Baseline typical day screening method based on BP neural network and related equipment

Abstract

The invention belongs to the technical field of virtual power plants, and discloses a baseline typical day screening method based on a BP neural network and related equipment. Wherein the baseline typical day screening method based on the BP neural network comprises the steps of inputting load data into a constructed baseline typical day screening model and outputting a screening result, the screening result comprises a production day or an abnormal day, and the construction step of the baseline typical day screening model comprises the steps of constructing a feature matrix grey-scale map; training a BP neural network model through the preprocessed feature matrix grey-scale map to obtain a baseline typical day screening model; according to the method, typical production days and abnormal days of an industrial enterprise can be identified and used for replacing workdays and rest days in a traditional baseline model, so that the virtual power plant can more accurately calculate the baseline of the industrial user, the baseline is more representative, the recent reference level of the industrial load is reflected, and the baseline calculation accuracy is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of virtual power plant technology, specifically to a baseline typical day screening method and related equipment based on a BP neural network. Background Technology

[0002] Virtual power plants (VPPs), as an emerging model that aggregates distributed energy resources to participate in power system operation and market transactions, have the core function of coordinating and controlling internal distributed power sources, energy storage systems, and flexible loads to externally function as a controllable and dispatchable power source. In this process, accurate baseline calculation is a key technical foundation for measuring the regulation capacity of virtual power plants (such as peak shaving and valley filling) and for conducting market settlement and compensation assessments.

[0003] Currently, the mainstream method for calculating the baseline of virtual power plants in scenarios such as commercial buildings and residential communities is the historical averaging method. This method typically categorizes historical load data based on calendar attributes (such as weekdays, weekends, and holidays), and then calculates the average load for the same period within each category as the baseline. For example, a common practice is to take the average load at the same time of day for the five normal working days prior to the target date as the baseline load curve for that working day, and use this to calculate peak-shaving compensation costs or assess the effectiveness of market participation. The underlying logic of this method is the assumption that load patterns have significant, calendar-synchronized periodicity.

[0004] However, when the resources aggregated by the virtual power plant are primarily large industrial loads, the aforementioned traditional baseline calculation methods face significant limitations. First, production plans and load levels are mainly determined by order volume, rather than fixed calendar cycles. Second, to maintain the continuity and stability of the production process, and for economic reasons, such industries generally adopt a 24-hour shift system, and production lines may continue to operate even on statutory holidays, resulting in negligible load differences between "workdays" and "rest days" (typically less than 5%). Therefore, the periodic assumptions underlying the traditional historical averaging method, which relies on the "workday / rest day" division, are difficult to apply in large industrial scenarios.

[0005] More complexly, industrial loads are highly susceptible to interference from both internal production factors (such as planned equipment maintenance, unplanned shutdowns, and production line switching) and external factors (such as fluctuations in raw material supply and changes in market orders), resulting in non-periodic and sudden load fluctuations. For example, a steel company undergoing an annual overhaul of its blast furnace might experience a sudden drop in overall load of over 30% on that day. If such abnormal day data is included in baseline calculations without differentiation, the baseline results will be severely distorted, failing to accurately reflect the benchmark level of industrial load under normal production conditions. This, in turn, leads to incorrect assessments of the virtual power plant's regulation capacity and unfair market settlements.

[0006] In summary, existing methods for calculating historical average baselines based on calendar classification fail to effectively identify and eliminate abnormal production days in large industrial loads, and are also unsuitable for their order-driven and continuous production characteristics, resulting in insufficient accuracy and poor representativeness of the calculated virtual power plant baselines. Summary of the Invention

[0007] The purpose of this invention is to provide a baseline typical day screening method and related equipment based on BP neural network to overcome the problems existing in the prior art. This invention can identify typical production days and abnormal days of industrial enterprises to replace working days and rest days in the traditional baseline model, so that the virtual power plant can calculate the industrial user baseline more accurately and be more representative, thereby reflecting the recent benchmark level of industrial load, and also significantly improving the accuracy of baseline calculation.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows: In a first aspect, the present invention provides a baseline typical day screening method based on a BP neural network, comprising the following steps: Input the load data into the constructed baseline typical day screening model and output the screening results, which include production days or abnormal days. The construction steps of the baseline typical day screening model include constructing a feature matrix grayscale image, training the BP neural network model with the preprocessed feature matrix grayscale image, and obtaining the baseline typical day screening model.

[0009] In some embodiments, the construction steps of the baseline typical day screening model specifically include: Acquire historical load data and construct a feature matrix grayscale image from the historical load data; The feature matrix grayscale image is preprocessed to obtain the dataset, which is then divided into a training set and a test set. Construct the input layer, hidden layer, and output layer of a BP neural network model; The BP neural network model, consisting of an input layer, hidden layer, and output layer, is trained using a training set, a cross-entropy loss function, and an Adam optimizer to obtain an output set. The validation set is then compared with the output set, and the weights of the BP neural network model are updated based on the comparison results to obtain the constructed baseline typical daily screening model.

[0010] In some embodiments, constructing a feature matrix grayscale image from historical load data specifically includes: Construct a feature matrix including feature elements from historical load data. M The feature element is the battery level value for 15 minutes corresponding to the date and time. feature matrix MScaling and integerization are performed sequentially to obtain the integer feature matrix. M Convert it into a grayscale image of the feature matrix.

[0011] In some embodiments, the preprocessing of the feature matrix grayscale image to obtain the dataset specifically includes: Flatten the grayscale image of the feature matrix into a one-dimensional vector, and then normalize the one-dimensional vector. Based on the actual production situation, each date in the historical load data is labeled with a category label, resulting in several category labels, including production days or abnormal days; The normalized one-dimensional vector and several class labels are used as the dataset.

[0012] In some embodiments, the number of neurons in the input layer is equal to the dimension of a one-dimensional vector.

[0013] In some embodiments, the hidden layer uses the ReLU activation function; the output layer uses the Softmax activation function.

[0014] In some embodiments, the process of inputting load data into a pre-constructed baseline typical day screening model and outputting screening results, wherein the screening results include production days or abnormal days, specifically including: Input the load data into the constructed baseline typical day screening model to obtain the probability values ​​of production days and abnormal days; If the probability value of a production day is greater than the probability value of an abnormal day, then output the production day. If the probability value of a production day is less than the probability value of an abnormal day, then the abnormal day is output.

[0015] Secondly, the present invention provides a baseline typical day screening system based on a BP neural network, comprising: The data acquisition module is used to acquire load data; The filtering module is used to input load data into the constructed baseline typical day filtering model and output the filtering results, which include production days or abnormal days. The construction steps of the baseline typical day filtering model include constructing a feature matrix grayscale image, training the BP neural network model with the preprocessed feature matrix grayscale image, and obtaining the baseline typical day filtering model.

[0016] Thirdly, the present invention provides a computer device including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0017] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the above-described method.

[0018] The above technical solution has the following advantages or beneficial effects: Firstly, this invention provides a baseline typical day screening method based on a BP neural network, which effectively solves the problem that traditional baseline calculation methods fail in industrial load scenarios due to their reliance on calendar cycles. This method constructs a feature matrix grayscale image from historical industrial load data, intuitively displaying the characteristics of multi-day load patterns. It also utilizes the powerful nonlinear recognition capability of the BP neural network to automatically learn the complex difference patterns between production days and abnormal days, achieving accurate classification of the actual production status of industrial users. The typical production day data selected in this way can accurately reflect the recent normal baseline level of industrial load, so that the baseline calculation of virtual power plants is no longer affected by abnormalities such as holidays and equipment maintenance, greatly improving the accuracy and representativeness of the baseline, and providing more reliable technical support for virtual power plants to participate in electricity market regulation and settlement.

[0019] In some embodiments, by constructing historical load data into a feature matrix grayscale image and using a BP neural network to deeply learn its inherent patterns, intelligent and accurate identification of production days and abnormal days for industrial users is achieved. This method effectively overcomes the limitations of traditional baseline calculation relying on calendar division, eliminates interference from abnormal data, and enables baseline calculation to accurately reflect the recent true benchmark level of industrial load, significantly improving the fairness and reliability of virtual power plant regulation capacity assessment and market settlement.

[0020] In some embodiments, by constructing a feature matrix from historical load data and converting it into a grayscale image, multidimensional time series data is transformed into a two-dimensional image that can intuitively identify pattern changes, providing a high-quality feature representation for subsequent neural network learning. This method utilizes the high information density of images, enabling BP neural networks to more effectively capture subtle differences and macroscopic change patterns in industrial loads across different dates, significantly improving the model's accuracy and generalization ability in distinguishing between production days and abnormal days, and providing a reliable foundation for accurately screening benchmark data.

[0021] In some embodiments, image data is converted into standardized vectors suitable for neural network processing by flattening and normalizing the grayscale image of the feature matrix, while a high-quality training set is constructed by combining real production annotations. This preprocessing process not only significantly improves the availability of data and the convergence speed of the model, but also ensures that the model's learning objectives are strictly aligned with actual industrial working conditions, thereby enabling the trained classification model to have stronger generalization ability and practical application value.

[0022] In some embodiments, this technical feature ensures that all temporal features contained in historical load data can be input into the model without loss by strictly matching the structure of the neural network input layer with the dimension of the input data. This lays a precise data foundation for subsequent network layers to perform effective feature extraction and pattern recognition, and is a key prerequisite for ensuring that the model can accurately distinguish between production days and abnormal days.

[0023] In some embodiments, the hidden layer using the ReLU activation function can effectively introduce nonlinearity, enhance the network's ability to fit complex industrial load patterns, and avoid the gradient vanishing problem; the output layer using the Softmax function can transform the network output into a probability distribution, intuitively reflecting the confidence level of whether the target day belongs to a "production day" or an "abnormal day", giving the classification results a clear probabilistic interpretation and significantly improving the credibility and interpretability of the model in practical applications.

[0024] In some embodiments, this step compares two probability values ​​output by the model (probability of a production day and probability of an abnormal day) and uses a "winner-takes-all" decision rule to transform the probability output of the neural network into a clear and unique classification result. This method provides objective and reproducible standards for classification judgment, avoids fuzzy decision-making, ensures the stability and reliability of the virtual power plant in screening typical days in practical applications, and facilitates automated system execution and integration.

[0025] Secondly, this invention provides a baseline typical day screening system based on a BP neural network. This system accurately collects industrial load data through a data acquisition module and automatically classifies it using an intelligent model embedded in the screening module. This system transforms the traditional analysis process, which relies on manual calendar rules, into a data-driven, model-autonomous learning intelligent identification process, effectively adapting to the complex and ever-changing production patterns in different industrial scenarios. By accurately distinguishing between production days and abnormal days, this system provides high-quality data input for virtual power plant baseline calculations, significantly improving the accuracy, representativeness, and timeliness of the baseline, thereby enhancing the accuracy of virtual power plant regulation capacity assessment and operational economy in the electricity market.

[0026] Thirdly, the present invention provides a computer device that, through a processor executing a specific computer program, can efficiently implement the steps of the method of the present invention. When performing data processing tasks, the computer device can accurately perform numerical calculations and logical judgments, avoiding errors caused by human factors. At the same time, since the computer program has high stability and reliability, it can ensure the accuracy and consistency of the data processing results.

[0027] Fourthly, the present invention provides a computer-readable storage medium. By programming the steps of the method of the present invention into a computer program and storing it on the computer-readable storage medium, users can easily load these programs onto any compatible computer device and execute them without rewriting or converting the code, which greatly improves the convenience and flexibility of program execution. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the process of constructing a feature matrix grayscale image from historical load data according to some embodiments of this specification; Figure 2 This is a schematic diagram of the training process for a baseline typical day screening model according to some embodiments of this specification; Figure 3 This is a schematic diagram of the structure of a computer device according to some embodiments of this specification. Detailed Implementation

[0029] The present invention will be further described in detail below with reference to specific embodiments. These descriptions are for explanation purposes only and are not intended to limit the scope of the invention. To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. 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 scope of protection of the present invention. It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention 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 the invention 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 a 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.

[0030] Terminology Explanation: The baseline of a virtual power plant refers to the benchmark load / output level predicted by historical operating data, industry standard parameters, or dynamic models when it does not participate in ancillary services such as peak shaving and frequency regulation. It is the core reference for measuring its regulation capabilities (such as peak shaving and valley filling).

[0031] A back propagation neural network (BP neural network) is a multi-layer feedforward neural network trained using the backpropagation algorithm. It consists of an input layer, hidden layers, and an output layer, and achieves non-linear mapping of data by simulating the connections between neurons in the human brain. Its core idea is to continuously adjust the network weights and thresholds using gradient descent to minimize the sum of squared errors between the network's actual output and the expected output, thereby enabling learning and prediction of the input data.

[0032] The baseline is an important indicator for measuring the regulation capacity of a virtual power plant, representing the benchmark energy consumption curve for a certain industry. Traditional historical average baseline calculation methods often consider the load on weekdays and rest days separately, calculating the baseline for each type of date. For example, when using the historical average method, the baseline can be the average load for the same period of the previous 5 normal working days, used to settle peak-shaving compensation amounts or assess the effectiveness of market participation.

[0033] The order-driven nature and continuous production patterns of large-scale industrial production significantly limit traditional baseline calculation methods in this scenario. Traditional baseline calculation methods reference weekdays and rest days in a calendar, assuming that production occurs on weekdays and ceases on rest days. This approach is suitable for the loads of shopping malls and office buildings, but it deviates greatly from industrial production loads. Taking process industries such as steel and chemicals as examples, they generally employ 24-hour shifts, with production lines operating year-round. As long as production is in progress, the load difference between weekdays and rest days is typically less than 5%. In this case, the "periodic assumption" relied upon by traditional baseline calculation methods such as historical averaging is difficult to apply—industrial loads are more susceptible to unforeseen factors such as equipment maintenance and raw material supply, exhibiting non-periodic fluctuations (e.g., a steel company experiencing a 30% drop in daily load due to blast furnace maintenance). Ultimately, the calculated baseline cannot accurately reflect the recent benchmark level of industrial load.

[0034] This invention transforms historical industrial load data into images, and then constructs a typical day screening model based on a BP neural network. It proposes a baseline typical day screening method and related equipment based on a BP neural network, which is applicable to the elimination of abnormal days and the selection of normal days in the baseline calculation of virtual power plants under industrial load scenarios. By using recent historical data, it identifies whether the target day is a production day or an abnormal day (abnormal days include non-production days such as maintenance, rest, and shutdown), and uses them to calculate the baseline of the virtual power plant for this type of load, thus more accurately reflecting the recent benchmark level of industrial load.

[0035] Example: This embodiment provides a baseline typical day screening method based on a BP neural network, including the following steps: Step 1, Obtain load data; Step 2: Input the load data into the constructed baseline typical day screening model and output the screening results, which include production days or abnormal days; The construction steps of the baseline typical day screening model include constructing a feature matrix grayscale image, training the BP neural network model with the preprocessed feature matrix grayscale image, and obtaining the baseline typical day screening model.

[0036] In some embodiments, the steps for constructing the baseline typical day screening model are described in [reference needed]. Figure 2 Specifically, it includes: Step 2.1: Obtain historical load data and construct a feature matrix grayscale image to represent the load changes of the day to be determined and the previous few days using the historical load data.

[0037] Specifically, the construction of the feature matrix grayscale image using historical load data includes: Construct a feature matrix including feature elements from historical load data. M The feature element is the battery level value for 15 minutes corresponding to the date and time. feature matrix M Scaling and integerization are performed sequentially to obtain the integer feature matrix. M Convert it into a grayscale image of the feature matrix.

[0038] Specifically, historical load data consists of electricity consumption data collected at a time resolution for constructing a feature matrix. The data body of historical load data is positive active power, which is the core data in electricity metering and represents the actual electricity consumed by users. The time granularity of historical load data is one data point every 15 minutes, which is a common metering and settlement cycle in virtual power plants and power systems. The data format of historical load data is a 96-point load curve.

[0039] Specifically, this embodiment uses a backpropagation (BP) neural network to identify information in the grayscale image of the feature matrix, thereby determining whether the target date is a production day or an abnormal day. First, the grayscale image of the feature matrix must be constructed.

[0040] A load curve is typically constructed from 96 discrete points, each representing the positive active power generated in the preceding 15 minutes. Let's assume the date we want to determine is a production day is... D day, D The electricity consumption at a certain point in the day for 15 minutes is Therefore, when constructing the matrix, it is necessary to refer to D Recently n The data for that day, i.e., the reference range: Construct the feature matrix M : ; In this embodiment,n Take 10, that is, for reference. D A feature matrix is ​​constructed using data from the current day and the previous 10 days to determine... D Is the date a production day or an abnormal day?

[0041] After constructing the feature matrix, the feature matrix is ​​then processed according to the maximum value of the training set in the dataset. M Scale to the range [0, 255] and quantize to integer, then use the imshow() method in Matplotlib to display the feature matrix. M Convert to a feature grayscale image. The size of the feature grayscale image is related to the feature matrix. M Completely consistent, each pixel in the feature grayscale image represents the battery level in 15 minutes. The darker the color in the feature grayscale image, the lower the battery level; conversely, the lighter the color, the higher the battery consumption. The first row of the feature grayscale image represents... The last line represents , and the characteristic matrix M The order is consistent.

[0042] The logic diagram for this step is shown below. Figure 1 ,Depend on Figure 1 It can be seen D The overall daily variation is small, which is significantly different from the previous 10 days, and may indicate that this is an abnormal day.

[0043] After obtaining the grayscale image of the feature matrix, this embodiment constructs a classification model for production days and abnormal days by training a BP neural network for screening. The following steps are required to construct the model.

[0044] Step 2.2: Preprocess the grayscale image of the feature matrix to obtain the dataset, and divide the dataset into training set and test set.

[0045] Specifically, the preprocessing of the feature matrix grayscale image to obtain the dataset includes: Flatten the grayscale image of the feature matrix into a one-dimensional vector, and then normalize the one-dimensional vector. Based on the actual production situation, each date in the historical load data is labeled with a category label, resulting in several category labels, including production days or abnormal days; The normalized one-dimensional vector and several class labels are used as the dataset.

[0046] Specifically, the 11×96 two-dimensional feature matrix grayscale image array is flattened into a 1056-dimensional vector (11×96=1056), ensuring the data type is floating-point. If NaN values ​​exist, they are filled with 0 (to be confirmed according to requirements). Pixel values ​​are mapped to the 0-1 range using min-max scaling.

[0047] The actual production days and abnormal days of the load were obtained through interviews and other means. Each feature grayscale image was labeled to obtain several category labels. The scaled one-dimensional vector and several category labels were used as the dataset. The dataset was randomly divided into a training set (80 samples) and a test set (20 samples) in an 8:2 ratio.

[0048] Step 2.3: Construct the input layer, hidden layer, and output layer of the BP neural network model.

[0049] Specifically, the number of neurons in the input layer is equal to the dimension of a one-dimensional vector; the hidden layer uses the ReLU activation function; and the output layer uses the Softmax activation function.

[0050] Specifically, the number of neurons in the input layer equals the feature dimension of 1056 (corresponding to the flattened grayscale vector, i.e., the dimension of a one-dimensional vector); the hidden layer has 128 neurons and uses the ReLU activation function to enhance nonlinear fitting ability; the output layer has 2 neurons and uses the Softmax activation function to output the probability distribution.

[0051] Specifically, the output layer has two neurons, one for production days and one for abnormal days.

[0052] Step 2.4: Train the BP neural network model after constructing the input layer, hidden layer, and output layer using the training set, cross-entropy loss function, and Adam optimizer to obtain the output set. Compare the validation set with the output set and update the weights of the BP neural network model based on the comparison results to obtain the constructed baseline typical daily screening model.

[0053] Specifically, cross-entropy loss is used to measure the difference between the predicted probability and the true label. The Adam optimizer is selected, with a learning rate of 0.001, and weights are updated via gradient descent. Training parameters: batch size of 32, number of iterations of 100. A backpropagation (BP) neural network model is trained based on these parameters, recording the training set loss and accuracy in each round until training ends. Accuracy is calculated on the test set until the model converges and the accuracy is acceptable.

[0054] In some embodiments, the process of inputting load data into a pre-constructed baseline typical day screening model and outputting screening results, wherein the screening results include production days or abnormal days, specifically including: Input the load data into the constructed baseline typical day screening model to obtain the probability values ​​of production days and abnormal days; If the probability value of a production day is greater than the probability value of an abnormal day, then output the production day. If the probability value of a production day is less than the probability value of an abnormal day, then the abnormal day is output.

[0055] Specifically, the sum of the probability values ​​of abnormal days and abnormal days is 1.

[0056] Specifically, after training the baseline typical day screening model based on BP neural network, it can be built into the virtual power plant system. After collecting industrial load data, the model can be used to infer whether the date is a production day or an abnormal day, and then used to calculate different types of baselines.

[0057] Specifically, in this embodiment, the baseline typical day screening model output is [0.08, 0.92], indicating that the probability of the model judging the day as a production day is 8%, and the probability of it being an abnormal day is 92%.

[0058] In one embodiment of the present invention, a baseline typical day screening system based on a BP neural network is provided, comprising: The data acquisition module is used to acquire load data; The filtering module is used to input load data into the constructed baseline typical day filtering model and output the filtering results, which include production days or abnormal days. The construction steps of the baseline typical day filtering model include constructing a feature matrix grayscale image, training the BP neural network model with the preprocessed feature matrix grayscale image, and obtaining the baseline typical day filtering model.

[0059] Traditional baseline calculation methods filter weekdays and rest days based on calendar days. This method is not suitable for order-driven industrial load production. This invention transforms historical daily loads into feature matrices and grayscale images to intuitively represent historical load conditions and trains a BP neural network to achieve the filtering function for typical days. By using production days and abnormal days separately in the baseline calculation process, the virtual power plant can more accurately calculate the industrial user baseline, thereby reflecting the recent benchmark level of industrial load.

[0060] See Figure 3In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, and is suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to realize a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used for the operation of a baseline typical day screening method based on a BP neural network.

[0061] In one embodiment of the present invention, a computer-readable storage medium is provided, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the operating system of the terminal; and the storage space also stores one or more instructions suitable for loading and execution by a processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be a high-speed RAM memory or a non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the corresponding steps of a baseline typical day screening method based on a BP neural network in the embodiment.

[0062] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present invention can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0063] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0064] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0065] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A baseline typical day screening method based on BP neural network, characterized in that, Includes the following steps: Input the load data into the constructed baseline typical day screening model and output the screening results, which include production days or abnormal days. The construction steps of the baseline typical day screening model include constructing a feature matrix grayscale image, training the BP neural network model with the preprocessed feature matrix grayscale image, and obtaining the baseline typical day screening model.

2. The baseline typical day screening method based on a BP neural network according to claim 1, characterized in that, The steps for constructing the baseline typical day screening model specifically include: Acquire historical load data and construct a feature matrix grayscale image from the historical load data; The feature matrix grayscale image is preprocessed to obtain the dataset, which is then divided into a training set and a test set. Construct the input layer, hidden layer, and output layer of a BP neural network model; The BP neural network model, consisting of an input layer, hidden layer, and output layer, is trained using a training set, a cross-entropy loss function, and an Adam optimizer to obtain an output set. The validation set is then compared with the output set, and the weights of the BP neural network model are updated based on the comparison results to obtain the constructed baseline typical daily screening model.

3. The baseline typical day screening method based on a BP neural network according to claim 2, characterized in that, The construction of the feature matrix grayscale image using historical load data specifically includes: Construct a feature matrix including feature elements from historical load data. M The feature element is the battery level value for 15 minutes corresponding to the date and time. feature matrix M Scaling and integerization are performed sequentially to obtain the integer feature matrix. M Convert it into a grayscale image of the feature matrix.

4. The baseline typical day screening method based on a BP neural network according to claim 2, characterized in that, The step of preprocessing the grayscale image of the feature matrix to obtain the dataset specifically includes: Flatten the grayscale image of the feature matrix into a one-dimensional vector, and then normalize the one-dimensional vector. Based on the actual production situation, each date in the historical load data is labeled with a category label, resulting in several category labels, including production days or abnormal days; The normalized one-dimensional vector and several class labels are used as the dataset.

5. The baseline typical day screening method based on a BP neural network according to claim 4, characterized in that, The number of neurons in the input layer is equal to the dimension of a one-dimensional vector.

6. The baseline typical day screening method based on a BP neural network according to claim 2, characterized in that, The hidden layer uses the ReLU activation function; the output layer uses the Softmax activation function.

7. The baseline typical day screening method based on a BP neural network according to claim 1, characterized in that, The process involves inputting load data into a pre-constructed baseline typical day screening model and outputting screening results. These results include production days and abnormal days, specifically: Input the load data into the constructed baseline typical day screening model to obtain the probability values ​​of production days and abnormal days; If the probability value of a production day is greater than the probability value of an abnormal day, then output the production day. If the probability value of a production day is less than the probability value of an abnormal day, then the abnormal day is output.

8. A baseline typical day screening system based on a BP neural network, characterized in that, include: The data acquisition module is used to acquire load data; The filtering module is used to input load data into the constructed baseline typical day filtering model and output the filtering results, which include production days or abnormal days. The construction steps of the baseline typical day filtering model include constructing a feature matrix grayscale image, training the BP neural network model with the preprocessed feature matrix grayscale image, and obtaining the baseline typical day filtering model.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method as described in any one of claims 1-7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method as described in any one of claims 1-7.

Images