Timing sequence prediction method, system and device based on CEEMDAN enhanced TCN and QLSTM fusion framework, and medium
By using a fusion framework of CEEMDAN, QLSTM, and TCN to decompose and predict power load data, the problems of model complexity and high computational cost are solved, achieving efficient multi-scale information processing and accurate power load forecasting.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- GUIZHOU POWER GRID CO LTD
- Filing Date
- 2025-12-18
- Publication Date
- 2026-05-05
AI Technical Summary
Existing power load forecasting models are highly complex and computationally expensive, making it difficult to simultaneously extract features that account for both short-term drastic fluctuations and long-term trend dependence. Furthermore, traditional decomposition methods are prone to mode aliasing, resulting in low accuracy.
The original load data is decomposed using Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN). The high-frequency and low-frequency component sequences are processed separately by combining Quantum Long Short-Term Memory Network (QLSTM) and Temporal Convolutional Network (TCN), and the results are fused by a hybrid prediction model.
It reduces the arbitrariness and variance of load data, improves the ability to capture long-term dependencies and short-term dynamic changes, reduces the demand for computing resources, and lowers the energy consumption and cost of power load forecasting.
Smart Images
Figure CN121980490A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of power load forecasting, and in particular to a time-series forecasting method, system, device, and medium based on a CEEMDAN-enhanced TCN and QLSTM fusion framework. Background Technology
[0002] Load forecasting is a crucial component of power system planning and operation management, addressing key issues such as power supply and demand balance, grid planning, economic dispatch, energy cost management, system reliability, and stability. It is essential for ensuring the economic feasibility, reliability, and environmental sustainability of the power system. With technological advancements, load forecasting methods have evolved to meet the dynamic demands of modern power systems. Traditional LSTM-TCN hybrid models demonstrate significant capabilities in handling multi-temporal scale characteristics. However, computational costs increase exponentially with model depth and complexity.
[0003] Therefore, this invention conceives of a QLSTM-TCN hybrid module, cleverly utilizing quantum properties such as superposition, entanglement, and quantum parallelism. This module seamlessly integrates QLSTM and TCN, fully leveraging the advantages of both components. Inspired by these advancements, a novel neural network architecture is introduced, called CEEMADAN (Complete Ensemble Empirical Mode Decomposition with Adaptive Noise)-QLSTM (Quantum Long Short-Term Memory Network)-TCN (Temporal Convolutional Network). QLSTM has already demonstrated its potential in sequence prediction and text classification tasks, highlighting not only the feasibility of combining quantum computing with neural networks but also significantly reducing model complexity and accelerating convergence. By utilizing quantum mechanics and classical computing techniques, this invention aims to revolutionize the field of load forecasting, providing a more efficient and scalable solution to address the challenges of complex power systems. Summary of the Invention
[0004] In view of the problems of existing models being complex, computationally expensive, and having limited accuracy, this invention is proposed.
[0005] Therefore, this invention provides a time series forecasting method, system, device, and medium based on the CEEMDAN-enhanced TCN and QLSTM fusion framework to address the problems that existing raw power load data contains a large amount of randomness and non-stationarity, resulting in low accuracy when directly predicted, and that traditional decomposition methods are prone to mode aliasing; single models are difficult to simultaneously take into account the feature extraction of short-term drastic fluctuations (high frequency) and long-term trend dependence (low frequency); and the failure to effectively utilize the parameter advantages of quantum computing in processing high-dimensional features to optimize the time series forecasting model.
[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a time series prediction method based on a CEEMDAN-enhanced TCN and QLSTM fusion framework, comprising: The original load data was acquired and decomposed using the CEEMDAN method of adaptive noise complete set empirical mode decomposition to obtain high-frequency component sequences and low-frequency component sequences. A hybrid prediction model is constructed, which includes a parallel temporal convolutional network (TCN) module and a quantum long short-term memory (QLSTM) module. The high-frequency component sequence is predicted using the TCN module to obtain the high-frequency prediction result; the low-frequency component sequence is predicted using the QLSTM module to obtain the low-frequency prediction result. The final prediction result is obtained by fusing the high-frequency prediction results with the low-frequency prediction results.
[0007] As a preferred embodiment of the time-series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework described in this invention, the method involves: acquiring raw load data and decomposing it using the CEEMDAN method (Adaptive Noise Complete Ensemble Empirical Mode Decomposition) to obtain high-frequency component sequences and low-frequency component sequences, including: Paired adaptive noise is added to the original load data, and empirical mode decomposition is performed on the load data after adding noise to obtain multiple intrinsic mode function (IMF) components and residual terms. Multiple subsequences are constructed by using multiple intrinsic mode function (IMF) components and residual terms. These subsequences are then reconstructed to obtain high-frequency component sequences and low-frequency component sequences.
[0008] As a preferred embodiment of the time-series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework described in this invention, the quantum long short-term memory network (QLSTM) module includes: The Quantum Long Short-Term Memory (QLSTM) module comprises multiple stacked QLSTM units. A single QLSTM cell utilizes multiple variable quantum circuits (VQCs) to construct forget gates, input gates, and output gates.
[0009] As a preferred embodiment of the timing prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework described in this invention, the variable quantum circuit (VQC) is constructed from a feature coding layer, trainable parameters, an entanglement layer, and a measurement layer, comprising: The multi-feature representation is encoded using a feature coding layer through unitary transformation. The encoded quantum state is evolved using multiple trainable parameters and entanglement layers; The output characteristics of the variable quantum circuit VQC are obtained by measuring the evolved quantum state through the measurement layer.
[0010] As a preferred embodiment of the timing prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework described in this invention, wherein: the processing logic of a single QLSTM unit includes: The hidden state from the previous time step is concatenated with the input vector from the current time step to obtain the connection input vector; The connection input vector is processed by the first VQC and then... The activation function generates a forgetting gate vector, which is used to filter the cell states from the previous time step. The connection input vector is processed by the second VQC and then... The activation function obtains the input gate vector. The third VQC process connects the input vector and passes it through the hyperbolic tangent Tanh activation function to obtain a new cell state candidate. The current cell state is updated based on the input gate vector and the new cell state candidate. The connection input vector is processed by the fourth VQC and then... The activation function produces an output gate vector, which is then combined with the current cell state to obtain the result to be processed. The fifth and sixth VQCs are used to process the results to obtain the hidden state and output results at the current time.
[0011] As a preferred embodiment of the temporal prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework described in this invention, the temporal convolutional network TCN module includes: The high-frequency component sequence is received as input data, and the input data is processed through the parallel backbone branch and residual branch in the temporal convolutional network TCN module. In the main branch, dilated convolution is used to perform convolution operations on the input data to extract temporal features, and ReLU activation and Dropout are performed on the convolutional data in sequence to obtain the first feature result; In the residual branch, pointwise convolution is used to adjust the channels of the input data to obtain the second feature result; The first feature result and the second feature result are added element-wise to obtain the high-frequency prediction result.
[0012] As a preferred embodiment of the time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework described in this invention, the hybrid prediction model further employs the Levenberg-Marquardt algorithm for parameter optimization, including: Calculate the error vector between the final predicted result and the actual load value; Based on the error vector, calculate the Jacobian matrix of the model parameters; Construct a system of linear equations containing the Jacobian matrix and damping factor, and solve it to obtain the parameter update step size; Based on the error change trend after parameter updates, the damping factor is dynamically adjusted, and the VQC and TCN parameters in the hybrid prediction model are updated until the convergence condition is met.
[0013] Secondly, this invention provides a time series prediction system based on a CEEMDAN-enhanced TCN and QLSTM fusion framework, comprising: The data decomposition module is used to acquire the raw load data and decompose it using the CEEMDAN method of adaptive noise complete set empirical mode decomposition to obtain high-frequency component sequences and low-frequency component sequences. The model building module is used to build a hybrid prediction model, which includes a parallel temporal convolutional network (TCN) module and a quantum long short-term memory (QLSTM) module. The prediction module is used to predict the high-frequency component sequence using the TCN module to obtain a high-frequency prediction result; and to predict the low-frequency component sequence using the QLSTM module to obtain a low-frequency prediction result. The fusion module is used to fuse high-frequency prediction results with low-frequency prediction results to obtain the final prediction result.
[0014] Thirdly, the present invention provides an electronic device, comprising: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of a timing prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework.
[0015] Fourthly, the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by a processor, implement the steps of the timing prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework.
[0016] Compared with existing technologies, the beneficial effects of this invention are as follows: It utilizes CEEMDAN mode decomposition technology to perform time-series decomposition on the original load data, reducing the arbitrariness and variance of the load data; it combines QLSTM with TCN to achieve efficient prediction of low-frequency and high-frequency information, enabling the model to simultaneously capture and process long-term dependencies and short-term dynamic changes in time-series data, thus enhancing the multi-scale information processing capability of the prediction model; furthermore, the architecture of this invention is flexible, allowing for adjustments to the combination of QLSTM and TCN according to different scenarios and needs, exhibiting wide applicability and strong adaptability, reducing computational resource requirements, and decreasing energy consumption and costs in the power load forecasting process. Attached Figure Description
[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0018] Figure 1 This is a schematic diagram of the overall process of a time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework according to an embodiment of the present invention.
[0019] Figure 2 This is a fusion framework diagram of a time series prediction method based on a CEEMDAN-enhanced TCN and QLSTM fusion framework according to an embodiment of the present invention.
[0020] Figure 3 This is a basic unit structure diagram of the QLSTM in the time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework according to an embodiment of the present invention.
[0021] Figure 4 This is a VQC circuit structure diagram of a timing prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework according to an embodiment of the present invention.
[0022] Figure 5 This is a block diagram of a time series prediction method based on a CEEMDAN-enhanced TCN and QLSTM fusion framework according to an embodiment of the present invention.
[0023] Figure 6 This is a structural block diagram of a computer device for a time-series prediction method based on a CEEMDAN-enhanced TCN and QLSTM fusion framework, according to an embodiment of the present invention. Detailed Implementation
[0024] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.
[0025] Example 1, referring to Figure 1 As an embodiment of the present invention, a time series prediction method based on a CEEMDAN-enhanced TCN and QLSTM fusion framework is provided, comprising: S100: Obtain the raw load data and decompose it using the CEEMDAN method of adaptive noise complete set empirical mode decomposition to obtain high-frequency component sequences and low-frequency component sequences. S200: Construct a hybrid prediction model, which includes a parallel temporal convolutional network (TCN) module and a quantum long short-term memory (QLSTM) module; S300: The high-frequency component sequence is predicted using the TCN module to obtain the high-frequency prediction result; the low-frequency component sequence is predicted using the QLSTM module to obtain the low-frequency prediction result. S400: The high-frequency prediction results are fused with the low-frequency prediction results to obtain the final prediction result.
[0026] It should be noted that electricity load data is affected by various factors such as meteorological conditions, economic activities, and user behavior. When faced with massive amounts of high-dimensional time-series data, traditional deep learning models often face a trade-off between depth and efficiency: increasing network depth can improve feature extraction capabilities, but it also leads to problems such as parameter explosion, training time consumption, and gradient vanishing. Furthermore, purely classical models struggle to simultaneously decouple and model high-frequency random noise and low-frequency long-term trends in the data with limited computing resources. Therefore, it is necessary to explore a method that can integrate the advantages of high-dimensional mapping in quantum computing with the time-series capture capabilities of classical deep learning.
[0027] Therefore, to address the aforementioned problems, through steps S100-S400, the original load data is decomposed into time series using CEEMDAN mode decomposition technology to reduce the arbitrariness and variance of the load data. Then, a load forecasting model based on QLSTM and TCN is established, and QLSTM and TCN are used to predict low-frequency and high-frequency information respectively to achieve the final load data forecast.
[0028] Example 2, refer to Figures 1-4 As an embodiment of the present invention, based on the above embodiment, a time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework is provided.
[0029] In this embodiment of the application, step S100 involves acquiring the original load data and decomposing it using the CEEMDAN (Complete Ensemble Empirical Mode Decomposition) method for adaptive noise to obtain high-frequency component sequences and low-frequency component sequences, including: Specifically, the CEEMDAN method, which uses adaptive noise to completely enumerate empirical mode decomposition, is used to add paired adaptive noise to the original load data, so that it has good continuity at different time scales. Empirical mode decomposition is then performed on the load data after adding noise to obtain multiple intrinsic mode function (IMF) components and residual terms.
[0030] Furthermore, multiple subsequences are constructed using multiple intrinsic mode function (IMF) components and residual terms. These subsequences are then reconstructed to obtain high-frequency component sequences and low-frequency component sequences.
[0031] In this embodiment of the application, step S200 involves constructing a hybrid prediction model, which includes a parallel temporal convolutional network (TCN) module and a quantum long short-term memory (QLSTM) module, comprising: Specifically, the temporal convolutional network (TCN) module includes parallel backbone branches and residual branches. The backbone branches utilize dilated convolution, ReLU activation, and Dropout processing; the residual branches utilize pointwise convolution processing; and finally, the results from the two branches are fused.
[0032] Furthermore, the quantum long short-term memory network (QLSTM) module includes multiple stacked quantum long short-term memory network (QLSTM) units; a single QLSTM unit utilizes multiple variable quantum circuits (VQCs) to construct forget gates, input gates, and output gates.
[0033] It should be noted that the quantum long short-term memory network (QLSTM) unit can be composed of a single unit or multiple units stacked in series.
[0034] For example, such as Figure 2 The diagram shows the CEEMDAN-enhanced QLSTM and TCN fusion framework. The temporal reconstruction part includes a parallel temporal convolutional network TCN module and a quantum long short-term memory network QLSTM module.
[0035] In this embodiment of the application, step S300 involves using a TCN module to predict the high-frequency component sequence to obtain a high-frequency prediction result; and using a QLSTM module to predict the low-frequency component sequence to obtain a low-frequency prediction result, including the following steps A1-A2: A1: The high-frequency component sequence is predicted using the TCN module to obtain high-frequency prediction results; Specifically, the high-frequency component sequence is received as input data, and the input data is processed through the parallel backbone branch and residual branch in the temporal convolutional network TCN module; Furthermore, in the main branch, dilated convolution is used to convolve the input data to extract temporal features. The convolved data is then sequentially processed with ReLU activation (Rectified Linear Unit) to add non-linearity and Dropout (random deactivation) to prevent overfitting, yielding the first feature result. In the residual branch, pointwise convolution is used to adjust the channels of the input data, obtaining the second feature result. Dropout is a classic and efficient regularization technique whose core objective is to address the overfitting problem during model training.
[0036] Furthermore, the first feature result and the second feature result are added element-wise to obtain the high-frequency prediction result.
[0037] A2: The low-frequency component sequence is predicted using a QLSTM module to obtain the low-frequency prediction result; Specifically, QLSTM can learn features with timing information, mainly including input gates, output gates, and forget gates.
[0038] For example, the basic unit structure of QLSTM is as follows: Figure 3 As shown. Each QLSTM cell contains 6 VQCs, where each VQC represents a variable quantum circuit. The tanh blocks and tanh blocks represent activation functions, respectively. It is time Input time, It is in a hidden state. It's the cell status. It is the output. and These represent element-wise multiplication and addition, respectively.
[0039] exist arrive The input is the hidden state of the previous time step and the current input vector. The data is concatenated, and the output is a vector obtained from the measurements of each VQC. These measurements represent the Pauli score for each qubit. The expected value is then used as the input and update block and the output block through a non-linear activation function.
[0040] Furthermore, the processing steps of a single QLSTM unit include the following sub-steps S1-S5: S1: Concatenate the hidden state from the previous time step with the input vector from the current time step to obtain the connection input vector; Specifically, receive the input vector And obtain the hidden state of the QLSTM unit output at the previous time step. ,Will and Concatenate the vectors to obtain the concatenated input vectors. .
[0041] It should be noted that, for a given problem size, the total number of qubits n required for a VQC is determined in order to encode the input vector into a quantum state, which serves as the input state for subsequent VQCs.
[0042] S2: Process the connection input vector through the first VQC and then... The activation function generates a forgetting gate vector, which is used to filter the cell states from the previous time step. Specifically, S2, or the forget gate, utilizes... Decide to forget the previous cell state Which information in it? The input vector is processed and then... The activation function is used to process the data to obtain the forgetting gate vector. ,in, The interval is [0,1], the purpose of which is to determine whether to "forget" or "retain" the cell state. The corresponding elements in.
[0043] S3: Process the connection input vector through the second VQC and then... The activation function obtains the input gate vector. The third VQC process connects the input vector and passes it through the hyperbolic tangent Tanh activation function to obtain a new cell state candidate. The current cell state is updated based on the input gate vector and the new cell state candidate. Specifically, S3, the input gate, determines what new information to add to the cell state, and involves two parallel VQC processes: through... deal with and through The activation function obtains the input gate vector. This is to determine which values will be added to the cell status; deal with New candidate cell states are obtained by applying the hyperbolic tangent Tanh activation function. .
[0044] Furthermore, for the old cell state Update to obtain the current cell state. .
[0045] S4: Process the connection input vector through the fourth VQC and then... The activation function produces an output gate vector, which is then combined with the current cell state to obtain the result to be processed. Specifically, S4 is the output gate. After updating the cell state, the QLSTM cell is ready to decide what to output. deal with and through The activation function yields the output gate vector. .
[0046] Furthermore, the current cell state Processed using the tanh function, and combined with... Multiply them to get the result to be processed.
[0047] S5: Utilize and Process the results of each step to obtain the hidden state at the current time. and output results . Furthermore, each variable quantum circuit VQC includes: Specifically, such as Figure 4 The diagram shows the VQC circuit structure. The Variable Quantum Circuit (VQC) is constructed from a feature encoding layer, trainable parameters, an entanglement layer, and a measurement layer. The VQC first initializes L qubits, then processes the initialized features... The quantum state is encoded by unitary transformation after passing through a feature encoding layer. Next, N trainable parameters and an entanglement layer are used to evolve the encoded quantum state. Finally, the evolved quantum state is measured through a measurement layer to obtain the output characteristics of the variable quantum circuit VQC. The measurement layer is where quantum information collapses into T (T and L values are not necessarily the same) classical output values.
[0048] In this embodiment of the application, step S400, which fuses the high-frequency prediction result with the low-frequency prediction result to obtain the final prediction result, includes: Specifically, the high-frequency prediction results are added to the low-frequency prediction results to obtain the final prediction result.
[0049] Furthermore, the hybrid forecasting model also employs the Levenberg-Marquardt algorithm for parameter optimization, including: calculating the error vector between the final forecast result and the actual load value; Based on the error vector, calculate the Jacobian matrix of the model parameters; Construct a system of linear equations containing the Jacobian matrix and damping factor, and solve it to obtain the parameter update step size; Based on the error change trend after parameter updates, the damping factor is dynamically adjusted, and the VQC and TCN parameters in the hybrid prediction model are updated until the convergence condition is met.
[0050] In summary, the CEEMDAN-enhanced QLSTM and TCN fusion framework proposed in this invention utilizes CEEMDAN mode decomposition technology to perform time-series decomposition on the original load data, reducing the arbitrariness and variance of the load data. By inputting low-frequency and high-frequency information into QLSTM and TCN respectively, efficient prediction of low-frequency and high-frequency information is achieved, enabling the model to simultaneously capture and process long-term dependencies and short-term dynamic changes in time-series data, significantly enhancing the multi-scale information processing capability of the prediction model. Furthermore, the flexible architecture of this invention allows for adjustments to the combination of QLSTM and TCN according to different scenarios and needs, exhibiting broad applicability and strong adaptability. This reduces computational resource requirements, thereby decreasing energy consumption and costs in the power load forecasting process.
[0051] Example 3 illustrates a time series prediction method based on a CEEMDAN-enhanced TCN and QLSTM fusion framework. It should be noted that the technical solution of this time series prediction system based on the CEEMDAN-enhanced TCN and QLSTM fusion framework is based on the same concept as the aforementioned time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework. Details not described in detail in the technical solution of the time series prediction system based on the CEEMDAN-enhanced TCN and QLSTM fusion framework in this example can be found in the description of the technical solution of the aforementioned time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework.
[0052] This embodiment also provides a time series prediction system based on a CEEMDAN-enhanced TCN and QLSTM fusion framework, such as... Figure 5 As shown, it includes: The data decomposition module is used to acquire the raw load data and decompose it using the CEEMDAN method of adaptive noise complete set empirical mode decomposition to obtain high-frequency component sequences and low-frequency component sequences. The model building module is used to build a hybrid prediction model, which includes a parallel temporal convolutional network (TCN) module and a quantum long short-term memory (QLSTM) module. The prediction module is used to predict the high-frequency component sequence using the TCN module to obtain a high-frequency prediction result; and to predict the low-frequency component sequence using the QLSTM module to obtain a low-frequency prediction result. The fusion module is used to fuse high-frequency prediction results with low-frequency prediction results to obtain the final prediction result.
[0053] Figure 6A structural block diagram of a computer device illustrated in an embodiment of this application is shown. This computer device can be implemented as a server as described in the above-described scheme of this application. The computer device includes a Central Processing Unit (CPU), a system memory including Random Access Memory (RAM) and Read-Only Memory (ROM), and a system bus connecting the system memory and the CPU. The computer device also includes a mass storage device for storing the operating system, applications, and other program modules. The mass storage device is connected to the CPU via a mass storage controller (not shown) connected to the system bus. The mass storage device and its associated computer-readable medium provide non-volatile storage for the computer device. That is, the mass storage device may include computer-readable media (not shown) such as a hard disk or a Compact Disc Read-Only Memory (CD-ROM) drive. The aforementioned system memory and mass storage device can be collectively referred to as memory.
[0054] This embodiment also provides an electronic device applicable to timing prediction based on the CEEMDAN enhanced TCN and QLSTM fusion framework, comprising: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the timing prediction method based on the CEEMDAN enhanced TCN and QLSTM fusion framework as proposed in the above embodiment.
[0055] This embodiment also provides a storage medium storing a computer program that, when executed by a processor, implements the timing prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as proposed in the above embodiments.
[0056] The storage medium proposed in this embodiment and the timing prediction method based on the CEEMDAN enhanced TCN and QLSTM fusion framework proposed in the above embodiments belong to the same inventive concept. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.
[0057] Based on the above description of the implementation methods, those skilled in the art can clearly understand that the present invention can be implemented using software and necessary general-purpose hardware, and of course, it can also be implemented using hardware. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory, hard disk, or optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of the various embodiments of the present invention.
[0058] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.
Claims
1. A time series prediction method based on a CEEMDAN-enhanced TCN and QLSTM fusion framework, characterized in that, include: The original load data was acquired and decomposed using the CEEMDAN method of adaptive noise complete set empirical mode decomposition to obtain high-frequency component sequences and low-frequency component sequences. A hybrid prediction model is constructed, which includes a parallel temporal convolutional network (TCN) module and a quantum long short-term memory (QLSTM) module. The high-frequency component sequence is predicted using the TCN module to obtain the high-frequency prediction result; the low-frequency component sequence is predicted using the QLSTM module to obtain the low-frequency prediction result. The final prediction result is obtained by fusing the high-frequency prediction results with the low-frequency prediction results.
2. The time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as described in claim 1, characterized in that, The raw load data is acquired and decomposed using the CEEMDAN method (Complete Ensemble Empirical Mode Decomposition for Adaptive Noise) to obtain high-frequency and low-frequency component sequences, including: Paired adaptive noise is added to the original load data, and empirical mode decomposition is performed on the load data after adding noise to obtain multiple intrinsic mode function (IMF) components and residual terms. Multiple subsequences are constructed by using multiple intrinsic mode function (IMF) components and residual terms. These subsequences are then reconstructed to obtain high-frequency component sequences and low-frequency component sequences.
3. The time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as described in claim 2, characterized in that, The quantum long short-time memory (QLSTM) network module includes: The Quantum Long Short-Term Memory (QLSTM) module comprises multiple stacked QLSTM units. A single QLSTM cell utilizes multiple variable quantum circuits (VQCs) to construct forget gates, input gates, and output gates.
4. The time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as described in claim 3, characterized in that, The variable quantum circuit (VQC) is constructed from a feature coding layer, trainable parameters and an entanglement layer, and a measurement layer, including: The multi-feature representation is encoded using a feature coding layer through unitary transformation. The encoded quantum state is evolved using multiple trainable parameters and entanglement layers; The output characteristics of the variable quantum circuit VQC are obtained by measuring the evolved quantum state through the measurement layer.
5. The time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as described in claim 4, characterized in that, The processing logic of a single QLSTM unit includes: The hidden state from the previous time step is concatenated with the input vector from the current time step to obtain the connection input vector; The connection input vector is processed by the first VQC and then... The activation function generates a forgetting gate vector, which is used to filter the cell states from the previous time step. The connection input vector is processed by the second VQC and then... The activation function obtains the input gate vector. The third VQC process connects the input vector and passes it through the hyperbolic tangent Tanh activation function to obtain a new cell state candidate. The current cell state is updated based on the input gate vector and the new cell state candidate. The connection input vector is processed by the fourth VQC and then... The activation function produces an output gate vector, which is then combined with the current cell state to obtain the result to be processed. The fifth and sixth VQCs are used to process the results to obtain the hidden state and output results at the current time.
6. The time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as described in claim 5, characterized in that, The temporal convolutional network (TCN) module includes: The high-frequency component sequence is received as input data, and the input data is processed through the parallel backbone branch and residual branch in the temporal convolutional network TCN module. In the main branch, dilated convolution is used to perform convolution operations on the input data to extract temporal features, and ReLU activation and Dropout are performed on the convolutional data in sequence to obtain the first feature result; In the residual branch, pointwise convolution is used to adjust the channels of the input data to obtain the second feature result; The first feature result and the second feature result are added element-wise to obtain the high-frequency prediction result.
7. The time series prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as described in claim 6, characterized in that, The hybrid prediction model also employs the Levenberg-Marquardt algorithm for parameter optimization, including: Calculate the error vector between the final predicted result and the actual load value; Based on the error vector, calculate the Jacobian matrix of the model parameters; Construct a system of linear equations containing the Jacobian matrix and damping factor, and solve it to obtain the parameter update step size; Based on the error change trend after parameter updates, the damping factor is dynamically adjusted, and the VQC and TCN parameters in the hybrid prediction model are updated until the convergence condition is met.
8. A time series prediction system based on a CEEMDAN-enhanced TCN and QLSTM fusion framework, employing the method as described in any one of claims 1-7, characterized in that, include: The data decomposition module is used to acquire the raw load data and decompose it using the CEEMDAN method of adaptive noise complete set empirical mode decomposition to obtain high-frequency component sequences and low-frequency component sequences. The model building module is used to build a hybrid prediction model, which includes a parallel temporal convolutional network (TCN) module and a quantum long short-term memory (QLSTM) module. The prediction module is used to predict the high-frequency component sequence using the TCN module to obtain a high-frequency prediction result; and to predict the low-frequency component sequence using the QLSTM module to obtain a low-frequency prediction result. The fusion module is used to fuse high-frequency prediction results with low-frequency prediction results to obtain the final prediction result.
9. An electronic device, characterized in that, include: Memory and processor; The memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions. When the computer-executable instructions are executed by the processor, they implement the steps of the timing prediction method based on the CEEMDAN enhanced TCN and QLSTM fusion framework as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, It includes the steps of storing computer-executable instructions that, when executed by a processor, implement the timing prediction method based on the CEEMDAN-enhanced TCN and QLSTM fusion framework as described in any one of claims 1 to 7.