Model preferential method and device based on feature weight, equipment and storage medium

Abstract

The invention discloses a model preferential method, device and equipment based on feature weight and a storage medium. The method comprises the following steps: acquiring historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, and historical real power data corresponding to the historical predicted power data; on the basis of the historical predicted power data, the historical model input data and the historical real power data, generating a performance feature vector corresponding to each candidate power prediction model; and determining a target power prediction model from at least two candidate power prediction models based on the performance feature vector and the to-be-optimized performance feature weight corresponding to each performance feature. According to the technical scheme, the prediction precision of power prediction is improved.

Description

Technical Field

[0001] This application relates to the field of computer technology, specifically to a model selection method, apparatus, device, and storage medium based on feature weights. Background Technology

[0002] As the global energy structure transitions towards cleaner and lower-carbon energy, the scale of new energy power generation, such as wind and solar power, continues to expand. However, the inherent intermittency and volatility of new energy power generation pose significant challenges to the safe and stable operation of the power system. High-precision power forecasting is a key technology for solving this problem, as it can provide forward-looking information for grid dispatching, thereby optimizing dispatching plans and improving the absorption of new energy sources.

[0003] Currently, AI-based power prediction models, such as neural networks and support vector machines, are widely used due to their powerful nonlinear data processing capabilities. However, practice shows that no single AI power prediction model can achieve optimal performance under all operating conditions. Different models have their own advantages and disadvantages, and their performance dynamically changes with factors such as weather conditions, seasonal variations, and equipment status. Summary of the Invention

[0004] This application provides a model selection method, apparatus, device, and storage medium based on feature weights to improve the prediction accuracy of power prediction models.

[0005] According to one aspect of this application, a model optimization method based on feature weights is provided, the method comprising:

[0006] Obtain historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, as well as historical actual power data corresponding to the historical predicted power data; wherein, the model structures corresponding to different candidate power prediction models are different from each other;

[0007] Based on the historical predicted power data, the historical model input data, and the historical real power data, a corresponding performance feature vector is generated for each candidate power prediction model; wherein, the performance feature vector is used to characterize the historical prediction performance of the candidate power prediction model.

[0008] Based on the performance feature vector and the weight of the performance feature to be optimized corresponding to each performance feature, a target power prediction model is determined from at least two candidate power prediction models; wherein, the target power prediction model refers to the candidate power prediction model with the best prediction performance.

[0009] According to another aspect of this application, a model optimization device based on feature weights is provided, the device comprising:

[0010] The data acquisition module is used to acquire historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, as well as historical real power data corresponding to the historical predicted power data; wherein, the model structures corresponding to different candidate power prediction models are different from each other;

[0011] The performance feature generation module is used to generate a corresponding performance feature vector for each candidate power prediction model based on the historical predicted power data, the historical model input data, and the historical real power data; wherein the performance feature vector is used to characterize the historical prediction performance of the candidate power prediction model.

[0012] The target model determination module is used to determine a target power prediction model from at least two candidate power prediction models based on the performance feature vector and the weight of the performance feature to be optimized corresponding to each performance feature; wherein, the target power prediction model refers to the candidate power prediction model with the best prediction performance.

[0013] According to another aspect of this application, an electronic device is provided, the electronic device comprising:

[0014] One or more processors;

[0015] Memory, used to store one or more programs;

[0016] When the one or more programs are executed by the one or more processors, the one or more processors implement any of the feature weight-based model selection methods provided in the embodiments of this application.

[0017] According to another aspect of this application, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements any of the feature weight-based model selection methods provided in the embodiments of this application.

[0018] According to another aspect of this application, a computer program product is provided, including a computer program that, when executed by a processor, implements any of the feature weight-based model selection methods provided in the embodiments of this application.

[0019] This application overcomes the subjectivity and instability of traditional single-model selection methods by determining the performance feature vector of at least one candidate power prediction model and selecting the best candidate power prediction model by using the weights of the performance features to be optimized corresponding to the performance feature vector. Through a data-driven and quantifiable approach, it automatically learns and determines the importance of various historical features for evaluating model performance, and finally determines the target power prediction model with the best prediction performance, thereby improving the prediction accuracy of power prediction. Attached Figure Description

[0020] Figure 1 This is a flowchart of a model selection method based on feature weights provided in Embodiment 1 of this application;

[0021] Figure 2 This is a flowchart of a model selection method based on feature weights according to Embodiment 2 of this application;

[0022] Figure 3 This is a schematic diagram of a model selection device based on feature weights according to Embodiment 3 of this application;

[0023] Figure 4 This is a schematic diagram of the structure of an electronic device that implements the feature weight-based model selection method of Embodiment 4 of this application. Detailed Implementation

[0024] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort should fall within the scope of protection of the present application.

[0025] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0026] Example 1

[0027] Figure 1 This is a flowchart of a model selection method based on feature weights according to Embodiment 1 of this application. This embodiment is applicable to determining the optimal power prediction model among multiple different power prediction models. It can be executed by a model selection device based on feature weights, which can be implemented in hardware and / or software and can be configured in a computer device, such as a server. Figure 1 As shown, the method includes:

[0028] S110. Obtain historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, as well as historical actual power data corresponding to the historical predicted power data.

[0029] The different candidate power prediction models have different model structures. Optionally, different candidate power prediction models can refer to prediction models trained using a model-specific algorithm.

[0030] Historical predicted power data refers to the predicted power data output by the candidate power prediction model within a historical prediction period, while historical actual power data refers to the corresponding actual power data within the historical prediction period. It should be noted that within the same historical prediction period, there is a one-to-one correspondence between historical predicted power data and historical actual power data; historical predicted power data is the output of the model prediction, while historical actual power data represents the actual existing attribute of the predicted object.

[0031] Historical model input data can refer to the model input data of the candidate power prediction model within the historical prediction period. Optionally, in this embodiment of the invention, the model input data can refer to the environmental meteorological data of the predicted object. For example, the predicted object can refer to a new energy power generation system, and the model input data can include the environmental temperature, environmental humidity, sunshine duration, and sunshine intensity of the area where the predicted object is located. It should be noted that the candidate power prediction model predicts power data for future time periods; therefore, the model input data should be the predicted environmental meteorological data corresponding to the future time period.

[0032] S120. Based on historical predicted power data, historical model input data, and historical real power data, generate a corresponding performance feature vector for each candidate power prediction model.

[0033] Among them, the performance feature vector can be used to characterize the historical prediction performance of the candidate power prediction model.

[0034] It should be noted that, assuming we have m candidate power prediction models, an n-dimensional performance feature vector is generated for each candidate power prediction model based on historical predicted power data, historical actual power data, and historical model input data. Optionally, the number of candidate power prediction models and the feature dimensions of the performance feature vector can be adaptively set according to those skilled in the art.

[0035] Optionally, the performance feature vector may include a prediction accuracy performance index and an error attribution index; the prediction accuracy performance index can be used to characterize the prediction accuracy of the candidate power prediction model in at least one historical prediction period; the error attribution index can be used to characterize the data quality of the apartment type input data of the candidate power prediction model in at least one historical prediction period.

[0036] It should be noted that the prediction accuracy performance metric can include the average prediction accuracy of the candidate power prediction model over at least one historical prediction period and the single-period prediction accuracy. Error attribution metrics can be used to measure the accuracy of the model's input data within the historical prediction period. For example, if the error attribution metric value is too low, it indicates that the medical errors in the candidate power prediction model may stem more from poor quality input data than from insufficient model capability. Optionally, the prediction accuracy can be determined by those skilled in the art, for example, by using MAPE (Mean Absolute Percentage Error) or Root Mean Squared Error (RMSE).

[0037] Optionally, based on historical predicted power data, historical model input data, and historical actual power data, a corresponding performance feature vector is generated for each candidate power prediction model, including: generating an accuracy performance index corresponding to the candidate power prediction model based on the historical predicted power data of the candidate power prediction model within the historical prediction period and the historical actual power data corresponding to the historical prediction period; and generating an error attribution index corresponding to the candidate power prediction model based on the prediction accuracy of the model input data of the candidate power prediction model within the historical prediction period.

[0038] In this embodiment of the invention, if the difference between the historical predicted power data and the historical actual power data corresponding to the historical prediction period is less than a first preset difference, it indicates that the historical predicted power data and the historical actual power data are consistent; conversely, it indicates that the historical predicted power data and the historical actual power data are inconsistent. Specifically, the accuracy performance index corresponding to a historical prediction period can be determined based on the data consistency between multiple historical predicted power data and multiple historical actual power data within a historical prediction period. The prediction accuracy of the model input data within the historical prediction period can be determined based on the consistency between the model input data and the real environmental meteorological data of the future time period corresponding to the model input data.

[0039] By analyzing the prediction accuracy of the candidate power prediction model within historical prediction periods and the accuracy of the model input data, the overall performance of the model can be evaluated.

[0040] Furthermore, based on the accuracy performance index and error attribution index corresponding to the candidate power prediction model, a historical performance evaluation matrix is ​​constructed for the candidate power prediction model, and the matrix elements are standardized to eliminate the influence between different units. Optionally, the standardization process can be adapted to the needs of those skilled in the art.

[0041] S130. Based on the performance feature vector and the weight of the performance feature to be optimized corresponding to each performance feature, determine the target power prediction model from at least two candidate power prediction models.

[0042] The target power prediction model can refer to the candidate power prediction model with the best prediction performance.

[0043] The weights of the performance features to be optimized can be used to measure the importance of each dimension of the performance feature vector.

[0044] This application embodiment determines the performance feature vector of at least one candidate power prediction model and selects the best candidate power prediction model by using the weights of the performance features to be optimized corresponding to the performance feature vector. This overcomes the subjectivity and instability of traditional single model selection methods. Through a data-driven and quantifiable approach, it automatically learns and determines the importance of various historical features for evaluating model performance, and finally determines the target power prediction model with the best prediction performance, thereby improving the prediction accuracy of power prediction.

[0045] Example 2

[0046] Figure 2 This is a flowchart of a model selection method based on feature weights according to Embodiment 2 of this application. Based on the technical solutions of the above embodiments, this embodiment further refines the step of "determining the target power prediction model from at least two candidate power prediction models based on the performance feature vector and the corresponding performance feature weights to be optimized for each performance feature." It should be noted that for parts not detailed in this embodiment, please refer to the relevant descriptions in other embodiments. Figure 2 As shown, the method includes:

[0047] S210. Obtain historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, as well as historical actual power data corresponding to the historical predicted power data.

[0048] S220: Based on historical predicted power data, historical model input data, and historical real power data, generate a corresponding performance feature vector for each candidate power prediction model.

[0049] S230. For each candidate power prediction model, determine the model prediction score of the candidate power prediction model based on its corresponding performance feature vector and the weight of the performance feature to be optimized.

[0050] Among them, the model prediction score can be used to comprehensively evaluate the model performance of candidate power prediction models.

[0051] Optionally, the model prediction score can be determined using the following formula:

[0052] ;

[0053] in, is the model prediction score of the i-th candidate power prediction model; n is the number of feature dimensions of the performance feature vector, that is, the number of performance features in the performance feature vector; Let j be the j-th performance feature of the candidate power prediction model; Let be the performance feature weights of the i-th candidate power prediction model on the j-th performance feature.

[0054] S240. Based on historical predicted power data and model prediction scores, construct an optimal objective loss function, and use the optimal objective loss function to optimize the weights of the performance features to be optimized.

[0055] The optimization objective loss function can be used to characterize the optimal performance of the performance feature weights on historical model input data. It should be noted that the smaller the value of the optimization objective loss function, the better the optimization performance of the performance feature weights.

[0056] Optionally, based on historical predicted power data and model prediction scores, an optimal objective loss function is constructed, and this function is used to optimize the weights of the performance features to be optimized. This includes: determining the corresponding value search space for each weight of the performance feature to be optimized; sampling and updating the weights of the performance feature to be optimized with a preset search step size to generate at least one set of weights of the performance feature to be optimized; for each set of weights of the performance feature to be optimized, determining the model prediction score corresponding to each candidate power prediction model under that set of weights, and selecting the candidate power prediction model with the highest model prediction score as the pre-selected power prediction model; generating an optimal objective loss function based on the difference between the optimal prediction accuracy of each candidate power prediction model in the historical prediction period and the historical prediction accuracy of the pre-selected power prediction model in the historical prediction period; for each set of weights of the performance feature to be optimized, determining the average loss function value of the optimal objective loss function value in each historical prediction period, and recombining the weights of the performance features to be optimized corresponding to the historical prediction period with the smallest average loss function value as the target performance feature weight recombination.

[0057] Optionally, before sampling and updating the weights of the performance features to be optimized with a preset search step size to generate at least one set of weights for the performance features to be optimized, the method further includes: using the analytic hierarchy process (AHP) to initialize the weights of the performance features to be optimized corresponding to each performance feature in the performance feature vector based on prior knowledge.

[0058] In this embodiment of the invention, a search space can be defined for each performance feature weight to be optimized. Based on the prior knowledge of domain experts (e.g., through the Analytic Hierarchy Process), a baseline weight for each performance feature weight to be optimized is determined. Using this baseline as the center, a reasonable value range and search step size are set for each optimized performance feature weight, thus forming a multi-dimensional weight grid. A sampling search is performed on each performance feature weight to be optimized within the search space to determine at least one combination of performance feature weights to be optimized in the multi-dimensional weight network. It should be noted that each sampling point in the search space can represent a combination of performance feature weights to be optimized. For each combination of performance feature weights to be optimized, the average optimization loss function value corresponding to that combination can be calculated on the historical model input dataset. Finally, after traversing all sampling points in the search space, the combination of performance feature weights to be optimized that minimizes the average optimization loss function value can be selected as the target performance feature weight reassembly. The historical model input dataset includes historical model input data from at least one historical prediction period.

[0059] Optionally, the optimal objective loss function can be expressed by the following formula:

[0060] ;

[0061] in, For the reorganization of performance feature weights to be optimized, This represents the highest prediction accuracy actually achieved by all candidate power prediction models during the historical prediction period. The prediction accuracy of the candidate power prediction model with the highest prediction score calculated by recombining the performance features to be optimized during this historical prediction period is given.

[0062] By transforming the model selection problem into a data-driven optimization problem, the objective, quantitative, and adaptive model selection process is achieved, thereby effectively improving the accuracy and reliability of new energy power prediction.

[0063] S250. Based on the performance feature vector and target performance feature weights corresponding to each candidate power prediction model, determine the target model prediction score corresponding to each candidate power prediction model, and take the candidate power prediction model with the highest target model prediction score as the target power prediction model.

[0064] Among them, the target performance feature weight can refer to the performance feature weight to be optimized after optimization.

[0065] This application embodiment improves the prediction accuracy of power prediction by scoring candidate power prediction models and optimizing the weights of performance features to be optimized. It dynamically determines the key features with the highest correlation to power prediction, and selects the best candidate power prediction models from two dimensions: predicted power data and metadata affecting model performance.

[0066] Example 3

[0067] Figure 3 This is a schematic diagram of a feature weight-based model selection device according to Embodiment 3 of this application. It is applicable to determining the optimal power prediction model among multiple different power prediction models. This feature weight-based model selection device can be implemented in hardware and / or software and can be configured in a computer device, such as a server. Figure 3 As shown, the device includes:

[0068] The data acquisition module 310 is used to acquire historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, as well as historical real power data corresponding to the historical predicted power data; wherein, the model structures corresponding to different candidate power prediction models are different from each other.

[0069] The performance feature generation module 320 is used to generate a corresponding performance feature vector for each candidate power prediction model based on the historical predicted power data, the historical model input data, and the historical real power data; wherein the performance feature vector is used to characterize the historical prediction performance of the candidate power prediction model.

[0070] The target model determination module 330 is used to determine a target power prediction model from at least two candidate power prediction models based on the performance feature vector and the weight of the performance feature to be optimized corresponding to each performance feature; wherein, the target power prediction model refers to the candidate power prediction model with the best prediction performance.

[0071] This application embodiment determines the performance feature vector of at least one candidate power prediction model and selects the best candidate power prediction model by using the weights of the performance features to be optimized corresponding to the performance feature vector. This overcomes the subjectivity and instability of traditional single model selection methods. Through a data-driven and quantifiable approach, it automatically learns and determines the importance of various historical features for evaluating model performance, and finally determines the target power prediction model with the best prediction performance, thereby improving the prediction accuracy of power prediction.

[0072] Optionally, the target model determination module 330 includes:

[0073] The prediction scoring unit is used to determine the model prediction score of each candidate power prediction model based on its corresponding performance feature vector and the weight of the performance feature to be optimized.

[0074] The weight optimization unit is used to construct an optimal objective loss function based on the historical predicted power data and the model prediction score, and to optimize the weights of the performance features to be optimized based on the optimal objective loss function.

[0075] The target prediction model determination unit is used to determine the target model prediction score for each candidate power prediction model based on the performance feature vector and target performance feature weights corresponding to each candidate power prediction model, and to take the candidate power prediction model with the highest target model prediction score as the target power prediction model; where the target performance feature weights refer to the optimized performance feature weights.

[0076] Optionally, the weight optimization unit includes:

[0077] The weight sampling subunit is used to determine the corresponding value search space for each performance feature weight to be optimized, and to sample and update the performance feature weight to be optimized with a preset search step size to generate at least one set of performance feature weights to be optimized.

[0078] The candidate model determination sub-unit is used to determine the model prediction score of each candidate power prediction model under the weight of each set of performance features to be optimized, and the candidate power prediction model with the highest model prediction score is selected as the pre-selected power prediction model.

[0079] The loss function sub-unit is used to generate the optimal target loss function based on the difference between the best prediction accuracy of each candidate power prediction model in the historical prediction period and the historical prediction accuracy of the pre-selected power prediction model in the same historical prediction period.

[0080] The target weight determination subunit is used to determine the average loss function value of the target loss function value in each historical prediction period for each set of performance feature weights to be optimized, and to reassemble the performance feature weights to be optimized corresponding to the historical prediction period with the smallest average loss function value as the target performance feature weight reassembly.

[0081] Optionally, the weight optimization unit may also include:

[0082] The weight initialization unit is used to initialize the weights of the performance features to be optimized corresponding to each performance feature in the performance feature vector based on prior knowledge before sampling and updating the weights of the performance features to be optimized with a preset search step size and generating at least one set of weights for the performance features to be optimized.

[0083] Optionally, the performance feature vector includes prediction accuracy performance metrics and error attribution metrics.

[0084] Optionally, the performance characteristic generation module 320 includes:

[0085] The accuracy performance index unit is used to generate the accuracy performance index corresponding to the candidate power prediction model based on the historical predicted power data of the candidate power prediction model in the historical prediction period and the historical real power data corresponding to the historical prediction period.

[0086] The error attribution index unit is used to generate the error attribution index corresponding to the candidate power prediction model based on the prediction accuracy of the model input data in the historical prediction period.

[0087] Optionally, the prediction accuracy performance index is used to characterize the prediction accuracy of the candidate power prediction model in at least one historical prediction period; the error attribution index is used to characterize the data quality of the household type input data of the candidate power prediction model in at least one historical prediction period.

[0088] The feature weight-based model selection device provided in this application can execute the feature weight-based model selection method provided in any embodiment of this application, and has the corresponding functional modules and beneficial effects for executing each feature weight-based model selection method.

[0089] According to embodiments of this application, this application also provides an electronic device, a readable storage medium, and a computer program product.

[0090] Example 4

[0091] Figure 4This is a schematic diagram of the structure of an electronic device 410 implementing the feature weight-based model selection method of this application embodiment. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices (such as helmets, glasses, watches, etc.), and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present application described and / or claimed herein.

[0092] like Figure 4 As shown, the electronic device 410 includes at least one processor 411 and a memory, such as a read-only memory 412 or a random access memory 413, communicatively connected to the at least one processor 411. The memory stores computer programs executable by the at least one processor. The processor 411 can perform various appropriate actions and processes based on the computer program stored in the read-only memory 412 or loaded from storage unit 418 into the random access memory 413. The random access memory 413 can also store various programs and data required for the operation of the electronic device 410. The processor 411, read-only memory 412, and random access memory 413 are interconnected via a bus 414. An input / output interface 415 is also connected to the bus 414.

[0093] Multiple components in electronic device 410 are connected to input / output interface 415, including: input unit 416, such as keyboard, mouse, etc.; output unit 417, such as various types of monitors, speakers, etc.; storage unit 418, such as disk, optical disk, etc.; and communication unit 419, such as network card, modem, wireless transceiver, etc. Communication unit 419 allows electronic device 410 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0094] Processor 411 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of processor 411 include, but are not limited to, central processing units, graphics processing units, various special-purpose artificial intelligence computing chips, various processors running machine learning model algorithms, digital signal processors, and any suitable processor, controller, microcontroller, etc. Processor 411 performs the various methods and processes described above, such as feature weight-based model optimization methods.

[0095] In some embodiments, the feature-weight-based model selection method may be implemented as a computer program tangibly contained in a computer-readable storage medium, such as storage unit 418. In some embodiments, part or all of the computer program may be loaded and / or installed on electronic device 410 via read-only memory 412 and / or communication unit 419. When the computer program is loaded into random access memory 413 and executed by processor 411, one or more steps of the feature-weight-based model selection method described above may be performed. Alternatively, in other embodiments, processor 411 may be configured for the feature-weight-based model selection method by any other suitable means (e.g., by means of firmware).

[0096] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays, application-specific integrated circuits (ASICs), application-specific standard products (ASICs), systems-on-a-chip (SoCs), payload programmable logic devices, computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0097] Computer programs used to implement the methods of this application may be written in any combination of one or more programming languages. These computer programs may be provided to the processor of a general-purpose computer, a special-purpose computer, or other programmable feature-weight-based model optimization device, such that when executed by the processor, the computer programs cause the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The computer programs may be executed entirely on a machine, partially on a machine, or as a standalone software package, partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0098] In the context of this application, a computer-readable storage medium can be a tangible medium that may contain or store a computer program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable storage medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. Alternatively, a computer-readable storage medium can be a machine-readable signal medium. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory, read-only memory, erasable programmable read-only memory, optical fibers, portable compact disk read-only memory, optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0099] To provide interaction with a user, the systems and techniques described herein can be implemented on an electronic device having: a display device (e.g., a cathode ray tube or liquid crystal display monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the electronic device. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0100] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include local area networks (LANs), wide area networks (WANs), blockchain networks, and the Internet.

[0101] A computing system can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or cloud host, which is a host product within the cloud computing service system to address the shortcomings of traditional physical hosts and virtual private servers, such as high management difficulty and weak business scalability.

[0102] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this application can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution of this application can be achieved, and this is not limited herein.

[0103] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A model optimization method based on feature weights, characterized in that, include: Obtain historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, as well as historical actual power data corresponding to the historical predicted power data; wherein, the model structures corresponding to different candidate power prediction models are different from each other; Based on the historical predicted power data, the historical model input data, and the historical real power data, a corresponding performance feature vector is generated for each candidate power prediction model; wherein, the performance feature vector is used to characterize the historical prediction performance of the candidate power prediction model. Based on the performance feature vector and the weight of the performance feature to be optimized corresponding to each performance feature, a target power prediction model is determined from at least two candidate power prediction models; wherein, the target power prediction model refers to the candidate power prediction model with the best prediction performance.

2. The method according to claim 1, characterized in that, The step of determining the target power prediction model from at least two candidate power prediction models based on the performance feature vector and the weight of the performance feature to be optimized corresponding to each performance feature includes: For each candidate power prediction model, the model prediction score of the candidate power prediction model is determined based on its corresponding performance feature vector and the weight of the performance feature to be optimized. Based on the historical predicted power data and the model prediction score, an optimal objective loss function is constructed, and the optimal objective loss function is used to optimize the weights of the performance features to be optimized. Based on the performance feature vector and target performance feature weights corresponding to each candidate power prediction model, the target model prediction score corresponding to each candidate power prediction model is determined, and the candidate power prediction model with the highest target model prediction score is taken as the target power prediction model; where the target performance feature weights refer to the optimized performance feature weights.

3. The method according to claim 2, characterized in that, The step of constructing an optimal objective loss function based on the historical predicted power data and the model prediction score, and then using the optimal objective loss function to optimize the weights of the performance features to be optimized, includes: For each performance feature weight to be optimized, determine its corresponding value search space, sample and update the performance feature weight to be optimized with a preset search step size, and generate at least one set of performance feature weights to be optimized. For each set of performance feature weights to be optimized, determine the model prediction score of each candidate power prediction model under the weights of that set of performance feature weights to be optimized, and take the candidate power prediction model with the highest model prediction score as the pre-selected power prediction model. Based on the optimal prediction accuracy of each candidate power prediction model within the historical prediction period, and the difference between the historical prediction accuracy of the pre-selected power prediction model within the same historical prediction period, an optimal target loss function is generated. For each set of performance feature weights to be optimized, the average loss function value of the target loss function value in each historical prediction period is determined, and the performance feature weights to be optimized corresponding to the historical prediction period with the smallest average loss function value are recombined as the target performance feature weights.

4. The method according to claim 3, characterized in that, Before sampling and updating the weights of the performance features to be optimized with a preset search step size to generate at least one set of weights for the performance features to be optimized, the process also includes: The Analytic Hierarchy Process (AHP) is used to initialize the weights of the performance features to be optimized for each performance feature in the performance feature vector based on prior knowledge.

5. The method according to claim 1, characterized in that, The performance feature vector includes prediction accuracy performance metrics and error attribution metrics; Accordingly, the step of generating a corresponding performance feature vector for each candidate power prediction model based on the historical predicted power data, the historical model input data, and the historical real power data includes: Based on the historical predicted power data of the candidate power prediction model within the historical prediction period and the historical actual power data corresponding to the historical prediction period, an accuracy performance index corresponding to the candidate power prediction model is generated. Based on the prediction accuracy of the candidate power prediction model in the historical prediction period, an error attribution index corresponding to the candidate power prediction model is generated.

6. The method according to claim 5, characterized in that, The prediction accuracy performance index is used to characterize the prediction accuracy of the candidate power prediction model in at least one historical prediction period; the error attribution index is used to characterize the data quality of the household type input data of the candidate power prediction model in at least one historical prediction period.

7. A model optimization device based on feature weights, characterized in that, include: The data acquisition module is used to acquire historical predicted power data and historical model input data corresponding to at least two candidate power prediction models, as well as historical real power data corresponding to the historical predicted power data; wherein, the model structures corresponding to different candidate power prediction models are different from each other; The performance feature generation module is used to generate a corresponding performance feature vector for each candidate power prediction model based on the historical predicted power data, the historical model input data, and the historical real power data; wherein the performance feature vector is used to characterize the historical prediction performance of the candidate power prediction model. The target model determination module is used to determine a target power prediction model from at least two candidate power prediction models based on the performance feature vector and the weight of the performance feature to be optimized corresponding to each performance feature; wherein, the target power prediction model refers to the candidate power prediction model with the best prediction performance.

8. An electronic device, characterized in that, include: One or more processors; Memory, used to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the feature weight-based model selection method as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by the processor, the program implements the feature weight-based model selection method as described in any one of claims 1-6.

10. A computer program product comprising a computer program that, when executed by a processor, implements the feature weight-based model selection method according to any one of claims 1-6.

Images