Large user load prediction method and system based on big data mining

Abstract

The invention discloses a big user load prediction method and system based on big data mining, and the method comprises the steps: collecting the historical record of user load data, and obtaining the power purchasing and selling data of a user; generating a dependency relationship between the user load data and the electricity purchasing and selling data by analyzing a fluctuation condition of the electricity purchasing and selling data; and according to the dependency relationship, generating a prediction result of the large user load for the historical record. According to the method, the precision and the real-time performance of load prediction are remarkably improved, the risk and the cost of power dispatching are reduced, powerful support is provided for decision making of power companies in the power market, and the method has wide application prospects.

Description

Technical Field

[0001] The present invention relates to the technical field of big data mining, and in particular to a large user load prediction method and system based on big data mining. Background Art

[0002] Power load forecasting is an important foundation for power system operation and power market operation. Accurate load forecasting helps power companies formulate reasonable power generation plans, optimize power dispatching, improve the safety and economy of power systems, and reduce the waste of power resources. With the complexity of modern power systems and the opening of power markets, the accuracy requirements of load forecasting are getting higher and higher. In particular, for the load forecasting of large users, due to their huge power consumption and significant impact on the power grid, high-precision and flexible forecasting methods are needed.

[0003] Traditional power load forecasting methods mainly rely on models such as time series analysis, linear regression, and support vector machines. These methods usually make predictions based on historical load data and assume that there is a certain linear or nonlinear relationship between future load and historical data. However, with the diversification of power user behavior patterns and the complexity of the market environment, traditional methods have shown obvious deficiencies in forecasting accuracy. Summary of the invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by the present invention is that the existing large user load prediction has the problem of low data quality.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions: a large user load forecasting method based on big data mining, comprising:

[0007] Collect historical records of user load data and obtain users' electricity purchase and sales data;

[0008] By analyzing the fluctuation of the power purchase and sale data, a dependency relationship between the user load data and the power purchase and sale data is generated;

[0009] A prediction result of large user load is generated for the historical records according to the dependency relationship.

[0010] As a preferred solution of the large user load prediction method based on big data mining described in the present invention, wherein: the user load data includes the user's power consumption data in different time periods, peak and valley power consumption data, and power consumption time distribution data;

[0011] The electricity purchase and sale data includes price data of users in historical electricity consumption periods.

[0012] As a preferred solution of the large user load prediction method based on big data mining described in the present invention, wherein: the fluctuation of the power purchase and sales data includes: in similar enterprises, users are divided into three levels: high, medium and low according to their annual power consumption; the fluctuation of the power purchase and sales data of users of the three levels is analyzed respectively;

[0013] Among them, the annual power consumption of high-level The annual power consumption of the medium level is The annual electricity consumption of the low level is (0, ω]; and ω represent the preset judgment thresholds,

[0014] For users of the same level, obtain the user load data before and after each electricity price change in the historical records, the purchase and sale electricity prices before and after the change, and the time of each electricity price change;

[0015] Under the same seasonal factors, the electricity consumption data of users in different time periods and peak and valley electricity consumption data are used as characteristic factors respectively; the inherent fluctuation is obtained by analyzing the volatility of characteristic factors under the same electricity price;

[0016] Under the same seasonal factor, the influence of the electricity price change on two characteristic factors is analyzed by a neural network to obtain characteristic fluctuations; the inherent fluctuations and the characteristic fluctuations are combined to generate a dependency relationship between the user load data and the electricity purchase data.

[0017] As a preferred solution of the large user load forecasting method based on big data mining described in the present invention, the same seasonal factor includes obtaining the seasonal change strategy of the power plant for the electricity selling price, and dividing the seasonal factors at each change time node according to the seasonal change strategy;

[0018] By dividing each year in the historical records, we get the division results for each year; in the seasonal change strategy of each year, the time nodes of each change are matched, and the historical data of each division result in the same time interval of the year are integrated to get the data set under the same seasonal factor.

[0019] As a preferred solution of the large user load forecasting method based on big data mining described in the present invention, wherein: the inherent fluctuation includes, in each data set, according to the change of electricity price, setting N price levels, aggregating the contents of the same electricity price, and obtaining the maximum and minimum values ​​of electricity consumption data and peak and valley electricity consumption data respectively;

[0020] Calculate using the maximum and minimum values ​​of power consumption data and peak and valley power consumption data:

[0021]

[0022] Among them, W 用 Indicates the inherent fluctuation upper limit ratio of electricity consumption data, W 谷 Indicates the inherent fluctuation upper limit ratio of peak and valley power consumption data, p max Indicates the maximum value of power consumption data, p min Indicates the minimum value of power consumption data, q max Indicates the maximum value of peak and valley power consumption data, q min Indicates the minimum value of peak and valley power consumption data;

[0023] The fixed fluctuation of each price level is obtained, where the fluctuation ratio range of electricity consumption data is [0, W n用 ], the fluctuation ratio range of peak and valley power consumption data is [0, W n谷 ]; W n用 represents the inherent fluctuation upper limit ratio of electricity consumption data in the nth price level, W n谷 It represents the inherent upper limit ratio of fluctuation of peak and valley electricity consumption data in the nth price level.

[0024] As a preferred solution of the large user load prediction method based on big data mining described in the present invention, wherein: the neural network includes training the data sets under different seasonal factors respectively to obtain the trained neural network, and using the neural network analysis for the seasonal factor at the current time t;

[0025] The input layer receives the power consumption U t , peak and valley power consumption data V t , and the electricity price P t As input;

[0026] X t0 =[U t ,V t ,P t ]

[0027] After considering the inherent fluctuation, the power consumption U t And peak and valley power consumption data V t Add random perturbations to update the input;

[0028] X t =[U t (1+∈),V t (1+δ),P t ]

[0029] Among them, X t0 represents the original input, X t represents the updated input; ∈ represents the disturbance factor of power consumption, ranging from [-W n用 , W n用]; δ represents the disturbance factor of peak and valley power consumption, ranging from [-W n谷 , W n谷 ]; at this time, n represents the price level at the current moment;

[0030] Extract time series features:

[0031] 1D-CNN: used to extract short-term feature patterns;

[0032] C t =ReLU(Conv1D(X t ,filters=32,kernel size =3))

[0033] LSTM: used to capture long-term dependencies;

[0034] H t =LSTM(X t ,units=50)

[0035] Among them, ReLU is the activation function, Conv1D is the one-dimensional convolution operation, filters is the number of convolution kernels, kernel size is the size of the convolution kernel, units represents the number of LSTM units, C t represents the time series features extracted by the one-dimensional convolutional neural network, H t Represents the time series features extracted by the long short-term memory network;

[0036] Fusion layer: After extracting the time series features, the features are fused with the original input features to ensure that the model can take into account both short-term and long-term feature effects:

[0037] F t =Concatenate([C t ,H t ])

[0038] Among them, F t Represents the fused feature vector, Concatenate is a concatenation operation;

[0039] Multi-task learning layer: Using the multi-task learning mechanism, by sharing some hidden layers, it is possible to simultaneously learn the characteristics of power consumption and peak and valley power consumption:

[0040] H shared =ReLU(W1·F t +b1)

[0041] Then set up separate hidden layers for power consumption and peak and valley power consumption:

[0042] HU =ReLU(W U ·H shared +b U )

[0043] H V =ReLU(W V ·H shared +b V )

[0044] Output layer: Output results for power consumption and peak-valley power consumption prediction:

[0045]

[0046] Among them, H shared represents the shared hidden layer features, W1 is the weight matrix of the fully connected layer, b1 is the bias vector, and H U represents the hidden layer features used to predict future electricity consumption, H V represents the hidden layer features used to predict future peak and valley electricity consumption, W U is the weight matrix of the fully connected layer, b U is the bias vector, W V is the weight matrix of the fully connected layer, b V is the bias vector, Represents the predicted electricity consumption data at time t+T, Represents the predicted peak and valley electricity consumption data at time t+T; Linear is a linear transformation, representing a fully connected layer;

[0047] The loss function of the neural network model is expressed as:

[0048]

[0049] Among them, θ represents the set of all parameters of the model, F represents the number of training samples, and U t+T and V t+T They are the actual power consumption and peak-valley power consumption at time t+T, and are the electricity consumption predicted by the model and the peak-valley electricity consumption, respectively. λ is the weight factor used in the loss function to balance the electricity consumption prediction error and the peak-valley electricity consumption prediction error.

[0050] As a preferred solution of the large user load prediction method based on big data mining described in the present invention, wherein: the prediction result of the large user load includes, assuming that the price level changes with the increase of electricity consumption, when the price level and seasonal factors do not change at the prediction time t+T, the result is directly output;

[0051] When the price level changes at the forecast time t+T, the time step is increased from the current time t until the price level just changes, and the forecast result Q1 for the current electricity consumption and peak-valley electricity consumption is output. Q1 is used as input, and the random disturbance ∈ and δ caused by the fixed fluctuation in the next price level are used to update the input. The neural network trained with the current seasonal factors is used to re-forecast, and the updating process of the disturbance factor is repeated until the time step reaches T;

[0052] When the seasonal factor changes at the prediction time t+T, the time step is increased from the current time t until the seasonal factor just changes, and the prediction result QS1 for the electricity consumption and peak-valley electricity consumption at this time is output; QS1 is used as input, and the neural network trained with the next seasonal factor and the disturbance factors ∈, δ in the price range corresponding to QS1 are used to re-predict, and the updating process of the neural network and the disturbance factor is repeated until the time step reaches T;

[0053] By analyzing the changes in price levels and seasonal factors, the neural network and interference factors are continuously updated. When the time step reaches T, the prediction result Qd at time t+T is output;

[0054] Where Qd represents the output of the d-th neural network, d ≥ 1.

[0055] A large user load forecasting system based on big data mining using the method described in the present invention is characterized by:

[0056] The collection unit collects the historical records of user load data and obtains the user's electricity purchase and sales data;

[0057] An analysis unit, generating a dependency relationship between the user load data and the power purchase and sale data by analyzing fluctuations in the power purchase and sale data;

[0058] The prediction unit generates a prediction result of the large user load for the historical records according to the dependency relationship.

[0059] A computer device comprises: a memory and a processor; the memory stores a computer program, wherein the processor implements the steps of any one of the methods of the present invention when executing the computer program.

[0060] A computer-readable storage medium stores a computer program, wherein the computer program, when executed by a processor, implements the steps of any one of the methods of the present invention.

[0061] Beneficial effects of the present invention: The large-user load forecasting method based on big data mining provided by the present invention realizes accurate forecasting of large-user loads by comprehensively utilizing user load data and electricity purchase and sales data, especially by analyzing the impact of electricity price fluctuations on user loads. Compared with traditional methods, the present invention adopts neural network technology, especially combines the advantages of 1D-CNN and LSTM, which can effectively capture the complex nonlinear relationship between load and electricity price, and dynamically adapt to market changes in different time periods. This method significantly improves the accuracy and real-time performance of load forecasting, reduces the risk and cost of power dispatching, provides strong support for power companies' decision-making in the power market, and has broad application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0063] Figure 1 An overall flow chart of a large user load forecasting method based on big data mining provided for the first embodiment of the present invention. DETAILED DESCRIPTION

[0064] 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.

[0065] Example 1, reference Figure 1 , which is an embodiment of the present invention, provides a large user load forecasting method based on big data mining, including:

[0066] S1: Collect historical records of user load data and obtain user electricity purchase and sales data.

[0067] The user load data includes the user's electricity consumption data in different time periods, peak and valley electricity consumption data, and electricity consumption time distribution data; the electricity purchase and sales data includes the user's price data in historical electricity consumption periods.

[0068] By collecting historical records of user load data, we can cover the user's electricity consumption in different time periods, peak and valley electricity consumption data, and electricity consumption time distribution data. These data can fully reflect the user's electricity consumption behavior pattern, including their electricity demand in different time periods, and electricity consumption characteristics during peak and valley periods. This information is of great significance for understanding the user's load characteristics and identifying electricity consumption patterns.

[0069] At the same time, obtaining the price data of users during historical electricity consumption periods helps analyze the impact of electricity price fluctuations on users' electricity consumption behavior. Electricity prices are an important factor affecting users' electricity consumption decisions, especially in the electricity market, where users often adjust their electricity consumption plans based on electricity price fluctuations. By correlating electricity purchase and sales price data with user load data, we can deeply understand the impact mechanism of price changes on load demand.

[0070] S2: Generate a dependency relationship between the user load data and the power purchase and sale data by analyzing fluctuations in the power purchase and sale data.

[0071] The fluctuation of the electricity purchase and sales data includes: in the same type of enterprises, users are divided into three levels: high, medium and low according to their annual electricity consumption; the fluctuation of the electricity purchase and sales data of users of the three levels is analyzed respectively. The annual power consumption of the medium level is The annual electricity consumption of the low level is (0, ω]; and ω represent the preset judgment thresholds,

[0072] For users of the same level, obtain the user load data before and after each electricity price change in the historical records, the purchase and sale electricity prices before and after the change, and the time of each electricity price change. Under the same seasonal factors, the user's electricity consumption data in different time periods and peak and valley electricity consumption data are used as characteristic factors; by analyzing the volatility of characteristic factors under the same electricity price, the inherent fluctuation is obtained.

[0073] Under the same seasonal factor, the influence of the electricity price change on two characteristic factors is analyzed by a neural network to obtain characteristic fluctuations; the inherent fluctuations and the characteristic fluctuations are combined to generate a dependency relationship between the user load data and the electricity purchase data.

[0074] The same seasonal factor includes obtaining the seasonal change strategy of the power plant for the electricity price, and dividing the seasonal factors at the time node of each change according to the seasonal change strategy. By dividing each year in the historical records, the division results for each year are obtained; the time nodes of each change in the seasonal change strategy of each year are matched, and the historical data of each division result in the same time interval of the year are integrated to obtain the data set under the same seasonal factor.

[0075] First of all, it should be known that the analysis of the same seasonal change strategy is meaningful, so the present invention only targets fixed strategies. According to this strategy, the seasonal changes of a year can be divided into multiple time periods; set as A1A2A3A4A5. Then there will be another round of such seasonal changes in the next year. The law of this change is the same as the previous one, so the time period of change is corresponding, set as A6A7A8A9A10, and then the data content of each same time interval is put into a data set (for the assumption of this case, 5 sets will be obtained: the set of A1 and A6, the set of A2 and A7, the set of A3 and A8, the set of A4 and A9, and the set of A5 and A10). This can make the data in this data set more able to reflect the user's behavior (without the interference of seasonal policies, and in general, seasonal policies are closely related to the user's electricity usage habits, so through the seasonal division, the user's electricity usage behavior can be effectively distinguished). By analyzing the data in each set, more accurate predictions can be made.

[0076] In fact, the present invention provides a prediction method under the guidance of a fixed and unchanging electricity price.

[0077] The inherent fluctuation includes that in each data set, according to the change of electricity price, N price levels are set, the contents of the same electricity price are summarized, and the maximum and minimum values ​​of electricity consumption data and peak and valley electricity consumption data are obtained respectively. The maximum and minimum values ​​of electricity consumption data and peak and valley electricity consumption data are used for calculation:

[0078]

[0079] Among them, W 用 Indicates the inherent fluctuation upper limit ratio of electricity consumption data, W 谷 Indicates the inherent fluctuation upper limit ratio of peak and valley power consumption data, p max Indicates the maximum value of power consumption data, p min Indicates the minimum value of power consumption data, q max Indicates the maximum value of peak and valley power consumption data, q min Indicates the minimum value of peak and valley power consumption data.

[0080] The fixed fluctuation of each price level is obtained, where the fluctuation ratio range of electricity consumption data is [0, W n用 ], the fluctuation ratio range of peak and valley power consumption data is [0, W n谷 ]; W n用 represents the inherent fluctuation upper limit ratio of electricity consumption data in the nth price level, W n谷 It represents the inherent upper limit ratio of fluctuation of peak and valley electricity consumption data in the nth price level.

[0081] The neural network includes training data sets under different seasonal factors respectively to obtain a trained neural network, and using the neural network to analyze the seasonal factors at the current time t.

[0082] The input layer receives the power consumption U t , peak and valley power consumption data V t , and the electricity price P t as input.

[0083] X t0 =[U t ,V t ,P t ]

[0084] After considering the inherent fluctuation, the power consumption U t And peak and valley power consumption data V t Add random perturbations to update the input.

[0085] X t =[U t (1+∈),V t (1+δ),P t ]

[0086] Among them, X t0 represents the original input, X t represents the updated input; ∈ represents the disturbance factor of power consumption, ranging from [-W n用 , W n用 ]; δ represents the disturbance factor of peak and valley power consumption, ranging from [-W n谷 , W n谷 ]; here n represents the price level at the current moment.

[0087] Extract time series features:

[0088] 1D-CNN: used to extract short-term feature patterns.

[0089] C t =ReLU(Conv1D(X t ,filters=32,kernel size =3))

[0090] LSTM: used to capture long-term dependencies.

[0091] H t =LSTM(X t ,units=50)

[0092] Among them, ReLU is the activation function, Conv1D is the one-dimensional convolution operation, filters is the number of convolution kernels, kernel size is the size of the convolution kernel, units represents the number of LSTM units, C t represents the time series features extracted by the one-dimensional convolutional neural network, H t Represents the time series features extracted by the long short-term memory network.

[0093] Fusion layer: After extracting the time series features, the features are fused with the original input features to ensure that the model can take into account both short-term and long-term feature effects:

[0094] F t =Concatenate([C t ,H t ])

[0095] Among them, F t Represents the fused feature vector, and Concatenate is a concatenation operation.

[0096] Multi-task learning layer: Using the multi-task learning mechanism, by sharing some hidden layers, it is possible to simultaneously learn the characteristics of power consumption and peak and valley power consumption:

[0097] H shared =ReLU(W1·F t +b1)

[0098] Then set up separate hidden layers for power consumption and peak and valley power consumption:

[0099] H U =ReLU(W U ·H shared +b U )

[0100] H V =ReLU(W V ·H shared +b V )

[0101] Output layer: Output results for power consumption and peak-valley power consumption prediction:

[0102]

[0103] Among them, H shared represents the shared hidden layer features, W1 is the weight matrix of the fully connected layer, b1 is the bias vector, and H U represents the hidden layer features used to predict future electricity consumption, H V represents the hidden layer features used to predict future peak and valley electricity consumption, WU is the weight matrix of the fully connected layer, b U is the bias vector, W V is the weight matrix of the fully connected layer, b V is the bias vector, Represents the predicted electricity consumption data at time t+T, Represents the predicted peak and valley electricity consumption data at time t+T; Linear is a linear transformation, representing a fully connected layer.

[0104] The loss function of the neural network model is expressed as:

[0105]

[0106] Among them, θ represents the set of all parameters of the model, F represents the number of training samples, and U t+T and V t+T They are the actual power consumption and peak-valley power consumption at time t+T, and are the electricity consumption predicted by the model and the peak-valley electricity consumption, respectively. λ is the weight factor used in the loss function to balance the electricity consumption prediction error and the peak-valley electricity consumption prediction error.

[0107] S3: Generate a prediction result of the large user load for the historical records according to the dependency relationship

[0108] The prediction result of the large user load includes: assuming that the price level changes with the increase of electricity consumption, when the price level and seasonal factors do not change at the prediction time t+T, the result is directly output.

[0109] When the price level changes at the prediction time t+T, the time step is increased starting from the current time t until the price level just changes, and the prediction result Q1 for the electricity consumption and peak-valley electricity consumption at this time is output. Q1 is used as input, and the random disturbance ∈ and δ caused by the fixed fluctuation in the next price level are used to update the input. The neural network trained with the current seasonal factors is used to re-predict, and the updating process of the disturbance factor is repeated until the time step reaches T.

[0110] When the seasonal factors change at the prediction time t+T, the time step is increased starting from the current time t until the seasonal factors just change, and the prediction result QS1 for the electricity consumption and peak-valley electricity consumption at this time is output; QS1 is used as input, and the neural network trained with the next seasonal factor and the disturbance factors ∈, δ in the price range corresponding to QS1 are used to re-predict, and the updating process of the neural network and the disturbance factors is repeated until the time step reaches T.

[0111] By analyzing the changes in price levels and seasonal factors, the neural network and interference factors are continuously updated. When the time step reaches T, the prediction result Qd at time t+T is output. Qd represents the output of the dth neural network, d≥1.

[0112] Changes in price levels often have a direct impact on users' electricity consumption behavior. When price levels change, users may adjust their electricity consumption plans (such as reducing electricity consumption during peak hours or increasing electricity consumption during off-peak hours). By capturing the moments when price levels change and reflecting these changes in the forecasting model, users' loads can be more accurately forecasted. When the price level changes, by increasing the time step until the change occurs, the starting point of the load change can be accurately located, and this information can then be used for subsequent forecasts. This approach ensures that the forecasting model can adapt to changes in price levels in a timely manner and avoid forecast bias caused by price fluctuations.

[0113] After the price level or seasonal factors change, new input data (including data after random perturbation) is used to re-forecast to ensure that the forecast model can be dynamically adjusted to reflect the new market environment and user behavior. By updating and perturbing the input data multiple times, the adaptability of the forecast model to the new environment is enhanced, making the forecast results more robust. At the same time, through iterative updates, the model can gradually approach the real user behavior and reduce the forecast error.

[0114] Seasonal factors (such as temperature, weather, season, etc.) have a significant impact on electricity demand. Users' electricity habits and needs change in different seasons. By updating the model and re-forecasting when seasonal factors change, the impact of seasonal changes on load can be more accurately reflected. When seasonal factors change, the forecast model is adjusted in a timely manner to reflect the new seasonal conditions, thereby improving the accuracy of the forecast. By combining changes in price levels and seasonal factors, the forecast model can more comprehensively reflect the complex situation in reality.

[0115] By combining the combined effects of price levels, seasonal factors, and random disturbances, we ensure that the forecast results can fully reflect all possible market changes. This comprehensive consideration makes the forecast model more universal, able to maintain high forecast accuracy under different market conditions, and provide strong support for power companies to formulate dispatch plans and electricity price strategies.

[0116] On the other hand, this embodiment also provides a large user load forecasting system based on big data mining, which includes:

[0117] The collection unit collects the historical records of user load data and obtains the user's electricity purchase and sales data.

[0118] The analysis unit generates a dependency relationship between the user load data and the power purchase and sale data by analyzing fluctuations in the power purchase and sale data.

[0119] The prediction unit generates a prediction result of the large user load for the historical records according to the dependency relationship.

[0120] If the above functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0121] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0122] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0123] It should be understood that the various parts of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, a plurality of steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0124] Example 2 is an embodiment of the present invention, which provides a large user load forecasting method based on big data mining. In order to verify the beneficial effects of the present invention, scientific demonstration is carried out through economic benefit calculation and simulation experiments.

[0125] The experiment selected six typical high-energy-consuming enterprises in an industrial park as test subjects. The electricity demand of these enterprises fluctuates with changes in market electricity prices and seasonal factors. The main purpose of the experiment is to verify the innovation and advantages of the new method by comparing the prediction accuracy, adaptability and real-time performance of the load forecasting method based on big data mining with the traditional load forecasting method in different scenarios.

[0126] The historical load data of the six enterprises over the past six years were collected, including hourly electricity consumption data, peak and valley electricity consumption data, and electricity consumption time distribution data. In addition, the purchase and sale price data in the same time period were also obtained. In order to more comprehensively reflect the market fluctuations, the experiment paid special attention to the data at key moments such as seasonal changes and drastic fluctuations in market electricity prices. After cleaning and preprocessing, these data were used as input data sets.

[0127] In the load forecasting process, for each forecast moment, it is determined whether the current price level and seasonal factors have changed. If there is a change, the model increases the time step to capture the specific moment when the change occurs, and re-forecasts based on the input data at that moment. If there is no change, the forecast result is directly output. Through repeated iterations, the actual electricity demand is gradually approached, and finally the forecast results of electricity consumption and peak and valley electricity consumption for the next month are output. The recorded data are shown in Table 1.

[0128] Table 1 Experimental data table

[0129]

[0130] Through the analysis of the experimental data table, it can be clearly seen that the load forecasting method based on big data mining has significant advantages in accuracy and adaptability. Traditional time series analysis methods (such as ARIMA) rely on the stationarity and linear assumptions of historical data when predicting electricity consumption. Although they can reflect the electricity consumption trend of enterprises to a certain extent, their prediction accuracy is significantly reduced when faced with nonlinear and non-stationary characteristics such as price changes or seasonal factors. The data in the table show that the prediction errors of traditional methods are generally large, especially in enterprises such as Enterprise E and Enterprise C, where the electricity consumption fluctuates greatly, the errors reach 6.5% and 5.3% respectively.

[0131] In contrast, the method based on big data mining can better capture the complex nonlinear relationship between electricity consumption and electricity prices by combining neural network technology, especially the combination of 1D-CNN and LSTM. During the forecasting process, this method dynamically adapts to changes in price levels and seasonal factors, adds random disturbance factors to enhance the robustness of the model, and makes the forecast results closer to the actual situation. As can be seen from the table data, the forecast error of the big data method in all enterprises is lower than that of the traditional method, and the error range is controlled between 2.5% and 6.5%, which is significantly better than the traditional method, especially in enterprises with large fluctuations.

[0132] In addition, when dealing with changes in price levels and seasonal factors, the method based on big data mining can effectively capture the impact of these changes on electricity consumption by increasing the time step and repeatedly iterating and updating the input data, thereby ensuring the real-time and accuracy of the forecast results. This dynamic adjustment and step-by-step approach to forecasting enables the model to adapt to complex changes in the power market and significantly improves the accuracy and applicability of load forecasting.

[0133] In practical applications, this load forecasting method based on big data mining not only improves the accuracy of power companies in load forecasting, but also reduces the waste of power resources caused by forecasting deviations, providing reliable technical support for power dispatching and market operations. Compared with the prior art, the present invention breaks through the limitations of traditional methods through innovative technical means and complex model design, significantly improves the overall effect of load forecasting, and has broad application prospects and practical value.

[0134] 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 large user load forecasting method based on big data mining, characterized in that: include: Collect historical records of user load data and obtain users' electricity purchase and sales data; By analyzing the fluctuation of the power purchase and sale data, a dependency relationship between the user load data and the power purchase and sale data is generated; A prediction result of large user load is generated for the historical records according to the dependency relationship.

2. The large user load prediction method based on big data mining according to claim 1 is characterized by: The user load data includes the user's power consumption data in different time periods, peak and valley power consumption data, and power consumption time distribution data; The electricity purchase and sale data includes price data of users in historical electricity consumption periods.

3. The large user load prediction method based on big data mining according to claim 2 is characterized in that: The fluctuation of the electricity purchase and sales data includes: among similar enterprises, users are divided into three levels: high, medium and low according to their annual electricity consumption; and the fluctuation of the electricity purchase and sales data of users of the three levels are analyzed respectively; Among them, the annual power consumption of high-level The annual power consumption of the medium level is The annual power consumption of low-level and ω represent the preset judgment thresholds, For users of the same level, obtain the user load data before and after each electricity price change in the historical records, the purchase and sale electricity prices before and after the change, and the time of each electricity price change; Under the same seasonal factors, the electricity consumption data of users in different time periods and peak and valley electricity consumption data are used as characteristic factors respectively; the inherent fluctuation is obtained by analyzing the volatility of characteristic factors under the same electricity price; Under the same seasonal factor, the influence of the electricity price change on two characteristic factors is analyzed by a neural network to obtain characteristic fluctuations; the inherent fluctuations and the characteristic fluctuations are combined to generate a dependency relationship between the user load data and the electricity purchase data.

4. The large user load prediction method based on big data mining according to claim 3 is characterized in that: The same seasonal factor includes obtaining the seasonal change strategy of the power plant for the electricity selling price, and dividing the seasonal factors at the time node of each change according to the seasonal change strategy; By dividing each year in the historical records, we get the division results for each year; in the seasonal change strategy of each year, the time nodes of each change are matched, and the historical data of each division result in the same time interval of the year are integrated to get the data set under the same seasonal factor.

5. The large user load prediction method based on big data mining according to claim 4 is characterized in that: The inherent fluctuation includes that in each data set, according to the change of electricity price, N price levels are set, the contents of the same electricity price are aggregated, and the maximum and minimum values ​​of electricity consumption data and peak and valley electricity consumption data are obtained respectively; Calculate using the maximum and minimum values ​​of power consumption data and peak and valley power consumption data: Among them, W 用 Indicates the inherent fluctuation upper limit ratio of electricity consumption data, W 谷 Indicates the inherent fluctuation upper limit ratio of peak and valley power consumption data, p max Indicates the maximum value of power consumption data, p min Indicates the minimum value of power consumption data, q max Indicates the maximum value of peak and valley power consumption data, q min Indicates the minimum value of peak and valley power consumption data; The fixed fluctuation of each price level is obtained, where the fluctuation ratio range of electricity consumption data is [0, W n用 ], the fluctuation ratio range of peak and valley power consumption data is [0, W n谷 ]; W n用 represents the inherent fluctuation upper limit ratio of electricity consumption data in the nth price level, W n谷 It represents the inherent upper limit ratio of fluctuation of peak and valley electricity consumption data in the nth price level.

6. The large user load forecasting method based on big data mining according to claim 5 is characterized in that: The neural network includes training the data sets under different seasonal factors respectively to obtain the trained neural network, and using the neural network to analyze the seasonal factors at the current time t; The input layer receives the power consumption U t , peak and valley power consumption data V t , and the electricity price P t As input; X t0 =[U t ,V t ,P t ] After considering the inherent fluctuation, the power consumption U t And peak and valley power consumption data V t Add random perturbations to update the input; X t =[U t (1+∈),V t (1+δ),P t ] Among them, X t0 represents the original input, X t represents the updated input; ∈ represents the disturbance factor of power consumption, ranging from [-W n用 , W n用 ]; δ represents the disturbance factor of peak and valley power consumption, ranging from [-W n谷 , W n谷 ]; at this time, n represents the price level at the current moment; Extract time series features: 1D-CNN: used to extract short-term feature patterns; C t =ReLU(Conv1D(X t ,filters=32,kernel size =3)) LSTM: used to capture long-term dependencies; H t =LSTM(X t ,units=50) Among them, ReLU is the activation function, Conv1D is the one-dimensional convolution operation, filters is the number of convolution kernels, kernel size is the size of the convolution kernel, units represents the number of LSTM units, C t represents the time series features extracted by the one-dimensional convolutional neural network, H t Represents the time series features extracted by the long short-term memory network; Fusion layer: After extracting the time series features, the features are fused with the original input features to ensure that the model can take into account both short-term and long-term feature effects: F t =Concatenate([C t ,H t ]) Among them, F t Represents the fused feature vector, Concatenate is a concatenation operation; Multi-task learning layer: Using the multi-task learning mechanism, by sharing some hidden layers, it is possible to simultaneously learn the characteristics of power consumption and peak and valley power consumption: H shared =ReLU(W1·F t +b1) Then set up separate hidden layers for power consumption and peak and valley power consumption: A U =ReLU(W U H shared +b U ) A V =ReLU(W V H shared +b V ) Output layer: Output results for power consumption and peak-valley power consumption prediction: Among them, H shared represents the shared hidden layer features, W1 is the weight matrix of the fully connected layer, b1 is the bias vector, and H U represents the hidden layer features used to predict future electricity consumption, H V represents the hidden layer features used to predict future peak and valley electricity consumption, W U is the weight matrix of the fully connected layer, b U is the bias vector, W V is the weight matrix of the fully connected layer, b V is the bias vector, Represents the predicted electricity consumption data at time t+T, Represents the predicted peak and valley electricity consumption data at time t+T; Linear is a linear transformation, representing a fully connected layer; The loss function of the neural network model is expressed as: Among them, θ represents the set of all parameters of the model, F represents the number of training samples, and U t+T and V t+T They are the actual power consumption and peak-valley power consumption at time t+T, and are the electricity consumption predicted by the model and the peak-valley electricity consumption, respectively. λ is the weight factor used in the loss function to balance the electricity consumption prediction error and the peak-valley electricity consumption prediction error.

7. The large user load prediction method based on big data mining according to claim 6 is characterized by: The prediction result of the large user load includes: assuming that the price level changes with the increase of power consumption, when the price level and seasonal factors do not change at the prediction time t+T, the result is directly output; When the price level changes at the forecast time t+T, the time step is increased from the current time t until the price level just changes, and the forecast result Q1 for the current electricity consumption and peak-valley electricity consumption is output. Q1 is used as input, and the random disturbance ∈ and δ caused by the fixed fluctuation in the next price level are used to update the input. The neural network trained with the current seasonal factors is used to re-forecast, and the updating process of the disturbance factor is repeated until the time step reaches T; When the seasonal factor changes at the prediction time t+T, the time step is increased from the current time t until the seasonal factor just changes, and the prediction result QS1 for the electricity consumption and peak-valley electricity consumption at this time is output; QS1 is used as input, and the neural network trained with the next seasonal factor and the disturbance factors ∈, δ in the price range corresponding to QS1 are used to re-predict, and the updating process of the neural network and the disturbance factor is repeated until the time step reaches T; By analyzing the changes in price levels and seasonal factors, the neural network and interference factors are continuously updated. When the time step reaches T, the prediction result Qd at time t+T is output; Where Qd represents the output of the d-th neural network, d ≥ 1.

8. A large user load forecasting system based on big data mining using the method according to any one of claims 1 to 7, characterized in that: The collection unit collects the historical records of user load data and obtains the user's electricity purchase and sales data; An analysis unit, generating a dependency relationship between the user load data and the power purchase and sale data by analyzing fluctuations in the power purchase and sale data; The prediction unit generates a prediction result of the large user load for the historical records according to the dependency relationship.

9. A computer device comprising: Memory and processor; The memory stores a computer program, characterized in that: when the processor executes the computer program, the steps of a large user load prediction method based on big data mining 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 a large user load prediction method based on big data mining are implemented.

Images