CN112054869A - Wavelength allocation method based on continuous Hopfield neural network - Google Patents
Wavelength allocation method based on continuous Hopfield neural network Download PDFInfo
- Publication number
- CN112054869A CN112054869A CN202010779318.XA CN202010779318A CN112054869A CN 112054869 A CN112054869 A CN 112054869A CN 202010779318 A CN202010779318 A CN 202010779318A CN 112054869 A CN112054869 A CN 112054869A
- Authority
- CN
- China
- Prior art keywords
- matrix
- output state
- state matrix
- row
- neural network
- 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.)
- Granted
Links
Images
Classifications
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04J—MULTIPLEX COMMUNICATION
- H04J14/00—Optical multiplex systems
- H04J14/02—Wavelength-division multiplex systems
- H04J14/0227—Operation, administration, maintenance or provisioning [OAMP] of WDM networks, e.g. media access, routing or wavelength allocation
-
- 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
- G06N3/044—Recurrent networks, e.g. Hopfield networks
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04J—MULTIPLEX COMMUNICATION
- H04J14/00—Optical multiplex systems
- H04J14/02—Wavelength-division multiplex systems
- H04J14/0227—Operation, administration, maintenance or provisioning [OAMP] of WDM networks, e.g. media access, routing or wavelength allocation
- H04J14/0254—Optical medium access
- H04J14/0267—Optical signaling or routing
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04J—MULTIPLEX COMMUNICATION
- H04J14/00—Optical multiplex systems
- H04J14/02—Wavelength-division multiplex systems
- H04J14/0278—WDM optical network architectures
- H04J14/0283—WDM ring architectures
Landscapes
- Engineering & Computer Science (AREA)
- Signal Processing (AREA)
- Computer Networks & Wireless Communication (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Biomedical Technology (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Optical Communication System (AREA)
Abstract
The invention discloses a wavelength allocation method based on a continuous Hopfield neural network, which comprises the following steps: acquiring an initial traffic matrix of the ring network and a mapping vector of a communication path; mapping the initial traffic matrix according to the mapping vector to obtain a mapped traffic matrix; acquiring an output state matrix according to the mapped traffic matrix; and updating the output state matrix by utilizing a continuous Hopfield neural network to obtain the optimal distribution wavelength. The wavelength allocation method of the invention uses the continuous Hopfield neural network to allocate the wavelength, which can effectively accelerate the speed of finding the optimal wavelength allocation and improve the reliability of the network.
Description
Technical Field
The invention belongs to the technical field of communication, and particularly relates to a wavelength allocation method based on a continuous Hopfield neural network.
Background
The wavelength Division multiplexing (wdm) (wavelength Division multiplexing) technology is widely applied to network transmission nowadays because it can realize waveguide sharing and improve the transmission bandwidth and utilization rate of a physical link. However, in a single waveguide ring-structured optical network-on-chip, as the overlap ratio of optical links increases, crosstalk between signals of different wavelengths is also increased.
The existing research on the network on optical chip mainly aims at reducing the communication blocking rate and searching an effective wavelength allocation method. As a wavelength assignment method for improving reliability based on the ant colony algorithm, there has been proposed a method in which communication paths are arranged mainly based on communication path information, and then wavelengths are assigned to the arranged paths by the ant colony algorithm based on the probability of pheromone generation. Although the method can find a better wavelength allocation method to improve the reliability, the complexity is higher, the search time is longer, in the practical calculation, the method is difficult to tend to the theoretically required optimal line under the condition of a certain cycle number, and meanwhile, the method is easy to generate the stagnation phenomenon and sometimes can not search the solution space further.
Disclosure of Invention
In order to solve the above problems in the prior art, the present invention provides a wavelength allocation method based on a continuous hopfield neural network. The technical problem to be solved by the invention is realized by the following technical scheme:
the invention provides a wavelength allocation method based on a continuous Hopfield neural network, which comprises the following steps:
s1: acquiring an initial traffic matrix of the ring network and a mapping vector of a communication path;
s2: mapping the initial traffic matrix according to the mapping vector to obtain a mapped traffic matrix;
s3: acquiring an output state matrix according to the mapped traffic matrix;
s4: and updating the output state matrix by utilizing a continuous Hopfield neural network to obtain the optimal distribution wavelength.
In an embodiment of the present invention, the S1 includes:
s11: acquiring communication path information of the ring network, and establishing a communication traffic matrix KNN between N nodes of the ring network to be optimized;
s12: determining a mapping vector MAP ═ { N ] that places the N nodes in a particular location in a ring network1,n2,...,ni,...,nNIn which n isiDenotes the number of the ith node in the mapped network, ni=1~N,i=1~N。
In an embodiment of the present invention, the S2 includes:
for the traffic matrix K according to the mapping vector MAPNNMapping to obtain a mapped traffic matrix Kmapped:
Kmapped[i,j]=KNN[MAP(i),MAP(j)]
Wherein i is 1 to N, j is 1 to N, Kmapped[i,j]Representing the i-th row and j-th column elements of the mapped traffic matrix.
In an embodiment of the present invention, the S3 includes:
s31: defining the post-mapping traffic matrix K using random numbersmappedAnd (3) the range of each random number of the corresponding initial output state matrix is 0-1, and the relationship between the output state matrix and the input state matrix satisfies the following conditions:
wherein, U (t) is the input state at the time t, V (t) is the output state at the time t, and q is the slope of the activation function;
s32: obtaining an input state expression at the time t according to the relation between the output state matrix and the input state matrix:
U(t)=arctanh(2×V(t)-1)×q。
in an embodiment of the present invention, the S4 includes:
s41: initializing a matrix cycle flag p and the maximum cycle number thereof, a program cycle flag r and the maximum cycle number thereof, an output state matrix detection flag, and a row vector SNR composed of the worst signal-to-noise ratiowcAnd SNRwcMedium maximum signal-to-noise ratio SNRoptimalA corresponding index;
s42: initializing the output state matrix and acquiring a corresponding input state matrix;
s43: determining an energy function weight of the continuous Hopfield neural network and calculating an output increment of the energy function;
s44: updating the output state matrix according to the output increment of the energy function, and adjusting the output state matrix according to a threshold to obtain a row vector SNR consisting of a wavelength distribution result and a corresponding worst signal-to-noise ratiowc;
S45: detecting whether the program loop flag r reaches the maximum loop times, if so, executing S46, otherwise, returning to the step S41;
s46: row vector SNR composed from said worst signal-to-noise ratiowcFinding the maximum signal-to-noise ratio SNRoptimalAnd SNRoptimalSelecting the SNR according to the index value indexoptimalCorresponding wavelength assignment results:
[SNRoptimal,index]=max(SNRwc)
Path=Path(index)
wherein, Path(index)And the index value in all the wavelength allocation results is the index result after the program reaches the maximum operation times.
In an embodiment of the present invention, the S43 includes:
s431: determining four energy function weights A, B, C and D of the continuous Hopfield neural network, wherein the four energy function weights correspond to three constraint conditions of an energy function and network cost information respectively;
s432: acquiring four increments of the four energy function weights A, B, C and D;
s433: obtaining a total increment function according to the four increments;
s434: and calculating the total output increment of the energy function according to the total increment function.
In an embodiment of the present invention, the S432 includes:
s4321: obtaining a first energy function m1With the constraint that each node in the ring network can only have one mapAnd, each row of the matrix can be represented as 1:
wherein v isxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixxjThe element representing the x row and j column in the output state matrix, m when satisfying the constraint condition1Is 0;
s4322: obtaining a second energy function m2The corresponding constraint condition is that each mapping in the ring network can only point to one node, and each column of the matrix represented in the output state matrix can only have one 1:
wherein v isxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixxjThe element representing the ith row and the ith column in the output state matrix, m when the constraint condition is satisfied2Is 0;
s4323: obtaining a third energy function m3The constraint condition is that each node in the ring network is mapped, and the sum of all elements in the matrix is represented as N:
wherein v isxiThe element representing the x row and i column in the output state matrix, m when the constraint condition is satisfied3Is 0;
s4324: obtaining a fourth energy function m4Energy information corresponding to the neural network:
wherein k ismappedxyElements, v, representing the x, y, columns of the mapped traffic matrixxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixy,i+1、vy,i-1Respectively representing the elements on the right and left sides of the ith column of the y-th row in the output state matrix.
In an embodiment of the present invention, the S433 includes:
four increments m according to the four energy function weights A, B, C and D1、m2、m3And m4Obtain the total delta function E:
in an embodiment of the present invention, the S44 includes:
s441: calculating an input state matrix U (t +1) and an output state matrix V (t +1) at the time t +1 according to the total output increment dE of the energy function, wherein:
U(t+1)=U(t)+dEΔt
wherein Δ t is the length of time for each increment generated;
s442: according to the output state matrix V (t +1) at the time t +1, defining the upper output threshold Gate of the neural network neuronupAnd lower threshold GatedownAnd adjusting the output state matrix:
wherein x is 1 to N, and y is 1 to N; v. ofxyTo representThe element in the x-th row and y-th column of the output state matrix, Vx, y]Representing the elements of the matrix Vx row and y column after threshold adjustment, wherein V is the output state matrix after adjustment;
s443: making a matrix cycle flag p equal to p +1, when p is greater than a set maximum value, judging whether an output state matrix meets a constraint condition of an energy function, if so, making flag equal to 1, exiting the cycle, obtaining a corresponding wavelength allocation result and calculating an SNR of the wavelength allocation result, if not, making flag equal to 0, and returning to the step S42, wherein the obtaining of the corresponding wavelength allocation result includes: the dimension of the output state matrix V in step S442 is N × N, and when the constraint condition is satisfied, each row has only one 1 element, the remaining elements are 0, each column has only one 1 element, the remaining elements are 0, and the sum of all the elements in the output state matrix V is N; obtaining the abscissa i and the ordinate j of the non-zero elements in each row of the output state matrix V which meet the constraint condition; and taking the abscissa i as the position of the wavelength in the wavelength allocation result Path, and taking the ordinate j as the subscript of the wavelength, wherein the Path is a row vector with the size of 1 × N.
In one embodiment of the present invention, in step S443, calculating the SNR of the wavelength assignment result includes:
calculating the received signal power of the destination nodes of n communication pairsWherein the content of the first and second substances,representing the received signal power of the destination node of the kth communication pair;
calculating the noise power received by the destination node of n communication pairsWherein the content of the first and second substances,representing the received noise power of the destination node of the kth communication pair;
signal power with kth communication pairAnd noise powerCalculate the SNR value for the kth communication pair:
wherein the content of the first and second substances,for the optical signal input power of the kth communication pair,waveguide propagation loss accumulated during transmission of the optical signal for the kth communication pair from the source node to the destination node,the micro-ring accumulated in the process of transmitting the optical signal of the kth communication pair from the source node to the destination node experiences loss,the micro-ring coupling loss accumulated during the transmission of the optical signal for the kth communication pair from the source node to the destination node,a noise power generated for the ith communication pair for the kth communication pair;
obtaining the minimum value from n SNR rules as SNR corresponding to the wavelength allocation result PathwcThe value is obtained.
Compared with the prior art, the invention has the beneficial effects that:
1. the invention adopts the continuous Hopfield neural network, converts the path information into a traffic matrix in the solving process, and feeds back the output matrix by using the energy function, thereby leading the output matrix to be rapidly converged, directly solving the optimal wavelength distribution result and improving the reliability of the network. Compared with the conventional random + selection method of randomly distributing the wavelength and then optimizing the wavelength, the method has the advantages of higher solving speed and lower complexity.
2. The continuous Hopfield neural network adopted by the invention has good fault tolerance, and has little influence on the final output result if sudden feedback error occurs in the optimization process of the network on the optical chip by using the method.
The present invention will be described in further detail with reference to the accompanying drawings and examples.
Drawings
Fig. 1 is a flowchart of a wavelength allocation method based on a continuous hopfield neural network according to an embodiment of the present invention;
fig. 2 is a sub-flowchart of calculating a wavelength assignment result using a hopfield neural network energy function according to an embodiment of the present invention.
Detailed Description
To further illustrate the technical means and effects of the present invention adopted to achieve the predetermined objects, a wavelength allocation method based on a continuous hopfield neural network according to the present invention is described in detail below with reference to the accompanying drawings and the detailed description.
The foregoing and other technical matters, features and effects of the present invention will be apparent from the following detailed description of the embodiments, which is to be read in connection with the accompanying drawings. The technical means and effects of the present invention adopted to achieve the predetermined purpose can be more deeply and specifically understood through the description of the specific embodiments, however, the attached drawings are provided for reference and description only and are not used for limiting the technical scheme of the present invention.
It is noted that, herein, relational terms such as first and second, and the like may be used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Also, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that an article or device that comprises a list of elements does not include only those elements but may include other elements not expressly listed. The term "comprising", without further limitation, means that the element so defined is not excluded from the presence of another, identical element in the article or device in which the element is included.
Referring to fig. 1, fig. 1 is a flowchart of a wavelength allocation method based on a continuous hopfield neural network according to an embodiment of the present invention. The method comprises the following steps:
s1: acquiring an initial traffic matrix of the ring network and a mapping vector of a communication path;
specifically, the S1 includes:
s11: obtaining communication path information of a ring network, and establishing a communication matrix K between N nodes of the ring network to be optimizedNN;
The communication path information mainly includes: communication path source node position number RsourceDestination node position number RdestinationAnd inter-node traffic l.
Traffic matrix KNNThe element in (1) is the traffic between nodes, and the traffic l is defined as follows:
wherein R issourceAnd RdestinationRespectively representing the location number of the source node and the location number of the destination node of the communication path in the ring network structure.
S12: determining a mapping vector MAP for placing N nodes at specific locations in a ring network { N }1,n2,...,ni,...,nNIn which n isiDenotes the number of the ith node in the mapped network, ni=1~N,i=1~N。
Exemplarily, assume that
Wherein, KNNWith the abscissa of (a) indicating the number of the source node, the ordinate indicating the number of the destination node, and MAP ═ 3, 1, 4, 2, the traffic matrix K is represented by the above-mentioned traffic matrix KNNAs can be seen from the expression of (a), the communication traffic of the node 1 and the nodes 2 and 3 is 64 and 128 respectively, the communication traffic of the node 2 and the node 3 is 64, and the communication traffic of the node 3 and the node 4 is 64; the meaning of MAP {3, 1, 4, 2} is that after mapping, the 1, 2, 3, 4 nodes in the original network become 3, 1, 4, 2 nodes in the mapped network.
S2: mapping the initial traffic matrix according to the mapping vector to obtain a mapped traffic matrix;
in particular, the traffic matrix K is mapped according to the mapping vector MAPNNMapping to obtain a mapped traffic matrix Kmapped:
Kmapped[i,j]=KNN[MAP(i),MAP(j)]
Wherein i is 1 to N, j is 1 to N, Kmapped[i,j]Representing the i-th row and j-th column elements of the mapped traffic matrix.
S3: acquiring an output state matrix according to the mapped traffic matrix;
specifically, the S3 includes:
s31: defining the post-mapping traffic matrix K using random numbersmappedA corresponding initial output state matrix, the size of the initial output state matrix and the traffic matrix KNNThe random numbers are in the range of 0-1, and the relationship between the output state matrix and the input state matrix satisfies the following conditions:
wherein, U (t) is the input state at the time t, V (t) is the output state at the time t, and q is the amplification factor of the neuron input signal and the slope of the activation function;
s32: obtaining an input state expression at the time t according to the relation between the output state matrix and the input state matrix:
U(t)=arctanh(2×V(t)-1)×q。
s4: and updating the output state matrix by utilizing a continuous Hopfield neural network to obtain the optimal distribution wavelength.
Specifically, the S4 includes:
s41: initializing a matrix cycle flag p and the maximum cycle number thereof, a program cycle flag r and the maximum cycle number thereof, an output state matrix detection flag, and a row vector SNR composed of the worst signal-to-noise ratiowcAnd SNRwcMedium maximum signal-to-noise ratio SNRoptimalA corresponding index;
it should be noted that the purpose of the matrix cycle flag p is to converge the output matrix through a certain number of iterations, the purpose of the program cycle flag r is to find the maximum SNR through a certain cycle, the purpose of the output state matrix detection flag is to determine whether the corresponding wavelength allocation method needs to be output by detecting whether the output matrix satisfies the constraint condition, and the purpose of the index is to find the maximum SNRoptimalCorresponding wavelength allocation results;
when the program loops once, let r be r +1, when r is greater than the set value, then at SNRwcTo find the SNR inoptimalIt is output as the maximum signal-to-noise ratio.
S42: initializing the output state matrix, and acquiring a corresponding input state matrix according to a relational expression of the output state matrix and the input state matrix;
specifically, the output state matrix V is initialized as follows:
assuming that V is an output state matrix of k × k, the embodiment generates a random number using a self-defined rand () function in a program, and a default value of the random number of the function is 0 to 1, so that the implementation manner of the output state matrix is as follows: v ═ rand (k, k).
S43: determining an energy function weight of the continuous Hopfield neural network and calculating an output increment of the energy function;
specifically, S431: and determining four energy function weights A, B, C and D of the continuous Hopfield neural network, wherein the weights correspond to three constraint conditions of an energy function and one network cost information respectively. The weight of the energy function represents the feedback strength of the four conditions in the energy function, and the feedback strength of each item can be modified according to the characteristics of the optimization target so as to improve the convergence speed.
S432: acquiring four increments of the four energy function weights;
defining four increments of the four energy function weights A, B, C and D as m respectively1,m2,m3,m4Initializing the increment; according to the condition of the output state matrix, there are three correspondences: each node in the ring network can only have one mapping, corresponding to the increment m1(ii) a Each mapping in the ring network can only point to one node, corresponding to an increment m2(ii) a Each node in the ring network needs to be mapped, corresponding to an increment m3。
Specifically, S4321: obtaining a first energy function m1The corresponding constraint condition is that each node in the ring network can only have one mapping, and each row of the matrix in the output state matrix can only have one 1:
wherein v isxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixxjRepresenting the elements in the output state matrix in the x row and j column, and n represents the dimension of the output state matrix.
When the constraint condition is satisfied, m1Is 0, i.e. there is no feedback to the energy function.
S4322: obtaining a second energyFunction m2The corresponding constraint condition is that each mapping in the ring network can only point to one node, and each column of the matrix represented in the output state matrix can only have one 1:
wherein v isxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixxjRepresenting the element in the ith row and column of the output state matrix.
When the constraint condition is satisfied, m2Is 0, i.e. there is no feedback to the energy function.
S4323: obtaining a third energy function m3The constraint condition is that each node in the ring network is mapped, and the sum of all elements in the matrix is represented as N:
wherein v isxiAnd the element of the ith row and the ith column in the output state matrix is represented, N represents the dimension of the output state matrix, N represents the number of nodes in the ring network, and the size of N is the same as that of N.
When the constraint condition is satisfied, m3Is 0, i.e. there is no feedback to the energy function.
S4324: obtaining a fourth energy function m4Energy information corresponding to the neural network:
wherein k ismappedxyElements, v, representing the x, y, columns of the mapped traffic matrixxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixy,i+1、vy,i-1Respectively representing the elements on the right and left sides of the ith column of the y-th row in the output state matrix.
S433: four increments m according to the four energy function weights A, B, C and D1、m2、m3And m4Obtain the total delta function E:
s434: calculating the total output increment of the energy function according to the total increment function:
where dE represents the total increase in energy function output.
S44: updating the output state matrix according to the output increment of the energy function, and adjusting the output state matrix according to a threshold to obtain a row vector SNR consisting of a wavelength distribution result and a corresponding worst signal-to-noise ratiowc;
Specifically, the S44 includes:
s441: calculating an input state matrix U (t +1) and an output state matrix V (t +1) at the time t +1 according to the total output increment dE of the energy function:
U(t+1)=U(t)+dEΔt
wherein Δ t is the length of time for each increment generated;
s442: according to the output state matrix V (t +1) at the time t + 1, defining the upper output threshold Gate of the neural network neuronupAnd lower threshold GatedownAnd adjusting the output state matrix:
wherein x ═ 1 &N,y=1~N;vxyRepresenting the elements of the x-th row and y-th column of the output state matrix, Vx, y]Representing the elements of the matrix Vx row and y column after threshold adjustment, wherein V is the output state matrix after adjustment;
s443: and after the output state matrix is adjusted, making a matrix cycle flag p equal to p +1, when p is greater than a set maximum value, judging whether the output state matrix meets the constraint condition of the energy function, if so, making the flag equal to 1, exiting the cycle, outputting a corresponding wavelength allocation result and calculating the SNR of the wavelength allocation result, otherwise, making the flag equal to 0, and returning to the step S42.
Specifically, the dimension of the matrix V in step S442 is N × N, and the matrix satisfies the constraint condition, that is: each row has only one 1 element, and the rest elements are 0; each column has only one 1 element, and the rest elements are 0; the sum of all elements in the matrix is N; obtaining the abscissa i and the ordinate j of the non-zero elements in each row of the output state matrix V which meet the constraint condition; the abscissa i is taken as the position of the wavelength in the wavelength assignment result Path, and the ordinate j is taken as the subscript of the wavelength, i.e. the serial number of the wavelength, wherein the Path is a row vector with the size of 1 × N.
Specifically, the wavelength assignment result is set to Path, where Path is a row vector with a size of 1 × N, and assuming that the ith row and jth column elements in the matrix V are 1, it means that the ith position element in the wavelength assignment result Path is λjAccording to the method, the corresponding wavelength distribution result Path can be obtained from the matrix V;
for example:
from the above matrix, the positions of 1 element in the matrix are: line 1, column 2, line 2, column 1, line 3, column 4, line 4, column 3;
the wavelength assignment result Path ═ λ thus obtained from the matrix V2,λ1,λ4,λ3];
Further, calculating the SNR of the wavelength assignment result includes:
first, the signal power received by the destination node of n communication pairs is calculatedWherein the content of the first and second substances,representing the received signal power at the destination node of the kth communication pair.
A communication between a source node and a destination node is called a communication pair.
Secondly, the noise power received by the destination node of the n communication pairs is calculatedWherein the content of the first and second substances,representing the noise power received by the destination node of the kth communication pair.
Thereafter, the signal power of the k communication pair is utilizedAnd noise powerCalculate the SNR value for the kth communication pair:
wherein the content of the first and second substances,for the optical signal input power of the kth communication pair,transmitting optical signals for the k-th communication pair from a source node to a destination nodeThe waveguide propagation loss accumulated during the spot process,the micro-ring accumulated in the process of transmitting the optical signal of the kth communication pair from the source node to the destination node experiences loss,the micro-ring coupling loss accumulated during the transmission of the optical signal for the kth communication pair from the source node to the destination node,a noise power generated for the ith communication pair for the kth communication pair;
finally, the minimum value is selected from the n SNR values and is used as the SNR corresponding to the wavelength allocation result PathwcThe value is obtained.
S45: detecting whether the program loop flag r reaches the maximum loop times, if so, executing S46, otherwise, returning to the step S41;
s46: row vector SNR composed from said worst signal-to-noise ratiowcFinding the maximum signal-to-noise ratio SNRoptimalAnd SNRoptimalSelecting the SNR according to the index value indexoptimalCorresponding wavelength assignment results:
[SNRoptimal,index]=max(SNRwc)
Path=Path(index)
wherein, Path(index)And the index value in all the wavelength allocation results is the index result after the program reaches the maximum operation times.
The technical idea of the invention is as follows: and establishing a traffic matrix according to path information of the ring network nodes, initializing an output state matrix and an input state matrix, performing feedback iteration according to an energy function of the Hopfield neural network, and finally obtaining an output state matrix meeting constraint conditions after multiple feedbacks to obtain a final wavelength distribution result.
The embodiment of the invention adopts the continuous Hopfield neural network, converts the path information into a traffic matrix in the solving process, and feeds back the output matrix by using the energy function, so that the output matrix is rapidly converged, the optimal wavelength distribution result is directly obtained, and the reliability of the network is improved. Compared with the conventional random + selection method of randomly allocating the wavelengths and then optimizing the wavelengths, the method provided by the embodiment of the invention has the advantages that the solving speed is higher, and the complexity is lower. In addition, the continuous Hopfield neural network adopted by the embodiment of the invention has good fault tolerance, and if a sudden feedback error condition occurs in the optimization process of the network on the optical chip by using the method, the influence on the final output result is small.
The foregoing is a more detailed description of the invention in connection with specific preferred embodiments and it is not intended that the invention be limited to these specific details. For those skilled in the art to which the invention pertains, several simple deductions or substitutions can be made without departing from the spirit of the invention, and all shall be considered as belonging to the protection scope of the invention.
Claims (10)
1. A wavelength allocation method based on a continuous Hopfield neural network is characterized by comprising the following steps:
s1: acquiring an initial traffic matrix of the ring network and a mapping vector of a communication path;
s2: mapping the initial traffic matrix according to the mapping vector to obtain a mapped traffic matrix;
s3: acquiring an output state matrix according to the mapped traffic matrix;
s4: and updating the output state matrix by utilizing a continuous Hopfield neural network to obtain the optimal distribution wavelength.
2. The method for wavelength allocation based on a continuous hopfield neural network as claimed in claim 1, wherein said S1 includes:
s11: obtaining communication paths of a ring networkPath information and establishing a traffic matrix K between N nodes of the ring network to be optimizedNN;
S12: determining a mapping vector MAP for placing N nodes at specific locations in a ring network { N }1,n2,...,ni,...,nNIn which n isiDenotes the number of the ith node in the mapped network, ni=1~N,i=1~N。
3. The method for wavelength allocation based on a continuous hopfield neural network as claimed in claim 2, wherein said S2 includes:
for the traffic matrix K according to the mapping vector MAPNNMapping to obtain a mapped traffic matrix Kmapped:
Kmapped[i,j]=KNN[MAP(i),MAP(j)]
Wherein i is 1 to N, j is 1 to N, Kmapped[i,j]Representing the i-th row and j-th column elements of the mapped traffic matrix.
4. The method for wavelength allocation based on a continuous hopfield neural network as claimed in claim 1, wherein said S3 includes:
s31: defining the post-mapping traffic matrix K using random numbersmappedAnd (3) the range of each random number of the corresponding initial output state matrix is 0-1, and the relationship between the output state matrix and the input state matrix satisfies the following conditions:
wherein, U (t) is the input state at the time t, V (t) is the output state at the time t, and q is the slope of the activation function;
s32: obtaining an input state expression at the time t according to the relation between the output state matrix and the input state matrix:
U(t)=arctanh(2×V(t)-1)×q。
5. the method for wavelength allocation based on a continuous hopfield neural network as claimed in claim 4, wherein said S4 includes:
s41: initializing a matrix cycle flag p and the maximum cycle number thereof, a program cycle flag r and the maximum cycle number thereof, an output state matrix detection flag, and a row vector SNR composed of the worst signal-to-noise ratiowcAnd SNRwcIndex corresponding to the medium maximum signal-to-noise ratio;
s42: initializing the output state matrix and acquiring a corresponding input state matrix;
s43: determining an energy function weight of the continuous Hopfield neural network and calculating an output increment of the energy function;
s44: updating the output state matrix according to the output increment of the energy function, and adjusting the output state matrix according to a threshold to obtain a row vector SNR consisting of a wavelength distribution result and a corresponding worst signal-to-noise ratiowc;
S45: detecting whether the program loop flag r reaches the maximum loop times, if so, executing S46, otherwise, returning to the step S41;
s46: row vector SNR composed from said worst signal-to-noise ratiowcFinding the maximum signal-to-noise ratio SNRoptimalAnd SNRoptimalSelecting the SNR according to the index value indexoptimalCorresponding wavelength assignment results:
[SNRoptimal,index]=max(SNRwc)
Path=Path(index)
wherein, Path(index)And the index value in all the wavelength allocation results is the index result after the program reaches the maximum operation times.
6. The method for wavelength allocation based on a continuous hopfield neural network as claimed in claim 5, wherein said S43 includes:
s431: determining four energy function weights A, B, C and D of the continuous Hopfield neural network, wherein the four energy function weights correspond to three constraint conditions of an energy function and network cost information respectively;
s432: acquiring four increments of the four energy function weights A, B, C and D;
s433: obtaining a total increment function according to the four increments;
s434: and calculating the total output increment of the energy function according to the total increment function.
7. The method for wavelength allocation based on continuous Hopfield neural network of claim 6, wherein the S432 comprises:
s4321: obtaining a first energy function m1The corresponding constraint condition is that each node in the ring network can only have one mapping, and each row of the matrix in the output state matrix can only have one 1:
wherein v isxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixxjThe element representing the x row and j column in the output state matrix, m when satisfying the constraint condition1Is 0;
s4322: obtaining a second energy function m2The corresponding constraint condition is that each mapping in the ring network can only point to one node, and each column of the matrix represented in the output state matrix can only have one 1:
wherein v isxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixxjThe element representing the ith row and the ith column in the output state matrix, m when the constraint condition is satisfied2Is 0;
s4323: obtaining a third energy function m3The constraint condition is that each node in the ring network is mapped, and the sum of all elements in the matrix is represented as N:
wherein v isxiThe element representing the x row and i column in the output state matrix, m when the constraint condition is satisfied3Is 0;
s4324: obtaining a fourth energy function m4Energy information corresponding to the neural network:
wherein k ismappedxyElements, v, representing the x, y, columns of the mapped traffic matrixxiRepresenting the element, v, of the ith column of the x-th row in the output state matrixy,i+1、vy,i-1Respectively representing the elements on the right and left sides of the ith column of the y-th row in the output state matrix.
9. the method for wavelength allocation based on a continuous hopfield neural network as claimed in claim 5, wherein said S44 includes:
s441: calculating an input state matrix U (t +1) and an output state matrix V (t +1) at the time t +1 according to the total output increment dE of the energy function, wherein:
U(t+1)=U(t)+dEΔt
wherein Δ t is the length of time for each increment generated;
s442: according to the output state matrix V (t +1) at the time t +1, defining the upper output threshold Gate of the neural network neuronupAnd lower threshold GatedownAnd adjusting the output state matrix:
wherein x is 1 to N, and y is 1 to N; v. ofxyRepresenting the elements of the x-th row and y-th column of the output state matrix, Vx, y]Representing the elements of the matrix Vx row and y column after threshold adjustment, wherein V is the output state matrix after adjustment;
s443: making a matrix cycle flag p equal to p +1, when p is greater than a set maximum value, judging whether an output state matrix meets a constraint condition of an energy function, if so, making flag equal to 1, exiting the cycle, obtaining a corresponding wavelength allocation result and calculating an SNR of the wavelength allocation result, if not, making flag equal to 0, and returning to the step S42, wherein the obtaining of the corresponding wavelength allocation result includes: the dimension of the output state matrix V in step S442 is N × N, and when the constraint condition is satisfied, each row has only one 1 element, the remaining elements are 0, each column has only one 1 element, the remaining elements are 0, and the sum of all the elements in the output state matrix V is N; obtaining the abscissa i and the ordinate j of the non-zero elements in each row of the output state matrix V which meet the constraint condition; and taking the abscissa i as the position of the wavelength in the wavelength allocation result Path, and taking the ordinate j as the subscript of the wavelength, wherein the Path is a row vector with the size of 1 × N.
10. The method for wavelength assignment based on continuous hopfield neural network as claimed in claim 9, wherein in step S443, calculating the SNR of the wavelength assignment result includes:
calculating the received signal power of the destination nodes of n communication pairsWherein the content of the first and second substances,representing the received signal power of the destination node of the kth communication pair;
calculating the noise power received by the destination node of n communication pairsWherein the content of the first and second substances,representing the received noise power of the destination node of the kth communication pair;
signal power with kth communication pairAnd noise powerCalculate the SNR value for the kth communication pair:
wherein the content of the first and second substances,for the optical signal input power of the kth communication pair,waveguide propagation loss accumulated during transmission of the optical signal for the kth communication pair from the source node to the destination node,the micro-ring accumulated in the process of transmitting the optical signal of the kth communication pair from the source node to the destination node experiences loss,the micro-ring coupling loss accumulated during the transmission of the optical signal for the kth communication pair from the source node to the destination node,a noise power generated for the ith communication pair for the kth communication pair;
obtaining the minimum SNR value of n SNR values as the SNR corresponding to the wavelength allocation result PathwcThe value is obtained.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010779318.XA CN112054869B (en) | 2020-08-05 | 2020-08-05 | Wavelength allocation method based on continuous Hopfield neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010779318.XA CN112054869B (en) | 2020-08-05 | 2020-08-05 | Wavelength allocation method based on continuous Hopfield neural network |
Publications (2)
Publication Number | Publication Date |
---|---|
CN112054869A true CN112054869A (en) | 2020-12-08 |
CN112054869B CN112054869B (en) | 2021-07-23 |
Family
ID=73601425
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202010779318.XA Active CN112054869B (en) | 2020-08-05 | 2020-08-05 | Wavelength allocation method based on continuous Hopfield neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN112054869B (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113162725A (en) * | 2021-05-17 | 2021-07-23 | 济南德达多网络科技有限公司 | Optical network operation method |
CN113986812A (en) * | 2021-09-07 | 2022-01-28 | 西安电子科技大学 | CHNN-based on-chip network mapping method and device |
CN117278463A (en) * | 2023-09-04 | 2023-12-22 | 三峡智控科技有限公司 | Path planning method, path planning device, electronic equipment and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104598971A (en) * | 2015-01-15 | 2015-05-06 | 宁波大学 | Radial basis function neural network based unit impulse response function extraction method |
CN105451103A (en) * | 2015-11-02 | 2016-03-30 | 西安电子科技大学 | Wavelength-allocation-based three-dimensional optical on-chip network router communication system and method |
CN106355245A (en) * | 2016-09-12 | 2017-01-25 | 哈尔滨工业大学 | Method for integrating array antenna directional images on basis of neural network algorithms |
US20170237495A1 (en) * | 2016-02-15 | 2017-08-17 | Hfr, Inc. | Method and system for compensating for latency difference due to switchover in fronthaul in ring topology form |
CN107592635A (en) * | 2017-09-05 | 2018-01-16 | 东南大学 | Malicious user method of discrimination based on SOM neutral nets in cognitive radio |
CN108737011A (en) * | 2018-06-15 | 2018-11-02 | 西安电子科技大学 | The Wavelength allocation method of reduction crosstalk based on ant group algorithm |
CN110177311A (en) * | 2019-06-03 | 2019-08-27 | 西安电子科技大学 | A kind of multi-wavelength distribution method based on multiple-objection optimization |
-
2020
- 2020-08-05 CN CN202010779318.XA patent/CN112054869B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104598971A (en) * | 2015-01-15 | 2015-05-06 | 宁波大学 | Radial basis function neural network based unit impulse response function extraction method |
CN105451103A (en) * | 2015-11-02 | 2016-03-30 | 西安电子科技大学 | Wavelength-allocation-based three-dimensional optical on-chip network router communication system and method |
US20170237495A1 (en) * | 2016-02-15 | 2017-08-17 | Hfr, Inc. | Method and system for compensating for latency difference due to switchover in fronthaul in ring topology form |
CN106355245A (en) * | 2016-09-12 | 2017-01-25 | 哈尔滨工业大学 | Method for integrating array antenna directional images on basis of neural network algorithms |
CN107592635A (en) * | 2017-09-05 | 2018-01-16 | 东南大学 | Malicious user method of discrimination based on SOM neutral nets in cognitive radio |
CN108737011A (en) * | 2018-06-15 | 2018-11-02 | 西安电子科技大学 | The Wavelength allocation method of reduction crosstalk based on ant group algorithm |
CN110177311A (en) * | 2019-06-03 | 2019-08-27 | 西安电子科技大学 | A kind of multi-wavelength distribution method based on multiple-objection optimization |
Cited By (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113162725A (en) * | 2021-05-17 | 2021-07-23 | 济南德达多网络科技有限公司 | Optical network operation method |
CN113162725B (en) * | 2021-05-17 | 2022-11-18 | 上海声赫致远科技集团有限公司 | Optical network operation method |
CN113986812A (en) * | 2021-09-07 | 2022-01-28 | 西安电子科技大学 | CHNN-based on-chip network mapping method and device |
CN117278463A (en) * | 2023-09-04 | 2023-12-22 | 三峡智控科技有限公司 | Path planning method, path planning device, electronic equipment and storage medium |
CN117278463B (en) * | 2023-09-04 | 2024-04-23 | 三峡智控科技有限公司 | Path planning method, path planning device, electronic equipment and storage medium |
Also Published As
Publication number | Publication date |
---|---|
CN112054869B (en) | 2021-07-23 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN112054869B (en) | Wavelength allocation method based on continuous Hopfield neural network | |
Aiello et al. | Approximate load balancing on dynamic and asynchronous networks | |
CN109039534B (en) | Sparse code division multiple access signal detection method based on deep neural network | |
CN108494710A (en) | Visible light communication MIMO anti-interference noise-reduction methods based on BP neural network | |
CN108737011B (en) | The Wavelength allocation method of reduction crosstalk based on ant group algorithm | |
Lalbakhsh et al. | Antnet with reward-penalty reinforcement learning | |
CN114666204A (en) | Fault root cause positioning method and system based on cause and effect reinforcement learning | |
CN113795050B (en) | Sum Tree sampling-based deep double-Q network dynamic power control method | |
CN114154685A (en) | Electric energy data scheduling method in smart power grid | |
CN112073983B (en) | Wireless data center network topology optimization method and system based on flow prediction | |
CN110336631B (en) | Signal detection method based on deep learning | |
CN110177311B (en) | Multi-wavelength distribution method based on multi-objective optimization | |
CN115276820A (en) | Method for setting power gradient of on-chip optical interconnection light source with mapping assistance | |
Vijayalakshmi et al. | Artificial immune based hybrid GA for QoS based multicast routing in large scale networks (AISMR) | |
CN112637812B (en) | Vehicle-mounted cooperative communication relay selection method based on supervised machine learning | |
KR102497362B1 (en) | System for multi-layered knowledge base and processing method thereof | |
Piotrowski et al. | The grouping differential evolution algorithm for multi-dimensional optimization problems | |
JP3757722B2 (en) | Multi-layer neural network unit optimization method and apparatus | |
Meybodi et al. | New Class of Learning Automata Based Schemes for Adaptation of Bachpropagation Algorithm Parameters | |
Guo et al. | Improved NSGA-II optimizing coding-link cost trade-offs for multicast routing in WDM networks | |
CN113900795A (en) | Impulse neural network mapping method | |
Zhou et al. | Deep learning-optical network routing algorithm based on wavelength continuity supervision | |
CN113784365B (en) | Communication resource management method for Internet of things | |
CN113037425B (en) | Multi-target controller placement method based on evolution perception in network | |
CN115296705B (en) | Active monitoring method in MIMO communication system |
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 |