Quantum echo state network model construction method for aero-engine fault early warning

Abstract

A kind of quantum echo state network model construction method for aero-engine fault early warning, first, the time series data of aero-engine is quantized;Then, quantum state data is used as the input of the first layer quantum network, after the action of quantum recursion circuit, the probabilistic output is obtained by quantum measurement, and then is transmitted to the reserve pool of the second layer echo state network to obtain a high-dimensional intermediate state;Finally, the output layer weight matrix can be obtained by least square method only once calculation, and the network training is completed.The trained quantum echo state network model is applied to aero-engine fault prediction, and the engine operation state prediction data with high accuracy at future time is given.The present application uses quantum controlled non-gate to realize entangled state, and approximately represents the coupling relationship that may exist between original input data, and quantum rotation gate is set as adjustable parameter of the first layer quantum network, so that the distribution of data can be adjusted;And echo state network layer can play the advantages of fast calculation, and one-step training can complete the calculation of final model.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of fault prediction of aero-engines, and a quantum echo state network model is designed by introducing a quantum computer mechanism to predict multiple operating state parameters of an aero-engine at a future time. BACKGROUND

[0002] With the development of the aerospace industry, the safety and stability of an aircraft in an operating state have become an important research topic. As the heart of an aircraft, an aero-engine is the most likely core component to fail, and once a failure occurs, it will at least affect the performance of the aircraft and at most cause the aircraft to crash and people to die. Therefore, it is necessary to predict the type of failure that may occur in the engine in advance and take appropriate protective measures to ensure the reliable operation of the aircraft. Whether the operating state of an aero-engine at a future time can be accurately and efficiently predicted is crucial to the safety and stability of the entire flight process. Among the many failure problems of an aero-engine, surge failure as a most common dangerous factor will directly affect the performance of the aero-engine and pose a great threat to the safety of the aircraft. An effective method to solve this problem is to propose a data-driven model for the past surge failure data of the aero-engine to predict whether the engine will fail at a future time, play an alarm role and take appropriate safety measures in time.

[0003] At present, there are several methods to predict the failure of an aero-engine, including:

[0004] 1) Fault tree analysis method.

[0005] This method mainly simulates real-time data in the operation process of an aircraft from the perspective of fault monitoring, analyzes and calculates the monitored data by using the fault tree method, and then judges the detailed causes of the failure. However, this fault tree analysis method requires that the personnel analyzing the failure must fully understand the object system being analyzed and be familiar with the application of the analysis method, which leads to different analysis personnel giving different fault tree analysis results. In addition, the calculation process of the fault tree is very complex, and it is difficult to accurately calculate the specific causes of the failure.

[0006] 2) Least squares support vector machine analysis method.

[0007] The method is to use the gas path parameters of the aero-engine to establish a least squares support vector machine model to monitor the state of the aero-engine. That is, according to the established model, the low-pressure rotor speed (n1), high-pressure rotor speed (n2) and tail nozzle outlet temperature (T6) and other parameters of the aero-engine are monitored, and the relative error rate of the predicted value and the true value is used to analyze the fault. However, it should be noted that the least squares support vector machine treats the data set as a vector pattern, which ignores the natural relationship of mutual coupling and mutual influence between data. In addition, if the data set is forcibly represented as a vector, the time sequence correlation of the original data will be destroyed, which will inevitably cause a large numerical error.

[0008] Based on the above discussion, a new quantum echo state network model for aero-engine fault prediction is proposed based on data-driven approach. The model mainly includes two network layers, the output of the first layer quantum heuristic neural network is transmitted to the reserve pool of the second layer echo state network, only one training is needed, and the final quantum echo state network model can effectively predict the running state data of the aero-engine.

[0009] This patent is supported by China Postdoctoral Science Foundation (2022TQ0179), National Natural Science Foundation of China (61890920, 61890921) and National Key R&D Program (2018YFB1700102). SUMMARY

[0010] The purpose of the present application is to design a quantum echo state network model which can be applied to the fault prediction of aero-engine. Since the aero-engine is a highly complex aerodynamic-thermal-mechanical system, the time series data generated by it has strong time sequence correlation, coupling and multi-modal characteristics, therefore, how to predict the fault of the aero-engine in the variable full envelope environment has always been a challenging problem.

[0011] In order to achieve the above purpose, the technical scheme adopted by the present application is:

[0012] The application discloses a quantum echo state network model for aero-engine fault early warning, and applies quantum computing theory to design a quantum circuit composed of a quantum rotation gate and a two-bit quantum controlled non-gate, to construct a first-layer quantum heuristic neural network as a front network of a second-layer echo state network. First, time sequence data of the aero-engine is quantized; then, quantum state data obtained is taken as input of the first-layer quantum network, and after the effect of a quantum recursion circuit, probability output is obtained through quantum measurement, and is transmitted to a reserve pool of the second-layer echo state network to obtain a high-dimensional intermediate state; finally, an output layer weight matrix is obtained through one-time calculation by using a least square method, and thus network training is completed. The trained quantum echo state network model can be applied to aero-engine fault prediction, and high-accuracy engine operation state prediction data at a future time is given. The specific steps are as follows:

[0013] Step 1: selecting data sample features

[0014] First, a plurality of sensor groups of the aero-engine are used to collect a plurality of operation state data of the aircraft, each kind of data represents a sample feature variable, and the data form is a group of discrete time sequence data, and the time sequence is recorded as t={1, 2, 3,..., T}, wherein t is a sampling time, and T is the last sampling time; then, data sample feature variables with high correlation degrees with aero-engine fault occurrence are selected from the plurality of operation state data as judgment criteria for fault prediction, such as environmental pressure, tail nozzle outlet temperature, engine combustion chamber temperature and the like. For example, after n data sample feature variables are selected, n*T initial data {x i (t)} are collected through the sensors, wherein i is from 1 to n, t is from 1 to T, and x i (t) represents a value of the i-th data sample feature variable (such as the tail nozzle outlet temperature) at the t time.

[0015] Step 2: initial data quantization

[0016] According to the quantum computing theory, the input of the quantum circuit layer must be a quantum state, so the initial data {x i (t)} collected in step 1 is subjected to quantumization processing to obtain quantum state form data |ψ>=a|0>+b|1>, wherein |ψ> represents a certain quantum state, |0>, |1> represent two kinds of base states of a binary quantum bit respectively, and a, b are corresponding probability amplitudes, and the physical meaning is that: when the quantum state |ψ> is subjected to quantum measurement, |0> is observed with a probability of |a| 2 , |1> is observed with a probability of |b| 2 , and |a| 2 +|b| 2 =1.

[0017] A set of n-dimensional initial data [x1(t), x2(t),..., x n (t)] Τ is collected at a moment t, the initial data {x i (t)} is first normalized according to formula (1), and formula (1) is as follows:

[0018]

[0019] Wherein, x i (t) and x respectively represent the initial data value and the data value after normalization of the i-th feature variable at the moment t, the value range of the normalized x belongs to [0, 1]; Min represents the minimum value taken by the i-th feature variable in the set of time series, j represents the moment corresponding to the minimum value, Max represents the maximum value taken by the i-th feature variable in the set of time series, k represents the moment corresponding to the maximum value, i = 1, 2,..., n represents that i takes from 1 to n, that is, the time series data of n feature variables are all normalized.

[0020] After normalizing the initial data, considering that the probability amplitude in front of the quantum bit |1> is used to represent the classical data information contained in the input quantum state, the normalized data can be prepared into quantum state form data according to the quantization rule given in formula (2), and formula (2) is as follows:

[0021]

[0022] Wherein, |x i (t)> represents the quantum state data prepared by the i-th feature variable at the moment t, cosθ i (t) and sinθ i (t) respectively correspond to the probability amplitude in front of the binary quantum bit ground state |0> and |1> of the quantum state |x i (t)>, and θ i (t) is obtained by taking the inverse sine of .

[0023] Using formula (2), i takes from 1 to n, and thus the quantum state data |x1(t)>, |x2(t)>,..., |x n (t)> at the moment t is obtained.

[0024] Step 3: Constructing a quantum circuit

[0025] The key to the advantage of the quantum heuristic neural network is how to construct the corresponding quantum circuit, and here a kind of quantum circuit similar to the classical neural network is given. Figure 1The quantum circuit construction method in this paper only uses two basic types of quantum rotation gates and two-qubit quantum controlled NOT gates, as described in detail below: First, n+1 quantum circuits are constructed. The input of the first n circuits (from top to bottom) at the initial time t=1 is the n quantum state data |x1(1)>,|x2(1)>,...,|x n (1)>;The (n+1)th quantum circuit is mainly used to calculate the output of the quantum circuit layer. Its input is an auxiliary quantum state |y(0)>, and its initial state is assumed to be |y(0)>=|0>. The first n quantum states are |x1(1)>,|x2(1)>,...,|x n (1)> passes through a quantum rotation gate in sequence, and then together with |y(0)> passes through n two-bit quantum controlled NOT gates from top to bottom. Note that a quantum rotation gate is also added to the (n+1)th quantum circuit to finally obtain |y(1)>. Then, |x1(2)>,|x2(2)>,...,|x n (2)>,|y(1)> are used as the inputs of the n+1 quantum states at the next time t=2, and the recursive calculation is performed until t=T.

[0026] In the above quantum circuit, the quantum rotation gate through which each quantum state input passes is denoted as R(θ), and its matrix form is:

[0027]

[0028] Where θ is the rotation angle, and its value ranges from [0, 2π].

[0029] The two-bit quantum controlled NOT gate that acts from top to bottom between every two adjacent quantum states is denoted as U. CN Its matrix form is as follows:

[0030]

[0031] Note that a total of n+1 quantum rotating gates are used, so n+1 rotation angles need to be given in advance. in As parameters of the quantum network layer. At time t, the input of the i-th quantum state |x i (t)> Through the quantum rotating gate The calculation formula is:

[0032]

[0033] The specific calculation process of the above quantum circuit is given below: Starting from time t=1, the above quantum circuit is used for recursive calculation. The input quantum states (from top to bottom) at time t=1 are |x1(t1)>,|x2(t1)>,...,|x n(t1)>,|y(t0)>, after one action of the entire quantum circuit, the quantum state |y(1)> is used as the input of the auxiliary quantum state position at the next moment.

[0034] Considering the effect of the quantum circuit at time t = r, the first n input quantum states pass through n quantum rotation gates, then through n quantum controlled NOT gates from top to bottom, and finally, another quantum rotation gate is applied at the location of the auxiliary quantum state. The formula for calculating the output quantum state |y(t = r)> is:

[0035]

[0036] in, Let y(r) represent the probability that the final quantum state at the i-th position from top to bottom in the quantum circuit is |1>, where i ranges from 1 to n; 2 This represents the probability that the quantum state obtained after the auxiliary quantum state position in the last row passes through the controlled NOT gate will take |1> when measured. Note that after the controlled NOT gate, the last row of quantum circuitry also undergoes a rotation gate. Therefore, the final output measured at time t=r is denoted as u(t). r ), which represents the probability of obtaining |1> by measuring |y(t=r)>. The calculation result is shown in formula (7):

[0037]

[0038] By recursively calculating according to formulas (6) and (7), and letting r range from 1 to T, we can obtain a set of time series outputs u(t) = [u(t1), u(t2), ..., u(t3)]. T )] Τ Next, we will take a new set of quantum rotating door rotation angle parameters. Using the same quantum circuit, a new set of outputs can be obtained. By randomly generating n sets of n+1 rotation angle parameters, n sets of T-dimensional output data can be obtained, denoted as matrix U(t)=[u1(t),u2(t),...,u n (t)].

[0039] Step 4: Build and train the quantum echo state network

[0040] As a novel recurrent neural network, the Echo State Network (ESN) is widely used in time series prediction due to its simple computational model (no need to solve for the gradient of the objective function, only a single linear regression is required for network training) and high stability. The ESN consists of three parts: an input layer, a reservoir, and an output layer. Its core structure is a randomly generated and immutable reservoir.

[0041] The specific construction method of the quantum echo state network in this invention is as follows:Figure 1 The output U(t) of the step 3 quantum network layer is taken as the input of the next layer of the echo state network, and the reservoir of the echo state network is designed as a sparse network with multiple neurons, including a high-dimensional sparse internal weight connection matrix W res and an input layer connection matrix W in for connecting U(t) res and W in are randomly generated and remain unchanged during the cyclic calculation of the echo state network. In this way, when the low-dimensional data U(t) obtained by the quantum layer enters the echo state network, it is projected into a sparse high-dimensional space and generates complex and diverse nonlinear states in the reservoir of the echo state network, thereby more rich and effective features can be extracted and the function of memorizing data can be realized. Finally, W out is the output weight matrix to be trained, which needs to be trained by the training set data given in the actual problem.

[0042] The cyclic calculation in the echo state network layer can be described according to formula (8):

[0043] x(t) = tanh(W res ·x(t-1) + W in ·U(t)), (8)

[0044] where U(t) is the time series input, x(t) is the reservoir state, and the dimension is much larger than n. The initial state is set as x(0) = 0, and x(t) is calculated recursively according to formula (8), where tanh is the activation function, W res and W in correspond to the internal weight connection matrix and the input layer connection matrix of the reservoir given above, respectively.

[0045] After the cyclic calculation, the output weight matrix W out is trained according to the training set data in the actual problem. The training process only needs to use the least square calculation in formula (9) once:

[0046] W out = YX Τ (XX Τ ) -1 , (9)

[0047] where X is the reservoir state storage matrix, and Y is the time series matrix of the real sample data of the training set.

[0048] W outAfter that, the whole quantum echo state network model training is completed. The trained quantum echo state network model is applied on the test set in the actual problem, which can be used to predict the n parameter data at the future time.

[0049] The beneficial effects of the present application are:

[0050] The present application uses quantum controlled non-gate to realize entangled state to approximately represent the coupling relationship that may exist between the original input data, sets the quantum rotation gate as the adjustable parameter of the first layer quantum network, and can adjust the distribution of data. The echo state network layer can play the advantages of fast calculation and one-step training to complete the calculation of the final model. Numerical experiments verify that the quantum echo state network model can accurately predict the state data of the aero-engine at the future flight time. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 is a quantum echo state network model under the condition of 3-dimensional data sample characteristic variables.

[0052] Figure 2 is a comparison chart of prediction data and real data on the training set, Figure 2(a) is a comparison chart of prediction data and real data on the training set, and Figure 2(b) is a point-by-point error chart of real value and prediction value on the training set.

[0053] Figure 3 is a comparison chart of prediction data and real data on the test set, Figure 3(a) is a comparison chart of prediction data and real data on the test set, and Figure 3(b) is a point-by-point error chart of real value and prediction value on the test set. DETAILED DESCRIPTION

[0054] The specific embodiments of the present application are further described below in combination with the drawings and technical solutions.

[0055] The present embodiment is a quantum echo state network model for aero-engine fault early warning, which applies a 3-dimensional input quantum echo state network model in the present application to predict the running parameter data of the engine at the future time, including the following steps. Figure 1

[0056] Step 1: Select data sample characteristics

[0057] ​For ease of description, consider a numerical example with n = 3 sample feature variables. In actual flight missions, it was found that the changes in ambient pressure, engine exhaust temperature, and engine combustion chamber temperature were significant when a fault occurred, indicating a high correlation between these three feature variables and the fault occurrence. Therefore, the following three sample features were selected: ambient pressure, engine exhaust temperature, and engine combustion chamber temperature. The data came from actual flight missions of a certain type of aero-engine, and data was collected using multiple sensors at a sampling time interval of 0.1 seconds. The last sampling time T = 60000, and the total time series record was t = 1:60000. Time series data was collected for each sample feature, resulting in 3 × 60000 sample data points, which were used for numerical simulation experiments.

[0058] Step 2: Splitting the training and test sets and preprocessing the data.

[0059] The sample data obtained in step 1 is divided into a training set and a test set. The time series data of the training set is selected as t = 40000:45000, and the time series data of the test set is selected as t = 46000:48000. The training set data is used to train the quantum echo state network model, and the test set data is used to verify the model's prediction effect. After dividing the training set and the test set, the data needs to be preprocessed. First, the initial data collected in step 1 is normalized according to formula (1), and then the quantum state input is prepared using the quantization rule described by formula (2). This completes the preprocessing of the initial data.

[0060] Step 3: Construction of Quantum Heuristic Neural Network Layers

[0061] The quantized training set data from step 2 is input into the first layer of the quantum heuristic neural network, and n=3 sets of rotation angles are randomly generated. Table 1 shows three different sets of quantum rotating gate rotation angles using the rand command in MATLAB.

[0062] Table 1: Rotation angles of three randomly generated quantum selection gates

[0063]

[0064] For each set of determined rotation angles, construct Figure 1 The quantum circuit in the middle lacks the support of quantum computer hardware, so it is calculated by classical computer simulation, that is, the output U(t) of the quantum heuristic neural network layer is obtained by recursively calculating according to formulas (6) and (7);

[0065] Step 4: Train the quantum echo state network on the training set

[0066] The output U(t) obtained by the quantum heuristic neural network layer on the training set is taken as the input of the second layer echo state network layer, and the related parameters in the reservoir pool are randomly generated, wherein the reservoir pool dimension is selected as 100, i.e., the input layer connection matrix W in is a randomly generated matrix with a dimension of 100*4, and the internal weight connection matrix W res is a randomly generated sparse matrix with a dimension of 100*100. After adjusting the related parameters to optimize the network, the prediction step is taken as 1, and the weight matrix W out of the output layer is calculated by formula (9). Thus, the training is completed.

[0067] Figure 1 A quantum echo state network with a three-dimensional sample input is given, and for the n-dimensional case, only similar generalization and construction are needed. The quantum network part gives the quantum circuit through which the input quantum state passes, and shows the recursive action at the target quantum state. U(t) is the time series output obtained by the quantum layer, W in and W res are randomly generated high-dimensional matrices in the echo state network reservoir pool, and W out is the output weight matrix, which can be obtained by one-time least square training.

[0068] Step 5: Predict data using the trained model and draw a graph for analysis

[0069] First, apply the trained model to the training set to obtain the predicted value. By observing the error between the predicted value and the true value, the training effect can be judged and a graph can be drawn for analysis. In FIG. 2, the comparison between the predicted value and the true value obtained at t=40010:43010 is plotted, and it is found that the predicted value given by the model is very close to the true data. The mean square error between the predicted value and the true value on the entire training set can be calculated as 0.0053344, indicating that the training effect is very good and can be used for testing on the test set. FIG. 2(a) shows the comparison between 3000 predicted values and true values on the training set. The solid line represents the true value on the training set, and the dashed line represents the predicted value given by the quantum echo state network model. FIG. 2(b) shows the error obtained by subtracting the predicted value from the true value at each time node.

[0070] Finally, the model is applied to the test set. The quantum state data on the test set obtained after preprocessing in step 2 is input into the trained Figure 1In the form of the quantum echo state network model, the prediction value at t = 46000:48000 time sequence is obtained. Figure 3 plots the error comparison between the prediction data and the real data given by the model at t = 46010:47010 time sequence, and the mean square error is calculated as 0.0065058, indicating that the prediction effect of the model is good. Figure 3(a) shows the comparison of 1000 prediction values and real values on the test set, the solid line represents the real value on the test set, and the dashed line represents the prediction value given by the quantum echo state network model, and Figure 3(b) shows the error obtained by subtracting the prediction value from the real value at each time node.

[0071] The quantum echo state network model is applied to the fault prediction of an aero-engine. When the prediction value and the real value differ greatly, it indicates that the engine may have a fault, thereby achieving the fault warning effect for the aero-engine.

[0072] Implementation results

[0073] 1) As can be seen from Figure 2, the quantum echo state network model is used for training on the training set, and good prediction effect can be achieved, the deviation between the prediction value and the real value is small, and the mean square error calculated is only 0.0053344.

[0074] 2) As can be seen from Figure 3, the prediction value obtained by using the quantum echo state network to predict the test set is highly consistent with the real value, the deviation between the two is small, and the mean square error calculated is only 0.0065058.

[0075] In summary, the numerical results of applying the quantum echo state network model to the aero-engine data prediction show that the quantum echo state network can indeed play the quantum advantage, and the prediction result is more accurate.

[0076] The above-described embodiments only express the implementation of the present application, but cannot be interpreted as a limitation on the scope of the patent of the present application. It should be pointed out that for those skilled in the art, without departing from the concept of the present application, a number of modifications and improvements can be made, which all belong to the protection scope of the present application.

Claims

1. A method for constructing a quantum echo state network model for aero-engine fault early warning, characterized in that, Comprise the following steps: Step 1: select the data sample characteristics of the aero-engine; Step 2: quantize the initial data selected in step 1; The initial data collected in step 1 Quantization is performed to obtain quantum state form data , wherein represents a certain quantum state, respectively represent two kinds of base states of binary quantum bits, is a corresponding probability amplitude, and its physical meaning is that, for the quantum state , a quantum measurement is performed, and will be observed with a probability of , will be observed with a probability of , and , and ; For a certain time A set of The initial data First, the initial data Normalization; After the initial data is normalized, the normalized data is prepared into a quantum state form data according to the quantization rule given by formula (2) to express the classical data information contained in the input quantum state by the preceding probability amplitude, and formula (2) is as follows: After the initial data is normalized, the normalized data is prepared into a quantum state form data according to the quantization rule given by formula (2) to express the classical data information contained in the input quantum state by the preceding probability amplitude, and formula (2) is as follows: （2）； wherein, represents the quantum state data prepared from the th feature variable at the time point , and correspond to the probability amplitudes of the quantum states before the binary quantum bit basis states and , are obtained by taking the inverse sine of . Using equation (2), let From 1 to , we get Quantum state data at time ; Step 3: build a quantum circuit; First construct A quantum circuit, the previous The line at the initial moment The input is the product prepared in step 2. quantum state data ;No. A quantum circuit is mainly used to calculate the output of a quantum circuit layer, and its input is an auxiliary quantum state. Its initial state is defaulted to ;forward a quantum state Passing through a quantum rotating gate in sequence, and then connected with Passing from top to bottom A two-qubit quantum controlled NOT gate, note that it is also necessary to... Adding a quantum rotating gate to a quantum circuit ultimately yields... , and then As the next moment of A quantum state input is used to perform recursive calculations until... ; In the above quantum circuit, each quantum state input through the quantum rotation gate is denoted as , wherein, is a rotation angle, and the value range belongs to ; each two adjacent quantum states between the two-bit quantum controlled non-gate acting from top to bottom is denoted as ; Since the quantum rotation gate is shared , the rotation angle , , , where is given as a parameter of the quantum network layer; at the th moment, the th quantum state input passes through the quantum rotation gate , and the calculation formula is as follows: （5）； Step 4: build and train a quantum echo state network template, apply the trained quantum echo state network model to aero-engine fault prediction, and give future engine operation state prediction data.

2. The method of claim 1, wherein, Comprise the following steps: Step 1: select the data sample characteristics First, a plurality of operating state data of the aircraft is collected by using a plurality of sensors of the aero-engine, each of the data representing a sample characteristic variable, and the data is in the form of a set of discrete time series data, and the time series is recorded as: wherein is a sampling time, is a last time of sampling; then, a data sample characteristic variable with a high correlation degree with the occurrence of the aero-engine fault is selected from the plurality of operating state data as a judgment standard for fault prediction, including environmental pressure, tail nozzle outlet temperature, engine combustion chamber temperature; Step 2: quantize the initial data selected in step 1; Step 3: build a quantum circuit From the time t = 0, the recursive computation is performed using the above quantum circuit, the input quantum state at the time t = 0 is After one action of the entire quantum circuit, the quantum state is taken as the input at the next time t = 1. Consider the quantum circuit action at the moment , the input quantum state after passing through quantum rotation gate, and then from top to bottom through quantum controlled non-gate, and finally acting a quantum rotation gate at the position of the auxiliary quantum state line, the output quantum state The calculation formula is: （6）； wherein, represents the probability of taking when measuring the final quantum state of the quantum circuit from top to bottom in the row position, from 1 to ; represents the probability of taking when measuring the quantum state obtained after the last row of auxiliary quantum states passes through the controlled non gate; the last row of quantum circuits also passes through a rotation gate after the controlled non gate, so the output at the moment is recorded as , which represents measuring and taking ; the calculation result is as formula (7): （7）； Recursive calculation is performed according to formulas (6) and (7), and let Take from 1 to , a set of time series output ; next take a new set of quantum rotation gate rotation angle parameters , , a new set of outputs can be obtained with the same quantum circuit, randomly generate groups of rotation angle parameters, and obtain groups of dimensional output data, denoted as matrix ; The specific calculation process of the above quantum circuit is as follows: Step 4: build and train a quantum echo state network Building quantum echo state network: the output of the quantum network layer calculated in step 3 is taken as the input of the next layer echo state network , the reservoir of the echo state network is designed as a sparse network with multiple neurons, including a high-dimensional sparse internal weight connection matrix of the reservoir and an input layer connection matrix used to connect , the two matrices and are randomly generated and remain unchanged during the loop calculation of the echo state network; when the low-dimensional data from the quantum layer enters the echo state network, it will be projected into a sparse high-dimensional space and generate complex and diverse nonlinear states in the reservoir of the echo state network, thereby extracting more rich and effective features and realizing the function of memorizing data; finally, is the output weight matrix to be trained, which needs to be trained by the training set data given in the actual problem; After the loop calculation, the output weight matrix is trained according to the training set data in the actual problem The training process only needs to use the least square calculation in formula (9) once: （9）； wherein, is a reserve pool state storage matrix, is a time series matrix of training set real sample data; By formula (9) get After that, the whole quantum echo state network model training is completed; the trained quantum echo state network model is applied on the test set in the actual problem to predict the parameter data at future time.

3. The method of claim 1, wherein, The quantum echo state network comprises an input layer, a reserve pool and an output layer, and the core structure is a randomly generated and unchanged reserve pool; In step 4, the loop calculation in the quantum echo state network layer is described according to formula (8): （8）； wherein, is a time series input, is a reservoir state, which has a dimension much larger than , the initial state is set to , is recursively calculated according to formula (8) wherein is an activation function, and correspond to the given reservoir internal weight connection matrix and input layer connection matrix, respectively.

Images