CN-121981317-A - Load time sequence mixed quantum fusion prediction method, system, equipment and medium

CN121981317ACN 121981317 ACN121981317 ACN 121981317ACN-121981317-A

Abstract

The invention is suitable for the field of power load prediction, and discloses a load time sequence mixed quantum fusion prediction method, a system, equipment and a medium, wherein the method comprises the steps of obtaining multi-channel historical load time sequence data, performing sequence embedding processing, obtaining an initial sequence, and inputting the initial sequence into a plurality of layers of star-shaped aggregation-distribution modules which are sequentially connected in series; the star aggregation-distribution module performs network processing on an initial sequence by adopting a first quantum parallel mixed network to obtain a multi-sequence representation, performs random pooling operation on the multi-sequence representation to obtain a core representation, performs fusion on the spliced core representation and the multi-sequence representation by adopting a second quantum parallel network to obtain an output sequence, and takes the output sequence of a previous module as input of a next module and takes a final output sequence as a power load prediction result. According to the invention, the parallel hybrid classical quantum regression network is introduced into the star aggregation-distribution module, so that complex time sequence data is effectively processed, and the performance of time sequence prediction is improved.

Inventors

XIAO YANHONG
WU XIN
Deng Yuedan
LI XINZHUO
ZHANG LI
XU KUI

Assignees

贵州电网有限责任公司

Dates

Publication Date: 20260505
Application Date: 20251222

Claims (10)

1. The load time sequence mixed quantum fusion prediction method is characterized by comprising the following steps of: acquiring multi-channel historical load time sequence data and performing sequence embedding processing to obtain an initial sequence; Inputting an initial sequence into a plurality of layers of star-shaped aggregation-distribution modules which are sequentially connected in series; The star aggregation-distribution module adopts a first quantum parallel hybrid network to perform network processing on an initial sequence to obtain a multi-sequence representation; Carrying out random pooling operation through multi-sequence representation to obtain a core representation; Fusing the core representation and the multi-sequence representation by adopting a second quantum parallel network to obtain an output sequence; The output sequence of the previous star aggregation-distribution module is taken as the input of the next module, and the final output sequence is taken as the power load prediction result.
2. The load timing hybrid quantum fusion prediction method of claim 1, wherein the star aggregation-distribution module comprises: adopting a plurality of layers of star-shaped aggregation-distribution modules which are sequentially connected in series, and utilizing a star-shaped structure to exchange information among different channels, wherein the output sequence of the former star-shaped aggregation-distribution module is used as the input of the latter module; And after the multi-layer star aggregation-distribution module is based, obtaining a power load prediction result by adopting a linear prediction head and inverse normalization through a final output sequence.
3. The load timing hybrid quantum fusion prediction method of claim 2, wherein the first quantum parallel hybrid network comprises: the first quantum parallel hybrid network comprises a quantum variation line VQC and a full-connection layer MLP; Performing Principal Component Analysis (PCA) dimension reduction through an initial sequence to obtain multi-feature representation; based on the multi-feature representation, the VQC and the MLP are respectively sent into, and the VQC output and the MLP output are subjected to two-to-one linear weight layer and linear layer to obtain multi-sequence representation.
4. The method for predicting load time sequence hybrid quantum fusion according to claim 2, wherein the quantum variation circuit VQC and full link layer MLP comprises: Based on multi-feature representation, processing by adopting a quantum variation line VQC to obtain a VQC output, wherein the VQC comprises a coding layer, a trainable parameter, an entanglement layer and a measuring layer; The coding layer comprises a multi-feature representation for performing feature coding on RY and RZ quantum revolving doors by adopting an H quantum logic gate initialization circuit; based on the multi-feature representation, the MLP output is obtained by processing the multi-feature representation by using the full-connection layer MLP, which comprises the steps of combining the multi-feature representation with a weight matrix and biasing, and transforming by an activation function.
5. The method of claim 4, further comprising encoding global information over multiple sequences using a random pooling operation based on the multiple sequence representation to obtain a core representation.
6. The method of claim 5, wherein the fusing of the core representation and the multi-sequence representation by the concatenation of the second quantum parallel network to obtain the output sequence comprises: splicing the core representation and the multi-sequence representation by Repeat Concat operations; And fusing and processing the spliced representation by adopting a second quantum parallel network, and obtaining an output sequence through a linear layer.
7. The method of load time sequence hybrid quantum fusion prediction according to claim 6, wherein the step of obtaining the multi-channel historical load time sequence data and performing the sequence embedding process to obtain the initial sequence by the pipeline fatigue model comprises the steps of: based on the multi-channel historical load time sequence data, adopting normalization to obtain load time sequence data; based on the load time sequence data, linear projection is adopted to conduct sequence embedding, and an initial sequence is obtained.
8. A load-timing hybrid quantum fusion prediction system employing the method of any of claims 1-7, comprising: the preprocessing module is used for acquiring multi-channel historical load time sequence data and performing sequence embedding processing to obtain an initial sequence; The information fusion module is used for inputting the initial sequence into the multi-layer star aggregation-distribution module which is sequentially connected in series, taking the output sequence of the former star aggregation-distribution module as the input of the latter module, and taking the final output sequence as the power load prediction result; and the star aggregation-distribution module is used for carrying out network processing on the initial sequence by adopting a first quantum parallel mixed network to obtain multi-sequence representation, carrying out random pooling operation by the multi-sequence representation to obtain core representation, splicing the core representation and the multi-sequence representation, and carrying out fusion by adopting a second quantum parallel network to obtain an output sequence.
9. An electronic device, comprising: A memory and a processor; The memory is configured to store computer-executable instructions that, when executed by a processor, implement the steps of the load-sequential hybrid quantum fusion prediction method of any one of claims 1 to 7.
10. A computer readable storage medium, comprising computer executable instructions stored thereon, which when executed by a processor, implement the steps of the load sequential hybrid quantum fusion prediction method of any one of claims 1 to 7.

Description

Load time sequence mixed quantum fusion prediction method, system, equipment and medium Technical Field The invention relates to the field of power load prediction, in particular to a load time sequence hybrid quantum fusion prediction method, a system, equipment and a medium. Background Power load prediction is a key component in power system planning and operation management, solving key problems such as power supply and demand balance, power grid planning, economic dispatch, energy cost management, system reliability and stability. Conventional load prediction techniques face the problems that channel independent methods lack utilization of correlation between channels, and channel dependent methods are not robust enough. In the field of multi-element time series, two modeling modes exist, one is a channel independent method and the other is a channel dependent method, wherein the channel independent method decomposes the multi-element time series into a plurality of single time series and applies a unified univariate prediction model to process. This approach is widely favored because of its strong robustness to non-stationary data, but it fails to take into account the inter-channel correlation, limiting further optimization of its performance. In contrast, channel dependent methods facilitate the exchange of information between channels by introducing specialized channel information fusion mechanisms. However, such methods face the dilemma that on the one hand they may rely too much on inter-channel correlation, thus lacking sufficient robustness in the face of sequence instability, and on the other hand they may employ complex relational modeling techniques such as attention mechanisms, resulting in increased computational complexity, which is difficult to scale up in large-scale applications. Therefore, how to utilize the robustness of channel independence and design a more robust and efficient channel interaction module is a problem that must be considered in academic circles to optimize a multi-component timing prediction method. Quantum machine learning is an emerging field aimed at enhancing classical machine learning models by utilizing quantum computing. Current machine learning models, such as deep neural networks, are computationally expensive and face limitations in scaling up. The advent of quantum computing provides opportunities to reduce the cost of these computations. Therefore, the technology introduces a centralized structure transfer and parallel quantum network framework, so that the model reduces the calculation cost, improves the robustness to abnormal channels, and obtains better performance with lower complexity. Disclosure of Invention The present invention has been made in view of the above problems of low robustness and high calculation cost existing in the prior art. Therefore, the invention provides a load time sequence hybrid quantum fusion prediction method, a system, equipment and a medium, which solve the problems that the existing channel independent method ignores the association among channels to cause precision bottleneck, the channel dependent method has high complexity of mutual computation and poor robustness to abnormal data, and the classical machine learning model is limited in performance when processing a complex time sequence mode. In order to solve the technical problems, the invention provides the following technical scheme: in a first aspect, the present invention provides a load timing hybrid quantum fusion prediction method, including: acquiring multi-channel historical load time sequence data and performing sequence embedding processing to obtain an initial sequence; Inputting an initial sequence into a plurality of layers of star-shaped aggregation-distribution modules which are sequentially connected in series; The star aggregation-distribution module adopts a first quantum parallel hybrid network to perform network processing on an initial sequence to obtain a multi-sequence representation; Carrying out random pooling operation through multi-sequence representation to obtain a core representation; Fusing the core representation and the multi-sequence representation by adopting a second quantum parallel network to obtain an output sequence; The output sequence of the previous star aggregation-distribution module is taken as the input of the next module, and the final output sequence is taken as the power load prediction result. As a preferable scheme of the load time sequence mixed quantum fusion prediction method, the star aggregation-distribution module comprises: adopting a plurality of layers of star-shaped aggregation-distribution modules which are sequentially connected in series, and utilizing a star-shaped structure to exchange information among different channels, wherein the output sequence of the former star-shaped aggregation-distribution module is used as the input of the latter module; And after the multi-layer star aggregation-distri