Virtual power plant edge gateway signal online maintenance method, system, device and medium

Abstract

The present invention discloses a method, system, device and medium for keeping a signal of a virtual power plant edge gateway online, and relates to the technical field of wireless communication signal detection. In response to a received online keeping request, the virtual power plant edge gateway is determined and a bit stream of information to be sent is obtained, the bit stream of information to be sent is mapped, a sending signal is generated, a decision tree is constructed according to a channel state matrix of the sending signal, the decision tree is searched, a heuristic value is obtained, the heuristic value is input into a preset path cost model, a path cost value is output, a memory space is searched using the path cost value, nodes that meet preset optimal node conditions are marked and input, a signal to be restored associated with the sending signal is formed, the signal to be restored is inversely mapped, and the restored bit stream of information to be sent and an index number are generated and output; the method solves the technical problem that traditional signal detection and optimal signal selection methods cannot ensure that the virtual power plant edge gateway device can work normally.

Description

Technical Field

[0001] The present invention relates to the technical field of wireless communication signal detection, and in particular to a method, system, device and medium for keeping edge gateway signals of a virtual power plant online. Background Art

[0002] In recent years, with the further development of various new technologies, communications, computing, and automation technologies have been widely applied in power grids and integrated with traditional power technologies, significantly enhancing the intelligence level of power grids. The application of sensor and information technology in power grids provides technical support for system status analysis and decision-making support. The mature development of dispatching, automation, and flexible transmission technologies provides a fundamental guarantee for the development and utilization of renewable energy and distributed power sources. The improvement of communication networks and the promotion and application of user information collection technologies have promoted two-way interaction between power grids and users. The continuous innovation of these technologies has made virtual power plant edge gateways a major development trend in the power industry.

[0003] Multiple-input, multiple-output (MIMO) communication systems, as a key component of next-generation wireless communication systems, utilize antenna arrays deployed at both the transmit and receive ends, significantly improving the transmission rate and spectrum efficiency of wireless links. However, in traditional MIMO systems (Multi-input, multiple-output (MIMO) virtual power plant edge gateway systems), system complexity and deployment costs increase exponentially with system scale. To address this issue, the concept of spatial modulation has been proposed, leading to the development of a series of spatial modulation schemes. Through spatial modulation, only a subset of transmit antennas can be activated for data transmission, while the combination of activated antennas carries more information. This maintains high data rates while reducing the number of deployed RF links, achieving a better balance between spectrum utilization and energy efficiency than traditional MIMO systems. Spatial modulation schemes have evolved from standard spatial modulation to generalized spatial modulation to dynamic spatial modulation schemes, with the number of active antennas also evolving from a single antenna to a fixed number of antennas to a dynamic number of antennas.

[0004] Spatial modulation has been recognized as a key technology for achieving ultra-high reliability and low-latency communications in future wireless networks. However, in MIMO systems based on dynamic spatial modulation schemes, the information bit stream is mapped into a three-dimensional constellation consisting of combinations of active antennas and two-dimensional symbol sets. While this three-dimensional mapping can alleviate detection ambiguity at the receiver, signal detection in MIMO systems based on the original spatial modulation scheme remains challenging and difficult due to the increased number of antennas. Furthermore, factors such as signal sparsity and dynamic antenna combinations further complicate signal detection. Specifically, brute-force methods for finding the maximum likelihood estimate (MLE) in MIMO systems based on the original spatial modulation scheme typically require searching all possible combinations of active antennas and two-dimensional symbol sets, which is infeasible in massive MIMO systems. Numerous studies have proposed low-complexity and suboptimal detection algorithms (such as Gaussian approximation and ordered block minimum mean square error) for MIMO systems based on the original spatial modulation scheme and generalized spatial modulation. While these algorithms offer lower computational complexity than brute-force methods, their bit error rate (BER) performance falls short of optimal performance.

[0005] To reduce computational complexity while pursuing optimal BER performance, researchers have recently focused on tree-search-based signal detection algorithms. Specifically, tree-search algorithms transform the detection problem into finding the shortest path along a constructed decision tree. Using this strategy, tree-search algorithms can significantly reduce average computational complexity, although their worst-case computational complexity remains exponential for all tree-search algorithms.

[0006] While many algorithms have been proposed to address signal detection in MIMO systems based on spatial modulation schemes and generalized spatial modulation schemes, effective signal detection methods in MIMO systems based on dynamic spatial modulation schemes remain underdeveloped. Furthermore, the dynamic nature of the number of active antennas in MIMO systems based on dynamic spatial modulation schemes prevents most existing signal detection and optimal signal selection methods from working properly. For example, existing message-passing-based detection methods typically assume a known, fixed sparsity constraint, but such algorithms fail in dynamic spatial modulation schemes due to variable sparsity. Summary of the Invention

[0007] The present invention provides a method, system, device and medium for keeping the signal of a virtual power plant edge gateway online, which solves the technical problem that in the existing virtual power plant edge gateway system based on a dynamic spatial modulation scheme, due to the dynamic change characteristics of the number of activated antennas, the traditional signal detection and optimal signal selection methods cannot ensure that the virtual power plant edge gateway device can work normally.

[0008] A first aspect of the present invention provides a method for keeping a virtual power plant edge gateway signal online, comprising:

[0009] In response to the received online maintenance request, determining the virtual power plant edge gateway corresponding to the online maintenance request and obtaining the corresponding information bit stream to be sent;

[0010] Mapping the information bit stream to be sent to generate a corresponding transmission signal;

[0011] Constructing a corresponding decision tree according to the channel state matrix of the transmitted signal;

[0012] Inputting the node vector associated with each tree node in the decision tree into the target deep neural network to obtain the corresponding heuristic value;

[0013] Input the heuristic value into a preset path cost model and output the corresponding path cost value;

[0014] Searching the memory space using the path cost value, marking and inputting nodes that meet the preset optimal node condition to form a signal to be recovered associated with the sent signal;

[0015] The signal to be restored is inversely mapped to generate and output a restored information bit stream to be sent and an index number.

[0016] Optionally, the step of mapping the information bit stream to be sent to generate a corresponding transmission signal includes:

[0017] Performing base-based conversion on the information bit stream to be sent to generate an initial mapped decimal bit stream;

[0018] Using the initial mapped decimal bit stream to input a preset intermediate decimal bit stream function to generate an intermediate mapped decimal bit stream;

[0019] Decoding the intermediate decimal bit stream to generate corresponding antenna combination data;

[0020] Using the intermediate mapping decimal bit stream to input a preset target decimal bit stream function to generate a target mapping decimal bit stream;

[0021] Performing base-to-base conversion on the target mapped decimal bit stream according to a preset length to generate a target binary sequence;

[0022] Converting the target binary sequence into constellation symbols in a preset two-dimensional constellation diagram;

[0023] A transmission signal is constructed using the antenna combination data and the constellation symbol.

[0024] Optionally, the step of constructing a corresponding decision tree according to the channel state matrix of the transmitted signal includes:

[0025] Receiving the transmitted signal at a receiving end;

[0026] Performing QR decomposition on the channel state matrix of the transmitted signal, and constructing a decision tree according to the orthogonal column vector group matrix in the decomposed channel state matrix.

[0027] Optionally, the training process of the target deep neural network includes:

[0028] Taking minimizing the likelihood function as the optimization goal, construct the initial deep neural network;

[0029] Randomly obtain bitstream training data and input it into the initial deep neural network for training to obtain the training path cost value of each tree node;

[0030] Calculating a loss function value corresponding to the bitstream training data based on the training path cost value; when the loss function value is greater than a preset standard loss function value, adjusting network parameters of the initial deep neural network using a mini-batch gradient descent method; and jumping to the step of randomly acquiring bitstream training data and inputting it into the initial deep neural network to obtain a training path cost value for each tree node, until the loss function value is less than or equal to the preset standard loss function value;

[0031] When the loss function value is less than or equal to the preset standard loss function value, the target deep neural network is output.

[0032] Optionally, the loss function value is expressed as follows:

[0033]

[0034] Where D represents the bitstream training data, R t represents the channel matrix, Represents all nodes on the entire path, z t represents the transmitted signal after QR decomposition, t represents the time slot number, where t = 1, 2, 3, ..., T;

[0035]

[0036] Where, Represents the loss function value, S K represents the kth node, Indicates the Nth t nodes, represents the true cost and f(S K\(\sum_{\theta} \mathcal{L}(\theta)\) represents the estimated cost sum of the initial deep neural network, and \(\theta\) represents the network parameters of the initial deep neural network;

[0037]

[0038] In the formula, \(\theta\) * represents the optimal network parameters.

[0039] Optionally, the preset path cost model is specifically:

[0040]

[0041] f * (s k ) = g(s k ) + h * (s k )

[0042] In the formula, f * (s k ) represents the path cost value, g(s k ) represents the \(s\) k -th cumulative cost value, h * (s k ) represents the optimal heuristic function, represents the remaining cost value from the \(s\) k -th node to -th node, b(s i ) represents the \(i\)-th incremental element in the upper triangular matrix, represents the -th cumulative cost value, s k represents a tree node in the \(k\)-th layer of the decision tree, z i represents the \(i\)-th transmitted signal after orthogonal transformation, r i,j represents the \((i, j)\)-th component element in the upper triangular matrix, s j represents the \(j\)-th component element.

[0043] Optionally, the preset optimal node condition is specifically the minimum path cost value and the tree height is greater than the preset tree height standard value;

[0044] Among them, the preset tree height standard value is specifically:

[0045]

[0046] In the formula, represents the preset tree height standard value, N t represents the number of transmit antennas, represents the memory space, and \(|S|\) represents the capacity of the constellation set.

[0047] Optionally, the step of performing inverse mapping on the signal to be recovered, generating and outputting a recovered information bit stream to be sent and an index number includes:

[0048] Acquiring antenna combination data associated with the signal to be recovered;

[0049] Encode the antenna combination data and output a corresponding index number;

[0050] Performing base-based conversion on the signal to be recovered to generate a corresponding initial inverse mapping binary sequence;

[0051] Using the initial inverse mapping binary sequence to input a preset target decimal bit stream function to generate a corresponding target inverse mapping decimal bit stream;

[0052] Using the target inverse-mapped decimal bit stream as input to a preset intermediate decimal bit stream function, to determine a corresponding initial inverse-mapped decimal bit stream;

[0053] The initial inverse-mapped decimal bit stream is converted into a decimal system according to a preset length, and a restored information bit stream to be sent and the index number are generated and output.

[0054] A second aspect of the present invention provides a virtual power plant edge gateway signal online maintenance system, comprising:

[0055] A response module, configured to respond to a received online maintenance request, determine the virtual power plant edge gateway corresponding to the online maintenance request and obtain a corresponding information bit stream to be sent;

[0056] A mapping module, configured to map the information bit stream to be sent and generate a corresponding transmission signal;

[0057] A decision tree construction module, configured to construct a corresponding decision tree according to the channel state matrix of the transmitted signal;

[0058] A search module is used to input the node vector associated with each tree node in the decision tree into the target deep neural network to obtain a corresponding heuristic value;

[0059] A path cost value module, configured to input the heuristic value into a preset path cost model and output a corresponding path cost value;

[0060] A signal module to be restored, configured to search the memory space using the path cost value, mark and input nodes that meet the preset optimal node conditions, and constitute a signal to be restored associated with the transmitted signal;

[0061] The inverse mapping module is used to perform inverse mapping on the signal to be recovered, generate and output the recovered information bit stream to be sent and the index number.

[0062] The third aspect of the present invention provides an electronic device, including a memory and a processor, wherein a computer program is stored in the memory. When the computer program is executed by the processor, the processor executes the steps of the method for keeping the virtual power plant edge gateway signal online as described in any one of the above items.

[0063] A fourth aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the method for keeping the virtual power plant edge gateway signal online as described in any one of the above items.

[0064] It can be seen from the above technical solutions that the present invention has the following advantages:

[0065] In response to a received online maintenance request, a virtual power plant edge gateway corresponding to the online maintenance request is determined and a corresponding information bit stream to be sent is obtained, the information bit stream to be sent is mapped, a corresponding transmission signal is generated, and a corresponding decision tree is constructed according to a channel state matrix of the transmission signal. The node vector associated with each tree node in the decision tree is input into a target deep neural network to obtain a corresponding heuristic value, the heuristic value is input into a preset path cost model, and a corresponding path cost value is output. The path cost value is used to search the memory space, and nodes that meet the preset optimal node conditions are marked and input to form a signal to be recovered associated with the transmission signal. The signal to be recovered is inversely mapped to generate and output the recovered information bit stream to be sent and an index number. The invention solves the technical problem that in the existing virtual power plant edge gateway system based on the dynamic spatial modulation scheme, due to the dynamic change characteristics of the number of activated antennas, the traditional signal detection and optimal signal selection methods cannot ensure that the virtual power plant edge gateway device can work normally.

[0066] In the multi-input multi-output (MIMO) virtual power plant edge gateway, the present invention designs a signal detection method with the lowest possible computational complexity and the detection performance as close to or as close to the optimal as possible, which helps to ensure that the virtual power plant edge gateway equipment is online in real time and meets the needs of high-speed real-time services such as virtual power plant frequency modulation. The present invention first proposes a dynamic spatial modulation DSM (CM-DSM) scheme based on combined mapping (CM), which solves the ambiguity problem of virtual power plant edge gateway signal detection based on the DSM scheme, and constructs a suitable decision tree for signal detection based on this scheme, thereby finding the maximum value in signal detection. The likelihood estimation problem is transformed into the problem of finding the minimum cost and path in a tree search; secondly, the present invention proposes a memory-efficiency balanced tree search (METS) algorithm, which is suitable for signal detection in multi-input multi-output systems based on the CM-DSM scheme, and the algorithm has the characteristics of low computational complexity and optimal detection performance while being memory-limited; finally, the present invention proposes a METS (DL-METS) algorithm accelerated by deep learning technology, which uses deep learning technology to accelerate the search for the optimal heuristic function, so that it can achieve almost optimal detection performance while the number of tree nodes visited is close to the minimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0067] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0068] Figure 1 A flowchart of a method for keeping a virtual power plant edge gateway signal online provided in Example 1 of the present invention;

[0069] Figure 2 A flowchart of a method for keeping a virtual power plant edge gateway signal online provided in the second embodiment of the present invention;

[0070] Figure 3 Schematic diagram of the MIMO virtual power plant edge gateway;

[0071] Figure 4 A schematic diagram of a decision tree for signal detection constructed in a MIMO system based on the CM-DSM algorithm;

[0072] Figure 5 This is a schematic diagram of the simulation results of bit error rate changing with signal-to-noise ratio in the Python simulation environment;

[0073] Figure 6This is a schematic diagram of the simulation results of the number of visited nodes changing with the signal-to-noise ratio in the Python simulation environment;

[0074] Figure 7 This is a structural block diagram of a virtual power plant edge gateway signal online maintenance system provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0075] An embodiment of the present invention provides a method, system, device and medium for keeping the signal of a virtual power plant edge gateway online, which is used to solve the technical problem that in the existing virtual power plant edge gateway system based on the dynamic spatial modulation scheme, due to the dynamic change characteristics of the number of activated antennas, the traditional signal detection and optimal signal selection methods cannot ensure that the virtual power plant edge gateway device can work normally.

[0076] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0077] See also Figure 1 , Figure 1 A flowchart of the steps of a method for keeping a virtual power plant edge gateway signal online provided in Example 1 of the present invention.

[0078] The present invention provides a method for keeping a virtual power plant edge gateway signal online, comprising:

[0079] Step 101: In response to a received online maintenance request, determine the virtual power plant edge gateway corresponding to the online maintenance request and obtain the corresponding information bit stream to be sent.

[0080] The online maintenance request refers to the request information for online maintenance of the multi-input multi-output virtual power plant edge gateway.

[0081] The information bit stream to be sent refers to the binary information bit stream that the transmitting end of the multi-input multi-output virtual power plant edge gateway will send to the receiving end.

[0082] In an embodiment of the present invention, in response to the received request information for online maintenance of a multi-input multi-output virtual power plant edge gateway, the virtual power plant edge gateway corresponding to the online maintenance request is determined, and the binary information bit stream to be sent from the transmitting end of the virtual power plant edge gateway to the receiving end is obtained.

[0083] Step 102: Map the information bit stream to be transmitted to generate a corresponding transmission signal.

[0084] Mapping refers to converting the information bit stream to be sent into a signal vector. The mapping in the present invention adopts a combined mapping (CM)-based dynamic spatial modulation (CM-DSM) algorithm.

[0085] In the embodiment of the present invention, the information bit stream to be sent is mapped and converted into a transmission signal vector.

[0086] Step 103: Construct a corresponding decision tree according to the channel state matrix of the transmitted signal.

[0087] In the embodiment of the present invention, a corresponding decision tree is constructed to perform signal detection according to a channel state matrix of a transmitted signal obtained by a dynamic spatial modulation (CM-DSM) algorithm.

[0088] Step 104: Input the node vector associated with each tree node in the decision tree into the target deep neural network to obtain the corresponding heuristic value.

[0089] The target deep neural network refers to the network model used to obtain the optimal path of the decision tree.

[0090] It is worth mentioning that in order to find the optimal path with the least possible number of searches in the tree search of the decision tree, the present invention proposes a DL-METS algorithm improved based on the METS algorithm for search. Furthermore, in order to train the proposed deep neural network based on the DL-METS algorithm, a training method suitable for DL-METS is proposed.

[0091] In an embodiment of the present invention, a decision tree is searched, and a node vector associated with each tree node in the decision tree is input into a target deep neural network, and a heuristic value corresponding to each tree node is output.

[0092] Step 105: Input the heuristic value into the preset path cost model and output the corresponding path cost value.

[0093] The preset path cost model refers to a network model used to obtain the path cost value associated with each tree node.

[0094] In the embodiment of the present invention, the heuristic value corresponding to each tree node is input into a preset path cost model, and the corresponding path cost value is output.

[0095] Step 106: Use the path cost value to search the memory space, mark and input the nodes that meet the preset optimal node conditions, and form the signal to be restored associated with the sent signal.

[0096] The preset optimal node condition refers to the minimum path cost value and the tree height is greater than the preset tree height standard value.

[0097] In an embodiment of the present invention, the nodes in the memory space are grouped according to their tree heights, and the nodes in each group are sorted in ascending order according to their path cost values. The first node in all groups (that is, the node with the smallest path cost value) and which satisfies the tree height not less than the preset tree height standard value is taken as the optimal node and marked as the target node. The nodes included in the minimum complete path from the root node to the lowest target node constitute the estimated signal vector of the transmitted signal, and the estimated signal vector is used as the signal to be recovered.

[0098] Step 107: Perform inverse mapping on the signal to be recovered, generate and output the recovered information bit stream to be sent and index number.

[0099] In an embodiment of the present invention, a dynamic spatial modulation (CM-DSM) algorithm is used at the receiving end to perform inverse mapping on the signal to be recovered, generate and output the recovered information bit stream and index number to be sent, and realize a one-to-one mapping between the information bit stream and the three-dimensional constellation diagram (composed of AACs and the traditional two-dimensional constellation diagram), thereby realizing signal detection and optimal signal selection, thereby ensuring that the device is online in real time.

[0100] In the present invention, in response to a received online maintenance request, a virtual power plant edge gateway corresponding to the online maintenance request is determined and a corresponding information bit stream to be sent is obtained, the information bit stream to be sent is mapped, a corresponding transmission signal is generated, and a corresponding decision tree is constructed according to a channel state matrix of the transmission signal. The node vector associated with each tree node in the decision tree is input into a target deep neural network to obtain a corresponding heuristic value, the heuristic value is input into a preset path cost model, a corresponding path cost value is output, the path cost value is used to search the memory space, and the nodes that meet the preset optimal node conditions are marked and input to form a signal to be recovered associated with the transmission signal, the signal to be recovered is inversely mapped, and the recovered information bit stream to be sent and an index number are generated and output; this solves the problem that in the existing virtual power plant edge gateway system based on the dynamic spatial modulation scheme, due to the dynamic change characteristics of the number of activated antennas, the traditional signal detection and optimal signal selection methods cannot ensure that the virtual power plant edge gateway device can The technical problem of being able to work normally; the present invention first proposes a dynamic spatial modulation DSM (CM-DSM) scheme based on combined mapping (CM), which solves the ambiguity problem of signal detection of virtual power plant edge gateway based on the DSM scheme, and constructs a suitable decision tree for signal detection based on this scheme, thereby converting the problem of finding the maximum likelihood estimate in signal detection into the problem of finding the minimum cost and path in tree search; secondly, the present invention proposes a memory-efficiency balanced tree search (METS) algorithm, which is suitable for signal detection of multi-input and multi-output systems based on the CM-DSM scheme, and the algorithm has the characteristics of low computational complexity and optimal detection performance while being memory-limited; finally, the present invention proposes a METS (DL-METS) algorithm accelerated by deep learning technology, which uses deep learning technology to accelerate the search for the optimal heuristic function, so that it achieves almost optimal detection performance while the number of tree nodes visited is close to the minimum.

[0101] See also Figure 2 , Figure 2 A flowchart of the steps of a method for keeping a virtual power plant edge gateway signal online provided in the second embodiment of the present invention.

[0102] The present invention provides a method for keeping a virtual power plant edge gateway signal online, comprising:

[0103] Step 201: In response to a received online maintenance request, determine the virtual power plant edge gateway corresponding to the online maintenance request and obtain the corresponding information bit stream to be sent.

[0104] In the embodiment of the present invention, the specific implementation process of step 201 is similar to that of step 101 and will not be repeated here.

[0105] Step 202: Map the information bit stream to be transmitted to generate a corresponding transmission signal.

[0106] Furthermore, step 202 may include the following sub-steps:

[0107] S11. Perform base-to-base conversion on the information bit stream to be sent to generate an initial mapped decimal bit stream.

[0108] In the embodiment of the present invention, since the information bit stream to be sent is a binary sequence b i =[b K ,b K-1 ,...,b1], therefore, it is necessary to convert the binary sequence of the information bit stream to be sent into the initial mapping decimal bit stream.

[0109] S12. Using the initial mapping decimal bit stream to input a preset intermediate decimal bit stream function, to generate an intermediate mapping decimal bit stream.

[0110] The preset intermediate decimal bit stream function is specifically:

[0111]

[0112] Where n represents the initial mapped decimal bit stream, |S| i represents the capacity of the i-th constellation set, r represents the intermediate variable, Indicates the Nth a The capacity of the constellation set, k represents the intermediate mapping decimal bit stream, N a Indicates N r The number of a-th activated antennas in the transmitter antenna count, K represents the length of data bits transmitted in each time slot, |X| Na represents the constellation set under quadrature amplitude modulation (QAM) or phase shift keying (PSK) modulation with the number of active antennas a. represents the binomial coefficient, N RF Indicates the number of RF links on the transmitter side.

[0113] S13. Decode the intermediate mapped decimal bit stream to generate corresponding antenna combination data.

[0114] According to the combination sorting, all antenna combinations (AACs) with the same number of activated antennas are sorted in lexicographic order, so that for a certain antenna combination data represented as activated c is a binary value of 0 or 1. This sequence indicates which of the Nt transmitting antennas is the active antenna. This represents the antenna combination, that is, for a total number of transmitting antennas (Nt), when the number of active antennas is Na, there are (Nt, Na) permutations and combinations. By traversing all Na possibilities, all antenna combinations can be obtained. The goal is to arrange antenna combinations with a small number of active antennas (such an arrangement can reduce energy consumption) for encoding the bit stream as much as possible (the remaining unused antenna combinations will have more active antennas than this number). This produces a one-to-one mapping (coding dictionary / table) from the bit stream to the symbols (represented jointly by AAC and the traditional two-dimensional constellation diagram).

[0115] In one example of the present invention, for ease of understanding, the present invention provides a pseudo code of an antenna combination (AAC) decoding algorithm:

[0116] proceduredecode(k);

[0117] Set c = [c o , c1, ..., C n ,...,C N ] is all 0;

[0118] r = k;

[0119] For k=N a to 1 do;

[0120] For n=N c +k-1 to k-1do;

[0121] then;

[0122] C n =1;

[0123]

[0124] break;

[0125] end if;

[0126] end for;

[0127] end for;

[0128] return c;

[0129] end procedure.

[0130] S14. Using the intermediate mapping decimal bit stream to input a preset target decimal bit stream function to generate a target mapping decimal bit stream.

[0131] The preset target decimal bit stream function is as follows:

[0132]

[0133] Where q represents the target mapping decimal bit stream.

[0134] S15. Convert the target mapped decimal bit stream into a base-1 value according to a preset length to generate a target binary sequence.

[0135] In the embodiment of the present invention, the target mapped decimal bit stream is converted into a base according to a preset length to generate a target binary sequence

[0136] S16. Convert the target binary sequence into constellation symbols in a preset two-dimensional constellation diagram.

[0137] In the embodiment of the present invention, the target binary sequence is converted into constellation symbols in a preset two-dimensional constellation diagram. This is a conventional symbol replacement and will not be described in detail.

[0138] S17. Use antenna combination data and constellation symbols to construct a transmission signal s.

[0139] In the embodiment of the present invention, antenna combination data is coupled with constellation symbols to construct a transmission signal.

[0140] In one example of the present invention, for ease of understanding, the present invention provides a pseudo code of an algorithm for mapping an information bit stream into a signal vector:

[0141] Input: The information bit stream to be sent is a binary sequence b i =[b K ,b K-1 ,...,b1], whose data bit length is K;

[0142] Output: send signal s;

[0143] 1: Take the binary sequence b i =[b K ,b K-1 ,...,b1] is converted into the initial mapping decimal bit stream n;

[0144] 2: for N a =1,2,..,N RF do;

[0145] 3: if

[0146] 4:

[0147] 5:

[0148]

[0149] 7: According to the above decoding procedure, the intermediate mapped decimal bit stream k is converted into the corresponding antenna combination data (AAC) 1;

[0150] 8: Convert the intermediate mapping decimal bit stream k into a length of N a The corresponding target binary sequence

[0151] 9: Then, according to the selected traditional constellation diagram, the target binary sequence Convert to the corresponding symbols in the constellation diagram;

[0152] 10: Construct the transmission signal s based on the AAC and symbols calculated above;

[0153] 11: return s;

[0154] 12: endif;

[0155] 13: end for.

[0156] Step 203: receiving the transmitted signal at the receiving end;

[0157] y=Hs+w

[0158] Where y represents the noisy transmitted signal received by the receiver, H represents the channel state matrix, s represents the transmitted signal, and w represents Gaussian white noise.

[0159] In the embodiment of the present invention, since the process of transmitting from the transmitting end to the receiving end may cause signal loss, the transmitted signal received at the receiving end is a transmitted signal containing noise.

[0160] in, represents the completely known channel state matrix, represents additive white Gaussian noise (AWGN) with mean 0 and variance 1. It should be noted that in the MIMO system based on the CM-DSM scheme, the valid constellation set is in is the alphabet of traditional two-dimensional modulation, where “0” is used to identify inactive antennas. In addition, this problem can be further expressed using the active subsystem:

[0161]

[0162] where h j Represents the j-th non-zero element s in H that corresponds to the signal vector s j The corresponding column vector, is a valid subchannel state matrix.

[0163] Mathematically speaking, the optimal way to estimate the signal is:

[0164]

[0165] This method exhaustively searches for the nearest Euclidean point within its search space. The computational complexity of finding the optimal estimate increases exponentially with the size of the problem. To reduce computational complexity, the optimal estimate problem for signal detection and optimal signal selection is decomposed using QR decomposition and converted into a decision tree mathematical model, as shown in step 204:

[0166] Step 204: Perform QR decomposition on the channel state matrix of the transmitted signal, and construct a decision tree based on the orthogonal column vector group matrix in the decomposed channel state matrix;

[0167]

[0168] Where Q1 and Q2 are orthogonal column vector matrix, R is the upper triangular matrix obtained by QR decomposition, N r Indicates the number of antennas at the receiving end, N t Indicates the number of transmitting antennas.

[0169] In the embodiment of the present invention, please refer to Figure 3 , Figure 3 Schematic diagram of MIMO virtual power plant edge gateway. It should be noted that the method proposed in the present invention is based on the MIMO virtual power plant edge gateway of the CM-DSM solution (the system schematic diagram is shown in the accompanying description). Figure 7 As shown in the figure), a decision tree is constructed to detect the signal. The orthogonal column vector matrix Q1 and Q2 obtained by decomposition are used to perform orthogonal transformation on the transmitted signal to obtain the number of layers N. t , a perfect multi-branch decision tree with S branches at each tree node;

[0170] Wherein, S is the modulation signal symbol table used by the system, |S| is the cardinality of the set, and in this embodiment, the mathematical model of the decision tree is:

[0171]

[0172] in, and It is composed of an orthogonal set of column vectors and is an upper triangular matrix. r i,j Represents the (i, j)th component in R counting from the bottom up, the jth increment From the partially known signal vector s k =[s k ,sk-1 ,…,s1] is uniquely determined. The cumulative cost can be expressed as Its recursive form is expressed as g(s k+1 )=g(s k )+b(s k+1 ). In this way, a decision tree corresponding to the signal detection problem is established.

[0173] It is worth mentioning that in the mathematical model, the detection process can be decoupled into a sequential decision process. Starting from the last component of the s vector, the components of the corresponding parts are solved from bottom to top to calculate their Euclidean distance, and the candidate signal with the smallest Euclidean distance is selected as the optimal estimate (equivalent to maximum likelihood estimation).

[0174] It should be noted that by constructing a decision tree, the problem is converted into a minimum cost and path search problem, so that some tree search algorithms such as the A* algorithm can be applied. The METS / DL-METS algorithm used in the present invention is an improved tree search algorithm.

[0175] See also Figure 4 , the relevant parameters are set to: N t =3, N RF =2, binary phase shift keying (BPSK) modulation, solid black nodes represent invalid nodes, solid white nodes represent valid nodes, and the nodes included in the path from the root node to the lowest layer target node constitute the estimated value of the transmitted signal.

[0176] Step 205: Input the node vector associated with each tree node in the decision tree into the target deep neural network to obtain the corresponding heuristic value.

[0177] In an embodiment of the present invention, a deep neural network is set at the signal receiving end to calculate the heuristic value corresponding to the tree node through the deep neural network, wherein the specific calculation formula of the heuristic value is as follows:

[0178]

[0179]

[0180] Where, g(s k ) indicates the sth k The cumulative cost value, h * (s k ) represents the optimal heuristic function, Indicates from s k Nodes to The remaining cost of each node, b(s i ) represents the i-th incremental element in the upper triangular matrix, Indicates the Cumulative cost value, sk represents a tree node in the kth layer of the decision tree, z i represents the transmitted signal after the i-th orthogonal transformation, r i,j Represents the (i, j)th component element in the upper triangular matrix, s j Represents the jth component element.

[0181] It should be noted that the present invention uses the METS or DL-METS algorithm to search:

[0182] For ease of understanding, the following is a pseudo code for a memory-efficiency balanced tree search METS algorithm:

[0183] Input: a completely known channel state matrix H and a noisy transmitted signal y received at the receiving end;

[0184] Output: signal to be restored and the number of nodes visited N v ;

[0185] 1: Initialize search radius R = ∞;

[0186] 2: Initialize the current available memory size to

[0187] 3: The pseudo root s 0 Put it into memory and set

[0188] 4: Set the height of the tree currently being visited to k = 0, and set the next node to be visited to s k =s 0 ;

[0189] 5: Set the number of nodes currently visited to N v =0;

[0190] 6: while k≠N t do;

[0191] 7: Visit s k , update N v =N v +1;

[0192] 8: Find s k All M (M≤|S|) child nodes, these child nodes not only fall on g(s k+1 )≤R, and the index number calculated by the inverse mapping algorithm is also less than 2 K ;

[0193] 9: Remove s from memory k, then add the M child nodes found above to the memory;

[0194] 10: Settings

[0195] 11: if any child node s k+1 is the target node and satisfies k=N t -1 and g(s k+1 )≤Rthen;

[0196] 12: Update search radius R = g(s k+1 );

[0197] 13: Prune the tree using the updated search radius and update the memory space accordingly;

[0198] 14: endif;

[0199] 15: The nodes in the memory space are grouped according to their tree height, and the nodes in each group are grouped according to their cumulative cost value g(s k ) are sorted in ascending order, then only the first node in all these groups and the tree height is not less than Among the nodes, find the best node and record it as s l ;

[0200] 16: Set the next node to be visited to s k =s l ;

[0201] 17: end while;

[0202] 18: return N v ;

[0203] It is worth mentioning that the METS algorithm receives the complete channel state matrix H, the receiving end signal y, the growth factor λ>1.0, and outputs the estimated signal vector Number of nodes visited N v , its memory space is limited, and the g-cost function and the search radius R are used to search for the complete path with the minimum g-cost function to estimate the signal, thereby realizing signal detection.

[0204] For ease of understanding, the following is a pseudo code for the DL-METS algorithm accelerated by deep learning technology:

[0205] Input: A completely known channel state matrix H and a noisy transmit signal y received at the receiver, with a growth factor λ > 1.0:

[0206] Output: signal to be restored and the number of visited nodes N v :

[0207] 1: Initialize the search radius R = f(s 0 |θ);

[0208] 2: Initialize the current available memory space size as

[0209] 3: Put the pseudo-root s 0 into the memory, and set

[0210] 4: Set the current tree layer being visited as k = 0, and set the next node to be visited as s k = s 0 ;

[0211] 5: Set the current number of visited nodes as N v = 0;

[0212] 6: while k ≠ N t do;

[0213] 7: Visit s k , and update N v = N v + 1;

[0214] 8: Find all M (M ≤ |S|) child nodes of s k , which not only fall within the spherical range of g(s k+1 ) ≤ R, but also the index numbers calculated according to the inverse mapping algorithm are less than 2 K ;

[0215] 9: Remove s k from the memory, and then add the above-found M child nodes to the memory;

[0216] 10: Set

[0217] 11: if any child node s k+1 is the target node and at the same time satisfies k = N t - 1 and g(s k+1 ) ≤ R then;

[0218] 12: Update the search radius R = f(s 0 |θ);

[0219] 13: Prune the tree with the updated search radius, and then update the memory space accordingly;

[0220] 14: endif;

[0221] 15: The nodes in the memory space are grouped according to their tree height, and the nodes in each group are grouped according to their cumulative cost value g(s k ) are sorted in ascending order, then only the first node in all these groups and the tree height is not less than Among the nodes, find the best node and record it as s l ;

[0222] 16: if s is found l then;

[0223] 17: Set the node s to be visited next k =s l ;

[0224] 18: else;

[0225] 19: Initialize search radius R = λR;

[0226] 20: Initialize the current available memory space size

[0227] 21: The pseudo root s 0 Add memory space, set

[0228] 22: Set the node s to be visited next k =s 0 ;

[0229] 23: end if;

[0230] 24: end while;

[0231] 25: return N v ;

[0232] It is worth mentioning that the DL-METS algorithm receives the complete channel state matrix H, the receiving end signal y, the growth factor λ>1.0, and outputs the estimated signal vector Number of nodes visited N v , its memory space is limited, and the f-cost function is represented by a deep neural network. The trained f-cost function and the search radius R are used to search for the complete path with the minimum f-cost function to estimate the signal, thereby realizing signal detection and optimal signal selection, thereby ensuring that the device is online in real time.

[0233] Furthermore, the training process of the target deep neural network includes:

[0234] Construct an initial deep neural network with minimizing the likelihood function as the optimization objective;

[0235] Randomly obtain bitstream training data and input it into the initial deep neural network for training to obtain the training path cost values of each tree node;

[0236] Calculate the loss function value corresponding to the bitstream training data according to the training path cost value. When the loss function value is greater than the preset standard loss function value, adjust the network parameters of the initial deep neural network using the mini-batch gradient descent method, and jump to execute the step of randomly obtaining bitstream training data and inputting it into the initial deep neural network to obtain the training path cost values of each tree node until the loss function value is less than or equal to the preset standard loss function value;

[0237] When the loss function value is less than or equal to the preset standard loss function value, output the target deep neural network.

[0238] In the embodiment of the present invention, a bitstream is randomly generated, and then the corresponding channel, noise, and random signal-to-noise ratio are used to simulate the communication process to generate the corresponding received signal. The training set consists of multiple (i.e., all nodes st^0,..., st^N on the entire path t and the received signal z after QR decomposition t , the channel matrix R t ), and at the same time, the mini-batch gradient descent method is adopted to minimize the average loss of the likelihood loss function on the subset D for supervised learning. The likelihood function is The optimization objective is That is, to minimize the mean square error between the cost sum f(S K 丨θ) estimated by the neural network and the true cost sum .

[0239] Furthermore, the expression of the loss function value is specifically:

[0240]

[0241] In the formula, D represents the bitstream training data, R t represents the channel matrix, represents all nodes on the entire path, z t represents the transmitted signal after QR decomposition, t represents the number of time slots, where t = 1, 2, 3,..., T;

[0242]

[0243] In the formula, represents the loss function value, S K represents the k-th node, Denote the N t th node, denote the sum of true costs, f(S K |θ) denote the estimated cost sum of the initial deep neural network, and θ denote the network parameters of the initial deep neural network;

[0244]

[0245] In the formula, θ * denote the optimal network parameters.

[0246] For ease of understanding, the following is the pseudo-code for the training of the neural network of a DL-METS algorithm:

[0247] 1: / / Randomly generate symbols for the training set;

[0248] 2: Set the total number of time slots to T;

[0249] 3: Initialize the data set

[0250] 4: for t = 1, 2,.., T do;

[0251] 5: Randomly generate z with a random signal-to-noise ratio (SNR) t , R t and

[0252] 6:

[0253] 7: / / Generate samples by enumerating the nodes on the transmitted signal GeneratesaMP23034173les byenumerating the nodes onthe transmitted signal;

[0254] 8: for k = 0, 1,.., N t -1 do;

[0255] 9: Select the node belonging to path s t , at the k-th layer

[0256] 10:

[0257] 11: end for;

[0258] 12: end for;

[0259] 13: / / Train the model with the generated samples;

[0260] 14: Randomly initialize network parameters θ;

[0261] 15: loop;

[0262] 16: From the entire dataset D all Randomly select a sample subset D from

[0263] 17: Calculate the average loss on dataset D according to the loss function formula;

[0264] 18: Apply stochastic gradient descent to update the network parameters θ;

[0265] 19: end loop.

[0266] It is worth mentioning that in the Python simulation environment, a computer is used to simulate the bit error rate of the method proposed in the present invention as the signal-to-noise ratio changes. In the simulation experiment, the channel adopts the "Kroneker" model, that is, the channel state matrix is represents the independent and identically distributed Gaussian matrix, R r Represents the covariance matrix of the receiving end, R t Represents the covariance matrix of the transmitter. Channel correlation occurs at the transmitter, so τ∈[0,1] represents the correlation coefficient and when τ=0, it means the channels are uncorrelated. The degree of channel correlation depends on τ. The parameters of the simulation experiment are set as: N r =N t =12, N RF =8, τ =0, quadrature phase shift keying (QPSK) modulation, the deep neural network consists of 4 fully connected layers, the parameters of these network layers are 128, 64, 32 and 16 respectively, and the learning rate is set to 10 during training. -6 The Adam optimizer is used, and the training set contains 10×10 6 Time slot samples and randomly collected training samples with a signal-to-noise ratio between 0dB and 30dB. The algorithms involved in the comparison include:

[0267] (1) METS(L), the memory-efficiency balanced tree search algorithm proposed in this paper, where L represents the available memory space;

[0268] (2) DL-METS-Basic(L,λ), the basic version of the DL-METS algorithm proposed in this paper, where L represents the available memory space and λ is the growth factor. The notation “Basic” indicates that the algorithm only estimates the initial radius and always uses the g-cost function to guide the search process;

[0269] (3) DL-METS(L,λ), which is the full version of DL-METS proposed in this paper. This algorithm not only estimates the initial radius but also uses the f-cost function to guide the search process;

[0270] (4) Maximum Likelihood Estimation (ML): The traditional maximum likelihood detection algorithm, whose bit error rate is regarded as the lower bound of all detection algorithms.

[0271] The simulation results are shown in the accompanying figure. Figure 5 As shown in the figure, after comparing these three methods, the METS algorithm and DL-METS-Basic algorithm included in the method proposed in the present invention accurately achieve the bit error rate performance of the best maximum likelihood estimation algorithm, and the DL-METS algorithm is also very close to the bit error rate performance of the best maximum likelihood estimation algorithm, verifying the effectiveness of this method.

[0272] In the Python simulation environment, a computer was used to simulate the change of the number of visited nodes with the signal-to-noise ratio of the method proposed in the present invention. The settings in the simulation experiment were consistent with the above. The simulation results are shown in the accompanying figure. Figure 6 As shown in the figure, after comparing these four methods, the number of visited nodes for the methods proposed in this invention, including the METS algorithm, the DL-METS-Basic algorithm, and the DL-METS algorithm, converges to its lower bound as the signal-to-noise ratio increases. Compared to brute force enumeration, these three METS-series algorithms require significantly fewer nodes to visit. Furthermore, the DL-METS_Basic algorithm visits fewer nodes than the METS algorithm, and when the signal-to-noise ratio is low, the DL-METS algorithm achieves a more significant reduction in computational complexity than the DL-METS-Basic algorithm. This demonstrates the efficiency and robustness of this method, particularly the effectiveness of using deep learning techniques to accelerate the search for the optimal heuristic function.

[0273] Step 206: Input the heuristic value into the preset path cost model and output the corresponding path cost value.

[0274] Furthermore, the preset path cost model is specifically:

[0275]

[0276]

[0277] Where, f * (s k ) represents the path cost, g(s k ) indicates the sth k The cumulative cost value, h * (s k ) represents the optimal heuristic function, Indicates from s k Nodes to The remaining cost of each node, b(s i ) represents the i-th incremental element in the upper triangular matrix, Indicates the Cumulative cost value, s k represents a tree node in the kth layer of the decision tree, z i represents the transmitted signal after the i-th orthogonal transformation, r i,j Represents the (i, j)th component element in the upper triangular matrix, s j Represents the jth component element.

[0278] Step 207: Use the path cost value to search the memory space, mark and input the nodes that meet the preset optimal node conditions, and form the signal to be restored associated with the sent signal.

[0279] Furthermore, the preset optimal node condition is specifically the minimum path cost value and the tree height is greater than the preset tree height standard value;

[0280] The preset tree height standard values ​​are as follows:

[0281]

[0282] Where, Indicates the preset tree height standard value, N t Indicates the number of transmitting antennas, Indicates memory space, Indicates the capacity of the constellation set.

[0283] In the embodiment of the present invention, the path cost value is used to search the memory space, and the nodes that meet the preset optimal node conditions are marked and input to form the signal to be restored associated with the sent signal.

[0284] Step 208: Perform inverse mapping on the signal to be recovered, generate and output the recovered information bit stream to be sent and index number.

[0285] Furthermore, step 208 may include the following sub-steps:

[0286] S21. Acquire antenna combination data associated with the signal to be recovered.

[0287] In the embodiment of the present invention, the antenna combination data AACc and the number of activated antennas N are obtained from the signal vector of the signal to be recovered. a ;

[0288] S22. Encode the antenna combination data and output the corresponding index number.

[0289] For ease of understanding, the following is a pseudo code of an antenna combination (AAC) coding algorithm:

[0290] 1: procedure code (c);

[0291] 2: Calculate the combined index k based on the encoding formula;

[0292] 3: return k;

[0293] 4: endprocedure;

[0294] The encoding formula is as follows:

[0295]

[0296] Among them, c i ∈{0,1} and This holds true for all n < k.

[0297] In the formula, k represents the index number, k n Represents the sum of the activated antenna combination data, where specifically c0, c1, ..., c n The sum of c n Indicates the nth activated antenna, n indicates the total number of antennas, c i represents the i-th active antenna, where i=1, 2, ..., n.

[0298] For ease of understanding, the following is a mapping table (part):

[0299]

[0300]

[0301] S23. Perform base-based conversion on the signal to be recovered to generate a corresponding initial inverse-mapped binary sequence.

[0302] In the embodiment of the present invention, since the signal to be recovered is in decimal form, it is necessary to perform base conversion on the signal to be recovered to generate the corresponding initial inverse mapping binary sequence.

[0303] S24 , using the initial inverse mapping binary sequence to input a preset target decimal bit stream function to generate a corresponding target inverse mapping decimal bit stream.

[0304] It is worth mentioning that the preset target decimal bit stream function here is consistent with the preset target decimal bit stream function in the mapping process. The only difference is the formula transformation, which is as follows:

[0305]

[0306] In the embodiment of the present invention, the initial inverse mapping binary sequence is input into a preset target decimal bit stream function to generate a corresponding target inverse mapping decimal bit stream. It should be noted that q in this step represents the target inverse mapping decimal bit stream.

[0307] S25 , using the target inverse-mapped decimal bit stream to input a preset intermediate decimal bit stream function, and determining a corresponding initial inverse-mapped decimal bit stream.

[0308] It is worth mentioning that the preset intermediate decimal bit stream function here is consistent with the preset intermediate decimal bit stream function in the mapping process. The only difference is the formula transformation, which is as follows:

[0309]

[0310] In an embodiment of the present invention, a preset intermediate decimal bit stream function is used to input the target inverse-mapped decimal bit stream. The intermediate inverse-mapped decimal bit stream k is first determined, and then converted into an initial inverse-mapped decimal bit stream n. It should be noted that n in this step represents the initial inverse-mapped decimal bit stream.

[0311] S26. Performing base-to-base conversion on the initial inverse-mapped decimal bit stream according to a preset length, generating and outputting a restored information bit stream to be sent and an index number.

[0312] In an embodiment of the present invention, the initial inverse-mapped decimal bit stream is converted into a corresponding binary sequence b of length K, that is, the recovered information bit stream to be sent, and the corresponding index number is output together. The mapping table can be searched according to the index number to determine the accuracy of the recovered information bit stream to be sent.

[0313] For ease of understanding, the following is a pseudo code for an algorithm that inversely maps a signal vector into an information bit stream:

[0314] Input: signal to be restored

[0315] Output: information bit stream b to be sent i =[b K ,b K-1 ,...,b1], whose length is K and the corresponding index number is n;

[0316] 1: Signal to be restored Take AACc and the number of active antennas N;

[0317] 2: Encode AAC c into prime number k according to the encoding formula;

[0318] 3: The signal to be restored N includeda Non-zero symbols are converted into a length of N a The initial inverse mapping binary sequence of log2(|S|);

[0319] 4: Calculate the corresponding target inverse mapping decimal bit stream based on the initial inverse mapping binary sequence obtained above;

[0320] 5:

[0321] 6:

[0322] 7: Convert the initial inverse-mapped decimal bit stream n into the corresponding recovered information bit stream b to be sent in the form of a binary sequence of length K;

[0323] 8: return b and n;

[0324] In the present invention, in response to a received online maintenance request, a virtual power plant edge gateway corresponding to the online maintenance request is determined and a corresponding information bit stream to be sent is obtained, the information bit stream to be sent is mapped, a corresponding transmission signal is generated, and a corresponding decision tree is constructed according to a channel state matrix of the transmission signal. The node vector associated with each tree node in the decision tree is input into a target deep neural network to obtain a corresponding heuristic value, the heuristic value is input into a preset path cost model, a corresponding path cost value is output, the path cost value is used to search the memory space, and the nodes that meet the preset optimal node conditions are marked and input to form a signal to be recovered associated with the transmission signal, the signal to be recovered is inversely mapped, and the recovered information bit stream to be sent and an index number are generated and output; this solves the problem that in the existing virtual power plant edge gateway system based on the dynamic spatial modulation scheme, due to the dynamic change characteristics of the number of activated antennas, the traditional signal detection and optimal signal selection methods cannot ensure that the virtual power plant edge gateway device can The technical problem of being able to work normally; the present invention first proposes a dynamic spatial modulation DSM (CM-DSM) scheme based on combined mapping (CM), which solves the ambiguity problem of signal detection of virtual power plant edge gateway based on the DSM scheme, and constructs a suitable decision tree for signal detection based on this scheme, thereby converting the problem of finding the maximum likelihood estimate in signal detection into the problem of finding the minimum cost and path in tree search; secondly, the present invention proposes a memory-efficiency balanced tree search (METS) algorithm, which is suitable for signal detection of multi-input and multi-output systems based on the CM-DSM scheme, and the algorithm has the characteristics of low computational complexity and optimal detection performance while being memory-limited; finally, the present invention proposes a METS (DL-METS) algorithm accelerated by deep learning technology, which uses deep learning technology to accelerate the search for the optimal heuristic function, so that it achieves almost optimal detection performance while the number of tree nodes visited is close to the minimum.

[0325] See also Figure 7 , Figure 7 This is a structural block diagram of a virtual power plant edge gateway signal online maintenance system provided in Example 3 of the present invention.

[0326] The present invention provides a virtual power plant edge gateway signal online maintenance system, comprising:

[0327] A response module 301 is configured to respond to a received online maintenance request, determine a virtual power plant edge gateway corresponding to the online maintenance request, and obtain a corresponding information bit stream to be sent;

[0328] A mapping module 302 is configured to map the information bit stream to be transmitted and generate a corresponding transmission signal;

[0329] A decision tree construction module 303 is used to construct a corresponding decision tree according to the channel state matrix of the transmitted signal;

[0330] A search module 304 is configured to input the node vector associated with each tree node in the decision tree into the target deep neural network to obtain a corresponding heuristic value;

[0331] The path cost value module 305 is used to input the heuristic value into the preset path cost model and output the corresponding path cost value;

[0332] The signal-to-be-recovered module 306 is configured to search the memory space using the path cost value, mark and input the nodes that meet the preset optimal node conditions, and form the signal-to-be-recovered associated with the transmitted signal;

[0333] The inverse mapping module 307 is configured to perform inverse mapping on the signal to be recovered, and generate and output a recovered information bit stream to be sent and an index number.

[0334] Furthermore, the mapping module 302 includes:

[0335] The initial decimal bit stream submodule is used to convert the bit stream of the information to be sent into a decimal system and generate an initial mapped decimal bit stream;

[0336] A first data processing submodule is configured to input an initial mapped decimal bit stream into a preset intermediate decimal bit stream function to generate an intermediate mapped decimal bit stream;

[0337] A decoding submodule, configured to decode the intermediate mapped decimal bit stream to generate corresponding antenna combination data;

[0338] The second data processing submodule is configured to use the intermediate mapping decimal bit stream as input into a preset target decimal bit stream function to generate a target mapping decimal bit stream;

[0339] A target binary sequence submodule is used to convert the target mapped decimal bit stream into a base-1 value according to a preset length to generate a target binary sequence;

[0340] A constellation symbol submodule is used to convert a target binary sequence into a constellation symbol in a preset two-dimensional constellation diagram;

[0341] The transmitting signal submodule is used to construct a transmitting signal using antenna combination data and constellation symbols.

[0342] Furthermore, the decision tree construction module 303 includes:

[0343] A receiving submodule, configured to receive a transmitted signal at a receiving end;

[0344] y=Hs+w

[0345] Where y represents the noisy signal received by the receiver, H represents the channel state matrix, s represents the transmitted signal, and w represents Gaussian white noise.

[0346] A construction submodule is used to perform QR decomposition on the channel state matrix of the transmitted signal and construct a decision tree based on the orthogonal column vector group matrix in the decomposed channel state matrix;

[0347]

[0348] Where Q1 and Q2 are orthogonal column vector matrix, R is the upper triangular matrix obtained by QR decomposition, N r Indicates the number of antennas at the receiving end, N t Indicates the number of transmitting antennas.

[0349] Furthermore, the training process of the target deep neural network includes:

[0350] Taking minimizing the likelihood function as the optimization goal, construct the initial deep neural network;

[0351] Randomly obtain bitstream training data and input it into the initial deep neural network for training, and obtain the training path cost value of each tree node;

[0352] Calculate the loss function value corresponding to the bitstream training data based on the training path cost value. When the loss function value is greater than the preset standard loss function value, use the mini-batch gradient descent method to adjust the network parameters of the initial deep neural network, and jump to the step of randomly obtaining bitstream training data and inputting it into the initial deep neural network to obtain the training path cost value of each tree node until the loss function value is less than or equal to the preset standard loss function value.

[0353] When the loss function value is less than or equal to the preset standard loss function value, the target deep neural network is output.

[0354] Furthermore, the expression of the loss function value is specifically:

[0355]

[0356] Where D represents the bitstream training data, R t represents the channel matrix, Represents all nodes on the entire path, z t represents the transmitted signal after QR decomposition, t represents the time slot number, where t = 1, 2, 3, ..., T;

[0357]

[0358] Where, represents the loss function, SK represents the k-th node, represents the N t -th node, represents the sum of true costs, f(S K |θ) represents the estimated cost sum of the initial deep neural network, and θ represents the network parameters of the initial deep neural network;

[0359]

[0360] In the formula, θ * represents the optimal network parameters.

[0361] Furthermore, the preset path cost model is specifically:

[0362]

[0363]

[0364] In the formula, f * (s k ) represents the path cost value, g(s k ) represents the cumulative cost value of the s k -th node, h * (s k ) represents the optimal heuristic function, represents the remaining cost value from the s k -th node to -th node, b(s i ) represents the i-th incremental element in the upper triangular matrix, represents the -th cumulative cost value, s k represents a tree node in the k-th layer of the decision tree, z i represents the i-th transmitted signal after orthogonal transformation, r i,j represents the (i, j)-th component element in the upper triangular matrix, s j represents the j-th component element.

[0365] Furthermore, the preset optimal node condition is specifically the minimum path cost value and the tree height is greater than the preset tree height standard value;

[0366] Among them, the preset tree height standard value is specifically:

[0367]

[0368] In the formula, represents the preset tree height standard value, N t represents the number of transmit antennas, represents the memory space, represents the capacity of the constellation set.

[0369] Furthermore, the inverse mapping module 307 includes:

[0370] Antenna combination data submodule, used to obtain antenna combination data associated with the signal to be recovered;

[0371] The preset index number function submodule is used to encode the antenna combination data and output the corresponding index number;

[0372] The initial binary sequence submodule is used to convert the base of the signal to be recovered and generate the corresponding initial inverse mapping binary sequence;

[0373] A third data processing submodule is configured to use an initial inverse mapping binary sequence as input to a preset target decimal bit stream function to generate a corresponding target inverse mapping decimal bit stream;

[0374] a fourth data processing submodule, configured to use a target inverse-mapped decimal bit stream as input to a preset intermediate decimal bit stream function to determine a corresponding initial inverse-mapped decimal bit stream;

[0375] The recovery submodule is used to convert the decimal bit stream of the intermediate inverse mapping according to the preset length, generate and output the recovered information bit stream to be sent and the index number.

[0376] In the present invention, in response to a received online maintenance request, a virtual power plant edge gateway corresponding to the online maintenance request is determined and a corresponding information bit stream to be sent is obtained, the information bit stream to be sent is mapped, a corresponding transmission signal is generated, and a corresponding decision tree is constructed according to a channel state matrix of the transmission signal. The node vector associated with each tree node in the decision tree is input into a target deep neural network to obtain a corresponding heuristic value, the heuristic value is input into a preset path cost model, a corresponding path cost value is output, the path cost value is used to search the memory space, and the nodes that meet the preset optimal node conditions are marked and input to form a signal to be recovered associated with the transmission signal, the signal to be recovered is inversely mapped, and the recovered information bit stream to be sent and an index number are generated and output; this solves the problem that in the existing virtual power plant edge gateway system based on the dynamic spatial modulation scheme, due to the dynamic change characteristics of the number of activated antennas, the traditional signal detection and optimal signal selection methods cannot ensure that the virtual power plant edge gateway device can The technical problem of being able to work normally; the present invention first proposes a dynamic spatial modulation DSM (CM-DSM) scheme based on combined mapping (CM), which solves the ambiguity problem of signal detection of virtual power plant edge gateway based on the DSM scheme, and constructs a suitable decision tree for signal detection based on this scheme, thereby converting the problem of finding the maximum likelihood estimate in signal detection into the problem of finding the minimum cost and path in tree search; secondly, the present invention proposes a memory-efficiency balanced tree search (METS) algorithm, which is suitable for signal detection of multi-input and multi-output systems based on the CM-DSM scheme, and the algorithm has the characteristics of low computational complexity and optimal detection performance while being memory-limited; finally, the present invention proposes a METS (DL-METS) algorithm accelerated by deep learning technology, which uses deep learning technology to accelerate the search for the optimal heuristic function, so that it achieves almost optimal detection performance while the number of tree nodes visited is close to the minimum.

[0377] An electronic device according to an embodiment of the present invention includes: a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes a method for keeping a virtual power plant edge gateway signal online as described in any of the above embodiments.

[0378] The memory can be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read-Only Memory), an EPROM, a hard disk, or a ROM. The memory has storage space for program codes for executing any of the method steps in the above method. For example, the storage space for program codes can include individual program codes for implementing the various steps in the above method. These program codes can be read from or written to one or more computer program products. These computer program products include program code carriers such as a hard disk, a compact disc (CD), a memory card, or a floppy disk. The program code can, for example, be compressed in an appropriate form. When these codes are run by a computing and processing device, they cause the computing and processing device to execute the various steps in the above-described method.

[0379] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed, a method for maintaining a virtual power plant edge gateway signal online as in any embodiment of the present invention is implemented.

[0380] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0381] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of the units is merely a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0382] The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0383] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0384] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0385] As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features thereof can be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for keeping a virtual power plant edge gateway signal online, characterized in that: include: In response to the received online maintenance request, determining the virtual power plant edge gateway corresponding to the online maintenance request and obtaining the corresponding information bit stream to be sent; Mapping the information bit stream to be sent to generate a corresponding transmission signal; Constructing a corresponding decision tree according to the channel state matrix of the transmitted signal; Inputting the node vector associated with each tree node in the decision tree into the target deep neural network to obtain the corresponding heuristic value; Input the heuristic value into a preset path cost model and output the corresponding path cost value; Searching the memory space using the path cost value, marking and inputting nodes that meet the preset optimal node condition to form a signal to be recovered associated with the sent signal; Performing inverse mapping on the signal to be recovered, generating and outputting a recovered information bit stream to be sent and an index number; The training process of the target deep neural network includes: Taking minimizing the likelihood function as the optimization goal, construct the initial deep neural network; Randomly obtain bitstream training data and input it into the initial deep neural network for training to obtain the training path cost value of each tree node; Calculating a loss function value corresponding to the bitstream training data based on the training path cost value; when the loss function value is greater than a preset standard loss function value, adjusting network parameters of the initial deep neural network using a mini-batch gradient descent method; and jumping to the step of randomly acquiring bitstream training data and inputting it into the initial deep neural network for training to obtain a training path cost value for each tree node, until the loss function value is less than or equal to the preset standard loss function value; When the loss function value is less than or equal to the preset standard loss function value, outputting the target deep neural network; The expression of the loss function value is specifically: ； Where, represents the bitstream training data, represents the channel matrix, Represents all nodes on the entire path, represents the transmitted signal after QR decomposition, t represents the time slot number, where t=1, 2, 3, ..., T; ； Where, represents the loss function value, represents the kth node, Indicates the nodes, represents the true cost and, represents the estimated cost and of the initial deep neural network, Representing network parameters of the initial deep neural network; ； Where, represents the optimal network parameters; The preset path cost model is specifically: ； ； ； ； ； Where, represents the path cost, Indicates the The cumulative cost value, represents the optimal heuristic function, Indicates from Nodes to The residual cost of each node is represents the i-th incremental element in the upper triangular matrix, Indicates the The cumulative cost value, represents a tree node in the kth layer of the decision tree, represents the transmitted signal after the i-th orthogonal transformation, represents the (i, j)th component element in the upper triangular matrix, Represents the jth component element.

2. The method for keeping the virtual power plant edge gateway signal online according to claim 1, characterized in that: The step of mapping the information bit stream to be sent to generate a corresponding transmission signal includes: Performing base-based conversion on the information bit stream to be sent to generate an initial mapped decimal bit stream; Using the initial mapped decimal bit stream to input a preset intermediate decimal bit stream function to generate an intermediate mapped decimal bit stream; Decoding the intermediate mapped decimal bit stream to generate corresponding antenna combination data; Using the intermediate mapping decimal bit stream to input a preset target decimal bit stream function to generate a target mapping decimal bit stream; Performing base-to-base conversion on the target mapped decimal bit stream according to a preset length to generate a target binary sequence; Converting the target binary sequence into constellation symbols in a preset two-dimensional constellation diagram; A transmission signal is constructed using the antenna combination data and the constellation symbol.

3. The method for keeping the virtual power plant edge gateway signal online according to claim 1, characterized in that: The step of constructing a corresponding decision tree according to the channel state matrix of the transmitted signal includes: Receiving the transmitted signal at a receiving end; Performing QR decomposition on the channel state matrix of the transmitted signal, and constructing a decision tree according to the orthogonal column vector group matrix in the decomposed channel state matrix.

4. The method for keeping the virtual power plant edge gateway signal online according to claim 1, characterized in that: The preset optimal node condition is specifically the minimum path cost value and the tree height is greater than the preset tree height standard value; The preset tree height standard value is specifically: ； Where, Indicates the preset tree height standard value, Indicates the number of transmitting antennas, represents the memory space, Indicates the capacity of the constellation set.

5. The method for keeping the virtual power plant edge gateway signal online according to claim 1, characterized in that: The step of performing inverse mapping on the signal to be recovered, generating and outputting a recovered information bit stream to be sent and an index number includes: Acquiring antenna combination data associated with the signal to be recovered; Encode the antenna combination data and output a corresponding index number; Performing base-based conversion on the signal to be recovered to generate a corresponding initial inverse mapping binary sequence; Using the initial inverse mapping binary sequence to input a preset target decimal bit stream function to generate a corresponding target inverse mapping decimal bit stream; Using the target inverse-mapped decimal bit stream as input to a preset intermediate decimal bit stream function, to determine a corresponding initial inverse-mapped decimal bit stream; The initial inverse-mapped decimal bit stream is converted into a decimal system according to a preset length, and a restored information bit stream to be sent and the index number are generated and output.

6. A virtual power plant edge gateway signal online maintenance system, characterized in that: include: A response module, configured to respond to a received online maintenance request, determine the virtual power plant edge gateway corresponding to the online maintenance request and obtain a corresponding information bit stream to be sent; A mapping module, configured to map the information bit stream to be sent and generate a corresponding transmission signal; A decision tree construction module, configured to construct a corresponding decision tree according to the channel state matrix of the transmitted signal; A search module, configured to input a node vector associated with each tree node in the decision tree into a target deep neural network to obtain a corresponding heuristic value; A path cost value module, configured to input the heuristic value into a preset path cost model and output a corresponding path cost value; A signal module to be restored, configured to search the memory space using the path cost value, mark and input nodes that meet the preset optimal node conditions, and constitute a signal to be restored associated with the transmitted signal; An inverse mapping module is used to perform inverse mapping on the signal to be recovered, generate and output a recovered information bit stream to be sent and an index number; The training process of the target deep neural network includes: Taking minimizing the likelihood function as the optimization goal, construct the initial deep neural network; Randomly obtain bitstream training data and input it into the initial deep neural network for training to obtain the training path cost value of each tree node; Calculating a loss function value corresponding to the bitstream training data based on the training path cost value; when the loss function value is greater than a preset standard loss function value, adjusting network parameters of the initial deep neural network using a mini-batch gradient descent method; and jumping to the step of randomly acquiring bitstream training data and inputting it into the initial deep neural network for training to obtain a training path cost value for each tree node, until the loss function value is less than or equal to the preset standard loss function value; When the loss function value is less than or equal to the preset standard loss function value, outputting the target deep neural network; The expression of the loss function value is specifically: ； Where, represents the bitstream training data, represents the channel matrix, Represents all nodes on the entire path, represents the transmitted signal after QR decomposition, t represents the time slot number, where t=1, 2, 3, ..., T; ； Where, represents the loss function value, represents the kth node, Indicates the nodes, represents the true cost and, represents the estimated cost and of the initial deep neural network, Representing network parameters of the initial deep neural network; ； Where, represents the optimal network parameters; The preset path cost model is specifically: ； ； ； ； ； Where, represents the path cost, Indicates the The cumulative cost value, represents the optimal heuristic function, Indicates from Nodes to The residual cost of each node is represents the i-th incremental element in the upper triangular matrix, Indicates the The cumulative cost value, represents a tree node in the kth layer of the decision tree, represents the transmitted signal after the i-th orthogonal transformation, represents the (i, j)th component element in the upper triangular matrix, Represents the jth component element.

7. An electronic device, characterized in that: It includes a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the virtual power plant edge gateway signal online maintenance method as described in any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed, the method for keeping the virtual power plant edge gateway signal online is implemented as described in any one of claims 1 to 5.

Images