Method and system for predicting carbon reduction amount of park carbon emission based on new energy

Abstract

The invention relates to the technical field of new energy and energy saving, and provides a new energy-based park carbon emission carbon reduction amount prediction method and system, which fuse and use a comprehensive similar day selection method of grey correlation analysis-dynamic time warping-Euclidean distance-included angle cosine combination, completely integrate empirical mode decomposition adaptive noise, and improve the prediction accuracy of the carbon reduction amount of park carbon emission. A high-precision photovoltaic short-term power generation power combined prediction model is established, and the problems that a similar day selection method of a traditional photovoltaic output prediction model excessively depends on subjective experience judgment, the screening standard is single, photovoltaic power generation data contains noise and multi-scale characteristics and the like are solved. The difficulty of accurate prediction of output caused by intermittency and volatility of photovoltaic power generation is overcome, and accurate prediction of photovoltaic power generation capacity is realized; and on the other hand, the photovoltaic output prediction result is innovatively combined with the electric power carbon emission factor, accurate measurement and calculation of the carbon reduction amount of the park new energy project are achieved, and powerful support is provided for achieving the green and low-carbon development target of the park.

Description

Technical Field

[0001] The present disclosure relates to the technical field of new energy and energy conservation, and in particular to a method and system for predicting carbon reduction of industrial park carbon emissions based on new energy. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] With the global emphasis on green and low-carbon development, industrial parks, as gathering places for industries and population, are also one of the main sources of carbon emissions, and their role in optimizing energy structure and controlling carbon emissions has become increasingly prominent. Existing data shows that the energy consumption of industrial parks accounts for 69% of the total energy consumption of the whole society, and carbon emissions account for 31% of the total emissions in the country. Therefore, accurately measuring carbon emissions in industrial parks and formulating corresponding carbon reduction measures are important basic work to promote low-carbon transformation.

[0004] Photovoltaic power generation has shown significant advantages in optimizing the energy structure of the park and reducing carbon emissions due to its clean, efficient and sustainable characteristics. By deploying distributed photovoltaic power generation systems, the park can increase its energy self-sufficiency rate, reduce its dependence on fossil energy, and improve the quality of the ecological environment. However, to achieve efficient operation of photovoltaic power generation systems and quantitative evaluation of carbon reduction effects, it is necessary to establish a high-precision photovoltaic power generation prediction model to provide technical support for the planning and implementation of new energy projects.

[0005] The inventors found in their research that existing research on photovoltaic power generation prediction has problems such as incomplete identification of influencing factors, strong subjectivity in similar day selection methods, low efficiency in hyperparameter optimization, and insufficient data noise processing, which leads to low accuracy and reliability of existing prediction models. In addition, the photovoltaic power generation prediction results have not been fully integrated into the carbon reduction calculation of new energy projects, resulting in inaccurate carbon reduction effect evaluation, making it difficult to provide effective guidance for the park's new energy construction plan. Summary of the invention

[0006] In order to solve the above problems, the present invention proposes a carbon reduction prediction method and system for carbon emissions in a park based on new energy. Based on the high-precision photovoltaic output prediction model of the Pearson correlation coefficient method, the comprehensive distance method, the CEEMDAN decomposition technology and the Informer model, the carbon emission factor of the power is combined to realize the accurate calculation of the carbon emission reduction of the park, providing strong technical support for the green and low-carbon development of the park.

[0007] In order to achieve the above objectives, the present disclosure adopts the following technical solutions:

[0008] One or more embodiments provide a method for predicting carbon reduction of a park's carbon emissions based on new energy, comprising the following steps:

[0009] Obtain the meteorological factor data of the park to be predicted;

[0010] Use correlation analysis method to screen out key meteorological factors that affect photovoltaic output;

[0011] After preprocessing the key meteorological factor data, the data are input into the trained photovoltaic output prediction model for prediction, and the photovoltaic output prediction result of the prediction day is obtained; the photovoltaic output prediction model adopts the comprehensive distance method integrating grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine fusion to construct a training set for training;

[0012] Based on the photovoltaic output forecast results and the carbon dioxide emission factor, the carbon emission reduction based on photovoltaic power generation is calculated.

[0013] One or more embodiments provide a carbon reduction prediction system for park carbon emissions based on new energy, including:

[0014] Data acquisition module: configured to acquire meteorological factor data of the park to be predicted;

[0015] Screening module: configured to use correlation analysis method to screen out key meteorological factor data that affect photovoltaic output;

[0016] Prediction module: configured to pre-process the acquired key meteorological factor data and input it into the trained photovoltaic output prediction model for prediction, so as to obtain the photovoltaic output prediction result of the prediction day; the photovoltaic output prediction model adopts the comprehensive distance method integrating grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine fusion to construct a training set for training;

[0017] Calculation module: configured to calculate the carbon emission reduction based on photovoltaic power generation according to the photovoltaic output prediction result and the carbon dioxide emission factor.

[0018] An electronic device includes a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are run by the processor, the steps in the above-mentioned method for predicting carbon reduction of park carbon emissions based on new energy are completed.

[0019] A computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the steps in the above-mentioned method for predicting carbon reduction of park carbon emissions based on new energy are completed.

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] The method disclosed in the present invention integrates the use of grey correlation analysis, dynamic time warping, a comprehensive similar day selection method combining Euclidean distance and angle cosine, complete ensemble empirical mode decomposition adaptive noise (CEEMDAN) and Informer model to establish a high-precision photovoltaic short-term power combination prediction model, solves the problems of the similar day selection method of the traditional photovoltaic output prediction model being too dependent on subjective experience judgment and a single screening standard, and the photovoltaic power generation data containing noise and multi-scale characteristics, overcomes the difficulties brought by the intermittent and volatile photovoltaic power generation to the accurate prediction of output, and realizes the accurate prediction of photovoltaic power generation; on the other hand, the photovoltaic output prediction results are innovatively combined with the power carbon emission factor to realize the accurate calculation of the carbon reduction of the park's new energy projects, and provide strong support for the park to achieve the green and low-carbon development goals.

[0022] The advantages of the present disclosure and the advantages of additional aspects will be described in detail in the following specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The accompanying drawings constituting a part of the present disclosure are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and the description thereof are used to explain the present disclosure but do not constitute a limitation of the present disclosure.

[0024] Figure 1 It is a schematic diagram of the training process of the photovoltaic output prediction model of the carbon reduction prediction method of Example 1 of the present disclosure;

[0025] Figure 2 is a flow chart of a method for decomposing photovoltaic power generation sequence using the CEEMDAN method according to Embodiment 1 of the present disclosure;

[0026] Figure 3 This is a schematic diagram of the main flow of the Informer model of Example 1 of the present disclosure;

[0027] FIG4( a ) is a schematic diagram of power generation in a typical summer month in an example experiment of Example 1 of the present disclosure;

[0028] FIG4( b ) is a temperature variation diagram of a typical summer month in an exemplary experiment of Example 1 of the present disclosure;

[0029] FIG4( c ) is a diagram showing changes in weather forecast irradiance for a typical summer month in an example experiment of Example 1 of the present disclosure;

[0030] FIG4( d ) is a diagram showing humidity changes in a typical summer month in an exemplary experiment of Example 1 of the present disclosure;

[0031] Figure 5 is the Person coefficient value of various meteorological factors and photovoltaic output in the example experiment of Example 1 of the present disclosure;

[0032] FIG6( a ) is a comparison diagram of photovoltaic outputs on similar days selected by different similar day selection methods in the example experiment of Example 1 of the present disclosure;

[0033] FIG6( b ) is a comparison diagram of the irradiance of similar days selected by different similar day selection methods in the example experiment of Example 1 of the present disclosure;

[0034] FIG6( c ) is a comparison diagram of similar day temperatures selected by different similar day selection methods in the example experiment of Example 1 of the present disclosure;

[0035] Figure 7 This is a diagram showing the decomposition results of the power generation on similar historical days in the example experiment of Example 1 of the present disclosure. DETAILED DESCRIPTION

[0036] The present disclosure is further described below in conjunction with the accompanying drawings and embodiments.

[0037] It should be noted that the following detailed descriptions are exemplary and are intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meanings as those commonly understood by those skilled in the art to which the present disclosure belongs.

[0038] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should be understood that when the terms "comprising" and / or "including" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof. It should be noted that, in the absence of conflict, the various embodiments in the present disclosure and the features in the embodiments can be combined with each other. The embodiments will be described in detail below in conjunction with the accompanying drawings.

[0039] Example 1

[0040] In the technical solutions disclosed in one or more embodiments, Figures 1 to 7 As shown, a method for predicting carbon reduction of a park's carbon emissions based on new energy includes the following steps:

[0041] Step 1: Obtain the meteorological factor data of the park to be predicted;

[0042] Step 2: Use correlation analysis method to screen out key meteorological factor data that affect photovoltaic output;

[0043] Step 3: After preprocessing the key meteorological factor data, the data is input into the trained photovoltaic output prediction model for prediction to obtain the photovoltaic output prediction result for the prediction day;

[0044] The photovoltaic output prediction model adopts a comprehensive distance method that combines grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine fusion to select similar days with similar meteorological characteristics and output curves to the day to be predicted from the historical day sample set, and uses the data of similar days as training sets to train multiple deep learning models, and selects the optimal model after training as the photovoltaic output prediction model;

[0045] Step 4: Calculate the carbon emission reduction based on photovoltaic power generation based on the photovoltaic output prediction results and the carbon dioxide emission factor.

[0046] The method in this embodiment integrates the use of grey correlation analysis, dynamic time warping, a comprehensive similar day selection method combining Euclidean distance and angle cosine, complete ensemble empirical mode decomposition adaptive noise (CEEMDAN) and the Informer model to establish a high-precision photovoltaic short-term power combination prediction model, which solves the problems of the similar day selection method of the traditional photovoltaic output prediction model being too dependent on subjective experience judgment and a single screening standard, and the photovoltaic power generation data containing noise and multi-scale characteristics, overcomes the difficulties in accurately predicting the output caused by the intermittent and volatile photovoltaic power generation, and realizes the accurate prediction of photovoltaic power generation; on the other hand, the photovoltaic output prediction results are innovatively combined with the power carbon emission factor to realize the accurate calculation of the carbon reduction of the park's new energy projects, providing strong support for the park to achieve the green and low-carbon development goals.

[0047] In step 2, the correlation analysis method may adopt the Pearson correlation coefficient method to calculate the Pearson correlation coefficient between the meteorological factor and the photovoltaic power generation power, and the meteorological factor with a coefficient greater than a set threshold is taken as the key meteorological factor.

[0048] A further technical solution, a method for training a photovoltaic output prediction model, includes the following steps:

[0049] Step S1, obtaining historical output data of the park's distributed photovoltaic power generation system and meteorological data;

[0050] Specifically, meteorological data may include data information of meteorological factors including irradiance, temperature, humidity, wind speed and wind direction;

[0051] Check the relevant information of the park's distributed photovoltaic power generation system, collect and organize the power generation data for the past two years, as well as meteorological parameters such as total radiation irradiance, temperature, humidity, wind direction and wind speed.

[0052] Step S2: Based on the data information obtained in step S1, the Pearson correlation coefficient method is used to identify key meteorological factors affecting photovoltaic output;

[0053] Step S3, using the comprehensive distance method of GRA-DTW-Euclidean distance-angle cosine combination, select similar days with similar meteorological characteristics and output curves to the day to be predicted from the historical day sample set;

[0054] Step S4: Based on the sequence reconstruction principle of sample entropy, the CEEMDAN method is used to decompose the output sequence of similar historical days to generate a model training sample set;

[0055] Step S5, using the meteorological data of the above-selected similar days and the power generation after steady-state decomposition as a training set, inputting them into the Informer model, and training to obtain a CEEMDAN-Informer model based on comprehensive distance similarity;

[0056] Step S6: Obtain a benchmark model, compare the prediction accuracy of the CEEMDAN-Informer model and the benchmark model with the set evaluation index, and select the optimal model as the photovoltaic output prediction model;

[0057] Specifically, the benchmark model may include GRU, LSTM, CNN-LSTM model and the Informer model without using the CEEMDAN method; using the GRU, LSTM, CNN-LSTM model and the Informer model without using the CEEMDAN method as the benchmark model, using the Mean Absolute Error (MAE), Root Mean Squared Error (RMSE) and Mean Absolute Percentage Error (MAPE) as evaluation indicators, comparing the prediction accuracy of the CEEMDAN-Informer model and the benchmark model, and determining the most suitable output prediction model for the target power station;

[0058] In step S2, based on the candidate influencing factors of photovoltaic output such as total radiation irradiance, temperature, humidity, wind direction and wind speed, the Pearson correlation coefficient method (PCC method) is used to determine the key meteorological influencing factors of photovoltaic output. The correlation coefficient calculation formula is:

[0059]

[0060] Where: r is the correlation coefficient between variables; X is the main influencing factor of photovoltaic output; Y is the short-term output data, in MW.

[0061] Furthermore, a correlation threshold is set, and when the calculated correlation coefficient is greater than the correlation threshold, the corresponding influencing factor is determined to be a key meteorological influencing factor of photovoltaic output; this embodiment adopts the PCC method to select two types of key meteorological influencing factors including total radiation irradiance and temperature;

[0062] In step S3, the degree of numerical similarity and shape similarity between the sequences are comprehensively considered, and the comprehensive distance method of GRA-DTW-Euclidean distance-angle cosine combination is used, that is, the Euclidean distance, grey relational degree, angle cosine and dynamic programming algorithm (DTW) are effectively combined, and the historical similarity of the day to be predicted is determined by calculating the comprehensive similarity between the day to be predicted and the historical days belonging to the same category; the steps specifically include:

[0063] Step S31: Calculate the Euclidean distance between the data of the day to be predicted and the historical day:

[0064] The Euclidean distance is a measure of the straight-line distance between the predicted date and the historical date in n-dimensional space, which is applicable to vector X = {x1, x2, ..., x N} and Y={y1,y2,…,y N}, the calculation formula is:

[0065]

[0066] Where: x i is the irradiance of the day to be predicted (W / m 2 ), temperature (℃) or short-term output data of the corresponding day; i is the historical daily irradiance (W / m 2 ), temperature (℃) or the short-term output data of the corresponding day, the unit of short-term output data is MW; vector X is the data of the day to be predicted; vector Y is the vector data of similar historical days

[0067] Step S32: Calculate the grey correlation between the day to be predicted and the historical day:

[0068] Step S321, normalize the data vectors X and Y to obtain a normalized data sequence X' and a reference sequence Y';

[0069] Step S322, calculating the absolute value of the difference between the data sequence X' and the reference sequence Y';

[0070] Step S323, calculate the correlation coefficient, i.e., the grey correlation degree, according to the absolute value of the difference between the sequences. The calculation formula is as follows:

[0071] Δ i =|x i '-y i '|(2b)

[0072]

[0073] Where: ξi is the grey relational degree, ρ is the resolution coefficient, which is used to control the sensitivity of the relational degree.

[0074] In formula (2c), the i below max and min represents the variable range for which the maximum or minimum value is sought;

[0075] Step S33: Calculate the cosine distance between the predicted day and the historical day:

[0076] The cosine distance measures the directional similarity by measuring the angle between two vectors. The calculation formula is as follows:

[0077]

[0078] Where: cos(θ) is the cosine angle between two vectors. The closer its value is to 1, the more similar the two vectors are.

[0079] Step S34: Based on the obtained Euclidean distance, the shortest distance between the predicted day and the historical day is obtained through a dynamic programming algorithm, that is, the optimal path is obtained. The specific algorithm for the shortest distance is as follows:

[0080]

[0081] Where: γ(i,j) represents the cumulative minimum distance between the i-th data point of the to-be-predicted day and the j-th data point of the historical day, starting from the starting point of the to-be-predicted day sequence and the historical day sequence. OD represents the Euclidean distance between the to-be-predicted day and the historical day.

[0082] Step S35: Using the weighted analysis method, the calculation results of the Euclidean distance, grey correlation, cosine distance and DTW value are weighted and combined to obtain the comprehensive similarity value between the predicted day and the historical day. The calculation formula is as follows:

[0083] μ=σ1·OD+σ2·ξ i +σ3·cos(θ)+σ4vγ(2f)

[0084] Where: μ is the comprehensive similarity value; σ1 is the weight of the Euclidean distance value; σ2 is the weight of the grey relational degree; σ3 is the weight of the cosine distance value; σ4 is the weight of the DTW value.

[0085] Step S36: Select similar days with similar meteorological characteristics and output curves to the day to be predicted according to the comprehensive similarity value; the smaller the comprehensive similarity value, the higher the similarity, and a threshold value can be set, and days less than the threshold value are determined as similar days;

[0086] In step S4, the CEEMDAN method is used to decompose the output sequence of similar historical days; while significantly reducing the reconstruction error and computational cost, it can effectively solve the problem of modal mixing and the difficulty in aligning the final ensemble average caused by the differences in the IMF decomposition results of each group.

[0087] Step S4, using the CEEMDAN method to decompose the output sequence, includes the following steps:

[0088] Step S41: Add Gaussian white noise to the original photovoltaic power signal s(t) of the similar day, and the noise level of the white noise is dynamically adjusted according to the volatility of the short-term output value; and the first IMF component of CEEMDAN is obtained by calculating the average value:

[0089]

[0090] Where: En(.) is the nth mode component of EMD.

[0091] Step S42: Calculate the first residual component based on the first IMF component:

[0092] r1(t)=s(t)-IMF1(3b)

[0093] Step S43: construct a new signal according to the first residual component, and obtain the second IMF component through EMD decomposition:

[0094] s1(t)=r1(t)+E1(ni(t))(3c)

[0095]

[0096] Step S44: loop through the process of step S41 to step S43 to obtain multiple residual signals and multiple IMFs, the nth residual signal and the (n+1)th IMF are:

[0097] r n (t) = r n-1 (t)-IMF n (t)(3e)

[0098]

[0099] Finally, the original signal can be expressed as:

[0100]

[0101] The end condition of the above decomposition is: until no meaningful IMF can be extracted from the residual signal, the judgment method is to average the modal components of the EMD decomposition results and judge whether the residual is a monotonic function. When it indicates that the signal decomposition has reached the limit;

[0102] In the above scheme, averaging is used to analyze the dynamic characteristics of the IMF components, ensure the accuracy of the decomposition results, and provide a basis for judging the monotonicity of the residual signal; judging that the residual is a monotonic function is a direct condition for the end of decomposition. By analyzing the dynamic behavior of the residual signal, it is ultimately decided whether to stop the decomposition.

[0103] In step S5, the meteorological data of similar days and the power generation decomposed by the CEEMDAN method are used as training sets and input into the Informer model to construct a CEEMDAN-Informer model based on comprehensive distance similarity. The steps specifically include:

[0104] Step S51: construct an Informer model, including an encoder, a decoder, and a fully connected layer; the encoder uses a multi-layer Transformer network;

[0105] Then, the meteorological data of similar days and the power generation after steady-state decomposition are used as training sets, and the Informer neural network is used to perform photovoltaic forecasting;

[0106] Step S52: pre-processing the steady-state decomposed power generation and meteorological data of similar days;

[0107] Extract power generation and meteorological series from the data set to construct the input data set, and perform the time series data X = {x1, x2, ..., x N}Standardization or normalization is performed to avoid the impact of different scales on model performance.

[0108] Step S53: The preprocessed historical time series data is used as the input of the encoder and first transmitted to the CNN network for feature extraction to capture the correlation and dynamic patterns over a long period of time;

[0109] The encoder input at time t includes temperature and irradiance, which can cover the historical data of the past 8 days to predict the photovoltaic output of the next day. The input data is expressed as:

[0110]

[0111] Where: h is the normalized dataset; m is the dimension of the historical dataset, which is set to 2.

[0112] Step S54: In the encoder, a sparse self-attention mechanism is used to filter out the data with the greatest relevance, avoiding processing all the data at each time step:

[0113]

[0114] Where: (·)att indicates that the attention module is used to perform the attention operation; Contains the basic variables of the attention module and the sparse attention mechanism; Convld(·) refers to a one-dimensional convolution operation on the time series, and Convld(·) represents the activation operation using exponential linear units (ELU) as the activation function; Maxpooling represents the maximum pooling operation with a step size of 2, which aims to halve the amount of computation per layer while retaining the information of long input time series.

[0115] Step S55: Use the multi-layer Transformer in the encoder to perform layered processing on the time series filtered by the sparse self-attention mechanism, that is, the distillation operation in the figure. The input and output of each layer of the Transformer are:

[0116]

[0117] Where: Z l,in is the input of layer l, Z l,out is the output of layer l.

[0118] Step S56: Decoding the encoder output and the conditional information after layered processing to obtain decoded output features;

[0119] The decoder input includes the encoder output and the layered processed conditional information Z l,out , the prediction process is:

[0120]

[0121] Where: X de is the output data of the decoder.

[0122] Step S57: The output data of the decoder is fused using a fully connected layer to obtain a prediction result;

[0123] The output data in the decoder is combined with the periodic load value of the load sequence to form a fully connected layer to achieve information fusion. Finally, the prediction result is generated in the output layer to predict the target long sequence in real time.

[0124] In power system load forecasting, the load sequence is the data sequence of power load over a period of time. The periodic load value is the part of this sequence that shows a periodic pattern. Taking the power load as an example, the daily peak and trough of electricity consumption may show a certain periodicity. For example, the residential electricity consumption is at its peak at 7-10 pm and at its trough at 2-4 am. These periodic load values ​​are periodic load values. The output data of the decoder contains some features and patterns learned by the model from the input data. The periodic load value of the load sequence carries the inherent periodic information of the data itself. In time series prediction, especially for the prediction of long target sequences with obvious periodicity, combining the two can enable the model to better utilize the two information sources.

[0125] Step S58: Adjust the parameters of the model by setting the evaluation index, and repeat the above training process until the required accuracy is reached to obtain the trained CEEMDAN-Informer model.

[0126] For further technical solutions, a single Informer, GRU, LSTM, and CNN-LSTM model are selected as the benchmark model, and the mean absolute error (MAE), root mean square error (RMSE), and mean absolute percentage error (MAPE) are used as evaluation indicators to compare the prediction accuracy of the CEEMDAN-Informer model and the basic model, and determine the most suitable output prediction model for the target power station. The specific calculation formulas of the three indicators are as follows:

[0127]

[0128]

[0129]

[0130] Where: N is the number of sampling points of photovoltaic output; p(n) and p'(n) are the actual value and predicted value of photovoltaic output respectively.

[0131] In a further technical solution, in step 4, the carbon emission reduction of the park's new energy investment project is calculated by combining the output prediction results of the selected model and the power carbon dioxide emission factor. The calculation formula specifically includes:

[0132] C(n)=EF×p′(n)(6a)

[0133] In the formula: C(n) is the carbon emission reduction, kg; EF is the electricity carbon dioxide emission factor, and its value is: 0.8426kgCO2 / kWh.

[0134] To illustrate the method of this embodiment, the model building process and prediction process of the method of this embodiment are explained below with a specific prediction example;

[0135] Step A: Check and investigate the relevant information of the distributed photovoltaic power generation system in the park. It can be seen that the location of the distributed photovoltaic power generation system is relatively concentrated. Collect and organize the power generation of the target park power generation system in the past two years, as well as meteorological parameters such as total radiation irradiance, temperature, humidity, wind direction and wind speed. Table 1 lists the meaning and influencing mechanism of meteorological factors affecting photovoltaic output; Figure 4 shows a schematic diagram of meteorological factors and power generation in a typical summer month.

[0136] Table 1 Summary of main meteorological factors affecting photovoltaic output

[0137]

[0138] Step B: To ensure the rationality and accuracy of the results, collect the actual monitoring data of five meteorological factors, namely, radiation intensity, wind speed, temperature, humidity and wind direction, and the corresponding power generation during the historical period. After eliminating redundant historical samples and data outliers, calculate the Pearson correlation coefficient between the five meteorological factors and the power generation to screen and determine the key influencing factors of photovoltaic output.

[0139] Figure 5 The Person coefficient heat map of candidate meteorological factors and photovoltaic output is displayed. Among them, the Person coefficients of total radiation irradiance and temperature with photovoltaic output are the highest, which are 0.968 and 0.722 respectively. Therefore, total radiation irradiance and temperature are selected as the key input variables of the photovoltaic output prediction model, laying the foundation for the selection of subsequent similar days and the construction of the optimization model.

[0140] Step C, randomly select August 22 to 28, 2023 as the days to be predicted, use historical samples from 2021 to 2023 as candidate days, and use the GRA-DTW-Euclidean-angle cosine comprehensive distance method to generate the historical similar day set of the above-mentioned days to be predicted. Table 2 lists the selection results of similar days. In order to better demonstrate the advanced nature of the combined similar day selection method of this embodiment, taking August 22, 2023, the day to be predicted, as an example, on the basis of selecting its historical similar days using a single GRA method and a comprehensive similarity method, the similar days selected by the two methods are compared and evaluated in terms of the degree of similarity between the key meteorological factors and power generation power of the days to be predicted, as shown in Figure 6 for details.

[0141] Table 2 Similar day selection results

[0142]

[0143]

[0144]

[0145] The total radiation irradiance and temperature were selected as the selection criteria. The historical similar days selected by the comprehensive similarity method of GRA-DTW-Euclidean distance-cosine angle combination had a high degree of consistency with the volatility of temperature, radiation and output on the day to be predicted. As shown in Table 3, the average comprehensive similarity between the target day to be predicted and the selected similar days was 0.9671. On the contrary, the single GRA method failed to fully consider the dynamic alignment of the time series and the comprehensive influence of multi-dimensional characteristics, resulting in the failure to accurately capture the dynamic change characteristics of multi-dimensional meteorological parameters at some moments, so the deviation was more significant, and the average comprehensive similarity between the selected day to be predicted and the selected similar days was only 0.9542.

[0146] Table 3 Comparison of similar day selection results of two methods

[0147]

[0148]

[0149] Step D: Since the original photovoltaic power sequence exhibits strong randomness and volatility, it is difficult to predict it directly, and it is easy to cause the recursive module to produce a gradient vanishing problem, so that the predicted value shows a significant lag relative to the true value.

[0150] Therefore, this experiment uses the short-term power output data of selected historical similar days as input variables and uses the CEEMDAN method to decompose them. The decomposition results are shown in Figure 7 .Depend on Figure 7 It can be seen that the decomposed trend components show different periodic characteristics and fluctuation degrees. Among them, the sample entropy values ​​of the first, second and third dimensional time series are 0.2286, 0.2128 and 0.09227 respectively, which show an obvious entropy reduction trend compared with the sample entropy value of 0.3374 of the original time series.

[0151] It can be seen that the CEEMDAN method can decompose the historical photovoltaic power generation series into multidimensional subsequences with regular changes, which significantly reduces the complexity and unpredictability of the original time series and optimizes the overall prediction performance of the model.

[0152] Step E: The meteorological data of the similar days selected above and the power generation after steady-state decomposition are used as training sets, input into the Informer model, and a CEEMDAN-Informer model based on comprehensive distance similarity is constructed.

[0153] Taking the prediction date of August 22, 2023, 10:00-12:00 as an example, the model prediction results are shown in Table 4.

[0154] Table 4 Prediction results of CEEMDAN-Informer model for some time periods

[0155] time 10:00 10:15 10:30 10:45 11:00 11:15 11:30 11:45 12:00 True value 19.79 21.46 22.9 24.34 25.77 27.21 27.93 28.66 29.38 Predicted value 20.04 21.86 23.94 25.33 26.48 27.31 27.83 27.62 27.43

[0156] Step F: Using single Informer, GRU, LSTM, and CNN-LSTM models as baseline models, and using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), and Mean Absolute Percentage Error (MAPE) as evaluation indicators, the prediction accuracy of the CEEMDAN-Informer model and the basic model is compared to determine the most suitable output prediction model for the target power station. Table 5 shows the differences in the average values ​​of the evaluation indicators (MAE, RMSE, and MAPE) of the five models for all predicted days.

[0157] Table 5 Comparison of model prediction results

[0158]

[0159]

[0160] In the table, LSTM (Long Short-Term Memory) is a long short-term memory neural network;

[0161] GRU (Gated Recurrent Unit) is a simplified version of the recurrent neural network, similar to LSTM, but with a simpler structure. It reduces the number of parameters by merging the forget gate and input gate of LSTM, thereby improving training efficiency.

[0162] CNN-LSTM (Convolutional Neural Network-LSTM) is a hybrid model that combines convolutional neural network (CNN) and LSTM. The CNN module is used to extract local features in time series (such as short-term fluctuations or periodic changes). The extracted features are passed as input to LSTM to further capture long-term dependencies.

[0163] Informer is a variant of the Transformer model designed specifically for processing long-sequence time series data. It is based on the Self-Attention Mechanism and can capture long-range dependencies in the sequence. By introducing the Sparse Attention Mechanism, the computational cost of the model is significantly reduced.

[0164] As shown in the table above, the MAE, RMSE, and MAPE values ​​of the single Informer model are 1.23, 1.60, and 0.06, respectively. Compared with the single Informer, GRU, LSTM, and CNN-LSTM models, the CEEMDAN-Informer model shows a higher prediction effect.

[0165] Step G: Calculate the carbon emission reduction of the park's new energy investment projects by combining the output prediction results of the selected model and the electricity carbon dioxide emission factor.

[0166] Based on the forecast results from August 22 to August 28, it can be seen that the total power generation from 8:00 to 16:00 every day is approximately 160.09MW, 173.40MW, 168.81MW, 169.43MW, 210.97MW, 216.19MW and 209.09MW. According to the national fossil energy power carbon dioxide emission factor of 0.8426kgCO2 / kWh in 2021, the specific carbon emission reduction value of the forecast day is calculated, and the results are shown in Table 6. By analogy, the carbon emission reduction of the park's future new energy investment projects can be accurately estimated.

[0167] Table 6 Carbon emission reduction on the forecasted day

[0168] MAE RMSE MAPE CEEMDAN-Informer 0.68 0.89 0.03 LSTM 1.45 1.90 0.15 GRU 1.64 1.87 0.10 CNN-LSTM 1.45 1.66 0.11 INFORMER 1.23 1.60 0.06

[0169] Step H. According to the above introduction to the photovoltaic output combination prediction method and carbon emission reduction calculation method of steps A to G, the data collection module, influencing factor identification module, similar day selection module, CEEMDAN decomposition module, prediction model construction module, prediction model performance evaluation module and carbon reduction calculation module are constructed in sequence to complete the deployment of the virtual device for accurate carbon reduction calculation in the park.

[0170] The prediction method of this embodiment first collects historical photovoltaic output data of the distributed photovoltaic power generation system and data on meteorological factors affecting photovoltaic output; secondly, based on the data collected in step A, the Pearson correlation coefficient method is used to perform correlation analysis on five types of meteorological factors including total radiation irradiance, wind speed, wind direction, temperature and humidity and photovoltaic short-term output, and extracts two types of factors that have a greater impact on photovoltaic output, namely total radiation irradiance and temperature; thirdly, the comprehensive distance method of GRA-DTW-Euclidean distance-angle cosine combination is used to classify historical days, and identify historical similar days with highly similar meteorological conditions to the day to be predicted; next, the CEEMDAN method is used to decompose the output sequence of historical similar days to generate a high-quality model training sample set; then, combined with key indicators such as MAE, RMSE and MAPE, a single neural network prediction model (LSTM, GRU, CNN-LST) is compared and analyzed. The performance indicators of the combined prediction model of CEEMDAN-Informer) and CEEMDAN-Informer are used to determine the most suitable PV output prediction model for the park, and the problems of the similar day selection method of the traditional PV output prediction model being overly dependent on subjective experience judgment and a single screening standard, as well as the noise and multi-scale characteristics contained in the PV power generation data are solved. The difficulties caused by the intermittent and volatile PV power generation in accurately predicting the output are overcome, and the accurate prediction of PV power generation is achieved. Furthermore, on the basis of selecting the actual carbon emissions of the park without implementing PV power generation projects as the baseline, the carbon emission reduction of the park's new energy investment projects is calculated by combining the output prediction results of the selected model and the national fossil energy power carbon dioxide emission factor. The PV output prediction results are innovatively combined with the power carbon emission factor to achieve accurate calculation of the carbon reduction of the park's new energy projects, providing strong support for the park to achieve its green and low-carbon development goals.

[0171] Example 2

[0172] Based on Example 1, this embodiment provides a carbon reduction prediction system for park carbon emissions based on new energy, including:

[0173] Data acquisition module: configured to acquire meteorological factor data of the park to be predicted;

[0174] Screening module: configured to use correlation analysis method to screen out key meteorological factor data that affect photovoltaic output;

[0175] Prediction module: configured to pre-process the acquired key meteorological factor data and input it into the trained photovoltaic output prediction model for prediction, so as to obtain the photovoltaic output prediction result of the prediction day; the photovoltaic output prediction model adopts the comprehensive distance method integrating grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine fusion to construct a training set for training;

[0176] Calculation module: configured to calculate the carbon emission reduction based on photovoltaic power generation according to the photovoltaic output prediction result and the carbon dioxide emission factor.

[0177] It should be noted here that the various modules in this embodiment correspond one-to-one to the various steps in Example 1, and the specific implementation process is the same, which will not be repeated here.

[0178] Example 3

[0179] Based on Example 1, this embodiment provides an electronic device, including a memory and a processor, and computer instructions stored in the memory and running on the processor. When the computer instructions are executed by the processor, the steps in the method for predicting carbon reduction of park carbon emissions based on new energy described in Example 1 are completed.

[0180] Example 4

[0181] Based on Example 1, this embodiment provides a computer-readable storage medium for storing computer instructions. When the computer instructions are executed by a processor, the steps in the method for predicting carbon reduction in a park based on new energy described in Example 1 are completed.

[0182] The above description is only a preferred embodiment of the present disclosure and is not intended to limit the present disclosure. For those skilled in the art, the present disclosure may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

[0183] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Technical personnel in the relevant field should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A method for predicting carbon reduction in a park based on new energy, characterized in that: The steps include: Obtain the meteorological factor data of the park to be predicted; Use correlation analysis method to screen out key meteorological factors that affect photovoltaic output; After preprocessing the key meteorological factor data, the data are input into the trained photovoltaic output prediction model for prediction, and the photovoltaic output prediction result of the prediction day is obtained; the photovoltaic output prediction model adopts the comprehensive distance method integrating grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine fusion to construct a training set for training; Based on the photovoltaic output forecast results and the carbon dioxide emission factor, the carbon emission reduction based on photovoltaic power generation is calculated.

2. The method for predicting carbon reduction of a park's carbon emissions based on new energy as claimed in claim 1, characterized in that: The comprehensive distance method that integrates grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine is used to select similar days with similar meteorological characteristics and output curves to the day to be predicted from the historical daily sample set. Multiple deep learning models are trained based on the data of similar days as training sets, and the optimal model after training is selected as the photovoltaic output prediction model.

3. The method for predicting carbon reduction of a park's carbon emissions based on new energy as claimed in claim 1, characterized in that: The correlation analysis method used was the Pearson correlation coefficient method.

4. The method for predicting carbon reduction of a park's carbon emissions based on new energy as claimed in claim 1, characterized in that: The training method of the photovoltaic output prediction model includes the following steps: Obtain historical output data of the park's distributed photovoltaic power generation system and meteorological data; The Pearson correlation coefficient method is used to identify key meteorological factors affecting photovoltaic output; The comprehensive distance method integrating grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine is used to select similar days with similar meteorological characteristics and output curves to the day to be predicted from the historical day sample set; Based on the principle of sequence reconstruction based on sample entropy, the CEEMDAN method is used to decompose the output sequence of similar historical days to generate a model training sample set; The meteorological data of similar days and the power generation decomposed by CEEMDAN method are used as training sets and input into the Informer model for training, and the CEEMDAN-Informer model based on comprehensive distance similarity is constructed; Obtain the benchmark model, compare the prediction accuracy of the CEEMDAN-Informer model and the benchmark model with the set evaluation indicators, and select the optimal model as the photovoltaic output prediction model.

5. The method for predicting carbon reduction of industrial park carbon emissions based on new energy as claimed in claim 4, characterized in that: The comprehensive distance method is used to select similar days with similar meteorological characteristics and output curves to the day to be predicted from the historical day sample set, including the following steps: Calculate the Euclidean distance between the data of the day to be predicted and the historical day; Calculate the grey correlation between the day to be predicted and the historical day; Calculate the cosine distance between the predicted day and the historical day; Based on the obtained Euclidean distance, the shortest distance between the predicted day and the historical day is obtained through the dynamic programming algorithm; Using the weighted analysis method, the calculation results of Euclidean distance, grey correlation, cosine distance and DTW value are weighted and combined to obtain the comprehensive similarity value between the predicted day and the historical day.

6. The method for predicting carbon reduction of a park's carbon emissions based on new energy as claimed in claim 5, characterized in that: Calculate the grey correlation between the day to be predicted and the historical day: The data vectors X and Y are standardized to obtain a standardized data sequence X' and a reference sequence Y'; Calculate the absolute value of the difference between the data sequence X' and the reference sequence Y'; The correlation coefficient, i.e., grey correlation degree, is calculated based on the absolute value of the difference between the sequences.

7. The method for predicting carbon reduction of a park's carbon emissions based on new energy as claimed in claim 4, characterized in that: The CEEMDAN method is used to decompose the output sequence, which includes the following steps: Step S41, adding Gaussian white noise to the original photovoltaic power signal s(t) of the similar day; the noise level of the white noise is dynamically adjusted according to the volatility of the short-term output value; and the first IMF component of CEEMDAN is obtained by calculating the average value; Step S42, calculating a first residual component based on the first IMF component; Step S43, constructing a new signal according to the first residual component, and obtaining a second IMF component through EMD decomposition; According to the decomposition process from step S41 to step S43, a plurality of residual signals and a plurality of IMF components are obtained; Alternatively, a CEEMDAN-Informer model based on comprehensive distance similarity is constructed. The training steps include: Construct an Informer model, including an encoder, a decoder, and a fully connected layer; the encoder uses a multi-layer Transformer network; Preprocess the steady-state decomposed power generation and meteorological data of similar days; The preprocessed historical time series data is used as the input of the encoder and first transmitted to the CNN network for feature extraction; The extracted features are input into the encoder, and the sparse self-attention mechanism is used to filter out the data with the greatest relevance. The multi-layer Transformer in the encoder is used to perform hierarchical processing on the time series filtered by the sparse self-attention mechanism. Decode the output of the encoder and the conditional information after layered processing to obtain the decoded output features; The decoder output data is fused using a fully connected layer to obtain the prediction result; By setting the evaluation index, adjusting the parameters of the model, the above training process is repeated until the required accuracy is met, and the trained CEEMDAN-Informer model is obtained.

8. A carbon reduction prediction system for park carbon emissions based on new energy, characterized in that: include: Data acquisition module: configured to acquire meteorological factor data of the park to be predicted; Screening module: configured to use correlation analysis method to screen out key meteorological factor data that affect photovoltaic output; Prediction module: configured to pre-process the acquired key meteorological factor data and input it into the trained photovoltaic output prediction model for prediction, so as to obtain the photovoltaic output prediction result of the prediction day; the photovoltaic output prediction model adopts the comprehensive distance method integrating grey correlation analysis, dynamic time warping, Euclidean distance and angle cosine fusion to construct a training set for training; Calculation module: configured to calculate the carbon emission reduction based on photovoltaic power generation according to the photovoltaic output prediction result and the carbon dioxide emission factor.

9. An electronic device, characterized in that: The method comprises a memory and a processor and computer instructions stored in the memory and executed on the processor. When the computer instructions are executed by the processor, the steps in the method for predicting carbon reduction of a park based on new energy are completed as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store computer instructions, which, when executed by a processor, complete the steps in the method for predicting carbon reduction in a park based on new energy as described in any one of claims 1 to 7.

Images