Implementation, application methods, devices, equipment, media, and products of prediction models

Abstract

This disclosure provides an implementation, application method, apparatus, device, medium, and product of a prediction model, relating to the field of communication technology. The implementation method of the network state prediction model includes: collecting training samples based on historical records corresponding to the network state, and inputting the training samples into an MPO network for modeling processing to obtain MPO branch sample vectors; converting the training samples into QNN branch sample vectors based on quantum state mapping; generating a fused feature vector based on the MPO branch sample vectors and the QNN branch sample vectors; inputting the fused feature vector into the network state prediction model for training; and deploying the trained network state prediction model on a server or terminal. Through the embodiments of this disclosure, the bottleneck of simultaneously extracting periodic and abrupt features is overcome, significantly reducing the model parameter size and inference overhead, providing stronger model expressive power and generalization ability, balancing prediction accuracy and efficiency, and improving robustness and scalability.

Description

Technical Field

[0001] This disclosure relates to the field of communication technology, and more specifically, to the implementation, application method, apparatus, device, medium, and product of a prediction model. Background Technology

[0002] In 5G wireless communication, a variety of prediction tasks with high-dimensional features and complex nonlinear correlations are often generated. For example, in typical scenarios such as cell-level fault detection, alarm prediction, and performance degradation diagnosis, the feature data in the above prediction tasks usually exhibit the characteristics of local mutation and periodicity coexisting. Moreover, the dimensionality is high and the data structure is complex. Traditional methods are difficult to simultaneously balance the accuracy and computational efficiency of the above prediction tasks.

[0003] It should be noted that the information disclosed in the background section above is only used to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0004] The purpose of this disclosure is to provide an implementation, application method, apparatus, device, medium, and product of a prediction model, which can at least to some extent overcome the problems of poor reliability of prediction tasks in communication scenarios caused by the limitations and defects of related technologies.

[0005] According to a first aspect of the present disclosure, a method for implementing a network state prediction model is provided, comprising: collecting training samples based on historical records corresponding to network states, and inputting the training samples into an MPO network for modeling processing to obtain MPO branch sample vectors; mapping the training samples into quantum states through a quantum neural network, and inputting the quantum states into a quantum circuit for processing to obtain QNN branch sample vectors; generating a fused feature vector based on the MPO branch sample vectors and the QNN branch sample vectors; inputting the fused feature vectors into the network state prediction model for training; and deploying the trained network state prediction model on a server or terminal.

[0006] In one exemplary embodiment of this disclosure, training samples are collected based on historical records corresponding to network states, and the training samples are input into the MPO network for modeling processing to obtain MPO branch sample vectors, including:

[0007] Feature vectors are extracted based on historical records corresponding to network states, and the feature vectors are converted into d-order tensors;

[0008] The d-th order tensor is input into the MPO network to obtain d core tensors output by the MPO network. The MPO network is used to decompose the weight matrix into d core tensors.

[0009] The core tensor is subjected to chain shrinkage calculation to obtain the MPO branch sample vector.

[0010] In one exemplary embodiment of this disclosure, mapping training samples to quantum states via a quantum neural network and inputting the quantum states into a quantum circuit for processing to obtain a QNN branch sample vector includes:

[0011] Training samples are mapped to quantum states using a quantum neural network, and these quantum states are then input into a quantum circuit for processing to obtain QNN branch sample vectors, including:

[0012] The training samples are mapped to quantum states through a quantum neural network; the quantum states are then input into a quantum circuit constructed by alternating evolution of encoded unitary operators and variational unitary operators to obtain the QNN branch sample vector.

[0013] In one exemplary embodiment of this disclosure, generating a fused feature vector based on the MPO branch sample vector and the QNN branch sample vector includes:

[0014] The MPO branch sample vector and the QNN branch sample vector are concatenated or weighted to generate the fused feature vector.

[0015] In one exemplary embodiment of this disclosure, inputting the fused feature vector into the network state prediction model for training includes:

[0016] The loss function of the network state prediction model is constructed based on the loss function corresponding to the MPO branch sample vector, the loss function corresponding to the QNN branch sample vector, and the loss function corresponding to the fused feature vector.

[0017] Based on the loss function of the network state prediction model, update the parameters corresponding to the core tensor and / or the parameters corresponding to the variable quantum circuit until the training of the network state prediction model is completed.

[0018] In one exemplary embodiment of this disclosure, updating the parameters corresponding to the core tensor and / or the parameters corresponding to the variable quantum circuit according to the loss function of the network state prediction model until the training of the network state prediction model is completed includes:

[0019] The gradient of the core tensor is obtained by calculating the gradient corresponding to the parameters of the core tensor through classical backpropagation.

[0020] The gradients corresponding to the parameters of the variable quantum circuit are estimated using the full-parameter migration method to obtain the quantum gradients;

[0021] A hybrid gradient flow is constructed based on the core tensor gradient and the quantum gradient;

[0022] Based on the loss function of the network state prediction model, update the parameters of the core tensor in the hybrid gradient flow and the parameters of the variable quantum circuit until the training of the network state prediction model is completed.

[0023] In one exemplary embodiment of this disclosure, deploying the trained network state prediction model on a server or terminal includes:

[0024] The trained network state prediction model is evaluated using test samples to obtain evaluation metrics, which include at least one of accuracy, precision, recall and F1 score.

[0025] In response to the evaluation index meeting the preset evaluation index range, the network state prediction model is deployed on the server or terminal.

[0026] In one exemplary embodiment of this disclosure, it further includes:

[0027] In response to the evaluation index not meeting the preset evaluation index range, the network state prediction model undergoes parameter adjustment and / or retraining.

[0028] The parameter adjustment process includes at least one of the following: changing the encoding method, adjusting the depth of the variable quantum circuit, increasing the MPO rank, adding a regularization term, using an attention mechanism for feature fusion, and optimizing the weights of the loss function.

[0029] According to a second aspect of the present disclosure, a method for applying a network state prediction model is provided, comprising:

[0030] Collect network status data indicators of the communication network;

[0031] The network status data indicators are converted into feature data;

[0032] The feature data is processed using a network state prediction model, which is obtained through the implementation method of the network state prediction model as described in any of the above technical solutions.

[0033] According to a third aspect of the present disclosure, an apparatus for implementing a network state prediction model is provided, comprising:

[0034] The acquisition module is configured to collect training samples based on the historical records corresponding to the network state, and input the training samples into the MPO network for modeling processing to obtain the MPO branch sample vector;

[0035] The conversion module is configured to map training samples to quantum states through a quantum neural network, and input the quantum states to a quantum circuit for processing to obtain QNN branch sample vectors;

[0036] The generation module is configured to generate a fused feature vector based on the MPO branch sample vector and the QNN branch sample vector.

[0037] The training module is configured to input the fused feature vector into the network state prediction model for training.

[0038] The deployment module is configured to deploy the trained network state prediction model on a server or terminal.

[0039] According to a fourth aspect of the present disclosure, a method for applying a network state prediction model is provided, comprising:

[0040] The data acquisition module is configured to collect network status data indicators of the communication network.

[0041] The conversion module is configured to convert the network status data indicators into feature data.

[0042] The prediction module is configured to process the feature data using a network state prediction model, wherein the network state prediction model is obtained through the implementation method of the network state prediction model as described in any of the above technical solutions.

[0043] According to a fifth aspect of this disclosure, an electronic device is provided, comprising: a memory; and a processor coupled to the memory, the processor being configured to perform the method as described in any one of the preceding methods based on instructions stored in the memory.

[0044] According to a sixth aspect of this disclosure, a computer-readable storage medium is provided having a program stored thereon that, when executed by a processor, implements the method as described in any of the preceding claims.

[0045] According to a seventh aspect of this disclosure, a computer program product is provided, comprising a computer program, characterized in that the computer program, when executed by a processor, implements the method as described in any of the preceding claims.

[0046] In this embodiment, training samples are collected based on historical records corresponding to network states, and these training samples are input into an MPO network for modeling processing to obtain MPO branch sample vectors. This not only reduces the computational load through tensor decomposition but also improves the accuracy of extracting local mutation features. It avoids the tendency of high-dimensional spaces to get trapped in local optima. Furthermore, the training samples are mapped to quantum states through a quantum neural network, and the quantum states are input into a quantum circuit for processing to obtain QNN branch sample vectors. The QNN branch sample vectors can fully reflect mutation features.

[0047] Furthermore, generating a fused feature vector based on the MPO branch sample vector and the QNN branch sample vector not only achieves efficient global optimization but also takes into account the impact of mutation features on the prediction results.

[0048] Based on this, the fused feature vector is input into the network state prediction model for training. Finally, the trained network state prediction model is deployed on a server or terminal, which balances the accuracy and computational efficiency of network state prediction. It can be extended to prediction tasks with high-dimensional and complex data structures in various scenarios, thus improving the robustness and scalability of network state prediction.

[0049] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure. Attached Figure Description

[0050] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this disclosure and, together with the description, serve to explain the principles of this disclosure. It is obvious that the drawings described below are merely some embodiments of this disclosure, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0051] Figure 1 This is a flowchart of an implementation method for a network state prediction model in an exemplary embodiment of this disclosure;

[0052] Figure 2 This is a flowchart of an implementation method for a network state prediction model in an exemplary embodiment of this disclosure;

[0053] Figure 3 This is a flowchart of another implementation method of a network state prediction model in an exemplary embodiment of this disclosure;

[0054] Figure 4 This is a flowchart of another implementation method of a network state prediction model in an exemplary embodiment of this disclosure;

[0055] Figure 5 This is a flowchart of another implementation method of a network state prediction model in an exemplary embodiment of this disclosure;

[0056] Figure 6 This is a flowchart of another implementation method of a network state prediction model in an exemplary embodiment of this disclosure;

[0057] Figure 7 This is a flowchart of another implementation method of a network state prediction model in an exemplary embodiment of this disclosure;

[0058] Figure 8 This is a flowchart of another implementation method of a network state prediction model in an exemplary embodiment of this disclosure;

[0059] Figure 9 This is a flowchart of another implementation method of a network state prediction model in an exemplary embodiment of this disclosure;

[0060] Figure 10 This is a flowchart of an application method of a network state prediction model in an exemplary embodiment of this disclosure;

[0061] Figure 11 This is a flowchart of an implementation scheme of a network state prediction model in an exemplary embodiment of this disclosure;

[0062] Figure 12 This is a block diagram of an implementation apparatus for a network state prediction model in an exemplary embodiment of the present disclosure;

[0063] Figure 13 This is a block diagram of an application apparatus for a network state prediction model in an exemplary embodiment of the present disclosure;

[0064] Figure 14 This is a block diagram of an electronic device according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0065] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this disclosure more comprehensive and complete, and to fully convey the concept of the example embodiments to those skilled in the art. The described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a full understanding of embodiments of this disclosure. However, those skilled in the art will recognize that the technical solutions of this disclosure can be practiced with one or more of the specific details omitted, or other methods, components, apparatus, steps, etc., can be employed. In other instances, well-known technical solutions are not shown or described in detail to avoid obscuring various aspects of this disclosure.

[0066] Furthermore, the accompanying drawings are merely illustrative of this disclosure, and the same reference numerals in the drawings denote the same or similar parts, thus repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0067] Figure 1 A schematic diagram of an exemplary system architecture for implementing a network state prediction model that can be applied to embodiments of the present disclosure is shown.

[0068] like Figure 1 As shown, system architecture 100 may include one or more of terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0069] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, there can be any number of terminal devices, networks, and servers. For example, server 105 could be a server cluster composed of multiple servers.

[0070] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Terminal devices 101, 102, and 103 can be various electronic devices with displays, including but not limited to smartphones, tablets, laptops, and desktop computers, etc.

[0071] In some embodiments, the implementation method of the network state prediction model provided in this disclosure is generally executed by server 105, and correspondingly, the implementation device of the network state prediction model is generally located in terminal device 103 (or terminal device 101 or 102). In other embodiments, some terminals may have functions similar to those of the server device to execute this method.

[0072] The exemplary embodiments of this disclosure will now be described in detail with reference to the accompanying drawings.

[0073] Figure 2 This is a flowchart of a method for implementing a network state prediction model in an exemplary embodiment of this disclosure.

[0074] refer to Figure 2 The implementation methods of network state prediction models can include:

[0075] Step S202: Collect training samples based on the historical records corresponding to the network state, and input the training samples into the MPO network for modeling processing to obtain the MPO branch sample vector.

[0076] In one embodiment of this disclosure, a dataset containing multiple training samples is obtained, denoted as... .

[0077] Each of them A one-dimensional vector representing a training sample can represent multiple performance metrics of a communication cell within the i-th time period, such as PRB utilization, call drop rate, and handover failure rate. There are n eigenvalues, where n is a positive integer equal to or equal to 2.

[0078] In one embodiment of this disclosure, the training samples with missing or outlier values ​​are preprocessed (e.g., normalized, interpolated, standardized, etc.), and a label set corresponding to the training samples is constructed. 0 represents normal and 1 represents fault, thus obtaining feature-label training sample pairs. .

[0079] In one embodiment of this disclosure, the matrix product operator (MPO) provides an effective means for compressed representation of high-dimensional features and modeling of local strong dependencies. Its low-rank tensor chain structure can significantly reduce the number of parameters while maintaining model accuracy.

[0080] Step S204: The training samples are mapped to quantum states through a quantum neural network, and the quantum states are input to a quantum circuit for processing to obtain QNN branch sample vectors.

[0081] In one embodiment of this disclosure, a one-dimensional vector This is mapped to quantum states that can be processed by quantum computers.

[0082] In one embodiment of this disclosure, depending on different encoding strategies, angle encoding or amplitude encoding can be used for quantum state initialization.

[0083] In one embodiment of this disclosure, if the encoding strategy employs angle encoding, then one qubit is assigned to each feature in the input vector, requiring a total of N qubits, and a parameterized rotation gate (e.g.) is used. Eigenvalues ​​are embedded in quantum states.

[0084] In one embodiment of this disclosure, if the encoding strategy employs amplitude encoding, then the entire bit vector is embedded in the amplitude coefficients of the quantum state, requiring a total of Each quantum bit carries N-dimensional information.

[0085] In one embodiment of this disclosure, a quantum neural network (QNN) constructs a nonlinear high-dimensional mapping space through the superposition and entanglement of quantum states, which has a powerful feature representation capability and can break through the local optimum limitation in the optimization process of classical neural networks.

[0086] Step S206: Generate a fused feature vector based on the MPO branch sample vector and the QNN branch sample vector.

[0087] In one embodiment of this disclosure, the MPO branch output vector is... (Including local mutation structural tensor features) and QNN branch output vector (Including periodic nonlinear quantum features) are spliced ​​or weighted to construct a fused feature vector y. fuse The expression is shown in the following formula (I):

[0088] ,or Formula (I).

[0089] Here, Concat indicates concatenation, and α and β are the preset weights for the weighted combination.

[0090] Step S208: Input the fused feature vector into the network state prediction model for training.

[0091] In one embodiment of this disclosure, the MPO branch sample vector and the QNN branch sample vector are used for local mutation feature compression extraction and periodic high-dimensional feature modeling, respectively. QNN constructs a nonlinear feature expression space through quantum superposition and entanglement, effectively breaking through the optimization bottleneck of classical neural networks in non-convex spaces. MPO achieves parameter compression with a low-rank tensor chain structure, which is not only suitable for local strong correlation scenarios under high-dimensional input, but also has better scalability and computational efficiency.

[0092] Step S210: Deploy the trained network state prediction model on a server or terminal.

[0093] In this embodiment, training samples are collected based on historical records corresponding to network states, and these training samples are input into an MPO network for modeling processing to obtain MPO branch sample vectors. This not only reduces the computational load through tensor decomposition but also improves the accuracy of extracting local mutation features. It avoids the tendency of high-dimensional spaces to get trapped in local optima. Furthermore, the training samples are mapped to quantum states through a quantum neural network, and the quantum states are input into a quantum circuit for processing to obtain QNN branch sample vectors. The QNN branch sample vectors can fully reflect mutation features.

[0094] Furthermore, generating a fused feature vector based on the MPO branch sample vector and the QNN branch sample vector not only achieves efficient global optimization but also takes into account the impact of mutation features on the prediction results.

[0095] Based on this, the fused feature vector is input into the network state prediction model for training. Finally, the trained network state prediction model is deployed on a server or terminal, which balances the accuracy and computational efficiency of network state prediction. It can be extended to prediction tasks with high-dimensional and complex data structures in various scenarios, thus improving the robustness and scalability of network state prediction.

[0096] The following section details each step of the implementation method for the network state prediction model.

[0097] In one exemplary embodiment of this disclosure, in Figure 2 Based on what is shown, as Figure 3 As shown, training samples are collected based on historical records corresponding to network states, and these training samples are input into the MPO network for modeling processing to obtain the MPO branch sample vectors, including:

[0098] Step S302: Extract feature vectors based on historical records corresponding to network states, and convert the feature vectors into d-order tensors.

[0099] In one embodiment of this disclosure, a one-dimensional vector Convert to d-order tensor To better represent the local feature structure of the original data explicitly, the target dimension of the transformation is... , This represents the size of the i-th dimension.

[0100] In one embodiment of this disclosure, it is assumed that If the length is 128, it can be converted into a 4th order quantum tensor (i.e., d=4), then the target dimension can be 2×4×4×4=128, and the product of the target dimensions must be equal to the length of the original vector. The converted d-order tensor is input into a tensor network based on MPO (MPO can reduce the amount of computation through tensor decomposition, and is particularly suitable for extracting local mutation features) to extract high-order local structural features from the input data.

[0101] Step S304: Input the d-order tensor into the MPO network to obtain d core tensors output by the MPO network. The MPO network is used to decompose the weight matrix into d core tensors.

[0102] In one embodiment of this disclosure, if the dimension of the k-th core tensor is Rk, 1 < k < d, then The possible values ​​are from 1 to Rk. It means to take each Multiply the results of the calculations together This represents a four-dimensional tensor element of the k-th core tensor, whose value is determined by the input dimension. Output Dimensions The connection dimension between the two core tensors. , The decision is made to use the entire weight matrix. The expression formed by combining d core tensors is shown in Formula (II):

[0103] Formula (II).

[0104] Step S306: Perform chain shrinkage calculation on the core tensor to obtain the MPO branch sample vector.

[0105] In one embodiment of this disclosure, the MPO will use a weight matrix Decomposed into d core tensors The combined representation is used to compress the number of parameters, reduce computational complexity, and utilize the core tensor. The final output vector is calculated using chain contraction. .

[0106] in, This represents the local mutation features extracted from the MPO branch. It is the product of the output dimensions. It is the product of the input dimensions. express It is a matrix with input dimension × output dimension. This represents the k-th core tensor (out of a total of d). and The virtual dimension representing the connection between tensors is used to control the information transfer capacity between each core tensor, i.e., the size of each core tensor. Choosing an appropriate R value is the key to controlling the performance of MPO modeling. Connecting these core tensors together can restore the entire weight matrix.

[0107] Additionally, the input index is the weight matrix. The output index is Location Represents the internal connectivity dimension of MPO Summation, It is an index representing an element in the virtual dimension of the current core tensor.

[0108] In one exemplary embodiment of this disclosure, in Figure 2 Based on what is shown, as Figure 4 As shown, training samples are mapped to quantum states through a quantum neural network, and these quantum states are input into a quantum circuit for processing to obtain QNN branch sample vectors, including:

[0109] Step S402: Map the training samples to quantum states using a quantum neural network.

[0110] Step S404: The quantum state is input into a quantum circuit constructed by alternating evolution of encoded unitary operators and variational unitary operators to obtain the QNN branch sample vector.

[0111] In one embodiment of this disclosure, the initial state of the quantum system is set to a zero quantum state, i.e., a zero quantum state, based on the quantum state mapping. ,in, This represents the required number of qubits. Further, a VQC (Variable Quantum Circuit) is constructed based on the initialization. The VQC consists of two alternating parts:

[0112] (1) Encoding the unitary operator Embedding input data into a quantum state is specifically implemented as a single-bit rotation gate in angle encoding. Or the controlled gate sequence in amplitude encoding, the specific method depends on the selected encoding method, such as rotating door, controlled door, etc.;

[0113] (2) Variational unitary operator : Includes parameters Tunable quantum gates are used to learn the nonlinear mapping between input and output, employing different parameter vectors in each layer. Furthermore, the inter-layer door structure can be designed in a heterogeneous manner.

[0114] In one embodiment of this disclosure, the variable quantum circuit is constructed in the following ways:

[0115] Encode the unitary operator With variational unitary operator Alternating stacking forms an L-layer structure, wherein:

[0116] The encoded unitary operator The input vector is processed through a rotating gate or a controlled gate. Embedded quantum state;

[0117] The variational unitary operator Includes adjustable parameters It is used to learn the nonlinear mapping of input and output.

[0118] In one embodiment of this disclosure, the "encoding" is repeated L times. and "transformation" The operation involves using the same input for each layer, but with different parameters. Controlling the variational unitary operations forms the final variational quantum state. The expression is shown in formula (III):

[0119] Formula (III).

[0120] In one embodiment of this disclosure, a one-dimensional input vector The initialization steps for mapping to quantum states include: employing either angle encoding or amplitude encoding strategies, where angle encoding assigns qubits to each feature and passes through parameterized rotation gates. Embedding eigenvalues ​​requires N qubits; amplitude encoding embeds vectors into the quantum state amplitude coefficients, requiring log_N qubits; initializing the quantum system to a completely zero state. , denoted as the number of qubits.

[0121] In one embodiment of this disclosure, the step of constructing a variable quantum circuit is: alternately stacking L layers of encoded unitary operators. With variational unitary operator Generate the final variable quantum state ,in Choose between a revolving door or a controlled door based on the coding strategy. For those with adjustable parameters Quantum gates.

[0122] In one embodiment of this disclosure, after construction, the quantum state is measured to extract classical information for subsequent classification and regression tasks, and a set of observable operators is selected. Generally, it is the tensor product of Pauli operators, which yields an estimate of the expected value. The expression is shown in formula (IV):

[0123] Formula (IV).

[0124] In one embodiment of this disclosure, multiple measurements are performed in a practical quantum circuit. The process involves repeatedly executing the circuit and statistically analyzing the measurement results. The expected value is the weighted average of these possible outcomes, ultimately yielding the output vector. The output of the QNN branch, i.e., the input sample The nonlinear and periodic quantum features extracted after quantum variational processing are used for subsequent fusion layer, discrimination module and loss function calculation.

[0125] In one exemplary embodiment of this disclosure, in Figure 2 Based on what is shown, as Figure 5 As shown, generating the fused feature vector based on the MPO branch sample vector and the QNN branch sample vector includes:

[0126] Step S502: The MPO branch sample vector and the QNN branch sample vector are concatenated or weighted to generate the fused feature vector.

[0127] In one embodiment of this disclosure, the fused feature vector is input to a classical discriminant module (such as an MLP or Softmax classifier) ​​to complete a binary classification task (normal / fault) prediction as shown in formula (V):

[0128] Formula (5).

[0129] in, It is a weight matrix. It is a bias vector used to adjust the model output, making it more flexible. It can be either the sigmoid or softmax function, outputting the prediction result. This represents the predicted fault state of the current input sample. To enhance the model's expressive power, the fusion layer can also introduce an attention mechanism to dynamically weight the feature components of MPO and QNN.

[0130] In one exemplary embodiment of this disclosure, in Figure 2 Based on what is shown, as Figure 6 As shown, training the network state prediction model by inputting the fused feature vector includes:

[0131] Step S602: Construct the loss function of the network state prediction model based on the loss function corresponding to the MPO branch sample vector, the loss function corresponding to the QNN branch sample vector, and the loss function corresponding to the fused feature vector.

[0132] Step S604: Update the parameters corresponding to the core tensor and / or the parameters corresponding to the variable quantum circuit according to the loss function of the network state prediction model until the training of the network state prediction model is completed.

[0133] In one embodiment of this disclosure, the loss function Loss for constructing the network state prediction model is shown in Equation (VI):

[0134] Formula (VI).

[0135] in, Output for MPO branch With tags The differences between them Output for QNN branches With tags The differences between them To merge the output results Cross-entropy loss between labels, hyperparameters The contribution of each branch's loss to the total loss can be optimized through grid search.

[0136] In one exemplary embodiment of this disclosure, in Figure 2 Based on what is shown, as Figure 7 As shown, updating the parameters corresponding to the core tensor and / or the parameters corresponding to the variable quantum circuit according to the loss function of the network state prediction model until the training of the network state prediction model is completed includes:

[0137] Step S702: Calculate the gradient corresponding to the parameters of the core tensor through classical backpropagation to obtain the core tensor gradient.

[0138] Step S704: The gradients corresponding to the parameters of the variable quantum circuit are estimated using the full-parameter migration method to obtain the quantum gradient.

[0139] Step S706: Construct a hybrid gradient flow based on the core tensor gradient and the quantum gradient.

[0140] Step S708: Update the parameters of the core tensor in the hybrid gradient flow and the parameters of the variable quantum circuit according to the loss function of the network state prediction model until the training of the network state prediction model is completed.

[0141] In one embodiment of this disclosure, during training, the gradients of the core tensor parameters of the MPO branch are calculated using classical backpropagation, and the parameters of the VQC (variable quantum circuit) in the QNN are... Quantum gradient estimation is performed using the full-parameter transfer method. The two methods are combined to form a hybrid gradient flow, which is uniformly updated by a classical optimizer (such as Adam) for end-to-end training.

[0142] In one exemplary embodiment of this disclosure, in Figure 2 Based on what is shown, as Figure 8 As shown, deploying the trained network state prediction model on a server or terminal includes:

[0143] Step S802: The trained network state prediction model is evaluated using test samples to obtain evaluation metrics, which include at least one of accuracy, precision, recall and F1 score.

[0144] Step S804: In response to the evaluation index meeting the preset evaluation index range, the network state prediction model is deployed on the server or terminal.

[0145] In one embodiment of this disclosure, the trained model is evaluated using a test set of real fault samples that do not overlap with the training set. Evaluation metrics include, but are not limited to, accuracy, precision, recall, and F1 score.

[0146] In one exemplary embodiment of this disclosure, in Figure 2Based on what is shown, as Figure 9 As shown, the implementation methods of the network state prediction model also include:

[0147] Step S902: In response to the evaluation index not meeting the preset evaluation index range, the network state prediction model is subjected to parameter adjustment and / or retraining.

[0148] In one exemplary embodiment of this disclosure, the parameter adjustment process includes at least one of changing the encoding method, adjusting the depth of the variable quantum circuit, increasing the MPO rank, adding a regularization term, using an attention mechanism for feature fusion, and optimizing the weights of the loss function.

[0149] In one embodiment of this disclosure, if the performance does not meet the requirements, module adjustments and training improvements are required, including but not limited to: changing the encoding method, adjusting the VQC (variable quantum circuit) depth, increasing the MPO rank, adding regularization terms to prevent overfitting, adjusting the fusion method to an attention mechanism, optimizing the loss function weights, etc., and re-evaluating until the performance requirements are met.

[0150] In one embodiment of this disclosure, the above method is deployed and connected to the 5G wireless cell network management data stream. The online inference performance is observed and tested. The MPO branch can use the Tensor Network Computation Library to implement matrix multiplication operations, accelerating tensor decomposition. The QNN branch uses a quantum computing cloud service or local simulator that supports VQC (variable quantum circuits) to improve the inference capability of complex periodic features. The fusion layer can use quantization or pruning techniques to reduce the number of parameters and computational latency, improving the system response speed and ultimately completing the technical integration in the existing communication network.

[0151] Figure 10 This is a flowchart of an exemplary embodiment of the present disclosure, showing the application method of a network state prediction model.

[0152] refer to Figure 10 Application methods of network state prediction models include:

[0153] Step S1002: Collect network status data indicators of the communication network;

[0154] Step S1004: Convert the network status data indicators into feature data;

[0155] Step S1006: The feature data is processed using a network state prediction model, which is obtained by the implementation method of the network state prediction model as described in any of the above technical solutions.

[0156] The following examples illustrate the application scenarios of the network state prediction model.

[0157] Step 1: Feature data modeling: Select cell operation data collected from the 5G wireless access network in City A over 30 consecutive days, with a sampling period of 15 minutes.

[0158] In one embodiment of this disclosure, the following eight metrics are selected as features: PRB utilization, uplink interference power, RRC connection establishment failure rate, eNB handover failure rate, downlink packet loss rate, user disconnection rate, number of user context releases, and data throughput, but are not limited to these.

[0159] In one embodiment of this disclosure, the data for each time slice is organized into an 8-dimensional vector. The total number of samples is 20,000.

[0160] In one embodiment of this disclosure, Z-score normalization is used to process features of each dimension, and corresponding tag sets are generated by statistically analyzing cell equipment alarm records. , where 0 represents normal and 1 represents a fault.

[0161] In one embodiment of this disclosure, the ratio of positive to negative samples is approximately 1:4, and the data is subsequently balanced using the SMOTE oversampling method.

[0162] Step 2: MPO branch modeling: To adapt to the MPO structure, the original 8-dimensional vector is zero-padded and extended to a length of 16.

[0163] In one embodiment of this disclosure, it is reshaped into a 4th-order tensor. .

[0164] The MPO weight tensor is set as follows:

[0165] (1) MPO order d=4;

[0166] (2) Each output dimension Input dimension ;

[0167] (3) MPO rank setting That is, the core tensor dimension is The connection dimension of the first and last kernel tensors is 1;

[0168] (4) The output vector dimension is ;

[0169] (5) The MPO network uses layer-by-layer tensor contraction to calculate the product of all core tensors and finally outputs a high-order tensor local feature representation for fusion.

[0170] Step 3: QNN branch modeling.

[0171] In one embodiment of this disclosure, angle coding is used to assign one qubit to each feature, so this embodiment uses eight qubits to construct a quantum circuit.

[0172] In one embodiment of this disclosure, each bit is initialized to Input encoding uses The revolving door will be used.

[0173] In one embodiment of this disclosure, a 3-layer VQC (Variable Quantum Circuit) structure (L=3) is constructed, each layer including:

[0174] (1) Data encoding layer Apply to each quantum bit Encoding gate.

[0175] (2) Variable stratification An entangled state is constructed by connecting all bits using CNOT gates, and then parameterized parameters are added to each bit. and Gates (48 variational parameters in total).

[0176] In one embodiment of this disclosure, the observation operation selects the following four Pauli tensor product operators:

[0177]

[0178]

[0179]

[0180]

[0181] In one embodiment of this disclosure, each quantum circuit performs 1024 measurements, calculates the expected value of each observable, and obtains an output vector. .

[0182] Step 4: Fusion layer calculation and fault prediction output.

[0183] In one embodiment of this disclosure, the QNN output vector and the MPO output vector are concatenated to form a fused feature vector as shown in formula (VII):

[0184] Formula (VII).

[0185] In one embodiment of this disclosure, the input is fed into a fully connected MLP containing two hidden layers (64 neurons per layer, with ReLU activation function), and the last layer uses the sigmoid function for binary classification output as shown in Equation (VIII):

[0186] Formula (8).

[0187] Step 5: Loss function definition and training optimization.

[0188] In one embodiment of this disclosure, the total loss function is defined as shown in formula (IX):

[0189] Formula (IX).

[0190] In one embodiment of this disclosure, the hyperparameters are set as follows: MSE is the mean squared error, and BCE is the binary cross-entropy.

[0191] In one embodiment of this disclosure, the training uses the Adam optimizer with an initial learning rate of 0.001, a batch size of 64, and 100 training rounds.

[0192] In one embodiment of this disclosure, the gradient is calculated using a parameter shifting method in the QNN branch to update the variational parameters.

[0193] Step 6: Result verification and performance evaluation.

[0194] In one embodiment of this disclosure, a test set containing 3,000 records was used for validation, and the model ultimately achieved the following performance: accuracy of 94.3%, precision of 91.7%, recall of 92.5%, and F1 score of 0.921.

[0195] In one embodiment of this disclosure, ablation experiments show that: after removing the QNN branch, the F1 score decreased to 87.2%; after removing the MPO branch, the F1 score decreased to 88.6%; after replacing QNN with a classic MLP, the F1 score was 89.5%; and after replacing MPO with a 1D CNN, the F1 score was 89.1%. The dual-branch structure proposed in this embodiment has significant advantages in handling periodic and abrupt failure modes.

[0196] Step 7: Deployment and Optimization.

[0197] In one embodiment of this disclosure, the network state prediction model can be deployed in an enterprise 5G operation and maintenance system. The MPO branch is implemented using PyTorch and TensorNetwork libraries, and the QNN branch is implemented using the Tianyan Quantum Computing Cloud Platform simulator (supporting 8 qubits, with configurable noise model).

[0198] In one embodiment of this disclosure, the fusion layer is deployed to the ARM edge computing node after quantization and pruning using the ONNX model, and it is expected to complete a cell state prediction within 100ms.

[0199] Figure 11 This is a flowchart of an implementation scheme of a network state prediction model in an exemplary embodiment of this disclosure.

[0200] refer to Figure 11 An implementation scheme of a network state prediction model in an exemplary embodiment of this disclosure may include:

[0201] Step S1102: Obtain the 5G cell network status dataset, preprocess the missing values, construct a label set, and obtain feature-label training sample pairs.

[0202] Step S1104: MPO branch modeling, extracting local mutation features, and obtaining the input vector. QNN branch modeling is used to extract nonlinear and periodic features to obtain the output vector. .

[0203] Step S1106: Construct a fused feature vector and input it into the discrimination module to complete the prediction task.

[0204] Step S1108: Construct the total loss function and perform training optimization.

[0205] Step S1110: Use the test set to perform performance evaluation. If the performance evaluation meets the preset requirements, proceed to step S1112. If the performance evaluation does not meet the preset requirements, proceed to step S1104.

[0206] Step S1112: Deploy and test online capabilities.

[0207] Corresponding to the above method embodiments, this disclosure also provides an implementation apparatus for a network state prediction model, which can be used to execute the above method embodiments.

[0208] Figure 12 This is a block diagram of an implementation apparatus for a network state prediction model in an exemplary embodiment of the present disclosure.

[0209] refer to Figure 12 The network state prediction model implementation device 1200 may include:

[0210] The acquisition module 1202 is configured to acquire training samples based on the historical records corresponding to the network state, and input the training samples into the MPO network for modeling processing to obtain the MPO branch sample vector;

[0211] The conversion module 1204 is configured to map training samples to quantum states through a quantum neural network, and input the quantum states to a quantum circuit for processing to obtain QNN branch sample vectors;

[0212] The generation module 1206 is configured to generate a fused feature vector based on the MPO branch sample vector and the QNN branch sample vector;

[0213] Training module 1208 is configured to input the fused feature vector into the network state prediction model for training;

[0214] Deployment module 1210 is configured to deploy the trained network state prediction model on a server or terminal.

[0215] In one exemplary embodiment of this disclosure, the acquisition module 1202 is further configured to:

[0216] Feature vectors are extracted based on historical records corresponding to network states, and the feature vectors are converted into d-order tensors;

[0217] The d-th order tensor is input into the MPO network to obtain d core tensors output by the MPO network. The MPO network is used to decompose the weight matrix into d core tensors.

[0218] The core tensor is subjected to chain shrinkage calculation to obtain the MPO branch sample vector.

[0219] In one exemplary embodiment of this disclosure, the conversion module 1204 is further configured to:

[0220] The training samples are mapped to the quantum states using a quantum neural network;

[0221] The quantum state is input into a quantum circuit constructed by alternating evolution of encoded unitary operators and variational unitary operators to obtain the QNN branch sample vector.

[0222] In one exemplary embodiment of this disclosure, the generation module 1206 is further configured to:

[0223] The MPO branch sample vector and the QNN branch sample vector are concatenated or weighted to generate the fused feature vector.

[0224] In one exemplary embodiment of this disclosure, the training module 1208 is further configured as follows:

[0225] The loss function of the network state prediction model is constructed based on the loss function corresponding to the MPO branch sample vector, the loss function corresponding to the QNN branch sample vector, and the loss function corresponding to the fused feature vector.

[0226] Based on the loss function of the network state prediction model, update the parameters corresponding to the core tensor and / or the parameters corresponding to the variable quantum circuit until the training of the network state prediction model is completed.

[0227] In one exemplary embodiment of this disclosure, the training module 1208 is further configured as follows:

[0228] The gradient of the core tensor is obtained by calculating the gradient corresponding to the parameters of the core tensor through classical backpropagation.

[0229] The gradients corresponding to the parameters of the variable quantum circuit are estimated using the full-parameter migration method to obtain the quantum gradients;

[0230] A hybrid gradient flow is constructed based on the core tensor gradient and the quantum gradient;

[0231] Based on the loss function of the network state prediction model, update the parameters of the core tensor in the hybrid gradient flow and the parameters of the variable quantum circuit until the training of the network state prediction model is completed.

[0232] In one exemplary embodiment of this disclosure, the deployment module 1210 is further configured to:

[0233] The trained network state prediction model is evaluated using test samples to obtain evaluation metrics, which include at least one of accuracy, precision, recall and F1 score.

[0234] In response to the evaluation index meeting the preset evaluation index range, the network state prediction model is deployed on the server or terminal.

[0235] In one exemplary embodiment of this disclosure, the deployment module 1210 is further configured to:

[0236] In response to the evaluation index not meeting the preset evaluation index range, the network state prediction model undergoes parameter adjustment and / or retraining.

[0237] The parameter adjustment process includes at least one of the following: changing the encoding method, adjusting the depth of the variable quantum circuit, increasing the MPO rank, adding a regularization term, using an attention mechanism for feature fusion, and optimizing the weights of the loss function.

[0238] Figure 13 This is a block diagram of an implementation apparatus for a network state prediction model in an exemplary embodiment of the present disclosure.

[0239] refer to Figure 13 The application device 1300 for the network state prediction model may include:

[0240] The acquisition module 1302 is configured to acquire network status data indicators of the communication network;

[0241] The conversion module 1304 is configured to convert the network status data indicators into feature data;

[0242] The prediction module 1306 is configured to process the feature data using a network state prediction model, wherein the network state prediction model is obtained by the implementation method of the network state prediction model as described in any of the above technical solutions.

[0243] Since the functions of the network state prediction model implementation device 1200 and the network state prediction model application device 1300 have been described in detail in their respective method embodiments, they will not be repeated here.

[0244] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to embodiments of this disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0245] In an exemplary embodiment of this disclosure, an electronic device capable of implementing the above-described method is also provided.

[0246] Those skilled in the art will understand that various aspects of this disclosure can be implemented as a system, method, or program product. Therefore, various aspects of this disclosure can be specifically implemented in the following forms: a completely hardware implementation, a completely software implementation (including firmware, microcode, etc.), or a combination of hardware and software aspects, collectively referred to herein as a "circuit," "module," or "system."

[0247] The following reference Figure 14 To describe an electronic device 1400 according to such an embodiment of the present disclosure. Figure 14 The electronic device 1400 shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments disclosed herein.

[0248] like Figure 14 As shown, the electronic device 1400 is manifested in the form of a general-purpose computing device. The components of the electronic device 1400 may include, but are not limited to: at least one processing unit 1410, at least one storage unit 1420, and a bus 1430 connecting different system components (including storage unit 1420 and processing unit 1410).

[0249] The storage unit stores program code that can be executed by the processing unit 1410, causing the processing unit 1410 to perform the steps described in the "Exemplary Methods" section of this specification according to various exemplary embodiments of this disclosure. For example, the processing unit 1410 can perform the method shown in the embodiments of this disclosure.

[0250] Storage unit 1420 may include readable media in the form of volatile storage units, such as random access memory (RAM) 14201 and / or cache 14202, and may further include read-only memory (ROM) 14203.

[0251] Storage unit 1420 may also include a program / utility 14204 having a set (at least one) of program modules 14205, such program modules 14205 including but not limited to: operating system, one or more application programs, other program modules and program data, each or some combination of these examples may include an implementation of a network environment.

[0252] Bus 1430 can represent one or more of several types of bus structures, including a memory cell bus or memory cell controller, a peripheral bus, a graphics acceleration port, a processing unit, or a local bus using any of the various bus structures.

[0253] Electronic device 1400 can also communicate with one or more external devices 1440 (e.g., keyboard, pointing device, Bluetooth device, etc.), one or more devices that enable a user to interact with electronic device 1400, and / or any device that enables electronic device 1400 to communicate with one or more other computing devices (e.g., router, modem, etc.). This communication can be performed via input / output (I / O) interface 1450. Furthermore, electronic device 1400 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 1460. As shown, network adapter 1460 communicates with other modules of electronic device 1400 via bus 1430. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 1400, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0254] Compared with existing optimization methods, the innovations of the embodiments of this disclosure include, but are not limited to:

[0255] 1. Overcoming the challenge of simultaneously and efficiently modeling periodic and abrupt features: Existing fault identification methods often suffer from insufficient feature extraction and decreased accuracy when faced with coexisting periodic fluctuations and instantaneous abrupt signals due to their singular modeling architecture. This disclosure proposes a parallel QNN and MPO-based tensor network dual-branch structure to model periodic high-dimensional patterns and local abrupt signals respectively, achieving synergistic capture of periodic and non-stationary features, significantly enhancing the model's adaptability and robustness to complex fault modes;

[0256] 2. Significantly reduce model parameter size and inference overhead, providing stronger model expressive power and generalization performance: Compared with the parameter redundancy problem of traditional deep neural networks, this disclosure introduces a quantum heuristic tensor network based on MPO, which performs structural compression and feature representation of input data through a low-rank tensor chain structure, significantly reducing the number of model parameters and improving inference speed, alleviating computing power pressure, and is suitable for real-time tasks in large-scale 5G wireless network environments.

[0257] 3. A parallel quantum-classical hybrid network architecture integrating QNN and MPO-based quantum heuristic tensor network is proposed for the first time: This disclosure breaks through the limitation of the single path of traditional fault identification model and integrates QNN and MPO-based quantum heuristic tensor decomposition in parallel for the first time. They are used to process periodic global features and abrupt local features respectively, and a dual decoupling model of feature dimension and structural dimension is established, which greatly improves the identification accuracy and stability.

[0258] 4. Innovatively introduce the concept of tensor decomposition from quantum mechanics to achieve parameter compression modeling of high-dimensional inputs: By drawing on the idea of ​​tensor state decomposition in quantum mechanics, classical inputs are represented through a low-rank MPO chain structure, which enables the originally high-dimensional network inputs to be effectively compressed into a compact tensor form while maintaining key local dependencies, resulting in higher scalability and training efficiency.

[0259] 5. It does not rely on powerful quantum computing resources; QNN can be implemented using NISQ-era quantum computers or quantum computing cloud platforms.

[0260] From the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, terminal device, or network device, etc.) to execute the methods according to the embodiments of this disclosure.

[0261] In exemplary embodiments of this disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the methods described above is stored. In some possible implementations, various aspects of this disclosure may also be implemented as a program product including program code that, when the program product is run on a terminal device, causes the terminal device to perform the steps of the various exemplary embodiments of this disclosure described in the "Exemplary Methods" section above.

[0262] The program product for implementing the above-described method according to embodiments of the present disclosure may employ a portable compact disc read-only memory (CD-ROM) and include program code, and may run on a terminal device, such as a personal computer. However, the program product of the present disclosure is not limited thereto. In this document, the readable storage medium may be any tangible medium containing or storing a program that may be used by or in conjunction with an instruction execution system, apparatus, or device.

[0263] The readable medium can be a readable signal medium or a readable storage medium. A readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of readable storage media (a non-exhaustive list) include: an electrical connection having one or more wires, a portable disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof.

[0264] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A readable signal medium may also be any readable medium other than a readable storage medium, capable of sending, propagating, or transmitting programs for use by or in conjunction with an instruction execution system, apparatus, or device.

[0265] The program code contained on the readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof.

[0266] In exemplary embodiments of this disclosure, a computer program product is also provided. This computer program product can be loaded or stored on any combination of one or more readable media. It can be written in any combination of one or more programming languages ​​to perform the operations of this disclosure. The programming languages ​​include object-oriented programming languages—such as Java, C++, etc.—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on a user's computing device, partially on a user's computing device, as a standalone software package, partially on a user's computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving remote computing devices, the remote computing device can be connected to the user's computing device via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computing device (e.g., via the Internet using an Internet service provider).

[0267] Furthermore, the above figures are merely illustrative of the processes included in the method according to exemplary embodiments of this disclosure and are not intended to be limiting. It is readily understood that the processes shown in the above figures do not indicate or limit the temporal order of these processes. Additionally, it is readily understood that these processes may be executed synchronously or asynchronously, for example, in multiple modules.

[0268] Other embodiments of this disclosure will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this disclosure that follow the general principles of this disclosure and include common knowledge or customary techniques in the art not disclosed herein. The specification and examples are to be considered exemplary only, and the true scope and concept of this disclosure are indicated by the claims.

Claims

1. A method for implementing a network state prediction model, the method comprising: The method comprises the following steps: collecting training samples based on historical records corresponding to network states, and inputting the training samples into an MPO network for modeling processing to obtain an MPO branch sample vector; mapping the training samples into quantum states through a quantum neural network, and inputting the quantum states into a quantum circuit for processing to obtain a QNN branch sample vector; generating a fusion feature vector according to the MPO branch sample vector and the QNN branch sample vector; inputting the fusion feature vector into a network state prediction model for training; deploying the trained network state prediction model on a server or terminal.

2. The method of claim 1, wherein, Collecting training samples based on historical records corresponding to network states, and inputting the training samples into an MPO network for modeling processing to obtain an MPO branch sample vector comprises: extracting a feature vector based on historical records corresponding to network states, and converting the feature vector into a d-order tensor; inputting the d-order tensor into the MPO network to obtain d core tensors output by the MPO network, the MPO network being used to decompose a weight matrix into d core tensors; performing chain contraction calculation on the core tensors to obtain the MPO branch sample vector.

3. The method of claim 1, wherein, Mapping the training samples into quantum states through a quantum neural network, and inputting the quantum states into a quantum circuit for processing to obtain a QNN branch sample vector comprises: mapping the training samples into the quantum states through the quantum neural network; inputting the quantum states into a quantum circuit constructed by alternately evolving an encoding unitary operator and a variational unitary operator for processing to obtain the QNN branch sample vector.

4. The method of claim 1, wherein, Generating a fusion feature vector according to the MPO branch sample vector and the QNN branch sample vector comprises: splicing or weighted combining the MPO branch sample vector and the QNN branch sample vector to generate the fusion feature vector.

5. The method of claim 2, wherein the network state prediction model is implemented by a neural network. Inputting the fusion feature vector into a network state prediction model for training comprises: constructing a loss function of the network state prediction model according to a loss function corresponding to the MPO branch sample vector, a loss function corresponding to the QNN branch sample vector, and a loss function corresponding to the fusion feature vector; updating parameters corresponding to the core tensors and / or parameters corresponding to the variational quantum circuit according to the loss function of the network state prediction model until the training of the network state prediction model is completed.

6. The method of claim 5, wherein, Updating the parameters corresponding to the core tensors and / or the parameters corresponding to the variational quantum circuit according to the loss function of the network state prediction model until the training of the network state prediction model is completed comprises: calculating the gradient of the parameters of the core tensors through classical backpropagation to obtain a core tensor gradient; estimating the gradient of the parameters of the variational quantum circuit using a full-parameter transfer method to obtain a quantum gradient; constructing a hybrid gradient flow according to the core tensor gradient and the quantum gradient; updating the parameters of the core tensors and the parameters of the variational quantum circuit in the hybrid gradient flow according to the loss function of the network state prediction model until the training of the network state prediction model is completed.

7. The method of claim 1, wherein, The trained network state prediction model is deployed on a server or a terminal, including: The trained network state prediction model is evaluated by using test samples to obtain evaluation indexes, including at least one of accuracy, precision, recall, and F1 score; In response to the evaluation indexes meeting a preset evaluation index range, the network state prediction model is deployed on the server or the terminal.

8. The method of claim 7, wherein, Further comprising: In response to the evaluation indexes not meeting the preset evaluation index range, the network state prediction model is subjected to parameter adjustment processing and / or retraining, The parameter adjustment processing includes at least one of changing an encoding mode, adjusting a depth of a variational quantum circuit, increasing an MPO rank, adding a regularization term, using an attention mechanism for feature fusion, and optimizing a weight of a loss function.

9. A method for applying a network state prediction model, characterized by, Including: Collecting network state data indexes of a communication network; Converting the network state data indexes into feature data; Using a network state prediction model to process the feature data, the network state prediction model being obtained by the implementation method of the network state prediction model in any one of claims 1-8.

10. An apparatus for implementing a network state prediction model, characterized in that, Including: A collection module configured to collect training samples based on historical records corresponding to network states, and input the training samples into an MPO network for modeling processing to obtain MPO branch sample vectors; A conversion module configured to map the training samples into quantum states by using a quantum neural network, and input the quantum states into a quantum circuit for processing to obtain QNN branch sample vectors; A generation module configured to generate a fusion feature vector according to the MPO branch sample vectors and the QNN branch sample vectors; A training module configured to input the fusion feature vector into a network state prediction model for training; A deployment module configured to deploy the trained network state prediction model on a server or a terminal.

11. An application device for a network state prediction model, characterized in that, Including: A collection module configured to collect network state data indexes of a communication network; A conversion module configured to convert the network state data indexes into feature data; A prediction module configured to use a network state prediction model to process the feature data, the network state prediction model being obtained by the implementation method of the network state prediction model in any one of claims 1-8.

12. An electronic device, comprising: Including: A memory; And A processor coupled to the memory, the processor being configured to execute the implementation method of the network state prediction model in any one of claims 1-8 and / or the application method of the network state prediction model in claim 9 based on instructions stored in the memory.

13. A computer-readable storage medium having a program stored thereon, the program being executed by a processor to implement the implementation method of the network state prediction model in any one of claims 1-8 and / or the application method of the network state prediction model in claim 9.

14. A computer program product comprising a computer program, characterized in that, The computer program is executed by a processor to implement the implementation method of the network state prediction model in any one of claims 1-8 and / or the application method of the network state prediction model in claim 9.

Images