter acquisition platform based on a wireless sensor network and a building settlement intelligent early warning system. The building settlement parameter acquisition platform based on the wireless sensor network comprises a plurality of building settlement detection nodes 1 and an on-site monitoring terminal 2, which are constructed into a wireless measurement and control network in a self-organizing manner to realize wireless communication between the detection nodes 1 and the on-site monitoring terminal 2; the detection node 1 sends the detected building parameters to the field monitoring terminal 2 and performs primary processing on the sensor data; the on-site monitoring terminal 2 transmits the control information to the detection node 1. The whole system structure is shown in figure 1.
2. Design of detection node
A large number of detection nodes 1 based on a wireless sensor network are used as building parameter sensing terminals, and the detection nodes 1 and the field monitoring terminal 2 interact with each other through information between the self-organizing wireless network. The detection node 1 comprises sensors of temperature, humidity, sedimentation amount and water level parameters of a building, a corresponding signal conditioning circuit, an MSP430 microprocessor and an NRF2401 wireless transmission module; the software of the detection node mainly realizes wireless communication and the acquisition and pretreatment of the settlement parameters of the building. The software is designed by adopting a C language program, so that the compatibility degree is high, the working efficiency of software design and development is greatly improved, and the reliability, readability and transportability of program codes are enhanced. The structure of the detection node is shown in fig. 3.
3. Site monitoring terminal software
The on-site monitoring end 2 is an industrial control computer, the on-site monitoring end 2 mainly realizes the collection of building parameters and the prediction of building settlement, realizes the information interaction with the detection node 1, and the on-site monitoring end 2 mainly has the functions of communication parameter setting, data analysis and data management and an intelligent early warning system for building settlement. The management software selects Microsoft Visual + +6.0 as a development tool, calls the Mscomm communication control of the system to design a communication program, and the functions of the field monitoring end software are shown in figure 4. The intelligent early warning system for building settlement comprises a plurality of building settlement interval number neural network models, a plurality of building settlement prediction models and an interval number probability neural network classifier, wherein the outputs of a plurality of building settlement sensors are respectively used as the inputs of the corresponding building settlement interval number neural network models, the outputs of the building settlement interval number neural network models are respectively used as the inputs of the corresponding building settlement prediction models, the outputs of the building settlement prediction models are used as the inputs of the interval number probability neural network classifier, the outputs of the interval number probability neural network classifier are interval numbers representing the settlement and collapse dangerous degree of a building, and the intelligent early warning system for building settlement realizes the detection, prediction and early warning of the settlement amount of the building to be detected; the structure and the function of the intelligent early warning system for building settlement are shown in figure 2. The intelligent early warning system for building settlement is designed as follows:
(1) design of neural network model for number of settlement intervals of building
The building settlement interval number neural network model consists of an RR time recurrent neural network model, an interval number DRNN neural network model and 3 beat Delay lines TDL (tapped Delay line), wherein the building settlement interval number neural network model is based on the dynamics and the fuzziness of the detected building settlement amount sensed by a building settlement sensor, the measured building settlement amount sensed by the building settlement sensor in a period of time is converted into the dynamic interval value of the building settlement, the output of the building settlement sensor is the input of the RR time recurrent neural network model, the output of the RR time recurrent neural network model is the input of the corresponding beat Delay line TDL, and the output of the 3 beat Delay lines TDL is the output of the interval number DRNN neural network modelThe output of the DRNN neural network model of the input interval number is u1(k) And u2(k),u1(k) And u2(k) Respectively as inputs to corresponding 2 beat delay lines TDL, u1(k) And u2(k) The upper limit value and the lower limit value respectively represent the output of the neural network model of the building settlement interval number, and the output interval number value of the detected building settlement amount of the building settlement sensor in a period of time is [ u2,u1]The identification structure of the neural network model for the number of the subsidence intervals of the building is shown in FIG. 5, wherein X (k-l), …, X (k-m) is historical data output by the RR time recurrent neural network model, and U is1(k-1),…,U1(k-d) historical data of the upper limit value of the output value of the neural network model for the number of subsidence intervals of the building, U2(k-1),…,U2(k-d) historical data of the lower limit value of the output value of the neural network model for the number of subsidence intervals of the building, u1(k) And u2(k) In order that the output value of the interval number DRNN neural network model represents the output of the building settlement interval number neural network model, k represents the current time, m and d represent the lag points of X and U, respectively, the building settlement interval number neural network model can be described as follows:
U(k)=[u2(k),u1(k)]=F[X(k),X(k-1),…,X(k-m);u1(k),…,u1(k-d);u2(k),…,u2(k-d)] (1)
A. RNN time recursive neural network design
The RNN time recursive neural network can process the sequence information of the size of the building settlement, uses the output of the previous state of the size of the building settlement as a part of the input of the predicted size of the next settlement, and has the function of 'memorizing' the size of the building settlement in a general sense. The RNN time recursive neural network can reserve the previous sequence of building settlement as output, and the next sequence of building settlement input and the reserved previous sequence of settlement output are jointly calculated to obtain the next sequence of building settlement output. x is the number oftIs the input at time t, stRepresenting the state of a memory unit of the network at time t, stBy a preceding stepState st-1And input x at the current timetJointly calculating to obtain:
st=f(Uxt+Wst-1) (2)
the stimulus function f is a non-linear function tanh in the RNN neural network, usually the first hidden state st-1The value of (c) will be initialized with 0, but actually initializing with a minimum value will cause the gradient to drop faster. otIs the output at time t, typically a probability vector calculated by a normalized exponential function:
ot=softmax(Vst) (3)
B. interval number DRNN neural network model design
The interval number DRNN neural network model is a dynamic regression neural network with feedback and the ability of adapting to time-varying characteristics, the network can more directly and vividly reflect the dynamic change performance of the settlement of a building and can more accurately predict the interval value of the settlement of the building, and the layer 3 network structure of the interval number DRNN neural network model is an input layer, and a hidden layer of the interval number DRNN neural network model is a regression layer and an output layer. In the interval number DRNN neural network model of the present invention, let I ═ I1(t),I2(t),…,In(t)]Inputting a vector for the network, wherein Ii(t) is the input of the ith neuron of the input layer of the interval number DRNN neural network model at the t moment, and the output of the jth neuron of the regression layer is Xj(t),Sj(t) is the input sum of the jth regression neuron, f (·) is a function of S, and O (t) is the output of the interval number DRNN neural network. The output of the interval number DRNN neural network model is:
the output of the RNN time recursive neural network is used as the input of the DRNN neural network model, and the output of the DRNN neural network model is the number of intervals of the size of the settlement of the building; the value of the output interval of the building settlement sensor for detecting the settlement of the building in a period of time is [ u [ ]2,u1]. For a period of timeThe output of the RNN time recursive neural network is used as the input of a DRNN neural network model, and the output of the DRNN neural network model is the interval number of the size of the settlement of the building; the value of the output interval of the building settlement sensor for detecting the settlement of the building in a period of time is [ u [ ]2,u1]。
(2) Design of building settlement prediction model
The building settlement prediction model comprises 2 small wave decomposition models, 2 groups of a plurality of Elman neural network prediction models, 2 LSTM neural networks, 2 TDLs and an interval number ridge neural network; the upper limit value and the lower limit value of the interval number output by the building settlement interval number neural network model are respectively used as the input of 2 small wave decomposition models of the building settlement prediction model, the 2 small wave decomposition models respectively decompose the upper limit value and the lower limit value of the interval number output by the building settlement interval number neural network model into 2 groups of low-frequency trend parts and a plurality of high-frequency fluctuation part signals, the 2 groups of low-frequency trend parts and the plurality of high-frequency fluctuation part signals are respectively used as the input of 2 groups of a plurality of Elman neural network prediction models, the output of the 2 groups of a plurality of Elman neural network prediction models is respectively used as the input of corresponding LSTM neural networks, the output of the 2 LSTM neural networks is used as the input of the interval number ridge wave neural network, the output of the 2 TDLs is used as the input of the interval number ridge wave neural network, and the output of the interval number ridge wave neural network is used as the prediction interval number value of the settlement of the detected points of the building;
A. 2 Small wave decomposition model design
The upper limit value and the lower limit value of the number of output intervals of the building settlement interval number neural network model are respectively used as the input of a wavelet decomposition model 1 and a wavelet decomposition model 2, the wavelet decomposition model decomposes the upper limit value and the lower limit value into 2 groups of low-frequency components and a plurality of high-frequency components respectively, and the 2 groups of low-frequency components and the plurality of high-frequency components are respectively used as the input of corresponding 2 groups of a plurality of Elman neural network prediction models. The invention uses wavelet analysis method to decompose the upper and lower limit values of time sequence of interval value of building settlement parameter, the wavelet decomposition makes self-correlation and cross-correlation analysis to each layer information of building settlement interval number, the smoothing process is made to the building settlement interval number signal in the wavelet decomposition process, therefore, the data of building settlement interval number after wavelet processing is much smoother. And (3) respectively establishing 2 groups of corresponding upper limit and lower limit multiple Elman neural network prediction models according to the characteristics of the number of the building settlement intervals after each layer of signal analysis to respectively predict the upper limit value and the lower limit value of the number of the building settlement intervals, and finally respectively taking the prediction results of each layer as the input of the corresponding LSTM neural network and the output of 2 LSTM neural networks as the fusion value of the prediction value of the number of the building settlement intervals. The wavelet multi-resolution decomposition process generally adopts a Mallat algorithm, and the decomposition relationship of the algorithm is expressed as follows:
h in formula (5)0、h1A low-pass decomposition filter and a high-pass decomposition filter, respectively. m isp、npRespectively, resolution is 2-pLow frequency coefficients and high frequency coefficients. The algorithm reconstructs the relationship as follows:
in the formula (6), g0、g1A low-pass reconstruction filter and a high-pass reconstruction filter, respectively. A. thep、DpRespectively resolution 2-pA lower low frequency component and a high frequency component. The Mallat algorithm decomposes the decomposed low frequency signal part of each layer into high frequency and low frequency again, thus performing layer-by-layer decomposition. The result obtained after p-layer decomposition of the historical data X of the original test is as follows:
X=D1+D2+…Dp+Ap (7)
a in the formula (7)pFor the part of the low-frequency signal after the p-th layer decomposition, DpThe high-frequency part after the decomposition of the p-th layer. The 2 small wave decomposition models can output historical data sequences of the upper limit value and the lower limit value of the number of the output intervals of the neural network model of the number of the subsidence intervals of the buildingThe signals are decomposed into different resolution spaces, and the effect of the processing is that historical data sequences of the upper limit value and the lower limit value of the number of the output intervals of the neural network model of the number of the building settlement intervals decomposed into each resolution space are simpler than those of the historical data sequences and the upper limit value and the lower limit value of the number of the building settlement intervals are more accurately predicted.
B. Design of multiple Elman neural network prediction models
The Elman neural network prediction models input high-frequency and low-frequency values of the upper limit value and the lower limit value of the number of the output intervals of the building settlement interval number neural network model after wavelet decomposition to predict the future value of the number of the building settlement intervals, and each Elman neural network prediction model can be regarded as a forward neural network with a local memory unit and local feedback connection. The Elman neural network prediction model has a special correlation layer besides the hidden layer; the correlation layer receives the feedback signal from the hidden layer, and each hidden layer node is connected with the corresponding correlation layer node. The association layer takes the hidden layer state at the previous moment and the network input at the current moment as the input of the hidden layer, which is equivalent to state feedback. The transfer function of the hidden layer is generally a Sigmoid function, the output layer is a linear function, and the associated layer is also a linear function. In order to effectively solve the problem of approximation accuracy in the prediction of the detected parameters and enhance the function of a correlation layer, the number of an input layer, an output layer and a hidden layer of an Elman neural network prediction model is respectively m, n and r; w is a1,w2,w3And w4Respectively representing the connection weight matrixes from the structural layer unit to the hidden layer, from the input layer to the hidden layer, from the hidden layer to the output layer and from the structural layer to the output layer, and then the expressions of the hidden layer, the associated layer and the output layer of the Elman neural network prediction model are respectively as follows:
cp(k)=xp(k-1) (9)
the number of an input layer, an output layer and a hidden layer of each Elman neural network prediction model is 5,1 and 11 respectively, the input of the model is low-frequency and high-frequency historical parameters of the upper limit value and the lower limit value of the number of output intervals of the building settlement interval neural network model after wavelet decomposition, the output of the model is predicted values of high-frequency and low-frequency parameters corresponding to the upper limit value and the lower limit value of the number of the building settlement interval, the upper limit value and the lower limit value of the number of the building settlement interval are predicted respectively, and the prediction accuracy is improved. 2 groups of a plurality of Elman neural network prediction models of the upper limit value and the lower limit value of the number of the building settlement intervals are respectively input into 2 LSTM neural networks, and the output of the 2 LSTM neural networks is a fusion value of the prediction values of the upper limit value and the lower limit value of the number of the building settlement intervals.
C. LSTM neural network design
The output of the 2 groups of the Elman neural network prediction models of the upper limit value and the lower limit value of the number of the building settlement intervals is respectively the input of 2 LSTM neural networks, the output of the 2 LSTM neural networks is the re-prediction of the predicted values of the upper limit value and the lower limit value of the number of the building settlement intervals, and the output of the 2 LSTM neural networks is used as the input of the interval number ridge wave neural network. The temporal Recurrent Neural Network (RNN), which consists of Long Short Term Memory (LSTM) units, is called an LSTM temporal recurrent neural network, also commonly referred to as an LSTM network. The LSTM neural network residue introduces mechanisms of Memory cells (Memory cells) and hidden layer states (Cell states) to control the transfer of information between hidden layers. The memory unit of an LSTM neural network has 3 Gates (Gates) as Input Gate, forgetting Gate and Output Gate. Wherein, the input gate can control the adding or filtering of new information; the forgetting door can forget the information to be lost and keep the useful information in the past; the output gate enables the memory unit to output only information related to the current time step. The 3 gate structures carry out operations such as matrix multiplication, nonlinear summation and the like in the memory unit, so that the memory still cannot be attenuated in continuous iteration. The long-short term memory unit (LSTM) structure unit is composed of a unit (Cell), an Input Gate (Input Gate), an Output Gate (Output Gate) and a forgetting Gate (Forget Gate). The unit is responsible for at willKeeping in mind the values over the time interval, three gates can each be considered as conventional artificial neurons for computing a weighted sum of activation functions. The LSTM neural network is a model capable of lasting long-term short-term memory, is suitable for classification, prediction of time sequence and other work, the LSTM effectively prevents gradient disappearance during RNN training, and the long-term short-term memory (LSTM) network is a special RNN. The model can learn long-term dependency information while avoiding the gradient vanishing problem. LSTM adds a structure called a Memory Cell (Memory Cell) to a neural node of a hidden layer of an internal structure RNN of a neuron to memorize past information, and adds three kinds of gate structures (Input, form, Output) to control use of history information. Let the input LSTM neural network data sequence be (x)1,x2,…,xT) The hidden layer state is (h)1,h2,…,hT) Then, time t has:
it=sigmoid(Whiht-1+WxiXt) (11)
ft=sigmoid(Whfht-1+WhfXt) (12)
ct=ft⊙ct-1+it⊙tanh(Whcht-1+WxcXt) (13)
ot=sigmoid(Whoht-1+WhxXt+Wcoct) (14)
ht=ot⊙tanh(ct) (15)
wherein it、ft、otRepresenting input, forget and output doors, ctRepresenting a cell, WhRepresenting the weight of the recursive connection, WxSigmoid and tanh represent the weights from the input layer to the hidden layer, and are two activation functions. The method comprises the steps of firstly establishing an LSTM neural network model, establishing a training set by utilizing the upper and lower limit number predicted values of the preprocessed building settlement interval number and carrying out model prediction by using the LSTM neural network with long-term and short-term memoryTraining is carried out, the LSTM neural network considers the time sequence and nonlinearity of the building settlement interval number upper and lower limit value prediction value data, and the prediction accuracy is high.
D. Interval number ridge wave neural network design
The interval number ridge wave neural network comprises 2 TDLs and 1 ridge wave neural network, the output of 2 LSTM neural networks is used as the input of the ridge wave neural network, the output of 2 TDLs is used as the input of the ridge wave neural network, the upper limit value and the lower limit value of the output interval number of the ridge wave neural network are respectively used as the input of 2 TDLs, the output of the interval number ridge wave neural network is the value of the settlement interval of a building to be detected in a period of time, the interval number ridge wave neural network has an mxpx2 three-layer structure, m represents the number of input layer nodes of the ridge wave neural network, p represents the number of hidden layer nodes of the ridge wave neural network, 2 represents the number of output layer nodes, the neural network taking a ridge wave function as a hidden layer excitation function is the ridge wave neural network, and X ═ X [ X ] of the neural network1,x2,…,xm]Represents the input quantity of the interval number ridgelet neural network, y represents the output quantity of the interval number ridgelet neural network, and U is [ U ═ U [1,u2,…up]A direction matrix representing a spinal neural network, wherein Ui=[ui1,ui2,…uim]Represents the ridge direction vector corresponding to the ith hidden layer neuron, a ═ a1,a2,…ap]A ridge scale vector representing a ridge neural network, b ═ b1,b2,…bp]A ridge position vector representing a ridge neural network, w ═ w1,w2,…wp]Representing the connection weight vector between the hidden layer and the output layer. The interval number ridge neural network output is expressed as:
wherein, i is 2, and 2 represents two nodes with output being interval number of the output interval number of the ridge wave neural network;
the output of the jth hidden layer neuron is expressed as:
(3) interval number probability neural network classifier design
The interval number probability neural network classifier is a probability neural network which inputs a plurality of groups of interval numbers and outputs the interval numbers as a group of interval numbers, the output of the building settlement prediction models is the input of the interval number probability neural network classifier, the output of the interval number probability neural network classifier is the interval number representing the settlement and collapse risk degree of the building, and the interval number probability neural network classifier can accurately predict the settlement and collapse risk degree of the building to be detected; according to engineering practice experience of settlement and collapse of the building and building deformation measurement specifications (JGJ8-2007), the interval number probability neural network classifier quantifies the predicted value of the settlement interval number of the detected building into values corresponding to 5 settlement and collapse risk degrees of the building, the 5 settlement and collapse risk degrees of the building are respectively in a normal state, relatively dangerous, very dangerous and correspond to 5 different interval numbers, and the corresponding relation table of the 5 different interval numbers and the 5 settlement and collapse risk degrees of the building is shown in table 1. And calculating the similarity between the interval number output by the interval number probability neural network classifier and 5 interval numbers representing different degrees of settlement and collapse of the building, wherein the building settlement and collapse risk degree corresponding to the interval number with the maximum similarity is determined as the building settlement and collapse risk degree.
TABLE 1 corresponding relationship table of settlement and collapse risk of building and number of intervals
Serial number
|
Sedimentation collapseDegree of risk of collapse
|
Number of intervals
|
1
|
Normal state
|
[0.00,0.20]
|
2
|
Is relatively dangerous
|
[0.20,0.40]
|
3
|
Danger of
|
[0.40,0.60]
|
4
|
Is very dangerous
|
[0.60,0.80]
|
5
|
Is very dangerous
|
[0.80,1.0] |
The interval number probability neural network is an artificial neural network based on statistical principle, and is a feedforward network model taking a Parzen window function as an activation function. The interval number Probability Neural Network (PNN) absorbs the advantages of the radial basis function neural network and the classical probability density estimation principle, and particularly has more remarkable advantages in mode classification compared with the traditional feed-forward neural network. The PNN is a network with two hidden layers. Wherein the first layer and the last layer are an input layer and an output layer respectively, and the middle two layers are hidden layers. The first hidden layer is a mode unit layer, a Parzen window function is used as an activation function, the second hidden layer is a summing unit layer, and the outputs of the first hidden layer are selectively summed. The PNN training method mainly comprises a recursive orthogonal least square algorithmAnd a recursive least squares approach. Both methods have faster convergence rate, compared with the latter method which has faster training speed and higher training precision. For an input vector x, the output value Y of the ith neuron of the interval number probability neural network output layerjCan be expressed as:
an x-dimensional input vector; hk(x) -the output of the second hidden layer second unit; w is ajk-connection weights of the second hidden layer first neurons and the output layer first neurons; p (-) -Parzen window kernel; c. Cki-a first hidden center vector of a first class of the first hidden layer; n isk-the number of hidden layer central vectors of the first hidden layer second class; | | · | -euclidean norm; m-number of neurons in the output layer; the input of the interval number probability neural network classifier is an interval number predicted value output by a plurality of building settlement prediction models, and the output is an interval number representing the collapse risk degree of a building.
4. Design example of building settlement parameter acquisition platform based on wireless sensor network
According to the shape of a detected building, a system designs a plane layout installation diagram of a detection node 1 and a field monitoring terminal 2, wherein the detection node 1 is arranged in the detected building in a balanced manner to realize the detection of the settlement of the building, and the system realizes the collection of building parameter parameters and the intelligent prediction of the settlement of the building.
The technical means disclosed in the invention scheme are not limited to the technical means disclosed in the above embodiments, but also include the technical scheme formed by any combination of the above technical features. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and such improvements and modifications are also considered to be within the scope of the present invention.