A multi-energy coordinated fast scheduling method based on quantum neural network

Abstract

The present application relates to a kind of quantum neural network-based multi-energy coordination fast scheduling method, belong to microgrid operation technical field.Clear the multi-energy coordination economic dispatching model considering water power, wind energy and solar energy, consider the multiple operating constraints of cascade hydropower.Construct the synchronous machine information fast prediction model based on QNN, model according to new energy and load measured data, realized the fast prediction of synchronous machine state and output in day.Construct the closed-loop warm start framework of QNN and multi-energy coordination optimization scheduling model, the decision information of QNN is passed to initial optimization model, the optimal solution of solution is fed back to QNN learning evolution, and then high-quality optimization scheduling result is quickly obtained.

Description

Technical Field

[0001] This invention belongs to the field of microgrid operation technology, specifically relating to a multi-energy coordinated fast scheduling method based on quantum neural networks. Background Technology

[0002] For the coordinated scheduling of complementary power generation systems with multidimensional uncertainties, including wind, solar, hydropower, and other renewable energy sources, stochastic optimization and robust optimization methods are typically used for modeling and processing. However, stochastic optimization methods have low reliability and computational efficiency, while robust optimization suffers from overly conservative optimization results. Modeling multi-energy complementary systems introduces a large number of variables and constraints, resulting in high model complexity. Solution methods based on C&CG (Column-and-Constraint Generation), Benders, and heuristic algorithms cannot achieve fast solutions.

[0003] In recent years, numerous scholars have applied neural networks to short-term load forecasting, short-term economic scheduling, and power flow forecasting. Neural network methods possess powerful capabilities for handling high-dimensional nonlinear data, and well-trained models exhibit strong online application capabilities. However, as problem complexity increases and the amount and quantity of sample information grow, neural network methods suffer from information loss. Furthermore, multi-energy complementary scheduling models involve a large number of state variables, which can lead to information loss when the neural network is transformed into a differentiable form. All of these factors limit the model's accurate mapping capabilities.

[0004] Quantum computing, by introducing the concept of quantum superposition, possesses parallel computing capabilities and, compared to traditional neural network models, has a higher memory capacity, eliminating catastrophic amnesia and making it suitable for demanding computations involving massive amounts of data. Current research utilizes radial basis function neural network models optimized by quantum particle swarm optimization to automatically optimize network structure configuration and acquire model parameters; there are quantum transient stability evaluation methods based on quantum machine learning design; the parallelism and efficiency of quantum neural networks (QNNs) facilitate faster search of the solution space, rapid coordination of multi-energy utilization, and real-time decision-making based on ultra-short-term prediction data. Simultaneously, quantum neural networks have higher memory capacity and better processing capabilities for large-scale state spaces, minimizing information loss when training input-output mapping models for multi-energy complementary systems, thus ensuring accurate model mapping capabilities.

[0005] Therefore, at this stage, it is necessary to design a multi-energy coordinated fast scheduling method based on quantum neural networks to solve the above problems. Summary of the Invention

[0006] The purpose of this invention is to provide a multi-energy coordinated rapid scheduling method based on quantum neural networks (QNNs) to address the technical problems existing in the prior art. It clarifies a multi-energy coordinated economic scheduling model that considers hydropower, wind power, and solar power, and takes into account various operational constraints of cascade hydropower. A rapid prediction model for synchronous machine information based on QNNs is constructed. This model, based on measured data of new energy sources and loads, achieves rapid intraday prediction of synchronous machine status and output. A closed-loop hot-start framework for QNNs and the multi-energy coordinated optimization scheduling model is constructed. QNN decision information is passed to the initial optimization model, and the optimal solution is fed back to the QNN for learning and evolution, thereby rapidly obtaining high-quality optimized scheduling results.

[0007] To achieve the above objectives, the technical solution of the present invention is as follows:

[0008] A fast multi-energy coordinated scheduling method based on quantum neural networks includes the following steps:

[0009] Step S1: Propose a scheduling model that takes into account the coordination of multiple energy sources; consider the coordinated operation cost of hydropower, wind power and solar power, as well as the electricity purchase cost for operation;

[0010] Step S2: By constructing a power generation prediction model based on quantum neural networks, the QNN model is trained using a large number of training samples containing historical renewable energy output level measurement data, load level, generator output level, and unit start-up and shutdown status data features. A mapping model is established with real-time wind, solar, and load measurements as inputs and unit output and unit combination as outputs.

[0011] Step S3: Construct a closed-loop hot-start framework for the QNN and multi-energy system optimization model. First, using an offline-trained QNN model, based on measured data of new energy sources and loads, predict the active power output and start-up / shutdown combinations of the synchronous machine, using this as the initial solution for the multi-energy complementary coordinated scheduling model; the optimal solution obtained through optimization is fed back to the QNN model to encourage further learning and evolution, thereby improving prediction accuracy.

[0012] Furthermore, step S1 is detailed as follows:

[0013] The specific form of the objective is as follows:

[0014]

[0015] in and These are the electricity purchase cost for time period t, the operating costs of the cascade hydropower system, and the wind / solar power station, respectively.

[0016] The constraints include predictive information from cascade hydropower systems and wind / solar systems to form operational constraints, including power balance constraints, water balance constraints, unit operation constraints, and grid constraints.

[0017] Equation (2) represents the power balance constraint of the power grid:

[0018]

[0019] In the formula, Ω represents the power purchase status parameter during time t, and P o,t , These represent purchased power, cascade hydropower output, wind / solar generator output, and active power of the load, respectively.

[0020] The reservoir's water storage capacity constraint is shown in equation (3);

[0021]

[0022] In the formula, V h,t , These represent the current water volume, minimum water storage capacity, and maximum water storage capacity of the reservoir, respectively.

[0023] The active power output constraint of cascade hydropower is shown in equation (4);

[0024]

[0025] In the formula, These represent the current output, minimum active power output, and actual maximum output level of the cascade hydropower, respectively.

[0026] The discharge capacity constraint of the cascade hydropower is shown in equation (5);

[0027]

[0028] In the formula, These represent the current discharge volume, minimum discharge volume, and maximum discharge volume of the cascade hydropower project, respectively.

[0029] The operational constraints of the cascade hydropower system are shown in equations (6) to (8), which are the water balance constraint, inter-stage hydraulic connection constraint and hydropower unit ramp rate constraint, respectively.

[0030]

[0031]

[0032]

[0033] In the formula, I h,t , These represent the reservoir inflow, power generation flow, and outflow of the h-th reservoir within time t; τ and L h,t It is the delay factor for inter-stage flow and inter-stage inflow in a cascade hydropower system; δ L δ UThese represent the minimum and maximum power change rates of the cascade hydropower system, respectively.

[0034] The output constraints of wind / solar new energy units are shown in the following formula;

[0035]

[0036] In the formula, Actual measured data on the contribution of new energy sources;

[0037] Power flow constraints in power grids involve power flow calculations. The DC power flow method is used to solve for the power flow of cross-sectional tie lines, and the constraints are as follows:

[0038] P tline =B diag LB -1 (P t +P t H +P t IIG -P t L (10)

[0039]

[0040]

[0041] In the formula, P tline P represents the DC power flow of each branch; B and L are the admittance coefficient matrix and node connection matrix of the branches in the network, respectively; t , These are the vector forms of power purchased within period t, output power of the cascade hydropower system, output power of the wind / solar power station, and active power demand of the load; It is the maximum transmittable branch power; x l N is the reactance of the branch; N is the branch number in the network.

[0042] Furthermore, step S2 is detailed as follows:

[0043] In quantum theory, the state of a qubit can be represented as:

[0044] |Ψ>=a0|0>+a1|1> (13)

[0045] Where a0 and a1 are arbitrary complex numbers and satisfy the normalization requirement |a0| 2 +|a1| 2 =1, |a0| 2 and |a1| 2Let |1> and |0> represent the probabilities of the qubit collapsing to |1> and |0>, respectively; taking the probability of the quantum state collapsing to |1> as the imaginary part and the probability of the qubit collapsing to |0> as the real part, we express it in complex form as:

[0046] f(θ)=cos(θ)+jsin(θ) (14)

[0047] Where, the probability amplitude of |0> corresponds to the square of the real part; the probability amplitude of |1> corresponds to the square of the imaginary part; j is the imaginary unit; θ is the phase angle, and different phase angles correspond to different quantum states;

[0048] Quantum gates are the foundation for realizing quantum computing. Quantum gates represent quantum computing and contain its characteristics. According to equation (2), the two types of quantum gates that constitute the most basic universal quantum gate set can be expressed as:

[0049] One phase shift door:

[0050]

[0051] Two controlled non-gatekeepers:

[0052]

[0053] Here, k is a control variable. When k = 1, the quantum state is reversed; when k = 0, it is not reversed. Different θ values ​​correspond to different quantum states. The two types of quantum gates achieve the evolution and transformation of quantum states by changing the value of θ.

[0054] In quantum computing, the final step, quantum measurement or quantum collapse, transforms a quantum state into classical information with a probability amplitude; |Ψ> with probability |a0| 2 As a quantum measurement of the quantum 0 state, |Ψ> with probability |a1| 2 Quantum measurement as quantum 1 state;

[0055] The structure of quantum neural networks can be generally described in the following form:

[0056]

[0057] In the formula, π(x,θ) is the output of the network, ψ(x) is the quantum state of the input, the input x is encoded into a quantum state and applied to a series of quantum gates U(θ), ​​Z represents the Z gate in quantum computing, and b is the bias term;

[0058] The quantum neural network model adopts a multi-layer activation function quantum network model. The input layer and output layer structure are the same as the classical BP neural network structure, while the hidden layer quantum neurons draw on the idea of ​​quantum state superposition and adopt a multi-quantum energy level transformation function. Each multi-energy level function is a series of activation functions generated by superimposing a multi-layer sigmoid function with a single linear function.

[0059] The output function of the hidden layer nodes in the network structure is:

[0060]

[0061] In the formula: θ r For quantum interval; n s α represents the number of quantum intervals, the size of which is related to the predicted number of classifications; n V is the steepness factor; T This is the transpose of the input layer weights of the neural network; X represents the input of the neural network, and X is described in the following form:

[0062]

[0063]

[0064] In the formula, Represents the input sample set, These represent the photovoltaic (PV) generator set, the wind turbine generator set, and the load set, respectively; x i Running data containing p-dimensional features, where feature P vre P D Q D These represent the measured power data of renewable energy sources and loads during the day, respectively.

[0065] The sample output feature y(x) represents the start-up and shutdown status and active power output level of the synchronous generator;

[0066] The training algorithm for the quantum neural network model with multi-layer activation functions uses gradient descent; the loss function is applied to the parameter θ. i The differential can be expressed as:

[0067]

[0068] Expanding the last term and differentiating it, then removing the constant term, we get:

[0069]

[0070] Further expansion and through Hermitian conjugation, it can be transformed into the following formula:

[0071]

[0072] U(θ) is composed of multiple gates, each controlled by different parameters, and U(θ) can be constructed as follows.

[0073] Differentiating the G-gate, the expression for the G-gate is as follows:

[0074]

[0075] In each training cycle, the training algorithm not only updates the connection weights between neurons in different layers, but also updates the quantum interval of each neuron in the hidden layer; the quantum interval update algorithm for hidden layer neurons is as follows;

[0076] By minimizing the total class variance, the hidden layer quantum spacing is obtained. The update equation; the r quantum intervals of the i-th neuron in the hidden layer are:

[0077]

[0078]

[0079]

[0080]

[0081] In the formula, η is the network learning rate; λ k n is the steepness factor; s The number of neuron outputs; The input vector is x k The output of the i-th neuron in the hidden layer; C m This is a pattern class vector.

[0082] Furthermore, step S3 is as follows:

[0083] In the forecasting phase, measured data of daily renewable energy and load levels are input into the QNN to achieve rapid forecasting of synchronous generator output information. Two QNNs are needed to perform classification and regression tasks respectively to predict the start-up and shutdown status of the synchronous generators. g And those who have contributed their efforts G ;

[0084] During the online phase, a closed-loop hot-start framework for QNN and multi-energy system optimization model is constructed. Synchronous machine prediction data obtained during the prediction phase is used as auxiliary decision information for intraday optimization scheduling and passed to the multi-energy complementary coordinated optimization scheduling model as the initial solution to achieve hot start of the model, accelerate and guide the model to find the optimal solution. The obtained optimal solution is fed back to the QNN model to promote further learning and evolution of QNN and improve the accuracy of prediction results.

[0085] The mathematical representation of the scheduling initial unmapping model built based on QNN is shown below:

[0086]

[0087]

[0088]

[0089] Among them, 'a' contains measurement data of new energy sources and load. s represents the start / stop state of the synchronous machine obtained through QNN prediction. With output data This includes hydropower and traditional synchronous generator units; IniS is the corresponding initial optimization solution; the start-up and shutdown status of the unit is reflected in the model by whether the output of the synchronous machine is 0, that is, 0 indicates that the unit is off, and 1 indicates the opposite.

[0090] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0091] One of the beneficial effects of this scheme is that it clarifies a multi-energy coordinated economic dispatch model that considers hydropower, wind power, and solar power, and takes into account various operational constraints of cascade hydropower. A rapid prediction model for synchronous machine information based on QNN is constructed. This model, based on measured data of new energy sources and loads, achieves rapid intraday prediction of synchronous machine status and output. A closed-loop hot-start framework for the QNN and multi-energy coordinated optimization dispatch model is constructed. The QNN decision information is passed to the initial optimization model, and the optimal solution is fed back to the QNN for learning and evolution, thereby quickly obtaining high-quality optimized dispatch results. Attached Figure Description

[0092] Figure 1 This is a schematic diagram of the multi-hidden-layer quantum neural network structure of the present invention.

[0093] Figure 2 This is a schematic diagram of the joint optimization method of the present invention.

[0094] Figure 3 This is a schematic diagram of the ZD near-zone equivalent system of the present invention.

[0095] Figure 4 This is a schematic diagram illustrating the fitting effects of the two neural network models of this invention.

[0096] Figure 5 This is a schematic diagram illustrating the scheduling error and prediction error analysis of the two models of this invention.

[0097] Figure 6 This is a schematic diagram of the model optimization scheduling results of the present invention. Detailed Implementation

[0098] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only for explaining the invention and are not intended to limit the invention; that is, the described embodiments are merely some embodiments of the invention, and not all embodiments. The components of the embodiments of the invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations.

[0099] Therefore, the following detailed description of the embodiments of the invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the invention without inventive effort are within the scope of protection of the invention. It should be noted that relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0100] Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0101] A fast multi-energy coordinated scheduling method based on quantum neural network is proposed, which is mainly described in three parts. The first part is the optimization scheduling model that takes into account the complementary coordination of multiple energy sources. The second part is the power generation prediction model based on quantum neural network. The third part is the closed-loop hot start framework of QNN and multi-energy coordinated optimization scheduling model.

[0102] An optimized scheduling model considering multi-energy complementarity and coordination: The scheduling model proposed in this invention should consider the coordinated operation costs of hydropower, wind power, and solar power, while also taking into account the procurement costs of operation.

[0103] The specific form of the model objective for electricity cost is as follows:

[0104]

[0105] in and These represent the electricity purchase cost for time period t, the operating costs of the cascade hydropower system, and the wind / solar power station, respectively.

[0106] The model's constraints include operational constraints formed by the predicted information of the cascade hydropower system and the wind / solar system, including power balance constraints, water balance constraints, unit operation constraints, and grid constraints.

[0107] Equation (2) represents the power balance constraint of the power grid:

[0108]

[0109] In the formula, Ω represents the power purchase status parameter during time t, and P o,t , These represent purchased power, cascade hydropower output, wind / solar generator output, and active power of the load, respectively.

[0110] The reservoir's water storage capacity constraint is shown in equation (3);

[0111]

[0112] In the formula, V h,t , These represent the current water volume, minimum water storage capacity, and maximum water storage capacity of the reservoir, respectively.

[0113] The active power output constraint of cascade hydropower is shown in equation (4);

[0114]

[0115] In the formula, These represent the current output, minimum active power output, and actual maximum output level of the cascade hydropower, respectively.

[0116] The discharge capacity constraint of the cascade hydropower is shown in equation (5);

[0117]

[0118] In the formula, These represent the current discharge volume, minimum discharge volume, and maximum discharge volume of the cascade hydropower project, respectively.

[0119] The operational constraints of the cascade hydropower system are shown in equations (6) to (8), which are the water balance constraint, inter-stage hydraulic connection constraint and hydropower unit ramp rate constraint, respectively.

[0120]

[0121]

[0122]

[0123] In the formula, I h,t , These represent the reservoir inflow, power generation flow, and outflow of the h-th reservoir within time t; τ and L h,t It is the delay factor for inter-stage flow and inter-stage inflow in a cascade hydropower system; δ L δ U These represent the minimum and maximum power change rates of the cascade hydropower system, respectively.

[0124] The output constraints of wind / solar new energy units are shown in the following formula;

[0125]

[0126] In the formula, Actual measured data on the contribution of new energy sources;

[0127] Power flow constraints in power grids involve power flow calculations, which can be categorized into AC power flow and DC power flow calculations in optimization. AC power flow calculations are complex and difficult to solve quickly in optimization. Since the resistance of tie lines in transmission lines is often much smaller than the line reactance, solving the power flow of tie lines in cross-sections using DC power flow methods is easier to implement. The constraints are shown below:

[0128] P tline =B diag LB -1 (P t +P t H +P t IIG -P t L (10)

[0129]

[0130]

[0131] In the formula, P tline P represents the DC power flow of each branch; B and L are the admittance coefficient matrix and node connection matrix of the branches in the network, respectively; t , These are the vector forms of power purchased within period t, output power of the cascade hydropower system, output power of the wind / solar power station, and active power demand of the load; It is the maximum transmittable branch power; x l N is the reactance of the branch; N is the branch number in the network.

[0132] A power generation prediction model based on quantum neural networks addresses the challenge of using optimization theory to formulate scheduling strategies, which struggles to handle the "dimensionality explosion" problem brought about by high-penetration renewable energy sources and cannot achieve rapid solutions for intraday scheduling strategies. Quantum computing can rapidly process large-scale data, and quantum neural networks far surpass traditional neural networks in their computational capabilities for high-dimensional features. To address this, this invention constructs a quantum neural network prediction model, establishing a mapping model with real-time measurements of wind, solar, and load as inputs and unit outputs as outputs, enabling real-time "hot start" optimization initialization of intraday scheduling decision variables.

[0133] In quantum theory, the state of a qubit can be represented as:

[0134] |Ψ>=a0|0>+a1|1> (13)

[0135] Where a0 and a1 are arbitrary complex numbers and satisfy the normalization requirement |a0| 2 +|a1| 2 =1, |a0| 2 and |a1| 2 Let |1> and |0> represent the probabilities of the qubit collapsing to |1> and |0>, respectively. Taking the probability of the quantum state collapsing to |1> as the imaginary part and the probability of the qubit collapsing to |0> as the real part, we express it in complex form as:

[0136] f(θ)=cos(θ)+jsin(θ) (14)

[0137] In this context, the probability amplitude of |0> corresponds to the square of the real part; the probability amplitude of |1> corresponds to the square of the imaginary part; j is the imaginary unit; and θ is the phase angle, with different phase angles corresponding to different quantum states.

[0138] Quantum gates are the foundation for realizing quantum computing; they represent quantum computing and encompass its characteristics. According to equation (2), the two types of quantum gates that constitute the most basic universal quantum gate set can be expressed as:

[0139] One phase shift door:

[0140]

[0141] Two controlled non-gatekeepers:

[0142]

[0143] Here, k is a control variable. When k = 1, the quantum state is reversed; when k = 0, it is not reversed. Different θ values ​​correspond to different quantum states. The two types of quantum gates achieve the evolution and transformation of the quantum state by changing the value of θ.

[0144] In quantum computing, the final step, quantum measurement or quantum collapse, transforms a quantum state into classical information (a definite value) through a probability amplitude. |Ψ> with probability |a0| 2 As a quantum measurement of the quantum 0 state, |Ψ> with probability |a1| 2 Quantum measurement as quantum 1 state.

[0145] The structure of quantum neural networks can be generally described in the following form:

[0146]

[0147] In the formula, π(x,θ) is the output of the network, ψ(x) is the input quantum state, the input x is encoded into a quantum state and applied to a series of quantum gates U(θ), ​​Z represents the Z gate in quantum computing, and b is the bias term.

[0148] In this invention, the quantum neural network model employs a multi-layer activation function quantum network model. The input and output layer structures of the QNN are identical to those of the classic BP neural network, while the hidden layer quantum neurons borrow the idea of ​​quantum state superposition, using multi-quantum level transformation functions. Each multi-level function is generated by superimposing a series of activation functions with multiple layers of sigmoid functions and a single layer of linear functions, as shown in the diagram. Figure 1 As shown.

[0149] The output function of the hidden layer nodes in the network structure is:

[0150]

[0151] In the formula: θ r For quantum interval; n s α represents the number of quantum intervals, the size of which is related to the predicted number of classifications; n V is the steepness factor; T X represents the transpose of the input layer weights of the neural network; X represents the input of the neural network, and in this invention, X is described in the following form:

[0152]

[0153]

[0154] In the formula, Represents the input sample set, These represent the sets of photovoltaic units, wind turbine units, and loads, respectively. i Running data containing p-dimensional features, where feature P vre P D Q D These represent the measured power data of renewable energy sources and loads during the day, respectively.

[0155] The sample output feature y(x) represents the start-up and shutdown status and active power output level of the synchronous generator.

[0156] The training algorithm for the quantum neural network model with multi-layer activation functions still employs gradient descent. The loss function is applied to the parameter θ. i The differential can be expressed as:

[0157]

[0158] Expanding the last term and differentiating it, then removing the constant term, we get:

[0159]

[0160] Further expansion and through Hermitian conjugation, it can be transformed into the following formula:

[0161]

[0162] U(θ) is composed of multiple gates, each controlled by different parameters. U(θ) can be constructed as a G-gate for differentiation. The expression for the G-gate is as follows:

[0163]

[0164] In each training cycle, the training algorithm updates not only the connection weights between neurons in different layers, but also the quantum spacing of neurons in the hidden layer. The former is the same as the update algorithm for a conventional backpropagation (BP) network, while the latter algorithm for updating the quantum spacing of hidden layer neurons is as follows.

[0165] By minimizing the total class variance, the hidden layer quantum spacing is obtained. The update equation is as follows. The r quantum intervals of the i-th neuron in the hidden layer are:

[0166]

[0167]

[0168]

[0169]

[0170] In the formula, η is the network learning rate; λ k n is the steepness factor; s The number of neuron outputs; The input vector is x k The output of the i-th neuron in the hidden layer; C m This is a pattern class vector.

[0171] A joint optimization method for multi-energy systems incorporating quantum neural networks: Joint optimization of multi-energy systems incorporating quantum neural networks first requires training a QNN model using a large number of training samples containing historical renewable energy output measurement data, load levels, generator output levels, and unit start-up and shutdown states. The input features of the training samples are historical wind / power active power output measurement data P. vre Load active / reactive power measurement data P D Q D The training label is the synchronous unit start-up and shutdown status. g With merit and effort P G .

[0172] In the forecasting phase, measured data on daily renewable energy and load levels are input into the QNN to achieve rapid forecasting of synchronous generator output information. It is worth noting that two QNNs are needed here to perform classification and regression tasks respectively, in order to predict the start-up and shutdown status of the synchronous generators. gAnd those who have contributed their efforts G .

[0173] During the online phase, a closed-loop hot-start framework for the QNN and the multi-energy system optimization model is constructed. Synchronous machine prediction data obtained during the forecast phase is used as auxiliary decision-making information for intraday optimal scheduling and passed to the multi-energy complementary coordinated optimization scheduling model as an initial solution, achieving a hot start for the model and accelerating and guiding it to find the optimal solution. The obtained optimal solution is fed back to the QNN model, prompting the QNN to further learn and evolve, improving the accuracy of the prediction results.

[0174] The specific process of joint optimization is as follows: Figure 2 As shown.

[0175] The mathematical representation of the scheduling initial unmapping model built based on QNN is shown below:

[0176]

[0177]

[0178]

[0179] Among them, 'a' contains measurement data of new energy sources and load. s represents the start / stop state of the synchronous machine obtained through QNN prediction. With output data This includes hydroelectric and traditional synchronous generating units; IniS

[0180] This is the corresponding initial optimization solution. It is worth noting that the unit's start-up and shutdown status in this invention's model is reflected by whether the synchronous machine output is 0; that is, 0 indicates the unit is off, and 1 indicates the opposite.

[0181] Since the QNN model used for synchronous machine information prediction is trained offline on a large amount of data, it can quickly obtain decision results on the output and state of synchronous generators based on measured data from new energy sources. This accelerates the scheduling of multiple energy sources that are not coordinated, avoids the adverse effects of new energy prediction errors on scheduling results, and enables rapid decision-making on intraday power generation strategies for multi-energy complementary systems. Currently, QNNs can be built with the assistance of various tools, including Qiskit, Pennylane, and Tensorflow based on Python. The QNN part in the example of this invention is built based on the Qiskit toolkit in Python, and the built QNN model has three quantum hidden layers.

[0182] Example Analysis:

[0183] The proposed QNN-based multi-energy complementary coordinated scheduling model was experimentally applied in the ZD near-region of the Sichuan power grid to verify the feasibility of the proposed method. The equivalent system in the ZD near-region is as follows: Figure 3 As shown, based on the topology and basic data, the region is simplified into a system of 29 nodes, including power plants such as Batang Power Plant, Suwalong, Kajiwa, Yangfanggou and Kala. The operation mode of Panzhihua New Energy Dafeng is used as the benchmark operation mode. Under this operation mode, Gannan grid-connected 1700MW, ZD collection station photovoltaic grid-connected 600MW, and Jinshang DC transmission power is 5 million kilowatts.

[0184] Analysis of the Classification and Regression Performance of Quantum Neural Networks:

[0185] Comparing the QNN and BP models for fitting the active power output of synchronous generators under different operating conditions, by Figure 4 (a)

[0186] As can be seen, the generator output curve obtained after fitting the BP network model has a large error compared with the actual curve. Figure 4 (b) It can be seen that the generator output curve obtained after fitting the QNN model has a high degree of fit with the actual curve, with small error and better fitting effect.

[0187] QNN achieves a classification accuracy of 99.68%, and can accurately determine the start-up and shutdown status of generator units when there is sufficient training data. Compared to the traditional BP algorithm, its accuracy is improved by 5.34%.

[0188] Table 1. Classification accuracy of CART decision trees and optimal decision trees

[0189]

[0190] As can be seen, the QNN model proposed in this invention performs better than the traditional BP neural network in both classification and regression tasks.

[0191] Model solution time:

[0192] The start-up and shutdown status and output prediction information of the synchronous generator set obtained by the QNN model are passed to the multi-energy complementary coordinated scheduling optimization model as the initial solution to achieve hot start and accelerate the optimization solution speed of the model. Compared with the model without embedding the optimal decision tree, the solution time is reduced by 396.7523 seconds and the solution efficiency is improved by 52.54%.

[0193] Table 2 Comparison of Model Solving Time

[0194]

[0195] Analysis of power output deviation in the dispatching of new energy generating units:

[0196] like Figure 5 As shown, the deviations between the day-ahead scheduling and the QNN-based multi-energy coordinated scheduling model of this invention for the optimized output of new energy units and the measured data are compared. Taking the wind turbine output capacity in the ZD region as an example, without considering the obstruction of new energy consumption,... Figure 5 (a) As can be seen, day-ahead dispatching typically optimizes the power generation plan for the next day based on the future day's renewable energy forecast curve. Therefore, the renewable energy forecast error is relatively large, resulting in a significant deviation from the actual measured renewable energy output level curve. For example... Figure 5 As shown in (b), the method of the present invention makes decisions based on the measured data of the output level of the new energy units at the current moment. The time scale is short and the prediction error and scheduling error are small.

[0197] Analysis of Multi-Energy Coordination Scheduling Results Based on QNN:

[0198] The results of coordinated scheduling of water, wind and solar power are as follows Figure 6 As shown, the model proposed in this invention can complete the coordinated and complementary scheduling of hydropower, wind power, and solar power in the ZD near-zone of the Sichuan power grid. Furthermore, during the period when the output level of new energy units is high (10:00-16:00), the synchronous generator units reduce their output or even shut down some units to make room for new energy units and prevent the situation of new energy consumption being hindered. This verifies the rationality and feasibility of the model in scheduling in a multi-energy complementary system. Table 3 shows some of the scheduling results.

[0199] Table 3 shows some scheduling results.

[0200]

[0201] Model cost analysis:

[0202] Table 4 compares the scheduling costs of the QNN-based multi-energy complementary coordinated scheduling model proposed in this invention with those of the traditional day-ahead unit combination model. The scheduling cost of the method proposed in this invention is 148.7216 / ￥ / MWh, which is much lower than the average unit scheduling cost of 185.5362 / ￥ / MWh of the traditional day-ahead unit combination. Therefore, the QNN-based multi-energy coordinated scheduling method proposed in this invention is more economical.

[0203] Table 4 Comparison of Model Scheduling Costs

[0204]

[0205] This invention proposes a multi-energy complementary coordinated scheduling model based on QNN, enabling rapid intraday power generation strategy decision-making with the goal of prioritizing the consumption of new energy sources. Furthermore, the effectiveness of this model was verified through experiments in the ZD near-field region of the Sichuan power grid. The model demonstrates good performance in terms of solution speed, decision accuracy, scheduling cost, security stability, and robustness, and can quickly provide reliable intraday scheduling auxiliary decision-making information.

[0206] The above are preferred embodiments of the present invention. Any changes made to the technical solution of the present invention that do not exceed the scope of the technical solution of the present invention shall fall within the protection scope of the present invention.

Claims

1. A fast multi-energy coordinated scheduling method based on quantum neural networks, characterized in that, Includes the following steps: Step S1: Propose a scheduling model that takes into account the coordination of multiple energy sources; consider the coordinated operation cost of hydropower, wind power and solar power, as well as the electricity purchase cost for operation; Step S2: By constructing a power generation prediction model based on quantum neural networks, the QNN model is trained using a large number of training samples containing historical renewable energy output level measurement data, load level, generator output level, and unit start-up and shutdown status data features. A mapping model is established with real-time wind, solar, and load measurements as inputs and unit output and unit combination as outputs. Step S3: Construct a closed-loop hot-start framework for QNN and multi-energy system optimization model: First, using an offline-trained QNN model, based on measured data of new energy sources and loads, predict the active power output and start-up / shutdown combinations of the synchronous machine, which serves as the initial solution for the multi-energy complementary coordinated scheduling model; the optimal solution obtained through optimization is fed back to the QNN model to encourage further learning and evolution, thereby improving prediction accuracy. Step S1 is as follows: The specific form of the objective is as follows: (1) in , and They are respectively t Time-of-use electricity purchase costs, operating costs of cascade hydropower systems and wind / solar power stations; The constraints include predictive information from cascade hydropower systems and wind / solar systems to form operational constraints, including power balance constraints, water balance constraints, unit operation constraints, and grid constraints. Equation (2) represents the power balance constraint of the power grid: (2) In the formula, Indicates time t Electricity purchase status parameters during the period , , , These represent purchased power, cascade hydropower output, wind / solar generator output, and active power of the load, respectively. The reservoir's water storage capacity constraint is shown in equation (3); (3) In the formula, , , These represent the current water volume, minimum water storage capacity, and maximum water storage capacity of the reservoir, respectively. The active power output constraint of cascade hydropower is shown in equation (4); (4) In the formula, , , These represent the current output, minimum active power output, and actual maximum output level of the cascade hydropower, respectively. The discharge capacity constraint of the cascade hydropower is shown in equation (5); (5) In the formula, , , These represent the current discharge volume, minimum discharge volume, and maximum discharge volume of the cascade hydropower project, respectively. The operational constraints of the cascade hydropower system are shown in equations (6) to (8), which are the water balance constraint, inter-stage hydraulic connection constraint and hydropower unit ramp rate constraint, respectively. (6) (7) (8) In the formula, , The first h The reservoir in time t The inflow, power generation, and outflow of the reservoir within the water body; and It is the delay factor for inter-stage water flow and inter-stage inflow in a cascade hydropower system; , These represent the minimum and maximum power change rates of the cascade hydropower system, respectively. The output constraints of wind / solar new energy units are shown in the following formula; (9) In the formula, Actual measured data on the contribution of new energy sources; Power flow constraints in power grids involve power flow calculations. The DC power flow method is used to solve for the power flow of cross-sectional tie lines, and the constraints are as follows: (10) (11) (12) In the formula, where This is the DC power flow of each branch; B , L These are the admittance coefficient matrix and node connection matrix of the branches in the network, respectively; , , , They are t The vector form of power purchased within the cycle, output power of cascade hydropower system, output power of wind / solar power station, and active power demand of load; It is the maximum transmittable branch power; It is the reactance of the branch circuit; N It is the branch number in the network.

2. The multi-energy coordinated fast scheduling method based on quantum neural networks according to claim 1, characterized in that, Step S2 is as follows: In quantum theory, the state of a qubit can be represented as: (13) in, and Let be any complex number and satisfy the normalization requirement , and These represent the qubit collapsing to... and The probability of collapsing a quantum state to The probability of is taken as the imaginary part, and the qubit collapses to The probability of is expressed in complex form as the real part: (14) in, The probability amplitude corresponds to the square of the real part; The probability amplitude corresponds to the square of the imaginary part; j is the imaginary unit; It is the phase angle; different phase angles correspond to different quantum states. Quantum gates are the foundation for realizing quantum computing. Quantum gates represent quantum computing and contain its characteristics. According to equation (2), the two types of quantum gates that constitute the most basic universal quantum gate set can be expressed as: One phase shift door: (15) Two controlled non-gatekeepers: (16) in, k It is a control variable, when k When =1, the quantum state is reversed; k When =0, it does not reverse; different θ Corresponding to different quantum states, the two types of quantum gates change θ Values ​​enable the evolution and transformation of quantum states; In quantum computing, a quantum state is finally transformed into classical information through quantum measurement, or quantum collapse, in the form of probability amplitude. With probability As a quantum measurement of the quantum 0 state, With probability Quantum measurement as quantum 1 state; The structure of quantum neural networks can be generally described in the following form: (17) In the formula, It is the output of the network. It is the input quantum state, the input After being encoded into quantum states, they are applied to a series of quantum gates. , Z Representing quantum computing Z Door, b It is a bias term; The quantum neural network model adopts a multi-layer activation function quantum network model. The input layer and output layer structure are the same as the classical BP neural network structure, while the hidden layer quantum neurons draw on the idea of ​​quantum state superposition and adopt a multi-quantum energy level transformation function. Each multi-energy level function is a series of activation functions generated by superimposing a multi-layer sigmoid function with a single linear function. The output function of the hidden layer nodes in the network structure is: (18) In the formula: For quantum intervals; The number of quantum intervals, the size of which is related to the predicted number of categories; Steepness factor; This is the transpose of the input layer weights of the neural network. X This represents the input to the neural network. X The description is in the following form: (19) (20) In the formula, Represents the input sample set, , , These represent sets of photovoltaic units, sets of wind turbine units, and sets of loads, respectively. Include p Dimensional feature running data, features , , These represent the measured power data of renewable energy sources and loads during the day, respectively. Sample output features It refers to the start-up and shutdown status and active power output level of the synchronous generator; The training algorithm for the quantum neural network model with multi-layer activation functions uses gradient descent; the loss function modulates the parameters. The differential can be expressed as: (21) Expanding the last term and differentiating it, then removing the constant term, we get: (22) Further expansion and through Hermitian conjugation, it can be transformed into the following formula: (23) in, It consists of multiple doors, each controlled by different parameters, which can... Constructed as G Differentiate the gate. G The gate expression is as follows: (24) In each training cycle, the training algorithm not only updates the connection weights between neurons in different layers, but also updates the quantum interval of each neuron in the hidden layer; the quantum interval update algorithm for hidden layer neurons is as follows; By minimizing the total class variance, the hidden layer quantum spacing is obtained. The update equation for the hidden layer; i one neuron r The quantum intervals are: (25) (26) (27) (28) In the formula, η For network learning rate; Steepness factor; The number of neuron outputs; The input vector is Hidden layer i The output of each neuron; This is a pattern class vector.

3. The multi-energy coordinated fast scheduling method based on quantum neural networks according to claim 2, characterized in that, Step S3 is as follows: In the forecasting phase, the measured data of daily renewable energy and load levels are input into the QNN to achieve rapid forecasting of synchronous generator output information. Two QNNs are needed to perform classification and regression tasks respectively to predict the start-up and shutdown status of the synchronous generators. and those who have contributed ; During the online phase, a closed-loop hot-start framework for QNN and multi-energy system optimization model is constructed. Synchronous machine prediction data obtained during the prediction phase is used as auxiliary decision information for intraday optimization scheduling and passed to the multi-energy complementary coordinated optimization scheduling model as the initial solution to achieve hot start of the model, accelerate and guide the model to find the optimal solution. The obtained optimal solution is fed back to the QNN model to promote further learning and evolution of QNN and improve the accuracy of prediction results. The mathematical representation of the scheduling initial unmapping model built based on QNN is shown below: in, Measurement data including new energy sources and load. ; Synchronous machine start / stop states obtained through QNN prediction With output data , Including hydropower and traditional synchronous generating units; This is the corresponding initial optimization solution; In the model, the start-up and shutdown status of the unit is reflected by whether the output of the synchronous machine is 0, that is, 0 means the unit is off, and 1 means the unit is on.

Images