Smart park power load prediction method and device, electronic equipment and storage medium

Abstract

The invention provides a smart park power load prediction method and device, electronic equipment and a storage medium, and relates to the technical field of power systems, and the method comprises the steps: predicting a power load prediction value at a certain time point in the future through employing the linear steady-state correlation between historical power utilization information concerned by ARIMA; a DCNN-LSTM model is utilized to pay attention to nonlinear correlation between historical electricity consumption information, and a residual error between an observation value at a certain time point in the future and a power load prediction value is predicted. And determining a final power load prediction result by combining the power load prediction value and the residual prediction value. When power load prediction is carried out, irregular changes caused by factors such as meteorological information and park capacity are considered while regular changes presented by historical data are considered, so that the accuracy of power load prediction is improved.

Description

Technical Field

[0001] The present application relates to the technical field of power systems, and in particular to a method, device, electronic device and storage medium for predicting power load in a smart park. Background Art

[0002] At present, load forecasting is an important research direction in the field of power system technology. Power load forecasting is widely used in power dispatching management, energy planning, energy conservation and emission reduction, etc. In practice, power load is affected by many factors, and power load data often presents nonlinearity. Traditional power load forecasting methods are difficult to capture the changing rules when processing such data, thus affecting the accuracy of the forecast.

[0003] Therefore, how to improve the accuracy of power load forecasting has become an urgent problem to be solved. Summary of the invention

[0004] Based on this, it is necessary to provide a smart park power load forecasting method, device, electronic device and storage medium to address the above technical problems, so as to improve the accuracy of power load forecasting.

[0005] A method for predicting electric load of a smart park is provided, wherein the historical electric consumption information of the smart park is obtained, wherein the historical electric consumption information includes electric load observation values ​​at historical moments, meteorological information observation values ​​at the historical moments, and park production capacity observation values ​​at the historical moments; the historical electric consumption information is preprocessed to obtain historical time series data; the historical time series data is input into a pre-trained hybrid prediction model to obtain electric load prediction results at future moments; the hybrid prediction model includes an electric load prediction module constructed based on an ARIMA model, and a residual prediction module constructed based on a DCNN model and an LSTM model, wherein the residual prediction module is used to predict the residual between the electric load observation value at the same moment and the electric load prediction value output by the electric load prediction module.

[0006] In an embodiment of the present application, the historical electricity consumption information is preprocessed to obtain historical time series data, including: preprocessing the power load observation value, the meteorological information observation value, and the park production capacity observation value respectively, the preprocessing including outlier removal, missing value interpolation, wavelet threshold denoising, Pearson correlation analysis, and data normalization; converting the preprocessed power load observation value, the meteorological information observation value, and the park production capacity observation value into time series data respectively as the historical time series data.

[0007] In an embodiment of the present application, the historical time series data is input into a pre-trained hybrid prediction model to obtain a power load forecast result at a future moment, including: using the power load prediction module to perform fitting processing on the historical time series data to obtain a power load prediction value sequence, the power load prediction value sequence including the power load prediction value at the historical moment and the power load prediction value at the future moment; calculating the residual between the power load observation value and the power load prediction value at the same historical moment to obtain a residual data set; inputting the residual data set, the meteorological information observation value in the historical time series data, and the park production capacity observation value into the residual prediction module to obtain the residual prediction value at the future moment; determining the power load prediction result at the future moment according to the power load prediction value at the future moment and the residual prediction value.

[0008] In an embodiment of the present application, the residual prediction module includes a deep convolution layer constructed based on DCNN, a flattening layer, a recurrent layer constructed based on LSTM, and a fully connected layer. The residual data set, the meteorological information observation values ​​in the historical time series data, and the park production capacity observation values ​​are input into the residual prediction module to obtain the residual prediction value at the future moment, including: inputting the residual data set, the meteorological information observation values ​​in the historical time series data, and the park production capacity observation values ​​into the deep convolution layer to obtain a first feature map; inputting the first feature map into the flattening layer to obtain a one-dimensional feature vector; inputting the one-dimensional feature vector into the recurrent recurrent layer to obtain the hidden unit state of the recurrent recurrent layer; inputting the one-dimensional feature vector and the hidden unit state into the fully connected layer for weighted combined prediction to obtain the residual prediction value.

[0009] In an embodiment of the present application, the power load prediction result is determined based on the power load prediction value at the future moment and the residual prediction value, including: calculating the sum of the power load prediction value and the residual prediction value; inputting the sum into a sigmoid function to obtain the power load prediction result.

[0010] In an embodiment of the present application, the residual data set, the meteorological information observation values ​​in the historical time series data, and the park production capacity observation values ​​are input into the deep convolution layer to obtain a first feature map, including: performing deep convolution on the residual data set, the meteorological information observation values, and the park production capacity observation values ​​respectively to obtain a second feature map; and performing point-by-point convolution on the second feature map to obtain the first feature map.

[0011] In an embodiment of the present application, the meteorological information observation values ​​include temperature observation values ​​and humidity observation values.

[0012] A smart park power load prediction device comprises: an acquisition module for acquiring historical power consumption information of the smart park, wherein the historical power consumption information comprises power load observation values ​​at historical moments, meteorological information observation values ​​at the historical moments, and park production capacity observation values ​​at the historical moments; a preprocessing module for preprocessing the historical power consumption information to obtain historical time series data; a prediction module for inputting the historical time series data into a pre-trained hybrid prediction model to obtain power load prediction results at future moments; the hybrid prediction model comprises a power load prediction module constructed based on an ARIMA model, and a residual prediction module constructed based on a DCNN model and an LSTM model, wherein the residual prediction module is used to predict the residual between the power load observation value at the same moment and the power load prediction value output by the power load prediction module.

[0013] In an embodiment of the present application, the preprocessing module is used to: preprocess the power load observation value, the meteorological information observation value, and the park production capacity observation value, respectively, and the preprocessing includes outlier removal, missing value interpolation, wavelet threshold denoising, Pearson correlation analysis, and data normalization; convert the preprocessed power load observation value, the meteorological information observation value, and the park production capacity observation value into time series data, respectively, as the historical time series data.

[0014] In an embodiment of the present application, the prediction module is used to: use the power load prediction module to fit the historical time series data to obtain a power load prediction value sequence, wherein the power load prediction value sequence includes the power load prediction value at the historical moment and the power load prediction value at the future moment; calculate the residual between the power load observation value and the power load prediction value at the same historical moment to obtain a residual data set; input the residual data set, the meteorological information observation value in the historical time series data, and the park production capacity observation value into the residual prediction module to obtain the residual prediction value at the future moment; determine the power load prediction result at the future moment based on the power load prediction value at the future moment and the residual prediction value.

[0015] In an embodiment of the present application, the residual prediction module includes a deep convolution layer constructed based on DCNN, a flattening layer, a recurrent layer constructed based on LSTM, and a fully connected layer. The prediction module is used to: input the residual data set, the meteorological information observation values ​​in the historical time series data, and the park production capacity observation values ​​into the deep convolution layer to obtain a first feature map; input the first feature map into the flattening layer to obtain a one-dimensional feature vector; input the one-dimensional feature vector into the recurrent layer to obtain the hidden unit state of the recurrent layer; input the one-dimensional feature vector and the hidden unit state into the fully connected layer for weighted combined prediction to obtain the residual prediction value.

[0016] In an embodiment of the present application, the prediction module is used to: calculate the sum of the power load prediction value and the residual prediction value; input the sum into a sigmoid function to obtain the power load prediction result.

[0017] In an embodiment of the present application, the prediction module is used to: perform deep convolution on the residual data set, the meteorological information observation value, and the park production capacity observation value, respectively, to obtain a second feature map; and perform point-by-point convolution on the second feature map to obtain the first feature map.

[0018] In an embodiment of the present application, the meteorological information observation values ​​include temperature observation values ​​and humidity observation values.

[0019] An electronic device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the smart park power load prediction method described in the above embodiment are implemented.

[0020] A computer-readable storage medium stores a computer program, which, when executed by a processor, implements the steps of the smart park power load forecasting method as described in the above embodiment.

[0021] In summary, this application proposes a method, device, electronic device and storage medium for power load forecasting in a smart park. This method uses ARIMA to consider the linear steady-state correlation between historical power consumption information to predict the power load forecast value at a certain point in the future. The DCNN-LSTM model is used to consider the nonlinear correlation between historical power consumption information to predict the residual between the observed value and the power load forecast value at a certain point in the future. The power load forecast value and the residual forecast value are combined to determine the final power load forecast result. When forecasting power load, while considering the regular changes presented in the historical power consumption information, the irregular changes caused by factors such as meteorological information and park production capacity are considered to improve the accuracy of power load forecasting, and it has strong generalization and reliability. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0023] Figure 1 is a flow chart of a method for predicting power load in a smart park according to an exemplary embodiment of the present application;

[0024] Figure 2 is a flow chart of a method for predicting power load in a smart park according to an exemplary embodiment of the present application;

[0025] Figure 3 is a flow chart of a method for predicting power load in a smart park according to another exemplary embodiment of the present application;

[0026] Figure 4 is a flow chart of a method for predicting power load in a smart park according to another exemplary embodiment of the present application;

[0027] Figure 5 is a flow chart of a method for predicting power load in a smart park according to another exemplary embodiment of the present application;

[0028] Figure 6 is a flow chart of a method for predicting power load in a smart park according to another exemplary embodiment of the present application;

[0029] Figure 7 is a schematic block diagram of a smart park power load prediction device according to another exemplary embodiment of the present application;

[0030] Figure 8 It is a schematic block diagram of an electronic device according to an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0031] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. The embodiments described by reference to the drawings are exemplary and are intended to be used to explain the present application, and cannot be understood as limiting the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

[0032] Figure 1 is a flow chart of a method for predicting power load in a smart park according to an exemplary embodiment of the present application. Figure 1 As shown, the method comprises the following steps:

[0033] S101, obtaining historical electricity consumption information of the smart park, wherein the historical electricity consumption information includes the power load observation value at the historical moment, the meteorological information observation value at the historical moment, and the park production capacity observation value at the historical moment.

[0034] In some embodiments, the smart park power system monitoring platform can be used as a data source to obtain the historical power consumption information of the smart park from the power data collected by the monitoring platform. The historical power consumption information may include but is not limited to the power load observation value at the historical moment, the meteorological information observation value at the historical moment, and the park capacity observation value at the historical moment.

[0035] Among them, the moments in "historical moments" and "future moments" in the embodiments of the present application can be understood as time points or timestamps, which can be divided into different granularities such as day / hour. For example, with the day as the division granularity, the collected historical electricity consumption information may include daily power load observations, daily meteorological information observations, and daily park production capacity observations in the past month. Among them, the park production capacity can be quantified with different parameters / indicators based on the different types of work involved in the smart park. For example, for the park responsible for product production and processing, it can be quantified according to the single-day benefit indicators of the smart park.

[0036] Among them, the observed value can be understood as the actual value measured or collected, and the meteorological information may include but is not limited to at least one of the following: temperature, humidity, wind direction, sunshine duration, weather phenomena (cloudy, sunny, rainy, snowy), etc.

[0037] In addition to data such as power load, weather information, and park capacity, historical power consumption information may also include historical moments and date types of historical moments. For example, the date type may be divided into two types: holidays and non-holidays. In addition, historical power consumption information may also include other relevant information that affects power load.

[0038] S102, pre-processing the historical electricity usage information to obtain historical time series data.

[0039] The historical electricity consumption information can be preprocessed using preprocessing methods in statistical methods to improve the quality of data input into the hybrid prediction model and to adapt to the algorithm requirements of the hybrid prediction model.

[0040] For example, operations such as removing outliers, filling missing values, and correcting data errors are performed on various data in historical electricity consumption information to ensure data quality.

[0041] The relevant function is called to convert the historical electricity consumption information into time series data. In the embodiment of the present application, the converted time series data is summarized as historical time series data. The relevant function called may be the datetime function in Pandas or the ts function / zoo function in R language.

[0042] After type conversion, the historical time series data still contains various data items in the historical electricity consumption information, that is, the historical time series data may include but is not limited to the power load observation values ​​at historical moments, the meteorological information observation values ​​at historical moments, the park production capacity observation values ​​at historical moments, etc.

[0043] S103, inputting the historical time series data into a pre-trained hybrid prediction model to obtain a power load prediction result at a future moment.

[0044] In the embodiment of the present application, the hybrid prediction model may include but is not limited to a power load prediction module constructed based on an ARIMA model, a residual prediction module constructed based on a DCNN model and an LSTM model, and the like;

[0045] The power load forecasting module built based on the ARIMA model is used to predict the power load at any time and output the power load forecast value at any time.

[0046] The residual prediction module is used to predict the residual between the power load observation value at the same time and the power load prediction value output by the power load prediction module.

[0047] For example, the historical time series data from January 1 to January 7 is input into the power load forecasting model built based on the ARIMA model to predict the power load forecast values ​​from January 2 to January 8.

[0048] The residuals between the power load forecast values ​​and the power load observation values ​​from January 2 to January 7 are calculated respectively to obtain a residual data set. The residual data set is input into the residual prediction module constructed based on the DCNN model and the LSTM model to obtain a residual prediction value sequence including the residual prediction value of January 8.

[0049] The embodiment of the present application uses ARIMA in a hybrid prediction model to focus on the linear changes in historical power load, and uses DCNN and LSTM to focus on the nonlinear changes in the residual, so as to improve the accuracy of power load prediction.

[0050] In the embodiment of the present application, historical data is fitted according to ARIMA, and then short-term correlation information and local dependencies between variables are extracted through DCNN, such as extracting local dependencies between meteorological information, park capacity and power load, and long-term correlation information between load data is captured through LSTM, thereby mining the dependency information and change trends between historical electricity consumption information in a multi-faceted manner, and establishing a prediction model with excellent performance. Based on objective information such as historical power load data, weather conditions, and park capacity at a daily scale, the power load of related scenarios in the short term in the future can be predicted, providing data support for the dispatch of regional microgrids at a daily scale, improving the stability and security of the power supply plan of the power grid, achieving the purpose of energy saving and cost reduction, and promoting the construction of smart park microgrids.

[0051] In summary, this application proposes a method for forecasting power load in a smart park, which uses ARIMA to consider the linear steady-state correlation between historical power consumption information, predicts the power load forecast value at a certain point in the future, and uses the DCNN-LSTM model to consider the nonlinear correlation between historical power consumption information, and predicts the residual between the observed value and the power load forecast value at a certain point in the future. The power load forecast value and the residual forecast value are combined to determine the final power load forecast result. When forecasting power load, while considering the regular changes presented by historical data, the irregular changes caused by factors such as meteorological information and park production capacity are considered to improve the accuracy of power load forecasting, which has strong generalization and reliability.

[0052] In the embodiment of the present application, the trained hybrid prediction model can be obtained by the following method:

[0053] Collect historical relevant data to construct training samples. The training samples include input data and expected output. The input data may include historical load information, temperature and humidity information within the sample day, and park production capacity information.

[0054] Perform data preprocessing on the training samples, such as outlier removal, missing value interpolation, wavelet threshold denoising, Pearson correlation analysis, data normalization, difference series stabilization, white noise test, etc.

[0055] Data normalization: preprocessing data of different dimensions or ranges to make them have similar scales, which is conducive to model training. Differential stationarization test: stripping off the influence of trend and seasonal factors on data fluctuations, making the data close to white noise, which is more suitable for modeling.

[0056] The preprocessed training sample data is converted into time series data, and the time series data is processed using the autocorrelation function ACF and the partial autocorrelation function PACF to determine the order of the ARIMA model. The ARIMA model with the determined order is used to fit the time series data to obtain the power load forecast value. The residual stabilization test is performed on the predicted value and the observed value, and the ARIMA model parameters are optimized based on the test results.

[0057] The residual between the predicted value and the observed value is constructed as a data set, and the deep separable convolution of DCNN is used to extract features from the data set and extract feature maps. After the feature map is flattened, it is input into the recurrent layer with LSTM as the network structure. LSTM captures the long-term dependencies in the time series by controlling the information flow between gated information units. The gated cell state of the hidden unit of the LSTM recurrent layer satisfies the information memory problem in the same update stage in adjacent cycles, and solves the gradient disappearance problem caused by long-term cross-transfer.

[0058] The flattened vector of the feature map and the hidden unit state of the LSTM (i.e., the hidden unit gated cell state mentioned above) are feature mapped and combined in the fully connected layer. The output of the fully connected layer can be processed by an activation function to improve the expressiveness of the model. The activation function finally outputs the residual prediction value. Here, the activation function can use the ReLU function to limit the range of the output value.

[0059] The final power load prediction value can be obtained by superimposing the power load prediction value and the residual prediction value as the power load prediction result; or the final power load prediction value can be input into the sigmoid activation function, and binary classification processing is performed to obtain the probability of power load exceeding the standard as the power load prediction result, and so on.

[0060] When training the hybrid prediction model, the embodiment of the present application may use the optimizer Adam to optimize the model, and select the mean square error loss function as the loss function for model training.

[0061] During model training, the validation set is used to verify the model in the training process, and the hyperparameters and network structure are adjusted based on the processing results of the validation set to optimize the model performance.

[0062] In addition, the test set is used to evaluate the performance of the model and calculate evaluation indicators such as prediction error, accuracy, MAE, and MAPE. Finally, the model with the best expression effect is selected as the trained hybrid prediction model.

[0063] In addition, the trained model can be encapsulated into an API service and deployed to a cloud server or embedded in an electric energy management system. Design a data stream interface to receive new input data in real time or regularly for online prediction. Automatically compare the prediction results with the actual load, analyze and evaluate the prediction effect of the model in different time periods or conditions, adjust the model structure, parameters or preprocessing steps according to the evaluation results, and iteratively optimize the prediction system.

[0064] Based on the above embodiments, Figure 2 As shown, in the above step S102, "preprocessing the historical electricity usage information to obtain historical time series data" includes the following steps:

[0065] S201, preprocessing the power load observation value, the meteorological information observation value, and the park production capacity observation value respectively, wherein the preprocessing includes outlier removal, missing value interpolation, wavelet threshold denoising, Pearson correlation analysis, and data normalization.

[0066] S202, converting the pre-processed power load observation value, the meteorological information observation value, and the park production capacity observation value into time series data respectively as the historical time series data.

[0067] Data normalization maps values ​​to the interval [0, 1] so that different types of data have similar scales, facilitating data analysis.

[0068] Wavelet threshold denoising is used to remove interference noise and can effectively identify abnormal values ​​in historical electricity consumption information. Its principle is to use wavelet transform to decompose the signal into components of different scales and frequencies, then suppress the noise part according to threshold processing, and finally reconstruct the signal after noise removal through inverse wavelet transform to make the data smooth.

[0069] Before inputting information data into the hybrid prediction model for load forecasting, the historical electricity consumption information is first subjected to correlation analysis, stabilization test, and white noise test to ensure that the data input into the hybrid prediction model have strong correlation and stabilization characteristics, so that the hybrid prediction model can extract related information from the input historical data for prediction.

[0070] In some embodiments, the correlation analysis may be Pearson correlation analysis, and the stationary test may be ADF test. If a group of data has a low correlation, or exhibits non-stationarity, or the group of data is tested to be white noise, the group of data may be discarded, or some parameter values ​​in the group of data may be removed and filled in combination with the analysis results, so that the data is correlated and stationary as a whole.

[0071] Finally, the preprocessed historical electricity consumption information is converted into time series data.

[0072] The embodiment of the present application pre-processes the historical electricity consumption information to ensure that the data input into the hybrid prediction model have strong correlation and stability, thereby improving the accuracy of the model prediction.

[0073] Based on the above embodiments, Figure 3 As shown, in the above step S103, "inputting the historical time series data into a pre-trained hybrid prediction model to obtain a power load prediction result at a future moment" includes the following steps:

[0074] S301, using the power load prediction module to perform fitting processing on the historical time series data to obtain a power load prediction value sequence, wherein the power load prediction value sequence includes the power load prediction values ​​at the historical moments and the power load prediction values ​​at the future moments.

[0075] S302, calculating the residual between the power load observation value and the power load forecast value at the same historical moment to obtain a residual data set.

[0076] S303, inputting the residual data set, the meteorological information observation value in the historical time series data and the park production capacity observation value into the residual prediction module to obtain the residual prediction value at the future moment.

[0077] S304: Determine the power load prediction result at the future time according to the power load prediction value at the future time and the residual prediction value.

[0078] In the power load forecasting module, the trained ARIMA model is used to fit the historical time series data to predict the power load forecast values ​​at multiple different historical moments. For example, the power load forecast value on January 3 is predicted based on the historical time series data on January 1 and January 2, the power load forecast value on January 4 is predicted based on the historical time series data on January 2 and January 3, and the power load forecast value on January 5 is predicted based on the historical time series data on January 3 and January 4. Among them, January 5 is a future moment.

[0079] Calculate the residual between the power load forecast value on January 3 and the power load observation value in the historical time series data on January 3; calculate the residual between the power load forecast value on January 4 and the power load observation value in the historical time series data on January 4; use the residual forecast module to predict the residual forecast value on January 5 based on the residuals on January 3 and January 4. Superimpose the residual forecast value on January 5 and the power load forecast value on January 5 to obtain the power load forecast result on January 5.

[0080] It should be noted that in actual applications, a large number of historical time series data from historical moments can be selected to participate in power load forecasting. The selection of the above historical moments is only used as an example and is not a limitation on the moments corresponding to the historical time series data selected in this application.

[0081] In some embodiments, only the residual data set can be used as the input data of the residual prediction module; or the residual data set and the meteorological information observation value at the time corresponding to the residual, and the park production capacity observation value at the time corresponding to the residual can be used together as the input data of the residual prediction model, so as to further consider the impact of meteorological information and park production capacity on the difference fluctuation between the power load prediction value and the observed value when performing residual prediction.

[0082] Based on the above embodiments, Figure 4 As shown, in the above step S303, "inputting the residual data set, the meteorological information observation value in the historical time series data and the park production capacity observation value into the residual prediction module to obtain the residual prediction value at the future moment" includes the following steps:

[0083] S401, input the residual data set, the meteorological information observation values ​​in the historical time series data, and the park production capacity observation values ​​into the deep convolution layer to obtain a first feature map.

[0084] S402: Input the first feature map into the flattening layer to obtain a one-dimensional feature vector.

[0085] S403: Input the one-dimensional feature vector into the recurrent layer to obtain the hidden unit state of the recurrent layer.

[0086] S404: Input the one-dimensional feature vector and the hidden unit state into the fully connected layer for weighted combination prediction to obtain the residual prediction value.

[0087] The residual prediction module includes a deep convolution layer built based on DCNN, a flattening layer, a cyclic recursive layer built based on LSTM, and a fully connected layer.

[0088] First, the deep convolution layer based on DCNN is used to perform deep separable convolution on the input data to extract the feature map. The extracted feature map is input into the flattening layer to flatten it into a one-dimensional feature vector. The flattened one-dimensional feature vector is processed using a recurrent layer to extract the long-term dependencies in the features.

[0089] The flattened one-dimensional feature vector and hidden unit state are input into the fully connected layer for weighted combination, and the fully connected layer outputs the residual prediction value.

[0090] In some embodiments, the output of the fully connected layer may be processed by an activation function, such as ReLU, to limit the range of the output value, and the activation function outputs the final residual prediction value.

[0091] In some embodiments, a pooling layer may be added after the DCNN deep convolutional layer and before the LSTM hidden recursive layer to reduce the feature dimension and reduce computational complexity.

[0092] In terms of space, the embodiments of the present application extract local features through convolution operations and pooling operations and gradually combine them into global features; in terms of time, the long-term dependent relationship of the time series is captured through a gate mechanism; the spatial and temporal correlations in residual data, meteorological information, and park production capacity data are effectively captured, and attention is paid to the random fluctuations and seasonal effects between the data ignored by ARIMA, and attention is paid to the nonlinear changes in power load.

[0093] Based on the above embodiments, Figure 5 As shown, the above step S304 of "determining the power load forecast result at the future moment according to the power load forecast value at the future moment and the residual forecast value" includes the following steps:

[0094] S501, calculating the sum of the power load prediction value and the residual prediction value.

[0095] S502, inputting the sum value into a sigmoid function to obtain the power load prediction result.

[0096] In some embodiments, the power load prediction value and the residual prediction value may be superimposed and then input into the activation function sigmoid for binary classification, and finally the probability of the power load exceeding the standard is output as the power load prediction result.

[0097] Based on the above embodiments, Figure 6 As shown, in the above step S401, "inputting the residual data set, the meteorological information observation value in the historical time series data, and the park production capacity observation value into the deep convolution layer to obtain a first feature map" includes the following steps:

[0098] S601, performing deep convolution on the residual data set, the meteorological information observation value, and the park production capacity observation value to obtain a second feature map.

[0099] S602: Perform point-by-point convolution on the second feature map to obtain the first feature map.

[0100] The residual prediction module can be directly understood as a DCNN-LSTM model. The first layer of the DCNN-LSTM model is a deep convolutional layer, which is used for deep separable convolution.

[0101] Depthwise separable convolution is an efficient convolution method that decomposes the traditional convolution operation into two steps: depthwise convolution and pointwise convolution.

[0102] Depthwise convolution: This step performs convolution operations on each channel of the input. Each channel of the input feature map is element-wise multiplied with the corresponding convolution kernel to obtain the convolution results of multiple channels. These convolution results are summed in the channel direction. In this way, the output feature map of the depthwise convolution has the same number of channels as the input feature map, but the spatial size (height and width) may change, depending on the size of the convolution kernel, the stride, and the padding method.

[0103] Point-by-point convolution: This step is to further convolve the output feature map of the depthwise convolution, but this time a 1x1 convolution kernel is used. The 1x1 convolution kernel can control the number of channels of the output feature map by changing the number of convolution kernels without changing the spatial size of the feature map. The main function of point-by-point convolution is to achieve information fusion between channels and improve feature extraction capabilities.

[0104] In summary, the process of depthwise separable convolution can be summarized as first performing depthwise convolution to process each input channel independently, and then performing pointwise convolution to achieve information fusion between channels through a 1x1 convolution kernel. This operation can significantly reduce the number of parameters and calculations while maintaining model performance, making the model more lightweight and suitable for use in resource-limited environments.

[0105] In some embodiments, the depthwise convolution parameters are set as follows: convolution kernel size kernel_size = (3,3), strides = (1,1), activation function activation = 'relu'; the point-by-point convolution parameters are as follows: filters = 32, convolution kernel size kernel_size = (1,1), strides = (1,1), activation function activation = 'relu'.

[0106] In some embodiments, a pooling layer may be set at the next layer of the deep convolutional layer, and its pooling type is maximum pooling, which is used to extract local features. The pooling window size is pool_size=2, and the step length is strides=1. The pooling layer reduces the amount of calculation by reducing the spatial dimension of the feature map.

[0107] The embodiment of the present application collects and obtains objective data such as historical power load data uploaded by the microgrid IOT, meteorological information data, and park capacity; uses statistical methods to clean and normalize abnormal data; converts the data into time series data and sends it to the trained ARIMA-DCNN-LSTM hybrid prediction model to predict the power load situation. The invention establishes a hybrid prediction model. ARIMA predicts power load based on the linear stability law in the time series. The network part sets a DCNN-LSTM with good performance. The DCNN-LSTM neural network is a deep learning model that combines a deep convolutional neural network (DCNN) and a long short-term memory network (LSTM). DCNN is used to extract short-term correlation information and local dependencies between variables, and uses deep separable convolution to reduce the amount of calculation and parameters and improve efficiency. LSTM is responsible for extracting long-term correlation information of power load, which can effectively capture complex patterns in time series data, and is particularly suitable for processing data with spatial and temporal dependencies. Compared with the prior art, this method has generalization and reliability, and can be used to improve the accuracy of power load prediction within the daily time scale, and provide strong data support for short-term power dispatching.

[0108] It should be understood that the size of the serial numbers of the steps in the above embodiments does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0109] Figure 7 is a block diagram of a smart park power load prediction according to an exemplary embodiment of the present application. Figure 7 As shown, the device 700 includes: an acquisition module 701, a preprocessing module 702 and a prediction module 703.

[0110] The acquisition module 701 is used to acquire the historical electricity consumption information of the smart park, wherein the historical electricity consumption information includes the power load observation value at the historical moment, the meteorological information observation value at the historical moment, and the park capacity observation value at the historical moment;

[0111] A preprocessing module 702 is used to preprocess the historical electricity usage information to obtain historical time series data;

[0112] The prediction module 703 is used to input the historical time series data into a pre-trained hybrid prediction model to obtain the power load prediction result at the future time; the hybrid prediction model includes a power load prediction module constructed based on the ARIMA model, and a residual prediction module constructed based on the DCNN model and the LSTM model. The residual prediction module is used to predict the residual between the power load observation value at the same time and the power load prediction value output by the power load prediction module.

[0113] In an embodiment of the present application, the preprocessing module is used to: preprocess the power load observation value, the meteorological information observation value, and the park production capacity observation value, respectively, and the preprocessing includes outlier removal, missing value interpolation, wavelet threshold denoising, Pearson correlation analysis, and data normalization; convert the preprocessed power load observation value, the meteorological information observation value, and the park production capacity observation value into time series data, respectively, as the historical time series data.

[0114] In an embodiment of the present application, the prediction module: uses the power load prediction module to perform fitting processing on the historical time series data to obtain a power load prediction value sequence, and the power load prediction value sequence includes the power load prediction value at the historical moment and the power load prediction value at the future moment; calculates the residual between the power load observation value and the power load prediction value at the same historical moment to obtain a residual data set; inputs the residual data set, the meteorological information observation value in the historical time series data, and the park production capacity observation value into the residual prediction module to obtain the residual prediction value at the future moment; determines the power load prediction result at the future moment according to the power load prediction value at the future moment and the residual prediction value.

[0115] In an embodiment of the present application, the residual prediction module includes a deep convolution layer constructed based on DCNN, a flattening layer, a recurrent layer constructed based on LSTM, and a fully connected layer. The prediction module is used to: input the residual data set, the meteorological information observation values ​​in the historical time series data, and the park production capacity observation values ​​into the deep convolution layer to obtain a first feature map; input the first feature map into the flattening layer to obtain a one-dimensional feature vector; input the one-dimensional feature vector into the recurrent layer to obtain the hidden unit state of the recurrent layer; input the one-dimensional feature vector and the hidden unit state into the fully connected layer for weighted combined prediction to obtain the residual prediction value.

[0116] In an embodiment of the present application, the prediction module is used to: calculate the sum of the power load prediction value and the residual prediction value; input the sum into a sigmoid function to obtain the power load prediction result.

[0117] In an embodiment of the present application, the prediction module is used to: perform deep convolution on the residual data set, the meteorological information observation value, and the park production capacity observation value, respectively, to obtain a second feature map; and perform point-by-point convolution on the second feature map to obtain the first feature map.

[0118] In an embodiment of the present application, the meteorological information observation values ​​include temperature observation values ​​and humidity observation values.

[0119] In summary, the embodiment of the present application proposes a method for predicting power load in a smart park. It uses ARIMA to consider the linear steady-state correlation between historical power consumption information to predict the power load prediction value at a certain point in the future. It uses the DCNN-LSTM model to consider the nonlinear correlation between historical power consumption information to predict the residual between the observed value and the power load prediction value at a certain point in the future. The power load prediction value and the residual prediction value are combined to determine the final power load prediction result. When predicting power load, while considering the regular changes presented by historical data, the irregular changes caused by factors such as meteorological information and park production capacity are considered to improve the accuracy of power load prediction, and it has strong generalization and reliability.

[0120] To implement the above embodiments, the embodiments of the present application also propose an electronic device 800, including a memory 801, a processor 802, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the steps of the method for obtaining the marketing strategy described in the above embodiments are implemented.

[0121] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0122] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0123] The embodiments described above are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application, and should all be included in the protection scope of the present application.

Claims

1. A method for predicting power load in a smart park, characterized in that: include: Acquire historical electricity consumption information of the smart park, the historical electricity consumption information including the power load observation value at the historical moment, the meteorological information observation value at the historical moment, and the park production capacity observation value at the historical moment; Preprocessing the historical electricity consumption information to obtain historical time series data; Inputting the historical time series data into a pre-trained hybrid prediction model to obtain a power load prediction result at a future moment; The hybrid prediction model includes a power load prediction module constructed based on the ARIMA model and a residual prediction module constructed based on the DCNN model and the LSTM model. The residual prediction module is used to predict the residual between the power load observation value at the same time and the power load prediction value output by the power load prediction module.

2. The method according to claim 1, characterized in that The preprocessing of the historical electricity usage information to obtain historical time series data includes: Preprocessing the power load observation value, the meteorological information observation value, and the park capacity observation value respectively, wherein the preprocessing includes outlier removal, missing value interpolation, wavelet threshold denoising, Pearson correlation analysis, and data normalization; The pre-processed power load observation value, the meteorological information observation value, and the park production capacity observation value are respectively converted into time series data as the historical time series data.

3. The method according to claim 1, characterized in that The step of inputting the historical time series data into a pre-trained hybrid prediction model to obtain a power load prediction result at a future time includes: Using the power load prediction module to perform fitting processing on the historical time series data to obtain a power load prediction value sequence, wherein the power load prediction value sequence includes the power load prediction values ​​at the historical moments and the power load prediction values ​​at the future moments; Calculating the residual between the power load observation value and the power load forecast value at the same historical moment to obtain a residual data set; Inputting the residual data set, the meteorological information observation value in the historical time series data and the park production capacity observation value into the residual prediction module to obtain the residual prediction value at the future moment; The power load prediction result at the future time is determined according to the power load prediction value at the future time and the residual prediction value.

4. The method according to claim 3, characterized in that The residual prediction module includes a deep convolution layer constructed based on DCNN, a flattening layer, a recurrent layer constructed based on LSTM, and a fully connected layer. The residual data set, the meteorological information observation value in the historical time series data, and the park production capacity observation value are input into the residual prediction module to obtain the residual prediction value at the future moment, including: Input the residual data set, the meteorological information observation value in the historical time series data, and the park production capacity observation value into the deep convolution layer to obtain a first feature map; Inputting the first feature map into the flattening layer to obtain a one-dimensional feature vector; Inputting the one-dimensional feature vector into the recurrent layer to obtain the hidden unit state of the recurrent layer; The one-dimensional feature vector and the hidden unit state are input into the fully connected layer for weighted combination prediction to obtain the residual prediction value.

5. The method according to claim 3, characterized in that The step of determining the power load prediction result according to the power load prediction value at the future moment and the residual prediction value comprises: Calculating the sum of the power load prediction value and the residual prediction value; The sum value is input into a sigmoid function to obtain the power load prediction result.

6. The method according to claim 4, characterized in that The residual data set, the meteorological information observation value in the historical time series data, and the park capacity observation value are input into the deep convolution layer to obtain a first feature map, including: Performing deep convolution on the residual data set, the meteorological information observation value, and the park production capacity observation value respectively to obtain a second feature map; Perform point-by-point convolution on the second feature map to obtain the first feature map.

7. The method according to any one of claims 1 to 6, characterized in that: The meteorological information observation values ​​include temperature observation values ​​and humidity observation values.

8. A smart park power load forecasting device, characterized in that: include: An acquisition module, used to acquire historical electricity consumption information of the smart park, wherein the historical electricity consumption information includes the power load observation value at the historical moment, the meteorological information observation value at the historical moment, and the park capacity observation value at the historical moment; A preprocessing module, used for preprocessing the historical electricity consumption information to obtain historical time series data; A prediction module, used to input the historical time series data into a pre-trained hybrid prediction model to obtain a power load prediction result at a future moment; The hybrid prediction model includes a power load prediction module constructed based on the ARIMA model and a residual prediction module constructed based on the DCNN model and the LSTM model. The residual prediction module is used to predict the residual between the power load observation value at the same time and the power load prediction value output by the power load prediction module.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, it implements the steps of the smart park power load forecasting method as described in any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the steps of the smart park power load forecasting method as described in any one of claims 1 to 7 are implemented.

Images