CNN-LSTM-based integrated energy system cooling, heating and power load prediction method

Abstract

The invention relates to a CNN-LSTM-based integrated energy system cooling, heating and power load prediction method. The method comprises the steps of S1, collecting cooling, heating and power load data of a park integrated energy system; step S2, preprocessing the collected data by adopting a Z-Score standardization method; s3, performing data correlation analysis on the preprocessed data; step S4, a CNN-LSTM model is constructed; and S5, adopting the constructed CNN-LSTM model to predict the cooling, heating and power load of the integrated energy system. Compared with the prior art, the method has the advantages of being beneficial to predicting the load of the park more accurately and more efficiently, accurately matching the electricity demand of the park, reducing the electric energy loss and the like.

Description

Technical Field

[0001] The present invention relates to a comprehensive energy system load forecasting technology, and in particular to a CNN-LSTM-based comprehensive energy system cooling, heating and electricity load forecasting method. Background Art

[0002] An integrated energy system refers to the use of advanced physical information technology and innovative management models in a certain area to integrate multiple energy sources such as coal, oil, natural gas, electricity, and thermal energy in the region, and realize coordinated planning, optimized operation, collaborative management, interactive response, and complementary assistance among multiple heterogeneous energy subsystems. Due to the complexity, randomness, and nonlinearity of cold, hot, and electric loads, traditional prediction methods such as time series analysis or regression model analysis often only consider the time series of the same load, while ignoring the impact between each load and related environmental factors. Machine learning algorithms such as neural networks, support vector machines, and random forests are also widely used in the prediction of cold, hot, and electric loads in integrated energy systems. These algorithms can predict future load changes by learning the characteristics and laws of historical data, but a single algorithm requires a large amount of historical data to train the model. If the data is insufficient or of low quality, it may affect the accuracy of the prediction results; for some complex and uncertain load change patterns, the prediction ability of machine learning algorithms may be limited. However, the combined model increases the complexity and training cost, and still faces challenges.

[0003] After searching, Chinese patent CN202311853939.8 proposes the technical field of integrated energy systems, and specifically discloses a load prediction model training method and a load prediction method for an integrated energy system. The MTL model based on the long short-term memory network extracts the coupling information between multiple loads and makes a preliminary prediction, and then uses the STL model based on the pre-double attention long short-term memory network to reduce input noise for secondary prediction; at the same time, Chinese patent CN202410141115.6 discloses a distributed load prediction method and system for an integrated energy system. The distributed load prediction method and system for an integrated energy system provided collects and analyzes the energy supply parameters and environmental parameters of each load in the energy supply area, thereby effectively improving the accuracy of the prediction values ​​of each load.

[0004] However, the above prior art has the following problems:

[0005] 1) Low prediction accuracy: Due to the complexity, randomness and nonlinearity of cold, hot and electric loads, traditional prediction methods such as sequence-to-sequence time series analysis or regression model analysis often only consider the time series of the same load, while ignoring the impact between each load and related environmental factors; this leads to low prediction accuracy for cold, hot and electric loads, which makes it difficult to meet the needs of efficient operation of the integrated energy system.

[0006] 2) Lack of research on multi-factor load forecasting: Compared with the extensive and in-depth research on electricity load forecasting at home and abroad, there are relatively few studies on multi-factor load forecasting for cooling, heating and electricity; this limits the development and application of cooling, heating and electricity load forecasting technology for integrated energy systems.

[0007] 3) Challenges brought by the diversity of energy forms: The integrated energy system involves multiple energy forms such as electricity, gas, cold and heat. The physical properties of various energy sources are different, and different energy loads will affect each other; this makes it difficult to directly apply the load forecasting method used in traditional single energy systems to integrated energy systems, increasing the difficulty of forecasting.

[0008] 4) Complexity of data collection and analysis: In order to accurately calculate the load, a series of data collection and analysis work is required, such as actual energy consumption data survey, regional load density analysis, typical industry energy consumption analysis, etc., which places high demands on prediction technology.

[0009] 5) Uncertainty factors: For large-scale integrated energy service projects at the park, city, and regional levels, the uncertainty of energy load increases the difficulty of project planning and design, which also puts higher requirements on the adaptability and accuracy of load forecasting technology.

[0010] Therefore, how to improve the load forecasting accuracy of the integrated energy system and optimize the coordinated control of cooling, heating and electricity has become a technical problem that needs to be solved. Summary of the invention

[0011] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a method for predicting the cooling, heating and electricity loads of an integrated energy system based on CNN-LSTM.

[0012] The purpose of the present invention can be achieved by the following technical solutions:

[0013] According to one aspect of the present invention, a method for predicting cooling, heating and electricity loads of an integrated energy system based on CNN-LSTM is provided, the method comprising:

[0014] Step S1, collecting cooling, heating and electricity load data of the park's comprehensive energy system;

[0015] Step S2, preprocessing the collected data using the Z-Score standardization method;

[0016] Step S3, performing data correlation analysis on the preprocessed data;

[0017] Step S4, constructing a CNN-LSTM model;

[0018] Step S5, using the constructed CNN-LSTM model to predict the cooling, heating and electricity loads of the integrated energy system.

[0019] As a preferred technical solution, the cooling, heating and electricity load data in step S1 include summer cooling load, summer electricity load, winter heating load and winter electricity load.

[0020] As a preferred technical solution, the load in step S1 adopts daily load, and the specific collected data includes:

[0021] Step S1.1, obtain the average daily load curve for a typical day

[0022] Step S1.2, obtaining a typical daily load curve δ;

[0023] Step S1.3, obtaining the typical daily load rate;

[0024] Step S1.4, obtaining the minimum load factor of a typical day.

[0025] As a preferred technical solution, the Z-Score standardization method in step S2 converts data of different magnitudes into Z-Score values ​​of uniform measurement for comparison, thereby unifying the data standards.

[0026] As a preferred technical solution, the Z-Score standardization calculation is as follows:

[0027]

[0028] where x in is the original load value; μ is the mean value of the population; δ is the standard deviation of the population.

[0029] As a preferred technical solution, the data correlation analysis in step S3 is:

[0030] Considering the correlation between summer cooling load and electric load and the correlation between winter heating load and winter electric load, the load correlation is analyzed and predicted by using the Pearson correlation coefficient.

[0031] As a preferred technical solution, the Pearson correlation coefficient calculation formula is as follows:

[0032]

[0033] Where:

[0034] r ij is the correlation coefficient between the i-th load and the j-th influencing factor;

[0035] x i ,y j is the i-th sampling point of load x and the j-th sampling point of influencing factor y;

[0036] is the average value of n sample points of load x and influencing factor y;

[0037] n: number of sample points.

[0038] As a preferred technical solution, the data correlation analysis in step S4 is:

[0039] Step S4.1, extracting features from the input load data through a convolutional neural network (CNN) model;

[0040] In step S4.2, the output of the CNN model is used as the input of the LSTM, and the LSTM is used for training and prediction analysis.

[0041] As a preferred technical solution, the convolutional neural network (CNN) model includes an input layer, a convolution layer, a pooling layer, and an excitation layer;

[0042] The input layer is used to convert the time series load data into a matrix vector and input it into the prediction model;

[0043] The convolution layer is used to extract characteristic information of the load, thereby obtaining global feedback of the characteristic information;

[0044] The pooling layer uses a maximum pooling layer to extract the maximum features from each feature vector;

[0045] The excitation layer is used to obtain load data for data feature extraction.

[0046] As a preferred technical solution, the method also includes: using mean absolute error MAE, mean relative percentage error MAPE and root mean square error RMSE to judge the effectiveness and optimization ability of the model.

[0047] Compared with the prior art, the present invention has the following advantages:

[0048] 1) The prediction dimension of the present invention is more comprehensive, and it directly predicts the cooling, heating and electricity loads of the integrated energy system, covering the load changes in the three dimensions of cooling, heating and electricity, and more comprehensively reflects the actual needs of the integrated energy system.

[0049] 2) By combining the characteristics of convolutional neural networks (CNN) and long short-term memory networks (LSTM), the present invention can more effectively capture the spatiotemporal characteristics in load data and significantly improve prediction accuracy.

[0050] 3) The present invention has made deeper innovations and optimizations in model architecture, data processing, and feature extraction to better meet the needs of cooling, heating, and electricity load forecasting in integrated energy systems.

[0051] 4) The CNN-LSTM model of the present invention combines the powerful feature extraction capabilities of convolutional neural networks (CNN) and the long-term dependency capture capabilities of long short-term memory networks (LSTM) when processing time series data. This enables the model to deeply explore the complex patterns and regularities in the cooling, heating and power load data, thereby providing more accurate prediction results. This not only helps system operators to make scheduling plans in advance, but also reduces energy waste and cost increases caused by prediction errors.

[0052] 5) In the present invention, accurate load forecasting is the key to achieving efficient operation of the system in terms of optimizing the cooling, heating and electricity coordinated control system. Load forecasting through the CNN-LSTM model can provide the park with reliable load change trend information, enabling it to flexibly adjust the cooling, heating and electricity supply strategy according to actual conditions, and achieve optimal configuration and efficient utilization of energy. This can not only improve the overall operating efficiency of the system, but also reduce energy consumption and environmental pollution, and achieve sustainable development. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a specific flow chart of the method of the present invention. DETAILED DESCRIPTION

[0054] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0055] The present invention is suitable for parks that want to improve the accuracy of load prediction of integrated energy systems, and helps to predict park loads more accurately and efficiently, accurately match park electricity demand, and reduce power losses.

[0056] like Figure 1 As shown, the method of the present invention specifically comprises:

[0057] Step 1: Collect the cooling, heating and electricity data of the park's integrated energy system, and divide them into daily load forecasting, monthly load forecasting and annual load forecasting according to the type of load forecasting required. The data indicators required for different forecast types of daily, monthly and annual forecasts are shown in Table 1. This invention is a short-term load forecasting study, so the daily load is analyzed in detail. a. Obtain the average daily load curve for a typical day

[0058]

[0059] Where:

[0060] Average daily load;

[0061] L(d,h): historical working day load;

[0062] d and h: represent date and hour;

[0063] N: The total number of working days in a season.

[0064] b. Select the typical daily load curve δ, as shown in the following formula. Select the load curve that is closest to the average daily load curve in all working days in summer and winter as the typical daily load curve. The similarity of the curves is measured by the sum of the Euclidean distances. The curve with the smallest sum of the Euclidean distances is the typical daily load curve. The definition is as follows:

[0065]

[0066] Where:

[0067] Average daily load;

[0068] L(d,h): historical working day load.

[0069] c. Typical day load rate: the ratio of the average load on a typical day to the maximum load on a typical day.

[0070] d. Typical day minimum load rate: the ratio of the minimum load on a typical day to the maximum load on a typical day.

[0071] Table 1

[0072]

[0073] Step 2: Data preprocessing. Using the Z-Score standardization method, data of different magnitudes are converted into Z-Score values ​​of uniform measurement for comparison, so that the data standards are unified, thereby improving data comparability and weakening data interpretability. Z-Score standardization calculation is as follows:

[0074]

[0075] Where:

[0076] x in : original load value;

[0077] μ: the mean value of the population;

[0078] δ: standard deviation of the population.

[0079] Step 3: Data correlation analysis. The correlation between summer cooling load and electric load and the correlation between winter heating load and winter electric load are mainly considered. The load correlation is analyzed and predicted by using the Pearson correlation coefficient.

[0080] The Pearson calculation formula is as follows:

[0081]

[0082] Where:

[0083] r ij : Correlation coefficient between the i-th load and the j-th influencing factor;

[0084] x i ,y j : The i-th sampling point of load x and the j-th sampling point of influencing factor y;

[0085] x, y: the average values ​​of n sample points of load x and influencing factor y;

[0086] n: number of sample points.

[0087] Step 4: CNN-LSTM model establishment

[0088] 4.1 Feature extraction of input load data Feature extraction is performed through convolutional neural network.

[0089] Adding a convolutional neural network (CNN) to the prediction model facilitates the extraction of local features from the input data and the capture of LSTM layer context information. CNN consists of an input layer, a convolution layer, a pooling layer, an excitation layer, a fully connected layer, and an output layer. ① Input layer: The input layer of CNN converts the load data of the time series into a matrix vector and inputs it into the prediction model, and then extracts features from the input load data, mainly through the convolution layer, the pooling layer, and the excitation layer. ② Convolution layer: Extract the characteristic information of the load, thereby obtaining global feedback of the characteristic information, avoiding the problem of information loss caused by the input of a long time series of the recurrent neural network and the difficulty in mining information between data. It is necessary to determine the size of the convolution kernel in the convolution layer, which determines the features of the input data extracted by the convolution kernel. ③ Pooling layer: The extracted data features are input into the pooling layer. The present invention selects the maximum pooling layer to facilitate the maximum feature extraction of each feature vector, which is conducive to better storage of data information. ④ Excitation layer: The feature data processed by the pooling layer obtains a new vector, and then passes through the excitation layer to obtain the load data with data feature extraction.

[0090] 4.2CNN-LSTM encoding. The preprocessed historical load data is input into the CNN model to extract feature information for output, and then the output of the CNN model is used as the input of the LSTM, which is trained for prediction and analysis. The LSTM update formula is as follows:

[0091] f t =σ(w f ·[ht-1 ,x t ]+b f )

[0092] i t =σ(w i ·[h t-1 ,x t ]+b i )

[0093] o t =σ(w o ·[h t-1 ,x t ]+b o )

[0094] c t % = tanh(w c ·[h t-1 ,x t ]+b c )

[0095] h t =o t e tanh(c t )

[0096] c t =f t ec t-1 +i t ec t %

[0097] Where:

[0098] w f , b f : weight matrix and bias vector of the forget gate;

[0099] w i , b i : weight matrix and bias vector of input gate;

[0100] w o , b o : The weight matrix and bias vector of the output gate;

[0101] σ: activation function, usually Tanh function or Sigmoid function;

[0102] x t 、c t 、h t : The data input to LSTM at time t, the memory state and the corresponding output data.

[0103] Step 5: Model evaluation and performance analysis. The effectiveness and optimization ability of the model are evaluated by the mean absolute error (MAE), mean relative percentage error (MAPE) and root mean square error (RMSE). ① MAE is the average value of the absolute error between the load forecast value and the true value. The smaller the value of the evaluation index, the better the prediction effect; ② MAPE is the average value of the load relative error. The smaller the value of the evaluation index, the better the prediction effect; ③ RMSE is the square root of the mean square error. The smaller the evaluation index, the better the prediction effect.

[0104]

[0105] Where: y i is the true value of the load at time i, It is the load forecast value at the corresponding moment in the forecast process, and N represents the sample data.

[0106] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A method for forecasting cooling, heating and electricity loads of an integrated energy system based on CNN-LSTM, characterized in that: The method includes: Step S1, collecting cooling, heating and electricity load data of the park's comprehensive energy system; Step S2, preprocessing the collected data using the Z-Score standardization method; Step S3, performing data correlation analysis on the preprocessed data; Step S4, constructing a CNN-LSTM model; Step S5, using the constructed CNN-LSTM model to predict the cooling, heating and electricity loads of the integrated energy system.

2. According to the CNN-LSTM-based cold, heat and electricity load prediction method for an integrated energy system according to claim 1, it is characterized in that: The cooling, heating and electricity load data in step S1 include summer cooling load, summer electricity load, winter heating load and winter electricity load.

3. According to the CNN-LSTM-based cold, heat and electricity load prediction method for an integrated energy system according to claim 1, it is characterized in that: The load in step S1 adopts daily load, and the specific collected data includes: Step S1.1, obtain the average daily load curve for a typical day Step S1.2, obtaining a typical daily load curve δ; Step S1.3, obtaining the typical daily load rate; Step S1.4, obtaining the minimum load factor of a typical day.

4. According to the CNN-LSTM-based cold, heat and electricity load prediction method for an integrated energy system according to claim 1, it is characterized in that: The Z-Score standardization method in step S2 converts data of different magnitudes into Z-Score values ​​of uniform measurement for comparison, thereby standardizing the data.

5. A method for predicting cooling, heating and electricity loads of an integrated energy system based on CNN-LSTM according to claim 1 or 4, characterized in that: The Z-Score standardization is calculated as follows: where x in is the original load value; μ is the mean value of the population; δ is the standard deviation of the population.

6. According to the CNN-LSTM-based cold, heat and electricity load prediction method for an integrated energy system according to claim 1, it is characterized in that: The data correlation analysis in step S3 is as follows: Considering the correlation between summer cooling load and electric load and the correlation between winter heating load and winter electric load, the load correlation is analyzed and predicted by using the Pearson correlation coefficient.

7. According to claim 6, a method for predicting cooling, heating and electricity loads of an integrated energy system based on CNN-LSTM, characterized in that: The Pearson correlation coefficient calculation formula is as follows: Where: r ij is the correlation coefficient between the i-th load and the j-th influencing factor; x i ,y j is the i-th sampling point of load x and the j-th sampling point of influencing factor y; is the average value of n sample points of load x and influencing factor y; n: number of sample points.

8. According to the CNN-LSTM-based cold, heat and electricity load prediction method for an integrated energy system according to claim 1, it is characterized in that: The data correlation analysis in step S4 is as follows: Step S4.1, extracting features from the input load data through a convolutional neural network (CNN) model; In step S4.2, the output of the CNN model is used as the input of the LSTM, and the LSTM is used for training and prediction analysis.

9. According to the CNN-LSTM-based cold, heat and electricity load prediction method for an integrated energy system as claimed in claim 8, it is characterized in that: The convolutional neural network (CNN) model includes an input layer, a convolution layer, a pooling layer, and an excitation layer; The input layer is used to convert the time series load data into a matrix vector and input it into the prediction model; The convolution layer is used to extract characteristic information of the load, thereby obtaining global feedback of the characteristic information; The pooling layer uses a maximum pooling layer to extract the maximum features from each feature vector; The excitation layer is used to obtain load data for data feature extraction.

10. The method for predicting cooling, heating and electricity loads of an integrated energy system based on CNN-LSTM according to claim 1, characterized in that: The method also includes: using mean absolute error MAE, mean relative percentage error MAPE and root mean square error RMSE to judge the effectiveness and optimization ability of the model.

Images