A power load optimization distribution method and system based on intelligent algorithm

Abstract

The application relates to the technical field of intelligent power system optimization scheduling, and discloses a power load optimization distribution method and system based on an intelligent algorithm, which comprises the following steps: collecting historical load data of a power system and preprocessing; calculating the fractal dimension of a load time sequence, identifying a load change critical point, and dynamically adjusting the size of a load prediction time window; inputting the dynamically adjusted time window data into an LSTM model for load prediction; and according to the load prediction result, intelligently optimizing and distributing the power load by using an optimization algorithm. The application makes the optimization distribution of the power load not only have high precision, but also be able to flexibly cope with various sudden fluctuations and complex constraint conditions in the power system. Through the steps, the application improves the operation efficiency, stability and economy of the power system, reduces the uncertainty and risk in the power system load distribution, and provides a new idea and technical support for future smart power grids and power dispatching systems.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent power system optimization scheduling, in particular to a power load optimization distribution method and system based on intelligent algorithm. BACKGROUND

[0002] With the continuous development of intelligentization and informatization, power load optimization distribution, as an important research field in power system scheduling, has received extensive attention in recent years. Traditional load forecasting methods are mostly based on linear regression, time series analysis and other statistical models, but with the variability and complexity of power demand, these traditional methods gradually become inadequate. In recent years, with the rapid development of artificial intelligence technology, especially the application of deep learning and intelligent optimization algorithms, the methods of power load forecasting and optimization distribution have gradually shifted from traditional statistical models to optimization methods based on intelligent algorithms. For example, the load forecasting method based on long short-term memory (LSTM) network can effectively capture the nonlinear trend of power load by using its powerful time series data modeling capability. In addition, particle swarm optimization (PSO), genetic algorithm (GA) and other intelligent optimization algorithms are also widely used in power load distribution and scheduling to improve the efficiency and stability of the power system.

[0003] Although the existing power load optimization distribution method introduces intelligent algorithms, especially the combination of load forecasting based on machine learning and intelligent optimization algorithms, it still has the following shortcomings, which cannot completely solve the power load optimization problem. First, traditional load forecasting methods mostly rely on fixed time windows, which cannot dynamically adjust the granularity of the forecast, resulting in low prediction accuracy when the load fluctuates sharply, thereby affecting the subsequent load distribution decision. This is because the fixed time window cannot adapt to the rapid changes in load demand, especially at the critical point of load change, it cannot accurately capture the fluctuations in a short period of time.

[0004] Secondly, although the existing intelligent optimization algorithm can solve the problem of power load distribution, it generally relies on fixed parameter settings and simple constraint conditions, and lacks the ability to flexibly adjust to the complex demands in the actual power system. Many algorithms are prone to fall into local optimal solution when facing complex power network structure and various real-time constraints (such as supply and demand balance, power system capacity, etc.), and cannot effectively find the global optimal solution.

[0005] In addition, although the existing method uses some evaluation indexes of prediction accuracy and load distribution efficiency, most of the methods fail to fully consider the direct impact of load forecasting error on power load distribution. Since the error of load forecasting may affect the accuracy of power load distribution to some extent, the existing technology often does not effectively constrain and optimize this problem, causing uncertainty in the process of load distribution and potential risks to the power system.

[0006] Therefore, the prior art has obvious deficiencies in accurately predicting load fluctuations, dynamically adjusting time windows, processing complex constraint conditions, ensuring global optimization and controlling prediction errors, and the present application is based on these deficiencies. A power load optimization distribution method based on intelligent algorithm is proposed, which aims to effectively improve the accuracy and efficiency of load distribution by dynamically adjusting the time window, combining fractal dimension analysis and intelligent optimization algorithm, and further improve the reliability and safety of distribution through load prediction accuracy constraints. SUMMARY

[0007] In view of the above problems, the present application is proposed.

[0008] Therefore, the present application solves the technical problems of the existing power load optimization distribution method, which has the problems of inaccurate load fluctuation capture, optimization algorithm easily falling into local optimal solution, lack of effective control of load prediction error, and how to accurately distribute power load and ensure the reliability of optimization by dynamically adjusting the time window, introducing fractal dimension analysis and combining intelligent optimization algorithm.

[0009] To solve the above technical problems, the present application provides the following technical scheme: a power load optimization distribution method based on intelligent algorithm, comprising: collecting historical load data of a power system and preprocessing.

[0010] Calculate the fractal dimension of the load time series, identify the load change critical point, and dynamically adjust the size of the time window for load prediction.

[0011] Using the dynamically adjusted time window data, input the LSTM model for load prediction.

[0012] According to the load prediction result, an optimization algorithm is used to intelligently optimize and distribute the power load.

[0013] As a preferred scheme of the power load optimization distribution method based on intelligent algorithm, wherein: the collection of historical load data of the power system and preprocessing includes:

[0014] Real-time collection of power load data from the SCADA system of the power system, including the power consumption, load change trend and power factor of each substation, distribution network and load point of the power system.

[0015] The preprocessing includes removing missing data, removing outliers, data smoothing and data normalization.

[0016] The missing data is checked, and if missing data is found, the missing data is filled by linear interpolation.

[0017] Outliers removal uses z-score to identify and remove obvious outliers.

[0018] Data smoothing smoothes the collected data, removes short-term fluctuations using the moving average method, and retains long-term trends.

[0019] Data normalization normalizes the load data, and uses the minimum-maximum normalization method to standardize the load data to the interval [0, 1].

[0020] As a preferred scheme of the power load optimization distribution method based on intelligent algorithm, wherein the calculation of the fractal dimension of the load time series comprises:

[0021] The original fractal dimension calculation formula of the load time series is represented as:

[0022]

[0023] Where D represents the fractal dimension, N(∈) represents the number of data points covered by a box with size ∈, and ∈ represents the box size.

[0024] In order to accurately capture the load fluctuation, a time window and a weighting coefficient are introduced, and the optimized fractal dimension formula is represented as:

[0025]

[0026] Where D adj (t) represents the dynamic fractal dimension at time t, considering the weighted influence of load fluctuation. N i (∈) represents the calculation of self-similarity in the i-th box, w i (t) represents the weighting coefficient, representing the weighted influence of load change, and is represented as:

[0027]

[0028] Where α represents the weighting coefficient adjustment parameter, controlling the sensitivity of load fluctuation. μ represents the threshold of load fluctuation, y i (t) represents the difference of the load sequence at time t, which is the load change amount of adjacent two time points.

[0029] As a preferred scheme of the power load optimization distribution method based on intelligent algorithm, wherein the identification of the load change critical point and the dynamic adjustment of the time window size of the load prediction comprises:

[0030] On the basis of fractal dimension, the critical point of load change is identified by observing the speed and fluctuation amplitude of load change, and is represented as:

[0031]

[0032] wherein, ΔD adj (t) represents the optimized fractal dimension change rate, indicating the amplitude of load fluctuation. adj (t-10) represents the dynamic fractal dimension at time t-1.

[0033] If ΔD adj (t) > T thresh , it is considered that the critical point of load change occurs at time t. The threshold value T thresh is obtained by analyzing historical data to distinguish important fluctuations from ordinary fluctuations.

[0034] The dynamic adjustment of the time window of load prediction is represented as:

[0035]

[0036] wherein, ∈ adj (t) represents the dynamic time window at time t, and β represents the adjustment parameter, controlling the response sensitivity of the time window.

[0037] As a preferred scheme of the power load optimization distribution method based on intelligent algorithm, wherein: the input of the LSTM model for load prediction using the dynamically adjusted time window data includes:

[0038] The calculation of each time step in the LSTM network includes the calculation of the forget gate, input gate, candidate layer and output gate.

[0039] As a preferred scheme of the power load optimization distribution method based on intelligent algorithm, wherein: the LSTM model includes:

[0040] According to the output h t of the LSTM, the final power load prediction value is represented as:

[0041]

[0042] wherein, represents the predicted load value at time t, and represents the prediction result output by the LSTM model. W y represents the prediction weight, and b y represents the prediction bias.

[0043] As a preferred scheme of the power load optimization distribution method based on intelligent algorithm, wherein: the intelligent optimization distribution of power load according to the load prediction result includes:

[0044] P pred,t represents the final power load prediction value of the LSTM model The optimization objective is to minimize the power distribution cost, denoted as:

[0045]

[0046] where F opt (x) represents the optimization objective function, denoted as the total cost of the power load optimization problem. Ch t represents the power cost at time period t, calculated based on the power market price and load demand fluctuations. P t (x t ) represents the amount of power load distributed at time period t. P pred,t represents the power load demand predicted by the LSTM model at time period t. λ t represents the penalty coefficient, used to balance the error between the distribution amount and the predicted value, reflecting the cost when the distribution amount deviates from the predicted value. γ t represents the penalty coefficient, used to handle power overload situations. represents the power overload indicator function, taking the value of 1 when P t (x t ) exceeds the system maximum load P max,t , otherwise 0.

[0047] As a preferred scheme of the power load optimization distribution method based on intelligent algorithm described in the present application, wherein: the intelligent optimization distribution of power load based on load prediction results using optimization algorithm further includes: using particle swarm optimization algorithm to optimize the objective function through the position and speed of particles, each particle of the particle swarm represents a power distribution scheme, and the update rule of the particle is represented as:

[0048]

[0049] wherein, represents the speed of particle i at time period t. represents the power distribution amount of particle i at time period t. w represents the inertia weight, controlling the influence degree of the previous speed of the particle. c1 and c2 represent the learning factor, respectively representing the learning degree of the particle to the individual optimal and group optimal. r1 and r2 represent random numbers, taking values in the range of [0, 1], used to introduce randomness. represents the personal optimal solution of particle i. g best represents the group optimal solution.

[0050] The constraint conditions of the particle swarm optimization algorithm include power compliance restrictions and load balance requirements.

[0051] The power load restriction is that the power distribution amount of each time period should be within the maximum load range, and the load balance requirement is that the total power demand should be equal to the total power distribution amount.

[0052] As a preferred scheme of the intelligent algorithm-based power load optimal allocation method, the constraint conditions further include:

[0053] A new load prediction accuracy constraint is added, which is expressed as:

[0054]

[0055] wherein, P t (x t ) represents the power load allocation amount of period t, which is the power load value calculated in the particle swarm optimization algorithm. pred,t represents the load prediction value of period t, which is calculated by the LSTM model. represents the maximum allowed prediction error threshold of period t, which is defined as the maximum difference between the load allocation amount and the load prediction value of the power load in period t, and is determined by the tolerance error of the power system and the change range of the power market.

[0056] An intelligent algorithm-based power load optimal allocation system, characterized in that it comprises,

[0057] A preprocessing module for collecting historical load data of the power system and performing preprocessing.

[0058] A prediction module for calculating the fractal dimension of the load time series, identifying the load change critical point, and dynamically adjusting the time window size of the load prediction. The dynamically adjusted time window data is input into the LSTM model for load prediction.

[0059] A calculation module for intelligently optimizing and allocating the power load by using an optimization algorithm according to the load prediction result.

[0060] A computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method as described above when executing the computer program.

[0061] A computer-readable storage medium having a computer program stored thereon, wherein the computer program is executed by a processor to implement the steps of the method as described above.

[0062] The beneficial effects of the present application: by introducing the step of dynamically adjusting the time window, the prediction granularity is dynamically adjusted according to the actual change of load fluctuation, which can accurately capture the instantaneous fluctuation of power load. In the traditional method, the size of the time window is fixed, which cannot be flexibly adjusted according to the change of actual load demand, resulting in a decrease in the accuracy of prediction in the period of severe load fluctuation, affecting the subsequent distribution decision. By dynamically adjusting the time window, the system can reduce the window size in the period of large load fluctuation and expand the window size in the period of small load fluctuation, thereby improving the accuracy of load prediction. The role of this step is to optimize the response speed of the prediction model, so that it can adapt to the change of load demand in time, ensuring the efficiency and accuracy of power load distribution. This adjustment method has important practical value in real-time scheduling of power systems, which can reduce the risk of power waste and overload, thereby improving the stability and economy of the system.

[0063] By introducing fractal dimension analysis to evaluate the self-similarity of load time series, the present application can more accurately identify the critical point of load change. Traditional load prediction methods often ignore the complexity of load fluctuation, easily missing some important changes in the short term, especially when the load fluctuation is severe. By using fractal dimension analysis, the change rule of load sequence can be measured more accurately, especially in periods of strong load volatility, which plays an important role in improving the accuracy of prediction. This step can automatically capture the time of sudden change in fluctuation amplitude by analyzing the self-similarity of load time series, so that the load prediction model can identify and adapt to sudden events in advance, thereby effectively avoiding the problem of load overload or power shortage caused by prediction lag. This technical innovation has significant practical value for improving the accuracy of load prediction, especially the prediction accuracy in periods of high volatility.

[0064] The present application limits the error between the prediction result and the actual load by introducing the load prediction accuracy constraint, ensuring the reliability of the power load distribution process. Traditional methods usually ignore the risks that may be caused by load prediction errors, and rely too much on the accuracy of prediction results, which can easily lead to uneven power distribution, resource waste and other problems. By setting the prediction error tolerance, the present application ensures the accuracy of load distribution, even in the presence of prediction errors, it can ensure the balance and stability of the power system. The innovation of this step lies in that it not only improves the stability of the load prediction of the power system, but also reduces the uncertainty and risk in power dispatching by controlling the prediction error, further improving the safety and reliability of the power system.

[0065] By using the particle swarm optimization algorithm (PSO) for intelligent optimization of power load distribution, the system can find the optimal power distribution scheme under multiple constraints. Traditional optimization algorithms usually use static, fixed parameter distribution strategies, which lack flexibility and adaptive ability. The particle swarm optimization algorithm can dynamically adjust the distribution strategy according to the actual demand of the power load and the system constraints, thereby maximizing the efficiency of power distribution and reducing operating costs. The role of this step is that PSO can consider power grid load balancing, power demand, cost and various constraints to achieve global optimal solution under more efficient conditions, avoiding the trap of local optimal solution. Through particle swarm optimization, the distribution of power load can be more flexible and accurate, greatly improving the economic benefit and operation stability of the power system. BRIEF DESCRIPTION OF DRAWINGS

[0066] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor. Among them:

[0067] Figure 1 The overall flowchart of a power load optimization distribution method and system based on intelligent algorithm provided for the first embodiment of the present application. DETAILED DESCRIPTION

[0068] In order to make the above-mentioned purposes, features and advantages of the present application more apparent and easy to understand, the specific embodiments of the present application will be described in detail below with reference to the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor should be within the scope of protection of the present application.

[0069] Embodiment 1, refer to Figure 1 For an embodiment of the present application, a power load optimization distribution method based on intelligent algorithm is provided, comprising:

[0070] S1: Collect the historical load data of the power system and preprocess it.

[0071] Real-time collection of power load data from the SCADA system of the power system, including the power consumption, load change trend and power factor of each substation, distribution network and load point of the power system.

[0072] It should be noted that the real-time data of the power system comes from the substation, distribution network and each load point, mainly including power consumption, load change trend, power factor and other data. High-precision smart meters, load monitoring equipment and distribution automation systems are used to collect relevant data through data acquisition terminals.

[0073] Further, in order to capture the rapid fluctuations of the load, the sampling frequency is set to per second, ensuring the real-time and accuracy of the data. Through modern communication networks, real-time transmission of data is realized, ensuring the rapid and accurate transmission of data flow to the centralized data processing platform.

[0074] Preprocessing includes removing missing data, removing outliers, data smoothing and data normalization.

[0075] Missing data removal checks the collected data, and if missing data is found, the missing data is filled by linear interpolation.

[0076] Outlier removal uses z-score to identify and remove obvious outliers.

[0077] Data smoothing performs smoothing processing on the collected data, using the moving average method to remove short-term fluctuations and retain long-term trends.

[0078] Data normalization normalizes the load data, using the min-max normalization method to standardize the load data to the interval [0, 1].

[0079] S2: Calculate the fractal dimension of the load time series, identify the load change critical point, and dynamically adjust the size of the time window for load prediction.

[0080] The original fractal dimension calculation formula of the load time series is represented as:

[0081]

[0082] Where D represents the fractal dimension, N(∈) represents the number of data points covered by a box of size ∈, and ∈ represents the size of the box.

[0083] To accurately capture load fluctuations, a time window and a weighting coefficient are introduced, and the optimized fractal dimension formula is represented as:

[0084]

[0085] Where D adj (t) represents the dynamic fractal dimension at time t, considering the weighted influence of load fluctuations. N i (∈)represents the calculation of self-similarity in the i-th box, w i (t) represents the weighting coefficient, representing the weighted influence of load changes, and is represented as:

[0086]

[0087] wherein a represents a weighting coefficient adjustment parameter, controlling the sensitivity of load fluctuation. μ represents the threshold value of load fluctuation, y i (t) represents the difference value of load sequence at time t, which is the load change amount of adjacent two time points.

[0088] It should be noted that in order to capture the change trend of load fluctuation, a time window is introduced in the original fractal dimension calculation of load time sequence. The size of this time window is automatically adjusted according to the characteristics of load fluctuation, historical data, and the change of fractal dimension. By focusing on the data in different time ranges through the time window, the detection accuracy of load fluctuation is improved. The introduction of weighting coefficient is to weight the calculation results according to the data at different time points, so as to improve the sensitivity to some important time periods (such as peak load period or load mutation period). The weighting coefficient is set according to the amplitude of load change, the urgency of time period and other factors.

[0089] Further, by introducing the time window, irrelevant or noise data can be effectively eliminated, ensuring more accurate fractal dimension calculation. The setting of weighting coefficient can be flexibly adjusted according to the influence of different data, avoiding too large deviation of single data point to the overall load prediction. The adaptive adjustment in the design idea can automatically optimize the size of time window when the load fluctuation is large, ensuring the adaptation to the changes in different time periods.

[0090] On the basis of fractal dimension, the critical point of load is identified by observing the speed and fluctuation amplitude of load change, which is represented as:

[0091]

[0092] wherein ΔD adj (t) represents the optimized fractal dimension change rate, representing the amplitude of load fluctuation. D adj (t-1) represents the dynamic fractal dimension at time t-1.

[0093] If ΔD adj (t) > T thresh , it is considered that the critical point of load change occurs at time t. The threshold value T thresh is the key fluctuation point obtained by historical data analysis, which is used to distinguish important fluctuation and ordinary fluctuation.

[0094] It should be noted that by calculating the rate of change of load data in a short time (i.e. the rate of increase and decrease of load), if the change speed suddenly increases, it may mean that the system will enter a new load state, or approach the critical point of load fluctuation. By observing the fluctuation amplitude of the load data, if the load fluctuation amplitude in a certain period is abnormally large (exceeding the normal fluctuation range), it indicates that the load in this period has a critical change, which may lead to overload or power imbalance of the system. Based on the calculation of fractal dimension, by observing the self-similarity change of the load time series, the mutation of load fluctuation can be detected. If the fractal dimension curve shows mutation or non-stationary characteristics, it can be used as a warning signal of the critical point.

[0095] The dynamic adjustment of the time window of load prediction is represented as:

[0096]

[0097] where ∈ adj (t) represents the dynamic time window at time t, and β represents the adjustment parameter, which controls the response sensitivity of the time window.

[0098] It should be noted that when the load critical point is identified, the time window size for prediction is automatically adjusted based on the suddenness and violent fluctuation of load change. In the period near the critical point, the time window will be reduced to capture the load fluctuation more finely. In the stable period of load, the time window can be appropriately relaxed to avoid excessive response to short-term fluctuations.

[0099] As the load data is continuously input, the system will monitor the fluctuation trend of the load in real time and dynamically adjust the size of the time window. For example, if the load changes rapidly and the fluctuation amplitude is large, the system will shorten the time window to ensure that the response of the prediction is faster. Conversely, if the load changes smoothly, the window can be relaxed to improve the calculation efficiency.

[0100] The size and adjustment frequency of the time window depend on the regularity of load change. The system will automatically optimize the adjustment process of the time window through the accuracy and error of load prediction, to maintain the real-time and accuracy of the system.

[0101] S3: Use the dynamically adjusted time window data to input the LSTM model for load prediction.

[0102] The calculation of each time step in the LSTM network includes the calculation of the forget gate, input gate, candidate layer and output gate.

[0103] The forget gate calculation process is represented as:

[0104] f t = σ(W f ·[h t-1 ,x t ]+bf )

[0105] The input gate computation process is represented as:

[0106] i t = σ(W i · [h t-1 , x t ] + b i )

[0107] The candidate cell state computation process is represented as:

[0108]

[0109] The cell state update is represented as:

[0110]

[0111] The output gate is represented as:

[0112] o t = σ(W o · [h t-1 , x t ] + b o )

[0113] The final output is represented as:

[0114] h t = o t · tanh(C t )

[0115] where f t represents the output of the forget gate, which determines the influence of the cell state at the previous time step on the current time step; i t represents the output of the input gate, which controls the degree of influence of the current input data; represents the candidate cell state, which represents the candidate update value at the current time step; C t represents the cell state, which preserves the long-term memory of the LSTM network; o t represents the output of the output gate, which determines the influence of the cell state on the output at the current time step; h t represents the output at the current time step, which represents the prediction of the LSTM network for the current load; σ represents the activation function; tanh represents the hyperbolic tangent activation function, which is used for the update of the cell state and the calculation of the output; x t represents the input data at time t, which contains the dynamically adjusted time window data; W f , W i , W C , W o represent weight matrices, which represent the training parameters for each gating operation; b f , bi , b C , b o represents a bias term, represents the training bias of each gating operation;

[0116] According to the output h t , the final power load prediction value is represented as:

[0117]

[0118] wherein, represents the predicted load value at time t, represents the prediction result output by the LSTM model. W y represents the prediction weight, b y represents the prediction bias.

[0119] S4: According to the load prediction result, an optimization algorithm is used to intelligently optimize the distribution of power load.

[0120] with P pred,t representing the final power load prediction value of the LSTM model The optimization goal is to minimize the power distribution cost, which is represented as:

[0121]

[0122] wherein, F opt (x) represents the optimization objective function, represents the total cost of the power load optimization problem. Ch t represents the power cost at period t, which is calculated according to the power market price and load demand fluctuations. P t (x t ) represents the power load amount allocated at period t. P pred,t represents the power load demand predicted by the LSTM model at period t. λ t represents the penalty coefficient, used to balance the error between the allocated amount and the predicted value, reflecting the cost when the allocated amount deviates from the predicted value. γ t represents the penalty coefficient, used to handle power overload situations. represents the power overload indicator function, which takes the value of 1 when P t (x t ) exceeds the maximum load P max,t of the system, otherwise it is 0.

[0123] It should be noted that the dynamic adjustment of the time window of load prediction is introduced into the optimization process, so that the system can automatically adjust the calculation granularity at different load demand periods, thereby more accurately controlling the distribution cost. Compared with the traditional fixed time window optimization method, the present application makes the optimization of power distribution cost more flexible and efficient. The dynamic time window can accurately adjust the calculation granularity under different load demand scenarios, thereby reducing unnecessary calculation and resource waste. The objective function can automatically adjust the optimization strategy as the load demand changes, and is more suitable for complex power load management requirements.

[0124] Using the particle swarm optimization algorithm, the objective function is optimized by the position and speed of the particles, and each particle of the particle swarm represents a power distribution scheme. The updating rule of the particle is represented as:

[0125]

[0126] wherein, represents the speed of particle i at period t. represents the power distribution amount of particle i at period t. w represents the inertia weight, which controls the influence degree of the previous speed of the particle. c1 and c2 represent the learning factor, which respectively represents the learning degree of the particle to the individual optimal and the group optimal. r1 and r2 represent random numbers, the value range of which is [0, 1], which is used to introduce randomness. represents the personal optimal solution of particle i. g best represents the group optimal solution.

[0127] The constraint conditions of the particle swarm optimization algorithm include the power compliance limit and the load balance requirement.

[0128] The power load limit is that the power distribution amount of each period should be within the maximum load range, and the load balance requirement is that the total power demand should be equal to the total power distribution amount.

[0129] The new load prediction accuracy constraint is represented as:

[0130]

[0131] wherein, P t (x t ) represents the power load distribution amount at period t, which is the power load value calculated in the particle swarm optimization algorithm. P pred,t represents the load prediction value at period t, which is calculated by the LSTM model. represents the maximum allowed prediction error threshold at period t, which is defined as the maximum difference between the load distribution amount and the load prediction value of the power load at period t, and is determined by the tolerance error of the power system and the change range of the power market.

[0132] It should be noted that the load prediction accuracy constraint ensures that the difference between the load prediction result and the actual load demand is within an acceptable range by limiting the upper limit of the load prediction error. This constraint condition is used to control the influence of prediction error in power load distribution, avoiding unreasonable distribution due to excessive error.

[0133] The load prediction accuracy constraint ensures that the predicted value based on which the load distribution is made is as accurate as possible, avoiding system instability caused by excessive prediction error. By setting the accuracy constraint, more precise load scheduling can be achieved, reducing uncertainty in power load distribution and improving the robustness of the system.

[0134] Further, the present application relies particularly on the LSTM model for power load prediction, and the output of the LSTM model may be affected by the training data and model optimization, therefore an additional constraint of error tolerance is needed to ensure the accuracy of load prediction meets the actual demand. In contrast, other traditional load prediction methods (such as linear regression, support vector machine, etc.) usually do not have such fine error control requirements, therefore this constraint condition is adapted to the deep learning load prediction process in the present application.

[0135] In the above embodiments, there is also an electric power load optimization distribution system based on intelligent algorithm, specifically:

[0136] The preprocessing module collects historical load data of the power system and performs preprocessing.

[0137] The prediction module calculates the fractal dimension of the load time series, identifies the load change critical point, and dynamically adjusts the size of the time window for load prediction. Using the dynamically adjusted time window data, the LSTM model is input for load prediction.

[0138] The calculation module uses an optimization algorithm to intelligently optimize and distribute the electric power load based on the load prediction result.

[0139] The computer device can be a server. The computer device comprises a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store data cluster data of a power monitoring system. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through network connection. The computer program is executed by the processor to implement a power load optimization distribution method based on an intelligent algorithm.

[0140] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned embodiments of each method. Any reference to memory, database or other medium used in each embodiment provided by the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The database involved in each embodiment provided by the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in each embodiment provided by the present application can be a general-purpose processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0141] Embodiment 2, as an embodiment of the present application, provides a power load optimization distribution method and system based on intelligent algorithm. In order to verify the beneficial effects of the present application, a simulation experiment is carried out for scientific demonstration.

[0142] This embodiment verifies the effect of the power load optimization distribution method based on intelligent algorithm through experimental data. The test mainly focuses on load prediction through the LSTM model and load distribution combined with the particle swarm optimization (PSO) algorithm to evaluate the advantages of the present application method in power load optimization distribution.

[0143] In this experiment, the power load distribution problem of a certain area is assumed, and the time period is set to 24 hours, and the load demand of each hour is predicted by the LSTM model. In order to verify the effectiveness of the method, the experiment is divided into two groups: the control group (traditional load prediction method) and the experimental group (LSTM prediction combined with PSO optimization distribution method). The control group uses the traditional regression analysis method for load prediction, and does not use dynamic adjustment of the time window, and the load distribution is directly optimized based on the prediction results. The experimental group uses the LSTM prediction model in the invention, and dynamically adjusts the time window, and combines the particle swarm optimization algorithm for load optimization distribution.

[0144] Collect the power load data of a certain area, including load demand, load change trend, power factor, etc. The data collection frequency is set to once every hour, and pre-processing (denoising, missing value filling, etc.) is performed.

[0145] Use the LSTM model to predict the power load demand of the next 24 hours. The control group uses the traditional linear regression method for prediction, and the experimental group uses the LSTM model.

[0146] For the experimental group, use fractal dimension analysis method, by dynamically adjusting the size of the time window, to ensure that the prediction accuracy is improved when the load fluctuates greatly.

[0147] Use the particle swarm optimization algorithm (PSO) to distribute the predicted results of the experimental group and the control group. In the distribution process, consider power demand, cost, constraint conditions, etc., and seek the optimal load distribution scheme through the optimization algorithm.

[0148] Compare the load prediction error, distribution cost and system load balancing of the two groups. The experimental results are shown in Table 1.

[0149] Table 1

[0150]

[0151] From the table, it can be seen that the error of the load value predicted by the experimental group is smaller than the actual load, and the error rate is continuously lower than that of the control group. This shows that the load prediction method based on LSTM can more accurately capture the nonlinear change trend of power load, and the prediction accuracy in periods of high fluctuation is significantly improved compared to traditional regression methods, especially in periods of severe load fluctuations (such as periods 7, 9 and 14), the error is smaller, indicating that the LSTM model can effectively deal with such complex fluctuations.

[0152] In addition, the effect of dynamically adjusting the time window is also reflected in the prediction accuracy. As the load changes progress, the adjustment of the time window ensures that the system can quickly respond to sudden load changes in a short period of time. This dynamic adjustment mechanism is crucial for power system scheduling, as it can significantly reduce the lag in load fluctuation prediction, thereby improving the accuracy of load distribution.

[0153] In terms of load distribution cost, the distribution cost of the experimental group is generally lower, reflecting that the optimization of load distribution through the particle swarm optimization algorithm (PSO) can effectively reduce the operating cost of the power system while meeting the load demand and system constraints. Although there is a slight gap in distribution cost between the experimental group and the control group, the overall difference is very small, and the optimization algorithm makes the system run more stably.

[0154] This embodiment verifies the innovation and effectiveness of the power load optimization distribution method based on intelligent algorithms by comparing the prediction error and load distribution cost of the experimental group and the control group. The combination of the LSTM model, dynamic time window adjustment, and particle swarm optimization algorithm makes the power load prediction more accurate, the load distribution cost lower, and the system better able to cope with the challenges brought by load fluctuations. Through these beneficial effects, the present invention demonstrates stronger adaptability and accuracy than traditional methods, especially in complex scenarios with large fluctuations in power demand, and has a significant advantage.

[0155] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present application and not to limit it. Although the present application 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 application can be modified or replaced by equivalents without departing from the spirit and scope of the present application, and they should be included in the scope of the claims of the present application.

Claims

1. A power load optimization allocation method based on intelligent algorithms, characterized in that, include: Collect historical load data from the power system and perform preprocessing; Calculate the fractal dimension of the load time series, identify the critical point of load change, and dynamically adjust the size of the time window for load forecasting; Load forecasting is performed by inputting dynamically adjusted time window data into an LSTM model. Based on the load forecast results, an optimization algorithm is used to intelligently optimize the allocation of power load; The fractal dimensions for calculating load time series include: The formula for calculating the original fractal dimension of the load time series is expressed as: Where D represents the fractal dimension, Indicates using size The number of data points covered by the box. Indicates the size of the box; To accurately capture load fluctuations, a time window and weighting coefficients are introduced. The optimized fractal dimension formula is expressed as follows: in, Represents the dynamic fractal dimension at time t, taking into account the weighted effect of load fluctuations; This represents the calculation of self-similarity in the i-th box. The weighting factor represents the weighted impact of load changes, expressed as: in, This represents the weighted coefficient adjustment parameter, which controls the sensitivity to load fluctuations; The threshold representing load fluctuations It represents the difference in the load sequence at time t, which is the change in load between two adjacent times; Identifying critical points of load changes and dynamically adjusting the time window size for load forecasting includes: Based on the fractal dimension, the critical point of the load is identified by observing the rate of change and the amplitude of fluctuations in the load, as expressed as: in, This represents the rate of change of the optimized fractal dimension, and the magnitude of the load fluctuation. Represents the dynamic fractal dimension at time t-1; like If , then it is considered that a critical point of load change has occurred at time t; where the threshold value is... These are key fluctuation points obtained through historical data analysis, used to distinguish between significant fluctuations and ordinary fluctuations; The time window for dynamically adjusting load forecasting is expressed as: in, Represents the dynamic time window at time t. This indicates the adjustment parameter, which controls the response sensitivity within the time window.

2. The power load optimization allocation method based on intelligent algorithms as described in claim 1, characterized in that: The preprocessing of the historical load data collected from the power system includes: Real-time power load data is collected from the SCADA system of the power system, including the power consumption, load change trend, and power factor of each substation, distribution network, and load point in the power system. Preprocessing includes removing missing data, removing outliers, data smoothing, and data normalization; Remove missing data by checking the collected data. If missing data is found, fill in the missing data using linear interpolation. Outlier removal uses methods such as z-score to identify and remove obvious outliers; Data smoothing involves smoothing the collected data by using a moving average method to remove short-term fluctuations while preserving long-term trends. Data normalization normalizes the load data by using the min-max normalization method to standardize the load data to the [0,1] interval.

3. The power load optimization allocation method based on intelligent algorithms as described in claim 2, characterized in that: The process of using dynamically adjusted time window data to input into the LSTM model for load forecasting includes: The computation of each time step in an LSTM network includes the computation of the forget gate, input gate, candidate layer, and output gate.

4. The power load optimization allocation method based on intelligent algorithms as described in claim 3, characterized in that: The LSTM model includes: Based on the output of LSTM The final power load forecast is expressed as follows: in, Represents the predicted load value at time t, and represents the prediction result output by the LSTM model; Indicates the prediction weights. This indicates the prediction bias.

5. The power load optimization allocation method based on intelligent algorithms as described in claim 4, characterized in that: The intelligent optimization allocation of power load based on load forecasting results and using optimization algorithms includes: use This represents the final power load forecast from the LSTM model. The optimization objective is confirmed to be minimizing the cost of electricity allocation, expressed as: in, Let represent the objective function, and let represent the total cost of the power load optimization problem. The electricity cost for time period t is calculated based on electricity market prices and load demand fluctuations. This represents the amount of electrical load allocated during time period t; This represents the electricity load demand predicted by the LSTM model for time period t. This represents the penalty coefficient, used to balance the error between the allocated amount and the predicted value, reflecting the cost when the allocated amount deviates from the predicted value; This represents the penalty coefficient, used to handle power overload situations; This represents the power overload indication function, when Exceeding the system's maximum load The value is 1 if it is true, and 0 otherwise.

6. The power load optimization allocation method based on intelligent algorithms as described in claim 5, characterized in that: The intelligent optimization allocation of power load based on load forecasting results and optimization algorithms also includes: The particle swarm optimization algorithm is used to optimize the objective function based on the position and velocity of the particles. Each particle in the swarm represents a power allocation scheme, and the particle update rule is expressed as follows: in, This represents the velocity of particle i during time interval t; This represents the amount of electricity allocated to particle i during time period t; Indicates the inertial weight, which controls the degree of influence of the particle's previous velocity; The learning factor represents the degree to which a particle learns from individual optimality and swarm optimality, respectively. This represents a random number, with a value range of [0,1], used to introduce randomness; This represents the individual optimal solution for particle i; Represents the group optimal solution; The constraints of the particle swarm optimization algorithm include power compliance constraints and load balancing requirements; The power load limit is that the power allocation for each time period should be within the maximum load range, and the load balance requirement is that the total power demand should be equal to the total power allocation.

7. The power load optimization allocation method based on intelligent algorithms as described in claim 6, characterized in that: The constraints also include: The new load forecast accuracy constraint is expressed as follows: in, The power load allocation for time period t is the power load value calculated in the particle swarm optimization algorithm. This represents the load forecast value for time period t, calculated by the LSTM model; The maximum allowable forecast error threshold for time period t is defined as the maximum difference between the load allocation and the load forecast value during time period t, which is determined by the tolerance error of the power system and the magnitude of changes in the power market.

8. A power load optimization and allocation system based on intelligent algorithms, employing the method described in any one of claims 1-6, characterized in that: The preprocessing module collects historical load data from the power system and performs preprocessing. The forecasting module calculates the fractal dimension of the load time series, identifies critical points of load change, and dynamically adjusts the size of the load forecasting time window. The dynamically adjusted time window data is then input into the LSTM model for load forecasting. The formula for calculating the original fractal dimension of the load time series is expressed as: Where D represents the fractal dimension, Indicates using size The number of data points covered by the box. Indicates the size of the box; To accurately capture load fluctuations, a time window and weighting coefficients are introduced. The optimized fractal dimension formula is expressed as follows: in, Represents the dynamic fractal dimension at time t, taking into account the weighted effect of load fluctuations; This represents the calculation of self-similarity in the i-th box. The weighting factor represents the weighted impact of load changes, expressed as: in, This represents the weighted coefficient adjustment parameter, which controls the sensitivity to load fluctuations; The threshold representing load fluctuations It represents the difference in the load sequence at time t, which is the change in load between two adjacent times; Identifying critical points of load changes and dynamically adjusting the time window size for load forecasting includes: Based on the fractal dimension, the critical point of the load is identified by observing the rate of change and the amplitude of fluctuations in the load, as expressed as: in, This represents the rate of change of the optimized fractal dimension, and the magnitude of the load fluctuation. Represents the dynamic fractal dimension at time t-1; like If , then it is considered that a critical point of load change has occurred at time t; where the threshold value is... These are key fluctuation points obtained through historical data analysis, used to distinguish between significant fluctuations and ordinary fluctuations; The time window for dynamically adjusting load forecasting is expressed as: in, Represents the dynamic time window at time t. This indicates the adjustment parameter, which controls the response sensitivity within the time window; The calculation module uses optimization algorithms to intelligently optimize and allocate power load based on load forecasting results.

Images