Quantum feature map circuit, quantum neural network, computing method thereof, and system and method for analyzing energy consumption data of vehicle

Abstract

This case discloses a quantum feature map circuit and its calculation method. The quantum feature map circuit includes a quantum gate that performs quantum gate operations on qubits and a rotating gate that then performs rotation gate operations on the qubits. The input variable value serves as the angle parameter value of the rotating gate in the quantum feature map circuit. This case also discloses a quantum neural network and its calculation method. The quantum neural network includes a quantum feature map circuit and a variational model circuit. The variational model circuit performs rotation gate operations on each qubit operated by the quantum feature map circuit, then performs quantum gate operations on any two qubits, and then performs rotation gate operations on each qubit. The weight variable value serves as the angle parameter value of each rotating gate in the variational model circuit. Furthermore, this case discloses a vehicle energy consumption data analysis system and method based on a quantum neural network.

Description

Technical Field

[0001] This case concerns quantum feature graph circuits and quantum neural networks; more specifically, it concerns a vehicle energy consumption data analysis system and method based on quantum feature graph circuits and their computational methods, and quantum neural networks and their computational methods. Prior Technology

[0002] One well-known energy consumption estimation technique can use genetic algorithms to analyze the relationship between driving behavior and fuel consumption. Its main technical feature is the genetic algorithm (GA).

[0003] Another known technique is to use decision tree methods to analyze the relationship between driving behavior and fuel consumption. Its main technical feature is the decision tree method.

[0004] Another known technique can calculate the linear relationship between vehicle driving time, vehicle waiting time, vehicle starting time, and fuel consumption. Its main technical feature is the linear correlation method.

[0005] While the aforementioned technologies can estimate fuel consumption, they still rely on traditional mathematical analysis or machine learning techniques, making it difficult to improve the speed and accuracy of the predictions. Summary of the Invention

[0006] To address the aforementioned and other issues, this case discloses quantum feature graph circuits and their computation methods, quantum neural networks and their computation methods, and vehicle energy consumption data analysis systems and methods.

[0007] A quantum feature map circuit includes: a qubit; a quantum gate coupled to the qubit for performing a quantum gate operation on the qubit; and a rotary gate coupled to the quantum gate for performing a rotary gate operation on the qubit operated by the quantum gate, wherein the input variable value is used as the angle parameter value of the rotary gate.

[0008] A quantum feature map circuit calculation method includes: performing a quantum gate operation on a qubit using a quantum gate of the quantum feature map circuit; and performing a rotation gate operation on a qubit that has undergone the quantum gate operation using a rotation gate, wherein the input variable value of the quantum feature map circuit is the angle parameter value of the rotation gate.

[0009] In the quantum feature map circuit and calculation method, the quantum gate is an Adama gate, and the rotary gate is an RY gate.

[0010] In the quantum feature map circuit and calculation method, the input variable value is mapped to the range of [-1.5, 1.5] after data preprocessing.

[0011] The quantum feature map circuit and calculation method are implemented on a real quantum computer or in a quantum computer simulator.

[0012] A quantum neural network includes: the aforementioned quantum feature map circuit; and a variational model circuit coupled to the quantum feature map circuit and including an RY gate to perform RY operations on qubits operated by the quantum feature map circuit, wherein the weight variable value is used as the angle parameter value of the RY gate of the variational model circuit.

[0013] A quantum neural network includes: a quantum feature map circuit, comprising quantum gates coupled to each qubit and rotation gates coupled to each quantum gate, for performing quantum gate operations on each qubit and then rotation gate operations, wherein the input variable values ​​serve as the angle parameter values ​​of each rotation gate in the quantum feature map circuit; and a variational model circuit, coupled to the multiple quantum feature map circuits and including quantum gates and multiple rotation gates, for performing rotation gate operations on each qubit operated by the quantum feature map circuit, then performing quantum gate operations on any two qubits, and then performing rotation gate operations on each qubit, wherein the weight variable values ​​serve as the angle parameter values ​​of each rotation gate in the variational model circuit.

[0014] A quantum neural network computation method includes: instructing each quantum gate of a quantum feature map circuit to perform quantum gate operations on each qubit; instructing each rotating gate of the quantum feature map circuit to perform rotating gate operations on each qubit, wherein the input variable value is the angle parameter value of each rotating gate in the quantum feature map circuit; instructing each rotating gate of a variational model circuit to perform rotating gate operations on each qubit; instructing the quantum gate of the variational model circuit to perform quantum gate operations on any two qubits; and instructing each rotating gate of the variational model circuit to perform rotating gate operations on each qubit, wherein the weight variable value is the angle parameter value of each rotating gate in the variational model circuit.

[0015] In the quantum neural network and its computation method, the quantum gate in the quantum feature map circuit is an Adama gate and the rotary gate is an RY gate, wherein the quantum gate in the variational model circuit is a controlled reverse gate and the rotary gate is an RY gate.

[0016] In the quantum neural network and its computation method, the quantum gate operation of the variational model circuit can be performed once or multiple times until all qubits perform controlled reverse gate operation.

[0017] In the quantum neural network and its computation method, the quantum neural network measures the quantum state of each qubit to calculate a predicted value based on the quantum state, wherein the predicted value is used as a regression application or a classification application, and wherein the predicted value is compared with the true value to generate a loss value, thereby performing a quantum optimization method to adjust the weight variable value based on the loss value.

[0018] A vehicle energy consumption data analysis system includes: a processing and storage device for storing vehicle driving records and energy consumption records; and a quantum computing device for executing the aforementioned quantum neural network, wherein the vehicle driving record is used as the input variable value to input the angle parameter value of the RY gate of the quantum feature map circuit, and the energy consumption record is used as the true value. The quantum neural network is trained based on the input variable value and the true value to adjust the weight variable value of the variational model circuit, thereby generating an energy consumption prediction value for an instantaneous vehicle driving record using the trained quantum neural network.

[0019] A method for analyzing vehicle energy consumption data includes: acquiring vehicle driving records and energy consumption records, using the vehicle driving records as input variable values, inputting the input variable values ​​into a quantum feature map circuit including an Adama gate and an RY gate to serve as the angle parameter value of the RY gate of the quantum feature map circuit, and using the energy consumption records as true values; training a quantum neural network including the quantum feature map circuit and a variational model circuit including a controlled reverse gate and an RY gate based on the input variable values ​​and the true values, to adjust the weight variable value of the RY gate of the variational model circuit; and generating an energy consumption prediction value for a real-time vehicle driving record using the trained quantum neural network.

[0020] In the aforementioned vehicle energy consumption data analysis system and method, the vehicle driving record is preprocessed to map to the value range of [-1.5, 1.5], and then used as the input variable value to serve as the angle parameter value of the RY gate of the quantum feature map circuit.

[0021] In the aforementioned vehicle energy consumption data analysis system and method, the energy consumption record is preprocessed to map to the value range of [-1, 1], and then used as the true value. The true value is then used to train the quantum neural network to adjust the weight variable value of the RY gate of the variational model circuit and generate regression prediction values.

[0022] In the aforementioned vehicle energy consumption data analysis system and method, the energy consumption record is preprocessed to map to the value range of [0,1], and then used as the true value. The true value is then used to train the quantum neural network to adjust the weight variable value of the RY gate of the variational model circuit and generate a classification prediction value.

[0023] In the aforementioned vehicle energy consumption data analysis system and method, the vehicle driving record includes the number of data entries for each speed range accumulated over a period of time, the energy consumption record includes the amount of fuel refueled during that period of time, and the predicted value is the predicted fuel consumption value.

[0024] In the aforementioned vehicle energy consumption data analysis system and method, the vehicle driving record includes the number of data entries for each speed range accumulated over a period of time, the energy consumption record includes the battery level at the start and end of the period of time, and the predicted value is the predicted energy consumption value.

[0025] Based on the above, the quantum feature graph circuit and method, the quantum neural network and method, and the vehicle energy consumption data analysis system and method disclosed in this case are based on the RY gate quantum feature graph circuit and combined with the variational model circuit to generate a quantum neural network, which is then applied to the estimation of energy consumption data of electric or fuel vehicles. Unlike traditional mathematical analysis or machine learning techniques, this method can significantly improve the speed and accuracy of the prediction. Simple Explanation of the Diagram

[0026] Figure 1A is a schematic diagram of the architecture of the quantum feature graph circuit in this case.

[0027] Figure 1B is a structural diagram of the quantum feature map circuit of Embodiment 1 of this case.

[0028] Figure 2 is a structural diagram of the variational model circuit of Embodiment 1 of this case.

[0029] Figure 3 is a structural diagram of the quantum neural network in Embodiment 1 of this case.

[0030] Figure 4 is a structural diagram of the quantum feature map circuit in Embodiment 2 of this case.

[0031] Figure 5 is a structural diagram of the variational model circuit in Embodiment 2 of this case.

[0032] Figure 6A is a schematic diagram of the architecture of the quantum neural network in this case.

[0033] Figure 6B is a structural diagram of the quantum neural network in Embodiment 2 of this case.

[0034] Figure 7 is a schematic diagram of the architecture of the vehicle energy consumption data analysis system in this case.

[0035] Figure 8 shows the energy consumption data analysis system applied to fuel vehicles in Embodiment 3 of this case.

[0036] Figure 9 is a structural diagram of the quantum feature map circuit in Embodiment 3 of this case.

[0037] Figure 10 is a structural diagram of the variational model circuit of Embodiment 3 of this case.

[0038] Figure 11 is a structural diagram of the quantum neural network in Embodiment 3 of this case.

[0039] Figure 12 shows the energy consumption data analysis system applied to electric vehicles in Embodiment 4 of this case.

[0040] Figure 13 is a structural diagram of the quantum feature map circuit in Embodiment 4 of this case.

[0041] Figure 14 is a structural diagram of the variational model circuit of Embodiment 4 of this case.

[0042] Figure 15 is a structural diagram of the quantum neural network in Embodiment 4 of this case.

[0043] Figure 16 is a schematic diagram of the architecture of the vehicle energy consumption data analysis system in this case.

[0044] Figure 17 is a schematic diagram of the architecture of the vehicle energy consumption data analysis system in this case.

[0045] Figure 18 is a flowchart illustrating the method for analyzing the energy consumption data of the vehicles in this case. Implementation

[0046] The technical content of this invention is described below through specific embodiments. Those skilled in the art can easily understand the advantages and effects of this invention from the content disclosed in this specification. However, this invention can also be implemented or applied through other different embodiments.

[0047] Please refer to Figures 1A and 1B, Figure 2, and Figure 3, which respectively illustrate the quantum feature map circuit, variational model circuit, and quantum neural network of Embodiment 1 of this case.

[0048] As shown in Figure 1A, the quantum feature map circuit includes a qubit 100, a quantum gate 101 coupled to the qubit 100 to perform quantum gate operation on the qubit 100, and a rotary gate 102 coupled to the quantum gate 101 to perform rotary gate operation on the quantum gate-operated qubit 100, wherein the input variable value is used as the angle parameter value of the rotary gate.

[0049] As shown in Figure 1B, the quantum feature map circuit includes a qubit |q〉, the initial quantum state system |0〉 of the qubit |q〉, a Hadamard gate operation is performed on the qubit |q〉, and an RY gate operation is performed on the qubit |q〉, and the input variable value x[0] is included as the angle parameter value of the RY gate.

[0050] In one embodiment, the quantum feature map circuit can be implemented in a real quantum computer or in a conventional computer using a quantum computer simulator. The computational logic of the quantum feature map circuit implemented in a quantum computer simulator is described below:

[0051] The qubit |q> has an amplitude vector matrix. This can express the quantum superposition system v0|0>+v1|1> of the qubit |q>. The squared probability matrix of the amplitude vector elements... The probability system v0 of the qubit |q〉 having |0〉 and the probability system v1 of having |1〉 can be expressed as follows: The initial quantum state system |0〉 of the qubit |q〉 is given by the amplitude vector matrix of the qubit |q〉. .

[0052] Hadamard gate calculation matrix system Perform a Hadamard gate operation on the qubit |q〉, which means calculating the matrix multiplication result of the Hadamard gate calculation matrix and the amplitude vector matrix of the qubit |q〉, H|q〉.

[0053] RY gate calculation matrix system The RY gate generates different phase differences by rotating the Y-axis of the Bloch sphere by an angle θ, where θ is the angle parameter value. The input variable value x[0] of the quantum feature map circuit is used as the angle parameter value of the RY gate. When the RY gate operation is performed on the quantum bit |q>, that is, the matrix multiplication result of the RY gate calculation matrix and the amplitude vector matrix of the quantum bit |q> is calculated, R(x[0])|q>.

[0054] The quantum feature map circuit performs a Hadamard gate operation on the qubit |q〉 and then an RY gate operation on the qubit |q〉, so the calculation result is R(x[0])H|q〉. The input variable value x[0] can be preprocessed before input to map the input variable value x[0] to the value range of [-1.5, 1.5].

[0055] As shown in Figure 2, the number of qubits in the variational model circuit is the same as the number of qubits in the quantum feature map circuit in Figure 1. The qubit |q〉 is subjected to a rotating gate (i.e., RY gate) operation, and the weight variable value w[0] can be used as the angle parameter value of these RY gates.

[0056] In one embodiment, the variational model circuit can be implemented on a real quantum computer or on a conventional computer using a quantum computer simulator. The computational logic of the variational model circuit implemented in a quantum computer simulator is described below:

[0057] RY gate calculation matrix system The RY gate generates different phase differences by rotating the Y-axis of the Bloch sphere by an angle θ, where θ is the angle parameter value. In the variational model circuit, the weight variable value w[0] is used as the angle parameter value of the RY gate. When the RY gate operation is performed on the qubit |q>, that is, the matrix multiplication result of the RY gate calculation matrix and the amplitude vector matrix of the qubit |q> is calculated, R(w[0])|q>.

[0058] As shown in Figure 3, the quantum neural network includes a quantum feature graph circuit and a variational model circuit. The quantum neural network first calculates the quantum feature graph circuit and then calculates the variational model circuit. That is, the quantum feature graph circuit propagates its calculation result R(x[0])H|q〉 to the variational model circuit. After passing through the variational model circuit, the forward propagation result is R(w[0])R(x[0])H|q〉.

[0059] The forward propagation result \(R(w[0])R(x[0])H|q\rangle\) measures the qubit \(|q\rangle\) to obtain the calculated amplitude vector matrix coefficient , in the regression application, \(\zeta_0^2 - \zeta_1^2\) can be calculated to obtain the predicted value \(y'\), and the squared error \((y' - y)^2\) can be calculated with the true value \(y\) as the loss value, and the quantum optimization method (algorithms (including Constrained Optimization By Linear Approximation (COBYLA), Simultaneous Perturbation Stochastic Approximation (SPSA), Analytic Quantum Gradient Descent (AQGD))) is used to adjust the weight variable value \(w[0]\) in the variational model circuit.

[0060] Iterate the quantum neural network multiple times. Each iteration calculates the quantum feature map circuit, calculates the variational model circuit, measures to obtain the predicted value, calculates the loss value, and adjusts the weight variable value \(w[0]\) in the variational model circuit. Stop when the iteration reaches \(I\) times. The value of \(I\) is set manually as the result of completing the training of the quantum neural network, and the trained weight variable value \(w[0]\) is obtained.

[0061] Please refer to Figures 4, 5, 6A, and 6B, which respectively illustrate the quantum feature map circuit, variational model circuit, and quantum neural network of the second embodiment of this case.

[0062] As shown in Figure 4, the quantum feature map circuit includes two qubits \(|q_0\rangle\) and qubit \(|q_1\rangle\). The initial quantum states of these qubits are all \(|0\rangle\). Therefore, the quantum state of the two qubits \(|q_1q_0\rangle\) can be expressed as \(|00\rangle\). The quantum gates perform Hadamard gate operations on these qubits respectively, and the rotation gates perform RY gate operations on these qubits respectively. And the two rotation gates respectively include the input variable values \(x[0]\) and \(x[1]\) as the angle parameter values of the RY gate.

[0063] In one embodiment, the quantum feature map circuit can be implemented on a real quantum computer or on a traditional computer using a quantum computer simulator. The calculation logic of the quantum feature map circuit implemented on a quantum computer simulator is described as follows:

[0064] The qubits \(|q_1q_0\rangle\) have an amplitude vector matrix This can express the quantum superposition system of these qubits |q1q0〉 as v00|00〉+v01|01〉+v10|10〉+v11|11〉. The probability matrix of the squared elements of this amplitude vector system... The probability system v00 can be used to express the probability of these qubits |q1q0> occurring |q00>, the probability system v01> occurring |q10>, the probability system v11> occurring |q10>, and the probability system v11> occurring |q11>. The initial quantum state system |q00> of these qubits |q1q0> is thus represented by the amplitude vector matrix of these qubits |q1q0>. .

[0065] Hadamard gate calculation matrix system When a Hadamard gate operation is performed on each of these qubits |q1q0>, that is, the matrix multiplication result of the outer product of the Hadamard gate computation matrix and the amplitude vector matrix of these qubits |q1q0> is calculated, This operator External accumulation.

[0066] RY gate calculation matrix system The RY gate generates different phase differences by rotating the Y-axis of the Bloch sphere by an angle θ, where θ is the angle parameter value. In the quantum feature map circuit, the input variable value x[0] is used as the RY gate angle parameter value for operating the qubit |q 0>, i.e., R(x[0]), and the input variable value x[1] is used as the RY gate angle parameter value for operating the qubit |q 1>, i.e., R(x[1]). When the RY gate operation is performed on each of these qubits |q 1 q 0>, that is, the matrix multiplication result of the outer product of the RY gate calculation matrix and the amplitude vector matrix of these qubits |q 1 q 0> is calculated, This operator External accumulation.

[0067] The quantum feature map circuit performs Hadamard gate operation on each of these qubits |q1q0> and then RY gate operation on each of these qubits |q1q0>. Therefore, the calculation result is (R(x[0]). R(x[1]))(H H)|q 1 q 0〉。 The input variable values ​​x[0] and x[1] can be preprocessed before input, and these input variable values ​​x[0] and x[1] can be mapped to the value range of [-1.5, 1.5].

[0068] As shown in Figure 5, the number of qubits in the variational model circuit is the same as the number of qubits in the quantum feature diagram circuit in Figure 4. Rotation gate (RY gate) operation is performed on each of these qubits |q 1 q 0>, and then quantum gate (CNOT gate) operation is performed on each of these qubits |q 1 q 0>. Rotation gate (RY gate) operation is performed on each of these qubits |q 1 q 0>. The four weighted variable values ​​w[0], w[1], w[2], and w[3] can be used as the angle parameter values ​​of these RY gates.

[0069] In one embodiment, the variational model circuit is implemented on a real quantum computer or on a conventional computer using a quantum computer simulator. The computational logic of the variational model circuit implemented in a quantum computer simulator is described below:

[0070] RY gate calculation matrix system The RY gate generates different phase differences by rotating the Bloch sphere by an angle θ along the Y-axis. θ is the angle parameter value. In the variational model circuit, the weight variable values ​​w[0] and w[1] are used as the angle parameter values ​​of the RY gate. When the RY gate operation is performed on the qubit |q0> and the qubit |q1> respectively, that is, the matrix multiplication result of the outer product of the RY gate calculation matrix and the amplitude vector matrix of the qubits |q1 and q0> is calculated, (R(w[0]) R(w[1]))|q 1 q 0〉.

[0071] CNOT gate calculation matrix system The control bit of the RY CNOT gate is the qubit |q0>, and the target bit is the qubit |q1>. It can be used to generate the interaction between these input variables. When the CNOT gate operation is performed on these qubits |q1q0>, that is, the matrix multiplication result of the CNOT gate calculation matrix and the amplitude vector matrix of these qubits |q1q0> is calculated, C(q0,|q1q0>)|q1q0>.

[0072] RY gate calculation matrix system The RY gate generates different phase differences by rotating the Bloch sphere by an angle θ along the Y-axis. θ is the angle parameter value. In the variational model circuit, the weight variable values ​​w[2] and w[3] are used as the angle parameter values ​​of the RY gate. When the RY gate operation is performed on the qubit |q0> and the qubit |q1> respectively, that is, the matrix multiplication result of the outer product of the RY gate calculation matrix and the amplitude vector matrix of the qubits |q1q0> is calculated, (R(w[2]) R(w[3]))|q 1 q 0〉.

[0073] As shown in Figure 6A, the quantum neural network includes a quantum feature map circuit 21 and a variational model circuit 22. The quantum feature map circuit 21 includes qubits 200 and 210, quantum gates 201 and 211 coupled to qubits 200 and 210, and rotation gates 202 and 212 coupled to quantum gates 201 and 211, to perform quantum gate operations on qubits 200 and 210 followed by rotation gate operations. The input variable values ​​serve as the angle parameter values ​​for rotation gates 202 and 212 in the quantum feature map circuit 21. The variational model circuit 22 includes quantum gates 23 and rotation gates 203 and 213, 204 and 214, to perform rotation gate operations on qubits 200 and 210, then perform quantum gate operations on any two qubits, and then perform rotation gate operations on each qubit. The weight variable values ​​serve as the angle parameter values ​​for rotation gates 203 and 213, 204 and 214 in the variational model circuit 22.

[0074] As shown in Figure 6B, the quantum neural network first calculates the quantum feature graph circuit, and then calculates the variational model circuit, meaning that the quantum feature graph circuit calculates the result (R(x[0]). R(x[1]))(H H)|q 1 q 0〉 is propagated to the variational model circuit, and the forward propagation result after passing through the variational model circuit is (R(w[2]). R(w[3]))C(q 0,|q 1 q 0〉)(R(w[0]) R(w[1]))(R(x[0]) R(x[1]))(H H)|q 1 q 0〉.

[0075] The forward propagation result (R(w[2])) R(w[3]))C(q 0,|q 1 q 0〉)(R(w[0]) R(w[1]))(R(x[0]) R(x[1]))(H H)|q 1 q 0〉After measuring these qubits|q 1 q 0〉, the calculated amplitude vector matrix is ​​obtained. In classification applications with two classes (Label'0' and Label'1'), we can calculate ζ00² + ζ01² to obtain the predicted value y0' as the probability of predicting Label'0', and calculate ζ10² + ζ11² to obtain the predicted value y1' as the probability of predicting Label'1'. Then, we calculate the cross-entropy value between these true values ​​y0 and y1. As loss values, and using a quantum optimization method (algorithm including Constrained Optimization By Linear Approximation (COBYLA), Simultaneous Perturbation Stochastic Approximation (SPSA), and Analytic Quantum Gradient Descent (AQGD))), the weight variable values ​​w[0], w[1], w[2], and w[3] in the variational model circuit are adjusted.

[0076] The quantum neural network is iterated multiple times. In each iteration, the quantum feature map circuit is calculated, the variational model circuit is calculated, the predicted value is measured, the loss value is calculated, and the weight variable values ​​w[0], w[1], w[2], and w[3] in the variational model circuit are adjusted. The iteration stops when I times. The value of I is set manually as the result of training the quantum neural network, and the trained weight variable values ​​w[0], w[1], w[2], and w[3] are obtained.

[0077] Referring to Figure 7, this invention proposes a vehicle energy consumption data analysis system, comprising a processing and storage device 31 and a quantum computing device 32. In one embodiment, the processing and storage device 31 is, for example, a conventional computer to perform data preprocessing, and the quantum computing device 32 is, for example, a real quantum computer or a quantum computer simulator to execute quantum neural networks to perform quantum neural network computation methods.

[0078] The processing storage device 31 stores vehicle driving records and energy consumption records. The vehicle driving records can be used as input variable values ​​to the angle parameter values ​​of the RY gate of the quantum feature map circuit. In addition, the data preprocessing method performed by the processing storage device 31 includes obtaining real-time vehicle driving records, using the real-time vehicle driving records as input variable values ​​to the angle parameter values ​​of the RY gate of the quantum feature map circuit, and then using a trained quantum neural network to generate energy consumption prediction values.

[0079] The energy consumption record can be used as a true value to train the quantum neural network and adjust the weight variables of the variational model circuit. Alternatively, the energy consumption record can undergo data preprocessing, mapping it to the range [-1, 1], and then be used as a true value. This true value can be used to train the quantum neural network, adjust the weight variables of the variational model circuit, and generate regression prediction values. Another method is to perform data preprocessing, mapping it to the range [0, 1], and then use this true value. This true value can be used to train the quantum neural network, adjust the weight variables of the variational model circuit, and generate classification prediction values.

[0080] Please refer to Figures 8, 9, 10, and 11, which illustrate the quantum neural network-based energy consumption data analysis system applied to fuel vehicles in Embodiment 3 of this case.

[0081] As shown in Figure 8, the energy consumption data analysis system includes a fuel vehicle 40, a processing and storage device 41, and a quantum computing device 42. The processing and storage device 41 is used to perform data preprocessing methods, and the quantum computing device 42 is used to perform quantum neural network computing methods to obtain fuel consumption prediction values.

[0082] Specifically, the fuel vehicle 40 can report a vehicle speed information to the processing and storage device 41 every 30 seconds. The processing and storage device 41 can store vehicle driving records and refueling records. The vehicle driving records include the number of data entries for each speed range accumulated within a time period, and the refueling records include the amount of fuel refueled accumulated within that time period.

[0083] This time period is, for example, one month. This speed range is, for example, every 10 km / h is a speed range, including speed range 0 (idle), speed range 1 (0 km / h to 10 km / h), speed range 2 (10 km / h to 20 km / h), ..., speed range 13 (above 120 km / h).

[0084] For each fuel vehicle, the total amount of fuel refueling recorded in 40 refueling records within the same month is calculated separately, and these fuel refueling amounts are normalized and mapped to the range [-1, 1] as the true value.

[0085] The number of data entries for each fuel vehicle's 40 driving records within the same month for each speed range was counted separately, and these data entries were normalized and mapped to the value range of [-1.5, 1.5]. The angle parameter values ​​of x[0], x[1], x[2], ..., x

[13] were then input as input variables to the RY gate of the quantum feature map circuit. The quantum feature map circuit contains 14 qubits, as shown in Figure 9.

[0086] As shown in Figure 10, the calculation results of the quantum feature map circuit can be propagated to a variational model circuit. The variational model circuit contains 14 qubits. An RY gate operation is performed on each qubit, and then a CNOT gate operation is performed on any two qubits until a CNOT gate operation is performed on each qubit. After that, an RY gate operation is performed on each qubit. There are a total of 28 weight variable values ​​that can be used as the angle parameter values ​​of these RY gates.

[0087] As shown in Figure 11, the quantum neural network includes the quantum feature map circuit and the variational model circuit. The quantum computing device executes the quantum neural network calculation method to train the quantum neural network (as described in Embodiment 1 above) and adjusts the weight variable values ​​of the variational model circuit.

[0088] Furthermore, the processing and storage device 41 can obtain real-time vehicle driving records from the fuel vehicle 40, acquire the cumulative number of data entries for each speed range over the past month, and input these data entries for each speed range as input variable values ​​to the angle parameter value of the RY gate of the quantum feature map circuit. This quantum neural network calculation method uses a trained quantum neural network to generate regression prediction values, and can denormalize these regression prediction values ​​into fuel consumption prediction values.

[0089] Please refer to Figures 12, 13, 14, and 15, which illustrate the quantum neural network-based energy consumption data analysis system applied to electric vehicles in Embodiment 4 of this case.

[0090] As shown in Figure 12, the energy consumption data analysis system includes an electric vehicle 50, a processing and storage device 51, and a quantum computing device 52. The processing and storage device 51 is used to perform data preprocessing methods, and the quantum computing device 52 is used to perform quantum neural network computing methods to obtain 2 energy consumption prediction values.

[0091] Specifically, the electric vehicle 50 can report vehicle speed information and battery level information to the conventional computer every 30 seconds; the processing storage device 51 can store vehicle driving records and battery level records, the vehicle driving records containing the number of data entries for each speed range accumulated within a time period, and the battery level records containing the battery level at the beginning of the time period and the battery level at the end of the time period.

[0092] This time period is like a day. This speed range is for example, every 10 km / h is a speed range, including speed range 0 (idle), speed range 1 (0 km / h to 10 km / h), speed range 2 (10 km / h to 20 km / h), ..., speed range 13 (above 120 km / h).

[0093] Each electric vehicle's power level records were obtained at the beginning and end of the same day, with 50 power levels. The power levels were divided into five categories, and these levels were normalized and mapped to five values ​​using one-hot encoding, with each value falling within the range [0,1]. The normalized values ​​of the initial power level, c[0], c[1], c[2], c[3], and c[4], were used as input variables to the angle parameter values ​​of the RY gate in the quantum feature map circuit. The normalized values ​​of the final power level, y[0], y[1], y[2], y[3], and y[4], were used as the true values.

[0094] The number of data entries for each of the 50 electric vehicles' driving records accumulated in each speed range on the same day was counted separately. These data entries were then normalized and mapped to the value range of [-1.5, 1.5]. The angle parameter values ​​of the RY gate of the quantum feature map circuit were obtained as input variables x[0], x[1], x[2], ..., x

[13] . A constant b[0] was added to the angle parameter value of the RY gate of the quantum feature map circuit. The quantum feature map circuit contains 20 qubits, as shown in Figure 13.

[0095] As shown in Figure 14, the calculation results of the quantum feature map circuit can be propagated to a variational model circuit. The variational model circuit contains 20 qubits. An RY gate operation is performed on each qubit, and then a CNOT gate operation is performed on any two qubits until a CNOT gate operation is performed on each qubit. After that, an RY gate operation is performed on each qubit. There are a total of 40 weight variable values ​​that can be used as the angle parameter values ​​of these RY gates.

[0096] As shown in Figure 15, the quantum neural network includes the quantum feature map circuit and the variational model circuit. The quantum computing device executes the quantum neural network calculation method to train the quantum neural network (as described in Embodiment 2 above) and adjusts the weight variable values ​​of the variational model circuit.

[0097] Furthermore, the processing and storage device 51 can obtain the electric vehicle's driving records from the electric vehicle 50, acquire the cumulative number of data entries for each speed range over the past day, and input these data entries for each speed range as input variable values ​​to the angle parameter value of the RY gate in the quantum feature map circuit. This quantum neural network calculation method uses a trained quantum neural network to generate classification prediction values, and can denormalize these classification prediction values ​​into energy level prediction values.

[0098] Please refer to Figure 16, a flowchart illustrating the quantum feature map circuit calculation method of this case. The quantum feature map circuit calculation method can be executed by the quantum feature map circuit shown in Figure 1A to be implemented on a real quantum computer or a quantum computer simulator.

[0099] In step S11, the quantum gate of the quantum feature map circuit is subjected to quantum gate operation on the qubit. In one embodiment, the quantum gate is an Adama gate.

[0100] In step S12, the rotating gate of the quantum feature map circuit performs a rotating gate operation on the qubits that have undergone quantum gate operation. In one embodiment, the rotating gate is an RY gate, and the input variable value of the quantum feature map circuit is the angle parameter value of the RY gate. The input variable value is preprocessed to map to the value range of [-1.5, 1.5].

[0101] Please refer to Figure 17, a schematic diagram of the quantum neural network computing method in this case. The quantum neural network computing method can be executed by the quantum feature graph circuit shown in Figure 6A to be implemented on a real quantum computer or a quantum computer simulator.

[0102] In step S21, each quantum gate of the quantum feature map circuit performs a quantum gate operation on each qubit. In one embodiment, the quantum gate is an Adama gate.

[0103] In step S22, each rotating gate of the quantum feature map circuit performs a rotating gate operation on each qubit. In one embodiment, the rotating gate is an RY gate, and the input variable value is the angle parameter value of each RY gate in the quantum feature map circuit.

[0104] In step S23, each rotating gate of the variational model circuit performs a rotating gate operation on each qubit. In one embodiment, the rotating gate is an RY gate.

[0105] In step S24, one or more quantum gates of the variational model circuit are subjected to quantum gate operation on any two qubits. In one embodiment, the quantum gate is a controlled inverse gate (CNOT).

[0106] In step S25, each rotating gate of the variational model circuit performs a rotating gate operation on each qubit. In one embodiment, the rotating gate is an RY gate, and the weight variable values ​​are the angle parameter values ​​of each RY gate in the variational model circuit.

[0107] The variational model circuit described in steps S23-S25 can be operated once or multiple times. Furthermore, the quantum states of each qubit can be measured to calculate predicted values ​​based on these quantum states. These predicted values ​​can be used for regression or classification applications, and can be compared with the true values ​​to generate loss values. A quantum optimization method (algorithm) is then executed to adjust the weight variable values ​​based on these loss values. In one embodiment, the quantum optimization method (algorithm) can be, for example, one of Constrained Optimization By Linear Approximation (COBYLA), Simultaneous Perturbation Stochastic Approximation (SPSA), or Analytic Quantum Gradient Descent (AQGD).

[0108] Please refer to Figure 18, a flowchart of the transportation energy consumption data analysis method in this case. The transportation energy consumption data analysis method can be executed by the transportation energy consumption data analysis system shown in Figure 7.

[0109] Step S31: Obtain vehicle driving records and energy consumption records.

[0110] Step S32: Use the vehicle driving record as an input variable value to serve as the angle parameter value of the RY gate in the quantum feature map circuit, and use the energy consumption record as the true value. In specific implementation, the vehicle driving record can be preprocessed to map to the value range of [-1.5, 1.5], and then used as the input variable value to serve as the angle parameter value of the RY gate in the quantum feature map circuit.

[0111] Step S33: Based on the input variable values ​​and the true value, train a quantum neural network including a quantum feature map circuit and a variational model circuit to adjust the weight variable values ​​of the RY gate of the variational model circuit. Specifically, the energy consumption record can be preprocessed to map to the range [-1, 1], then used as the true value, and subsequently used to train the quantum neural network to adjust the weight variable values ​​of the RY gate of the variational model circuit, generating regression prediction values. Alternatively, the energy consumption record can be preprocessed to map to the range [0, 1], then used as the true value, and subsequently used to train the quantum neural network to adjust the weight variable values ​​of the RY gate of the variational model circuit, generating classification prediction values.

[0112] Step S34: Generate an energy consumption prediction value from a trained quantum neural network based on a real-time vehicle driving record. For fuel-powered vehicles, the vehicle driving record includes the number of data points accumulated over a time period for each speed range, and the energy consumption record includes the amount of fuel refueled during that time period. The predicted value is the fuel consumption prediction value. For electric vehicles, the vehicle driving record includes the number of data points accumulated over a time period for each speed range, and the energy consumption record includes the battery level at the start and end of that time period. The predicted value is the electricity consumption prediction value.

[0113] In summary, the quantum feature map circuit and method, the quantum neural network and method, and the vehicle energy consumption data analysis system and method of this case have the following advantages:

[0114] I. The quantum range mapping method disclosed in this case can perform data preprocessing on the range of input variables to make it more suitable for mapping to a range that conforms to quantum circuit calculations.

[0115] Second, the quantum feature map circuit based on the RY gate disclosed in this case can make the input variables more suitable for mapping to the value range that conforms to quantum circuit calculation.

[0116] Third, the quantum neural network proposed in this case, which combines the quantum feature graph circuit based on the RY gate and the quantum variational model circuit, can be applied to regression and classification applications.

[0117] Fourth, the application of quantum neural networks disclosed in this case for estimating the power consumption of electric vehicles can quickly and accurately estimate power consumption.

[0118] Fifth, although the quantum neural network disclosed in this case can be executed on a quantum computer, it can also be executed on a traditional computer using a quantum computer simulator, producing feature extraction results that traditional neural networks cannot achieve.

[0119] The above embodiments are merely illustrative of the effects of this application and are not intended to limit the scope of this application. Anyone skilled in the art can modify and alter the above embodiments without departing from the spirit and scope of this application. Therefore, the scope of protection of this application should be as set forth in the following patent application claims.

[0120] 100, 200, 210: qubits 101,201,211: Quantum Gate 102,202,212,203,213,204,214: Rotary gates 21: Quantum Feature Map Circuit 22: Variational Model Circuit 23: Quantum Gate 31, 41, 51: Processing storage devices 32, 42, 52: Quantum computing devices 40: Fuel-powered vehicles 50: Electric vehicles S11~S12: Steps S21~S25: Steps S31~S34: Steps

Claims

1. A quantum feature map circuit, comprising: A qubit has an initial quantum state; The first quantum gate is coupled to the qubit to perform an Adama operation on the qubit; The second quantum gate is coupled to the first quantum gate to perform an RY rotation operation on the qubit after it has been operated on by the first quantum gate, wherein the input variable value is used as the angle parameter value of the RY rotation operation of the second quantum gate.

2. The quantum feature map circuit as described in claim 1, wherein, The input variable value is mapped to the range [-1.5, 1.5] after data preprocessing.

3. The quantum feature map circuit described in claim 1 is implemented by a real quantum computer or a quantum computer simulator.

4. A quantum neural network, comprising: Quantum feature map circuit as described in any one of claims 1 to 3; A variational model circuit, coupled to the quantum feature map circuit and including a third quantum gate, performs an RY rotation operation on the qubit after the first quantum gate operation and the second quantum gate operation, wherein the weight variable value is used as the angle parameter value of the RY rotation operation of the third quantum gate; and a measurement unit measures the quantum state of the qubit after passing through the variational model circuit, and calculates a prediction value based on the quantum state.

5. A quantum neural network, comprising: A quantum feature map circuit includes: a plurality of qubits; a plurality of first quantum gates, each coupled to one of the qubits, for performing an Adama operation on each qubit; and a plurality of second quantum gates, each coupled to one of the first quantum gates, for performing an RY rotation operation on each qubit after the operation on each of the first quantum gates, wherein a plurality of input variable values ​​are respectively used as angle parameter values ​​for the RY rotation operation of each of the second quantum gates; a variational model circuit is coupled to the quantum feature map circuit, the variational model circuit including: a plurality of third quantum gates, each coupled to one of the second quantum gates, for performing an RY rotation operation on each qubit after the operation on each of the first quantum gates. After the circuit operation, each qubit undergoes an RY rotation operation; a plurality of fourth quantum gates are coupled to the plurality of third quantum gates to perform a controlled reverse gate operation on any two qubits; a plurality of fifth quantum gates are coupled to the plurality of fourth quantum gates to perform an RY rotation operation on each qubit after the operation of the plurality of fourth quantum gates, wherein the plurality of weight variable values ​​are respectively used as the angle parameter values ​​of the RY rotation operation of each third quantum gate and each fifth quantum gate; and a measurement unit measures the quantum state of each qubit after passing through the variational model circuit to calculate a predicted value based on each quantum state.

6. The quantum neural network as described in claim 5, wherein, The predicted value is used for regression or classification applications.

7. The quantum neural network as described in claim 5, wherein, The predicted value is compared with the true value to generate a loss value, thereby performing a quantum optimization method to adjust the values ​​of the complex weight variables according to the loss value.

8. A vehicle energy consumption data analysis system, comprising: Processing and storage devices are used to store vehicle driving records and energy consumption records; And a quantum computing device, communicatively connected to the processing and storage device, for executing a quantum neural network as described in any one of claims 5 to 7, wherein the vehicle driving record is used as the plurality of input variable values ​​to be input into the quantum feature map circuit as angular parameter values ​​for the RY rotation operation of each of the second quantum gates, and the energy consumption record is used as the plurality of true values. The quantum neural network is trained based on the plurality of input variable values ​​and the plurality of true values ​​to adjust the plurality of weight variable values ​​of the variational model circuit, thereby generating an energy consumption prediction value for an instantaneous vehicle driving record using the trained quantum neural network.

9. The vehicle energy consumption data analysis system as described in claim 8, wherein, The energy consumption record is preprocessed to map to the range of [-1,1], and then used as the complex true values. These complex true values ​​are then used to train the quantum neural network to adjust the complex weight variable values ​​of the variational model circuit and generate regression prediction values.

10. The vehicle energy consumption data analysis system as described in claim 8, wherein, The energy consumption record is preprocessed to map to the value range of [0,1], and then used as the complex true values. These complex true values ​​are then used to train the quantum neural network to adjust the complex weight variable values ​​of the variational model circuit and generate classification prediction values.

11. The vehicle energy consumption data analysis system as described in claim 8, wherein, The vehicle driving record includes the number of data entries for each speed range accumulated over a period of time, the energy consumption record includes the amount of fuel refueled during that period of time, and the predicted value is the fuel consumption prediction value.

12. The vehicle energy consumption data analysis system as described in claim 8, wherein, The vehicle driving record includes the number of data entries accumulated over a period of time for each speed range, the energy consumption record includes the battery level at the start and end of the period, and the predicted value is the predicted energy consumption value.

13. The vehicle energy consumption data analysis system as described in claim 8, wherein, The processing and storage device is a computer, and the quantum computing device is a real quantum computer or a quantum computer simulator.

14. A method for calculating quantum feature graph circuits, comprising: The first quantum gate of the quantum feature graph circuit is subjected to an Adama operation on the qubit with the initial quantum state. And the second quantum gate of the quantum feature map circuit performs an RY rotation operation on the qubit after the operation of the first quantum gate, wherein the input variable value is the angle parameter value of the RY rotation operation of the second quantum gate.

15. The quantum feature map circuit calculation method as described in claim 14, wherein, The input variable value is mapped to the range [-1.5, 1.5] after data preprocessing.

16. The quantum feature map circuit calculation method as described in claim 14 is implemented on a real quantum computer or in a quantum computer simulator.

17. A quantum neural network computing method, comprising: Let the complex number of first quantum gates of the quantum feature graph circuit perform Adama operations on the complex number of qubits respectively; The process involves: 1) Performing an RY rotation operation on each qubit after operation by the first quantum gate using a plurality of second quantum gates of the quantum feature map circuit; 2) Performing an RY rotation operation on each qubit after operation by the quantum feature map circuit using a plurality of third quantum gates of the variational model circuit; 3) Performing a controlled inverse gate operation on any two of the plurality of qubits using a plurality of fourth quantum gates of the variational model circuit; 4) Performing an RY rotation operation on each qubit after operation by the plurality of fifth quantum gates of the variational model circuit using a plurality of fourth quantum gates; 5) Performing an RY rotation operation on each qubit after operation by the plurality of fourth quantum gates of the variational model circuit; 6) Performing an RY rotation operation on each qubit after operation by the plurality of fourth quantum gates; 7) Performing an RY rotation operation on each qubit after operation by the third quantum gates and each fifth quantum gate using a plurality of weight variables; and 8) Measuring the quantum state of each qubit after passing through the variational model circuit to calculate a predicted value based on each quantum state.

18. The quantum neural network computing method as described in claim 17, wherein, The operation of the variational model circuit is repeated once or multiple times, and the controlled reverse gate operation is continued until each of the qubits has performed the controlled reverse gate operation.

19. The quantum neural network computing method as described in claim 17, wherein, The predicted value is used for regression or classification applications.

20. The quantum neural network computing method as described in claim 17, wherein, The predicted value is compared with the true value to generate a loss value, thereby performing a quantum optimization method to adjust the values ​​of the complex weight variables according to the loss value.

21. A method for analyzing energy consumption data of transportation vehicles, comprising: Obtain vehicle driving records and energy consumption records; The vehicle driving record is used as the plurality of input variable values ​​to be input into a quantum neural network as described in any one of claims 5 to 7, the plurality of input variable values ​​being used as angular parameter values ​​for the RY rotation operation of each of the second quantum gates, and the energy consumption record being used as a plurality of true values; the quantum neural network is trained based on the plurality of input variable values ​​and the plurality of true values ​​to adjust the plurality of weight variable values ​​of the variational model circuit; and an energy consumption prediction value is generated for an instantaneous vehicle driving record using the trained quantum neural network.

22. The method for analyzing vehicle energy consumption data as described in claim 21, wherein, The vehicle driving record is preprocessed to map to the value range of [-1.5, 1.5], and then used as the values ​​of the multiple input variables to serve as the angle parameter values ​​for the RY rotation operation of each of the second quantum gates.

23. The method for analyzing vehicle energy consumption data as described in claim 21, wherein, The energy consumption record is preprocessed to map to the range of [-1,1], and then used as the complex true values. These complex true values ​​are then used to train the quantum neural network to adjust the complex weight variable values ​​of the variational model circuit and generate regression prediction values.

24. The method for analyzing vehicle energy consumption data as described in claim 21, wherein, The energy consumption record is preprocessed to map to the value range of [0,1], and then used as the complex true values. These complex true values ​​are then used to train the quantum neural network to adjust the complex weight variable values ​​of the variational model circuit and generate classification prediction values.

25. The method for analyzing vehicle energy consumption data as described in claim 21, wherein, The vehicle driving record includes the number of data entries for each speed range accumulated over a period of time, the energy consumption record includes the amount of fuel refueled during that period of time, and the predicted value is the fuel consumption prediction value.

26. The method for analyzing vehicle energy consumption data as described in claim 21, wherein, The vehicle driving record includes the number of data entries accumulated over a period of time for each speed range, the energy consumption record includes the battery level at the start and end of the period, and the predicted value is the predicted energy consumption value.

Images