Day-ahead load prediction method and system based on MIC theory and monotone value decomposition
Through the load prediction method based on MIC theory and monotonic value decomposition, combined with CNN, GRU and attention mechanism, the problem of insufficient load prediction accuracy in cascade hydropower scheduling is solved, efficient load prediction is achieved, and water resources and electricity supply are improved guarantee.
Patent Information
- Application Number
- CN202410785095.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-18
- Publication Date
- 2024-11-01
AI Technical Summary
The existing load prediction technology has insufficient accuracy and timeliness in cascade hydropower scheduling, which affects the rational use of water resources and the guarantee of power supply.
Using the pre-load prediction method based on MIC theory and monotonic value decomposition, a load prediction model is constructed by obtaining multi-source data sets for filling and feature parameter analysis, and a network architecture of CNN, GRU and attention mechanism is used to predict, combining moving average and Entropy regularization optimizes the loss function.
It improves the accuracy of data completion, captures local trends and periodic characteristics, effectively discovers the correlation between variables, realizes efficient load prediction, and improves the accuracy and stability of prediction.
Smart Images

Figure CN118889367A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of cascade hydropower dispatching, and in particular to a method and system for day-ahead load forecasting based on MIC theory and monotonic value decomposition. Background Art
[0002] Cascade hydropower can effectively utilize water resources and reduce carbon emissions in the power system, which plays a vital role in supporting the construction of new power systems and the "dual carbon" goals. In addition, pumped storage power stations in cascade hydropower can further smooth the volatility of renewable energy output and reduce wind and solar power abandonment. In cascade hydropower, water is an important power generation resource, and the reasonable scheduling of water storage is an important factor affecting its operation. For safety reasons, the water level of the reservoir needs to be kept within a safe range, so it is necessary to absorb the excess water, that is, to use it for power generation. However, power generation in the upper reservoir will cause the water level of the downstream reservoir to rise. When the water level exceeds the safety limit, the water abandonment operation will cause energy waste. In this case, by conducting load forecasting, on the one hand, arranging reservoirs for power generation during peak load periods can maximize the reduction of system power generation costs and make up for the waste of resources caused by water abandonment. On the other hand, when renewable energy is abundant and the load has an increasing trend, arranging storage power stations for storage can also improve the utilization rate of new energy and reduce system operating costs.
[0003] In summary, load forecasting plays a vital role in cascade hydropower dispatching. It not only affects the rational use of water resources and the guarantee of power supply, but also directly relates to the safety of the hydropower system. Therefore, in the operation of cascade hydropower systems, strengthening the research and application of load forecasting technology and improving the accuracy and timeliness of forecasting are of great significance for optimizing the allocation of hydropower resources and improving the operating efficiency of the power system. Summary of the invention
[0004] In view of the problems existing in the existing load forecasting, the present invention is proposed.
[0005] Therefore, the problem to be solved by the present invention is to strengthen the research and application of load forecasting technology.
[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:
[0007] In a first aspect, an embodiment of the present invention provides a method for day-ahead load forecasting based on MIC theory and monotonic value decomposition, which includes:
[0008] Acquire a multi-source data set of a power grid in a target area, and fill in gaps in the multi-source data set;
[0009] Analyze the characteristic parameters of the data after filling in the gaps, and select the characteristic parameters to construct the data set;
[0010] Constructing a load forecasting model, and using the data set to train the load forecasting model;
[0011] A loss function of the load forecasting model is defined, and the load forecasting model is used to perform load forecasting to obtain a load forecasting result.
[0012] As a preferred solution of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition of the present invention, the multi-source data set is supplemented with the following:
[0013] Decompose multi-source data sets into three components: trend, cycle, and random;
[0014] Fill in the gaps in the three components;
[0015] The components after the gaps are reintegrated to form a sequence in the same format as the multi-source data.
[0016] As a preferred solution of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition of the present invention, wherein: the characteristic parameter analysis is selected according to the MIC value, and the characteristic parameter with a MIC value greater than n is used as the parameter for the final power load forecasting;
[0017] The calculation of the MIC value is shown in the following formula:
[0018]
[0019] Among them, p(h li ,h oi ) represents the joint probability of Hl and Ho; p(h li ),p(h oi ) represents the marginal probability of Hl, Ho; Ho∈{H t ,H wv ,H wh ,H r ,H h ,H p ,H s ,H u}; N is the number of historical data; B(N) represents the partition constraint value, which is generally taken as N 0.6 ; m, n represent H l , H o The number of portions evenly divided from small to large; I(H l ,H o |m,n) represents the mutual information value between Hl and Ho under the m, n partition.
[0020] As a preferred solution of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition of the present invention, wherein: the load forecasting model includes a first network model, a second network model and a third network model;
[0021] The objective function of the load forecasting model is as follows:
[0022]
[0023] Among them, l cnn is the load prediction value of the CNN network; l gru is the load prediction value of the GRU network; w cnn ,w gru are the weight prediction values of the hybrid network respectively; B represents the amount of data in a batch used for training.
[0024] As a preferred solution of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition of the present invention, the loss function of the load forecasting model is as follows:
[0025]
[0026] Among them, λ is the regularization coefficient, θ1+θ2=1,θ1≥0,θ2≥0, y1 is the load prediction value of the CNN network, y2 is the load prediction value of the GRU network, and θ1, θ2 are the weights of the hybrid network output of the CNN network and the GRU network.
[0027] As a preferred solution of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition of the present invention, the trend component is supplemented as shown in the following formula:
[0028]
[0029] Among them, X t is the observed value at time t, t is time, W is the window size used to calculate the moving average, β is the smoothing coefficient, T t Estimated value of the trend component at time t, T t ′ is the completed value of the trend component at time t, which is an estimate of the missing trend component obtained using window moving average or exponential smoothing.
[0030] As a preferred solution of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition of the present invention, the load forecasting result is obtained including:
[0031] The data set is input into the load forecasting model for preliminary feature extraction to generate two feature sequences, and the feature sequences are sent to the third network model in parallel. The fully connected layers 1 and 2 of the third network model receive the feature sequences respectively, and generate two attention weight sequences, weight 1 and weight 2, after Sigmoid activation;
[0032] The weight 1 and the weight 2 are weightedly fused with the feature sequences of the first network model and the second network model respectively to generate a final weighted feature sequence, and the weighted feature sequence is passed through the output layer to obtain a final prediction result.
[0033] In a second aspect, an embodiment of the present invention provides a day-ahead load forecasting system based on MIC theory and monotonic value decomposition, which includes:
[0034] A data acquisition module is used to obtain historical data sets of the power grid in the target area;
[0035] The data preprocessing module is used to decompose the data sequence into three components: trend, cycle, and randomness, and then perform gap filling processing on these three components respectively;
[0036] The feature analysis module uses the MIC method to analyze the feature parameters of the preprocessed data set, screen key feature parameters, and construct the data set;
[0037] The model building prediction module is used to build a load prediction model. During the prediction, the input data is input into the model, and the weighted sum is performed according to the weight of the network output result in the model and the output of the third network to obtain the final load prediction result.
[0038] In a third aspect, an embodiment of the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, wherein: when the processor executes the computer program, it implements any step of the above-mentioned day-ahead load forecasting method based on MIC theory and monotonic value decomposition.
[0039] In a fourth aspect, an embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, any step of the above-mentioned day-ahead load forecasting method based on MIC theory and monotonic value decomposition is implemented.
[0040] The beneficial effect of the present invention is to decompose the original time series into three components: trend, cycle and random. Different completion strategies are adopted for missing values, which improves the accuracy of data completion. The introduction of methods such as moving average and contemporaneous mean can better capture local trends and periodic characteristics, and the estimation is closer to the actual situation. The MIC theory can effectively discover the linear and nonlinear associations between variables, so as to select meteorological parameters with high correlation with load as prediction features. An integrated network architecture is proposed, which integrates CNN, GRU and attention mechanism. The advantages of the three models complement each other, fully explore the different patterns and characteristics of the data, and obtain the final prediction value by weighted summation of CNN, GRU and attention weights. The joint loss function directly optimizes the mean square error between the final prediction value and the true value, and introduces entropy regularization to avoid weight imbalance. The three model parameters are jointly optimized to better capture the complex characteristic patterns of the data. The overall process achieves efficient load forecasting through the synergy of the three models. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work. Among them:
[0042] Figure 1 Schematic diagram of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition.
[0043] Figure 2 Schematic diagram of the CNN network for the day-ahead load forecasting method based on MIC theory and monotonic value decomposition.
[0044] Figure 3 The diagram of the GRU network for the day-ahead load forecasting method based on MIC theory and monotonic value decomposition.
[0045] Figure 4 Schematic diagram of a hybrid network based on the MIC theory and monotonic value decomposition method for day-ahead load forecasting.
[0046] Figure 5 Schematic diagram of load forecasting model based on MIC theory and monotonic value decomposition.
[0047] Figure 6 The forecast results of the power load of a certain place for two days based on the day-ahead load forecasting method of MIC theory and monotonic value decomposition, where Real (red line) represents the actual value and Pred (blue line) represents the predicted value Figure 1 .
[0048] Figure 7The forecast results of the power load of a certain place for two days based on the day-ahead load forecasting method of MIC theory and monotonic value decomposition, where Real (red line) represents the actual value and Pred (blue line) represents the predicted value Figure 2 . DETAILED DESCRIPTION
[0049] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.
[0050] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.
[0051] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.
[0052] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.
[0053] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.
[0054] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.
[0055] Example 1
[0056] Reference Figure 1 to Figure 4 , which is the first embodiment of the present invention, provides a method for day-ahead load forecasting based on MIC theory and monotonic value decomposition, including:
[0057] In this load forecasting method, it is first necessary to obtain the historical data set H of the target area power grid. H contains multiple load-related data sequences, including: H = {Hl, Ht, Hwv, Hwh, Hr, Hh, Hp, Hs, Hu}, where Hl represents load, Ht represents temperature, Hwv represents wind speed, Hwh represents wind direction, Hr represents rainfall, Hh represents relative humidity, Hp represents ground pressure, Hs represents total sunshine intensity, and Hu represents ultraviolet intensity.
[0058] The above data series are all derived from the actual monitoring data of the power grid in the target area. The time accuracy can be daily, hourly or finer. They reflect the various environmental meteorological factors that affect the load and are an important data basis for building a load forecasting model.
[0059] After obtaining the original historical data set H, it is necessary to perform necessary preprocessing on the data to eliminate noise and outliers in the data and to reasonably estimate and fill in missing values.
[0060] Preferably, the original data sequence is decomposed into three components: trend, period, and random. Then, the three components are filled in respectively, and finally, they are recombined into a completed sequence.
[0061] First, for the periodic component P, if L consecutive points are missing, we cannot accurately estimate the current periodic component. By introducing L as the period length parameter, we can calculate the average value of the previous period as an estimate of the current period. In addition, using a fixed window and period length can better capture local trends and periodic characteristics and improve the accuracy of completion.
[0062] Furthermore, for load data, if some data are missing due to cold weather in the past week, the average load value of the days before and after with normal temperature can be used as the trend estimate of the missing days in this week, which is closer to reality. That is, the moving average is used instead of the original exponential smoothing method, as shown in the following formula:
[0063]
[0064] Among them, X t is the observed value at time t, t is time, W is the window size used to calculate the moving average, β is the smoothing coefficient, T t Estimated value of the trend component at time t, T t ′ is the completed value of the trend component at time t, which is an estimate of the missing trend component obtained using window moving average or exponential smoothing.
[0065] Furthermore, for the periodic component, when missing values occur, the mean of the previous complete cycle is used instead of the original mean of the same period to better capture the overall characteristics of the cycle. For temperature data, if some days in the month are missing, the average temperature of the previous month is used as the cycle estimate for the missing days of the month, which is more in line with the actual law of seasonal changes, as shown in the following formula:
[0066]
[0067] Among them, P t is the estimated value of the periodic component at time t, P t ′ is the completion value of the period component at time t, which is an estimate of the missing period component obtained using the mean of the period in the same period or the mean of the previous period. s is the current period number, which is a non-negative integer and represents the period of the time series in which the current period is located.
[0068] Furthermore, for random components, when missing values occur, it is impossible to make a more accurate estimate of them, so they are directly replaced with 0 values to prevent the introduction of more noise:
[0069]
[0070] Among them, R t is the random component at time t, R t ′ is the random component completion value at time t, which takes 0 for missing values and retains the original random component for non-missing values.
[0071] By reintegrating the above improved components, we can get the data sequence after the original time series is completed. By introducing auxiliary parameters, we can distinguish between missing and non-missing situations, and use different completion strategies for the three components, thereby improving the overall accuracy of the completion:
[0072]
[0073] Furthermore, the MIC method is used to analyze the characteristic parameters of the preprocessed data set H, screen the key characteristic parameters, and form the data set D. The characteristic parameter analysis and screening method is based on the maximum information coefficient theory. The characteristic parameters with MIC values greater than 0.3 are selected as the final parameters for load forecasting, and the data set D is formed. The calculation of MIC is expressed by the following formula:
[0074]
[0075] Among them, p(h li ,h oi ) represents the joint probability of Hl and Ho; p(h li ),p(h oi ) represents the marginal probability of Hl, Ho; Ho∈{H t ,H wv ,H wh ,H r ,H h ,H p ,H s ,H u}; N is the number of historical data; B(N) represents the partition constraint value, which is generally taken as N 0.6 ; m, n represent H l , H o The number of portions evenly divided from small to large; I(H l ,H o |m,n) represents the mutual information value between Hl and Ho under the m, n division (that is, the different h li or oi are considered the same value).
[0076] Furthermore, the characteristic parameters with MIC values greater than 0.3 were normalized to form a data set D, that is, D = {D l ,D t ,D r ,D h ,D s}, where D l represents the normalized load, D t represents the normalized temperature, D r represents the normalized rainfall, D h represents the normalized relative humidity, D s Represents the normalized total solar intensity.
[0077] Among them, the normalization method is expressed by the following formula:
[0078]
[0079] Among them, H i ∈H,H i,min Represents H i The minimum value of H i,max Represents H i The maximum value of .
[0080] Furthermore, a first network model, a second network model and a third network model are constructed.
[0081] For example, in a feasible embodiment, the first network model, the second network model and the third network model may respectively adopt a CNN network, a GRU network and a hybrid network model.
[0082] Then, the objective function of the complete load forecasting model structure is as follows:
[0083]
[0084] Among them, l cnn is the load prediction value of the CNN network; l gru is the load prediction value of the GRU network; w cnn ,w gru are the weight prediction values of the hybrid network respectively; B represents the amount of data in a batch used for training.
[0085] Furthermore, based on the value decomposition method, a joint loss function of the three networks is defined, where the final load prediction value is the product of the load prediction value of the CNN network and the weight output by the hybrid network, and the sum of the load prediction value of the GRU network and the weight output by the hybrid network. The joint loss function is equal to the mean square error between the final load prediction value and the true load value.
[0086] The final load forecast value can be defined as:
[0087]
[0088] Among them, θ1+θ2=1,θ1≥0,θ2≥0, y1 is the load prediction value of the CNN network, y2 is the load prediction value of the GRU network, and θ1, θ2 are the weights of the hybrid network output of the CNN network and the GRU network.
[0089] Then, the joint loss function can be defined as the mean square error between the final load forecast value and the true load value:
[0090]
[0091] In order to avoid extreme weights (such as close to 0 or 1), the entropy regularization term is introduced:
[0092] Ω(θ1,θ2)=-θ1 logθ1-θ2log(θ2)
[0093] Then, the final joint loss function is:
[0094]
[0095] Among them, λ is the regularization coefficient, which is used to balance the prediction error term and the entropy regularization term.
[0096] During the training process, we need to optimize the parameters of the CNN network, GRU network, and hybrid network at the same time to minimize the joint loss function. Specifically, we need to calculate the loss function through back propagation. and And update the corresponding network parameters.
[0097] Furthermore, the three networks are trained using the data set D and tested using the test data set. When load forecasting is performed, the input data is input into the three networks to obtain the load forecasting results.
[0098] The training or testing process is as follows:
[0099] For the input 96-frame D sequence audio data, preliminary feature extraction is first performed through the CNN model to generate two feature sequences: convolution layer 1 and convolution layer 2.
[0100] The two CNN feature sequences are fed into the hybrid network model in parallel for further processing. The hybrid network model consists of fully connected layers 1 and 2 and Sigmoid activation. Fully connected layers 1 and 2 receive features from convolutional layers 1 and 2 respectively, and generate two attention weight sequences after Sigmoid activation: weight 1 and weight 2. The GRU network model processes the original 96-frame D sequence audio data separately to generate the GRU layer and fully connected layer outputs. The two attention weight sequences generated by the hybrid network model will be weighted fused with the feature sequences of the CNN model and the GRU model respectively to generate the final weighted feature sequence.
[0101] The weighted feature sequence passes through the output layer to obtain the final autoregressive prediction result, which is compared with the true value to complete the training or testing process.
[0102] Furthermore, this embodiment also provides a day-ahead load forecasting system based on MIC theory and monotonic value decomposition, including:
[0103] A data acquisition module is used to obtain historical data sets of the power grid in the target area;
[0104] The data preprocessing module is used to decompose the data sequence into three components: trend, cycle, and randomness, and then perform gap filling processing on these three components respectively;
[0105] The feature analysis module uses the MIC method to analyze the feature parameters of the preprocessed data set, screen key feature parameters, and construct the data set;
[0106] The model building prediction module is used to build a load prediction model. During the prediction, the input data is input into the model, and the weighted sum is performed according to the weight of the network output result in the model and the output of the third network to obtain the final load prediction result.
[0107] This embodiment also provides a computer device, which is applicable to the day-ahead load forecasting method based on MIC theory and monotonic value decomposition, and includes a memory and a processor; the memory is used to store computer executable instructions, and the processor is used to execute computer executable instructions to implement the day-ahead load forecasting method based on MIC theory and monotonic value decomposition as proposed in the above embodiment.
[0108] The computer device may be a terminal, and the computer device includes a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner can be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a key, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse, etc.
[0109] This embodiment also provides a storage medium on which a computer program is stored. When the program is executed by a processor, the method for day-ahead load forecasting based on MIC theory and monotonic value decomposition as proposed in the above embodiment is implemented.
[0110] The storage medium proposed in this embodiment and the data storage method proposed in the above embodiment belong to the same inventive concept. The technical details not fully described in this embodiment can be found in the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.
[0111] Example 2
[0112] Reference Figure 5-6, which is the second embodiment of the present invention, provides a day-ahead load forecasting method based on MIC theory and monotonic value decomposition. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.
[0113] In order to further verify the effectiveness of the day-ahead load forecasting method based on MIC theory and monotonic value decomposition, a simulation experiment was conducted. Specifically, the power load data of a certain power grid from 2019 to 2022 and other local data were used for load forecasting training and forecasting.
[0114] This experiment is based on Python and Tensorflow 2.0 framework, and the computer used is configured with Intel Corei5-9300H CPU@2.40GHz and 1 NVIDIA GTX 1660Ti GPU.
[0115] Figure 6 and Figure 7 The prediction results of the load forecasting model designed by the present invention for the day-ahead load are shown respectively, and it can be seen that the prediction error is relatively small and has sufficient accuracy.
[0116] Figure 6 The figure shows the prediction effect of the load forecasting method on the power load on a certain day. Overall, the trend of the predicted value curve (blue line) and the actual value curve (red line) are highly consistent, which can well capture the periodic change law of the load. Specifically, at around 0-30 time steps and 60 time steps, the prediction curve accurately captures the two peak periods of the load, and the height of the peak is very close to the actual value. In the valley stage between the peaks, such as 30-60 time steps, the prediction curve can also simulate the downward trend of the actual load more closely. It is worth mentioning that even at individual time points such as around 20 time steps, there are some deviations between the predicted value and the actual value, but the overall deviation value is very small and does not affect the capture of the overall law. Therefore, the method has a very high accuracy in the day-ahead prediction of the load on this day.
[0117] Figure 7 It shows the load forecast for another day. From a macro perspective, the trend of the forecast curve is basically consistent with the actual curve, and it can also well reflect the cyclical change law of the load. Figure 6 In comparison, there are some deviations between the predicted curve and the actual curve for this day. For example, in the 20-40 time step interval, although the predicted curve captures the overall upward trend of the actual curve, the peak value is still lower than the actual value. A similar situation also occurs at the peak stage around 60 time steps. However, it should be pointed out that despite the above deviations, the trend profile of the predicted curve is still highly consistent with the actual curve, and there is no serious fitting error.
[0118] The analysis of the two figures shows that the load forecasting method can accurately capture the trend of actual load changes in most cases, and the prediction of peak and valley values is also relatively accurate. Even if there are certain deviations in individual periods, the overall fitting effect is still good, which can meet the requirements of day-ahead load forecasting and provide valuable decision-making basis for cascade hydropower scheduling. At the same time, the two figures also reflect the stability and applicability of this method under different weather and load modes.
[0119] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.
Claims
1. A day-ahead load forecasting method based on MIC theory and monotonic value decomposition, characterized by: include, Acquire a multi-source data set of a power grid in a target area, and fill in gaps in the multi-source data set; Analyze the characteristic parameters of the data after filling in the gaps, and select the characteristic parameters to construct the data set; Constructing a load forecasting model, and using the data set to train the load forecasting model; A loss function of the load forecasting model is defined, and the load forecasting model is used to perform load forecasting to obtain a load forecasting result.
2. The method for day-ahead load forecasting based on MIC theory and monotonic value decomposition according to claim 1, characterized in that: Filling the gaps in the multi-source data set includes: Decompose multi-source data sets into three components: trend, cycle, and random; Fill in the gaps in the three components; The components after the gaps are reintegrated to form a sequence in the same format as the multi-source data.
3. The method for day-ahead load forecasting based on MIC theory and monotonic value decomposition as claimed in claim 2, characterized in that: The characteristic parameter analysis is performed based on the MIC value, and the characteristic parameter with a MIC value greater than n is used as a parameter for the final power load prediction; The calculation of the MIC value is shown in the following formula: Among them, p(h li ,h oi ) represents the joint probability of Hl and Ho; p(h li ),p(h oi ) represents the marginal probability of Hl, Ho; Ho∈{H t ,H wv ,H wh ,H r ,H h ,H p ,H s ,H u }; N is the number of historical data; B(N) represents the partition constraint value, which is generally taken as N 0.6 ; m, n represent H l , H o The number of portions evenly divided from small to large; I(H l ,H o |m,n) represents the mutual information value between Hl and Ho under the m, n partition.
4. The method for day-ahead load forecasting based on MIC theory and monotonic value decomposition as claimed in claim 3, characterized in that: The load forecasting model includes a first network model, a second network model and a third network model; The objective function of the load forecasting model is as follows: Among them, l cnn is the load prediction value of the CNN network; l gru is the load prediction value of the GRU network; w cnn ,w gru are the weight prediction values of the hybrid network respectively; B represents the amount of data in a batch used for training.
5. The method for day-ahead load forecasting based on MIC theory and monotonic value decomposition as claimed in claim 4, characterized in that: The loss function of the load forecasting model is shown as follows: Among them, λ is the regularization coefficient, θ1+θ2=1,θ1≥0,θ2≥0, y1 is the load prediction value of the CNN network, y2 is the load prediction value of the GRU network, and θ1, θ2 are the weights of the hybrid network output of the CNN network and the GRU network.
6. The method for day-ahead load forecasting based on MIC theory and monotonic value decomposition according to claim 5, characterized in that: The trend component is filled as follows: Among them, X t is the observed value at time t, t is time, W is the window size used to calculate the moving average, β is the smoothing coefficient, T t Estimated value of the trend component at time t, T t ′ is the completed value of the trend component at time t, which is an estimate of the missing trend component obtained using window moving average or exponential smoothing.
7. The method for day-ahead load forecasting based on MIC theory and monotonic value decomposition according to claim 6, characterized in that: Obtaining the load forecast result includes: The data set is input into the load forecasting model for preliminary feature extraction to generate two feature sequences, and the feature sequences are sent to the third network model in parallel. The fully connected layers 1 and 2 of the third network model receive the feature sequences respectively, and generate two attention weight sequences, weight 1 and weight 2, after Sigmoid activation; The weight 1 and the weight 2 are weightedly fused with the feature sequences of the first network model and the second network model respectively to generate a final weighted feature sequence, and the weighted feature sequence is passed through the output layer to obtain a final prediction result.
8. A day-ahead load forecasting system based on MIC theory and monotonic value decomposition, based on the day-ahead load forecasting method based on MIC theory and monotonic value decomposition according to any one of claims 1 to 7, characterized in that: include, A data acquisition module is used to obtain historical data sets of the power grid in the target area; The data preprocessing module is used to decompose the data sequence into three components: trend, cycle, and randomness, and then perform gap filling processing on these three components respectively; The feature analysis module uses the MIC method to analyze the feature parameters of the preprocessed data set, screen key feature parameters, and construct the data set; The model building prediction module is used to build a load prediction model. During the prediction, the input data is input into the model, and the weighted sum is performed according to the weight of the network output result in the model and the output of the third network to obtain the final load prediction result.
9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method for day-ahead load forecasting based on MIC theory and monotonic value decomposition according to any one of claims 1 to 7 are implemented.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method for day-ahead load forecasting based on MIC theory and monotonic value decomposition according to any one of claims 1 to 7 are implemented.
