Energy storage system-based demand control method, system, equipment and medium

Abstract

The invention relates to a demand control method, system and equipment based on an energy storage system, and a medium. The method comprises the following steps: collecting power load characteristics and related influence factor data thereof; analyzing the collected power load characteristics and related influence factor data by using a principal component-weighted grey relational analysis method to extract key input variables for load prediction; inputting the key input variable into a long and short term memory load prediction model based on an Attention mechanism, and training the long and short term memory load prediction model to output load prediction data in a period of time in the future; according to the predicted load prediction data, constructing a pre-month energy storage optimization model to determine a predicted maximum demand value; and according to the maximum demand value, carrying out rolling optimization on an intra-day scheduling scheme so as to carry out optimal charging and discharging operation of energy storage. According to the method, through optimal scheduling of the energy storage system, the accuracy of power load prediction and the efficiency of power cost control are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of power system control, and in particular to a demand control method, system, equipment and medium based on an energy storage system. Background Art

[0002] With the rapid development of industrialization and urbanization, the demand for electricity is growing, especially the power load characteristics of industrial users are highly dynamic and unpredictable. This instability is mainly due to the fact that power load is affected by many factors, including economic activities, policy adjustments, weather changes, time cycles (such as daily cycles and seasonal changes), and industry-specific factors. The interaction of these complex factors has led to huge challenges in the operation of the power system, especially in demand management and electricity price control.

[0003] In the existing technology, power demand management usually relies on traditional load forecasting techniques, such as linear regression, time series analysis and other methods. However, these methods often fail to accurately capture the complexity and dynamics of load changes, especially in the face of extreme weather events or sudden economic changes. In addition, these traditional technologies fail to effectively utilize the advantages provided by big data and machine learning technologies, and therefore have obvious deficiencies in forecasting accuracy and real-time response capabilities.

[0004] With the development of energy storage technology, using energy storage systems for demand control has become a potential solution. Energy storage systems can release energy during peak demand periods and store energy during low demand periods, thereby helping to balance grid loads and reduce electricity costs. However, how to optimize the operation strategy of energy storage systems to achieve optimal economic benefits and system stability remains a technical challenge. Existing energy storage demand control methods often lack a comprehensive consideration of the complex influencing factors of load characteristics and a refined real-time scheduling strategy. Summary of the invention

[0005] The present invention provides a demand control method, system, device and medium based on an energy storage system, the purpose of which is to solve the problems of low accuracy and insufficient responsiveness of power demand management in the prior art, and to improve the accuracy of power load forecasting and the efficiency of power cost control through optimized scheduling of the energy storage system.

[0006] To achieve the above object, the first aspect of the present invention provides a demand control method based on an energy storage system, comprising the following steps:

[0007] Collect data on power load characteristics and related influencing factors;

[0008] The collected data on power load characteristics and their related influencing factors are analyzed using principal component-weighted grey relational analysis to extract key input variables for load forecasting;

[0009] Inputting the key input variables into a long short-term memory load forecasting model based on the Attention mechanism, and outputting load forecasting data for a period of time in the future by training the long short-term memory load forecasting model;

[0010] Based on the load forecast data obtained, a month-ahead energy storage optimization model is constructed to determine the predicted maximum demand value;

[0011] According to the maximum demand value, the scheduling plan within the day is optimized on a rolling basis to perform optimal charging and discharging operations of energy storage.

[0012] Furthermore, before using the principal component-weighted grey correlation analysis method to analyze the collected data on power load characteristics and related influencing factors, it also includes preprocessing the collected data on power load characteristics and related influencing factors.

[0013] Furthermore, the principal component-weighted grey correlation analysis method is used to analyze the collected data on power load characteristics and related influencing factors to extract key input variables for load forecasting, including:

[0014] The collected data on power load characteristics and related influencing factors are averaged, and the average value of each indicator is calculated to form an averaged sequence;

[0015] Calculate the standard deviation sequence based on the mean value sequence to obtain the standard deviation of each indicator;

[0016] The data were standardized using the Z-score standardization method to generate a standardized sequence;

[0017] Calculate the correlation coefficient sequence based on the standardized sequence to determine the correlation between the data;

[0018] Solve the characteristic equation for the correlation coefficient sequence, calculate and sort the eigenvalues ​​and corresponding eigenvectors;

[0019] Sort by eigenvalues, calculate the contribution rate and cumulative contribution rate of each eigenvalue, and select the number of principal components whose cumulative contribution rate reaches a predetermined threshold;

[0020] Using the first several selected principal components, the score of each data sample on the principal component is calculated according to its eigenvector to form a new data sequence;

[0021] Grey relational analysis is performed on the new data series, and the time-weighted relation is calculated to extract the key input variables for load forecasting.

[0022] Furthermore, the method of performing grey relational analysis on the new data sequence includes:

[0023] The new data sequence and the load forecast target variable constitute a reference sequence and a comparison sequence;

[0024] Calculate the absolute value of the difference between the reference sequence and each comparison sequence to form a difference matrix;

[0025] Determine the maximum and minimum differences in the difference matrix;

[0026] According to the maximum difference and the minimum difference in the determined difference matrix, the correlation coefficient of each time point is calculated using the grey correlation coefficient formula, which is:

[0027]

[0028] Among them, W is the time-weighted correlation, ξ is the correlation coefficient sequence, T is the historical time length of the data taken, t is the historical moment, and w is the weight value;

[0029] The correlation coefficients at all time points are weighted and summed to calculate the time-weighted grey correlation degree of each input variable;

[0030] The input variables are sorted according to the calculated time-weighted grey correlation degree, and the variables with higher correlation are extracted as the key input variables for load forecasting.

[0031] Furthermore, the step of inputting the key input variables into the long short-term memory load prediction model based on the Attention mechanism includes:

[0032] Construct a long short-term memory load prediction model architecture based on the Attention mechanism, which includes the input layer, LSTM layer, Attention layer and output layer;

[0033] The extracted key input variables are organized into time series data and divided into training sets, validation sets, and test sets;

[0034] The training set is input into the long short-term memory load prediction model, the model parameters are adjusted using the back propagation algorithm, and the loss function is optimized through multiple rounds of iterations until convergence;

[0035] The Attention mechanism is applied after each LSTM layer output to calculate the weight of each time step on the final prediction result;

[0036] The time step output adjusted by the Attention mechanism is passed to the fully connected layer to generate the final load forecast data;

[0037] Using the validation set to verify the prediction accuracy of the long short-term memory load prediction model, and adjusting the model parameters as needed to obtain the trained long short-term memory load prediction model;

[0038] The test set is input into the trained long short-term memory load forecasting model to output load forecast data for a period of time in the future.

[0039] Furthermore, the steps of constructing a month-ahead energy storage optimization model based on the obtained load forecast data and determining the maximum demand value include:

[0040] Establish a month-ahead energy storage optimization model, which takes minimizing the peak demand as the objective function and includes the following constraints: energy storage system capacity, charge and discharge power limit, charge and discharge efficiency, and operating cost of energy storage equipment;

[0041] Using the optimization solution algorithm, the maximum demand peak is calculated based on the predicted load data and the parameters of the energy storage optimization model in the previous month;

[0042] By adjusting the charging and discharging strategy of the pre-month energy storage optimization model, the pre-month energy storage optimization model is used to reduce the predicted maximum demand peak and determine the final maximum demand reporting value.

[0043] Furthermore, according to the maximum demand value, the scheduling plan within the day is optimized in a rolling manner to perform the optimal charging and discharging operation of the energy storage. The specific steps include:

[0044] Based on the determined maximum demand value, an intraday dispatch model is established, which includes the time period electricity price in the power market and the operation restrictions of the energy storage system;

[0045] Use a rolling optimization strategy to update input data daily, including the latest forecast load data and electricity price information;

[0046] Through the optimization algorithm, the charging and discharging operations of the energy storage system are planned at a given time according to the power market conditions and the status of the energy storage system on that day;

[0047] Output daily charging and discharging instructions to maximize user economic benefits.

[0048] To achieve the above object, the second aspect of the present invention provides a demand control system based on an energy storage system, comprising the following modules:

[0049] A data collection module is used to collect data on power load characteristics and related influencing factors;

[0050] A data analysis module is used to analyze the collected data on power load characteristics and related influencing factors using a principal component-weighted grey correlation analysis method to extract key input variables for load forecasting;

[0051] The load forecasting module, based on the long short-term memory network of the Attention mechanism, is used to receive key input variables and output load forecast data for a period of time in the future;

[0052] The energy storage optimization module is used to build a month-ahead energy storage optimization model based on load forecast data to determine the predicted maximum demand value;

[0053] The dispatch optimization module is used to optimize the daily dispatch plan according to the maximum demand value, so as to achieve the optimal charging and discharging operation of the energy storage system.

[0054] To achieve the above objectives, the third aspect of the present invention provides an electronic device, including a processor and a memory, wherein the processor is used to implement the steps of the demand control method based on the energy storage system when executing a computer program stored in the memory.

[0055] To achieve the above objectives, a fourth aspect of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the demand control method based on the energy storage system are executed.

[0056] Beneficial effects of the present invention:

[0057] Compared with the prior art, the present invention provides a demand control method, system, device and medium based on an energy storage system. The present invention first collects data on user power load characteristics and related influencing factors. These data include historical power consumption data, weather conditions, production plans, etc. By collecting these data, the user's power consumption pattern and influencing factors can be fully understood, providing a basis for subsequent analysis and prediction. Next, the collected data is preprocessed. The preprocessing steps include removing noise and outliers, filling missing data, etc. After the data preprocessing is completed, the data is analyzed using a principal component-weighted grey correlation analysis method. Principal component analysis (PCA) can reduce the dimension of the data and extract the most important features, while weighted grey correlation analysis is used to determine the correlation between these features and the load. In this way, the key input variables for predicting the load can be determined. These key variables will be input into a long short-term memory (LSTM) load prediction model based on the attention mechanism. The LSTM model is good at processing time series data and can capture the time dependency in the data. The attention mechanism can dynamically adjust the model's attention to different time steps, thereby improving the accuracy of the prediction. Through this model, the load curve for a period of time in the future can be output. After obtaining the load forecast results, a two-stage optimization is carried out. The first stage is to optimize the maximum pre-month demand. By adjusting the charging and discharging strategy of the energy storage system, the peak demand is reduced and the demand cost is reduced. The goal of this step is to reduce electricity costs by smoothing the load curve and avoiding excessive peak demand. The second stage is to roll-optimize the intraday scheduling plan. According to real-time load data and electricity price information, the operating status of the energy storage system is dynamically adjusted to realize functions such as demand control, peak-valley arbitrage and real-time scheduling. The advantage of rolling optimization is that it can adjust the strategy in real time according to the latest data and information to ensure the optimal operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments.

[0059] Figure 1 It is a flow chart of a demand control method based on an energy storage system disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0061] According to an embodiment of the present invention, it should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the following production method, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0062] like Figure 1 As shown, the present invention provides a demand control method based on an energy storage system, comprising the following steps:

[0063] Step S100, collecting data on power load characteristics and related influencing factors;

[0064] Step S200: Analyze the collected data on power load characteristics and related influencing factors using principal component-weighted grey correlation analysis to extract key input variables for load forecasting;

[0065] Step S300: input the key input variables into the long short-term memory load forecasting model based on the Attention mechanism, and output the load forecasting data for a period of time in the future by training the long short-term memory load forecasting model;

[0066] Step S400: construct a month-ahead energy storage optimization model based on the load forecast data obtained by forecasting to determine the predicted maximum demand value;

[0067] Step S500: According to the maximum demand value, the scheduling plan within the day is optimized in a rolling manner to perform optimal charging and discharging operations of the energy storage.

[0068] In this embodiment, as described in step S100 above, the power load characteristics of industrial users show strong instability due to the influence of various emergencies and influencing factors. In order to achieve efficient and accurate prediction of power load characteristics, it is necessary not only to fully study the development law of power load characteristics themselves, but also to consider external influencing factors in all aspects, such as economic factors, power policy factors, meteorological factors, time factors, industry development factors and other random interference factors. In this step, various influencing factors related to user load and power load historical data are collected as the original data sequence, as shown in formula (1-1):

[0069]

[0070] In formula (1-1), X represents the data matrix; X ij It represents the element in the i-th row and j-th column, represents the j-th index value of the i-th sample, m is the number of indicators of power load characteristics and their influencing factors, and n is the number of a certain indicator variable.

[0071] In this embodiment, as described in step S200 above, the principal component-weighted grey relational analysis method (PAC-WGRA) can not only use the principal component analysis idea to solve the problem of complex sample structure and repeated information in the grey relational analysis method, but also use the grey relational analysis idea to quantitatively calculate the influence of each influencing factor after extracting the principal component. In addition, in order to make the analysis results more in line with the actual situation and meet the requirements of development, the data closer to the current time should account for a larger proportion, and the data farther away should be reduced in proportion, so that the analysis results cover the latest data information, so time-weighted processing is required. First, according to formula (1-2), the original data sequence is averaged to calculate the averaged sequence

[0072]

[0073] In formula (1-2), is the average value of the jth indicator.

[0074] The standard deviation sequence S is calculated using formula (1-3) based on the original data sequence and the mean value sequence.

[0075]

[0076] In formula (1-3), S j (j=1,2,…,m) is the standard deviation of the jth indicator.

[0077] Use formula (1-4) to perform Z-score standardization and calculate the standardized sequence Z.

[0078]

[0079] In formula (1-4), Z ij (i=1,2,…,n;j=1,2,…,m) is the i-th standard value of the j-th indicator.

[0080] According to the standardized sequence, the correlation coefficient sequence R is calculated using formula (1-5), where R ij (i=1,2,…,m-1; j=1,2,…,m-1).

[0081]

[0082] In formula (1-5), R is the correlation coefficient matrix between indicators, R ij is the correlation coefficient between the i-th and j-th indicators, Z represents the standardized data matrix, T is the transpose operation of the matrix, and k is the sample index.

[0083] Solve the characteristic equation |λI-R| corresponding to the correlation coefficient sequence, first calculate the eigenvalue λ j (j=1,2,…,m-1) and sort them by size; then calculate the corresponding eigenvector e j (j=1,2,…,m-1); finally, calculate the contribution rate of a certain eigenvalue according to formula (1-6), calculate the cumulative contribution rate of the eigenvalue according to formula (1-7), and judge whether the first a (a≤m-1) eigenvalues ​​are greater than 1, and whether the corresponding cumulative contribution rate exceeds 85%. If the conditions are met, it means that only the first a principal components can be used for subsequent analysis, otherwise it is necessary to reselect the analysis indicator system.

[0084]

[0085] In formula (1-6), λ j is the jth eigenvalue.

[0086]

[0087] According to formula (1-8), the coefficient l of the principal component expression is calculated, and the principal component expression is shown in formula (1-9), where e ij is the eigenvector e j The i-th component of i (i=1,2,…,a) is the score value of the first principal component.

[0088]

[0089] In formula (1-8), l ij Represents the coefficient of the i-th principal component on the j-th index in principal component analysis (PCA).

[0090]

[0091] Substitute the data into formula (1-9), calculate the score of each principal component, and redefine the comprehensive meaning of each principal component. Calculate the comprehensive score Y according to the contribution rate of each principal component, as shown in formula (1-10), where n is the number of an indicator variable Y k The j-th comprehensive score value of the principal component comprehensive index is used, and the comprehensive score is used to replace it to form a new data series.

[0092]

[0093] Carry out grey correlation analysis on the new data sequence and calculate the correlation sequence.

[0094] Separate the historical time corresponding to the historical data, and determine the fuzzy priority relationship matrix according to the principle of "near is big and far is small" Where T is the historical time length of the data, t1 and t2 are historical moments, is the comparison of the importance of a certain influencing factor index data at time t1 and time t2, and there is If t1>t2, then let This means that the data at time t1 is more important than the data at time t2. If t1<t2, then This means that the data at time t2 is more important than the data at time t1. If t1 = t2, then This means that the data at time t2 is as important as the data at time t1. According to formula (1-11), the fuzzy priority relationship matrix is ​​transformed into the fuzzy standard matrix

[0095]

[0096] In the formula, G t,q It represents the fuzzy priority relation matrix element. Is the calculation result, used to calculate is each element of the fuzzy criterion matrix.

[0097] According to formula (1-12), the weight value w of the correlation coefficient between the power load characteristics and a certain influencing factor is calculated, where the parameter

[0098]

[0099] Where T is the historical time length of the data taken.

[0100] The time-weighted correlation W is calculated according to formula (1-13), where ξ is the correlation coefficient sequence.

[0101]

[0102] Among them, W is the time-weighted correlation, ξ is the correlation coefficient sequence, T is the historical time length of the data taken, t is the historical moment, and w is the weight value.

[0103] The above time-weighted correlation is the calculation result based on PCA-WGRA quantitative analysis.

[0104] In this implementation, as described in step S300 above, it is first necessary to establish an LSTM demand forecasting model based on the Attention mechanism. The LSTM unit itself has a Gate mechanism, that is, the input gate selects to input the current information, and the forget gate selects to forget the past information, but the LSTM needs to capture the timing information step by step. As the step increases, the memory of the more distant data will gradually decay, making it difficult to retain all historical useful information. After the Attention mechanism is introduced, the hidden layer states of all steps of the LSTM will be weighted to focus on the more important hidden layer state information in the historical demand data. The long short-term memory neural network LSTM algorithm, the main steps include:

[0105] 1) After quantitative analysis by PAC-WGRA, determine the input eigenvalue x t .

[0106] 2) Build the LSTM algorithm architecture, including network node parameter settings, input gate, unit gate, forget gate, current state, output gate, and output layer design.

[0107] 2.1) Network node parameters, input node inputSize = K, K is the input feature item x t Including the total number of feature items, the preferred time step timeStep=60, the preferred hidden node hiddenSize=64, the connection layer node linearSize=64, the final output node outSize=1, and the preferred batch training size batchSize=50.

[0108] 2.2) The input gate is recorded as: Among them, net i represents the input gate network, σ represents the sigmoid activation function, x t Indicates the sample unit input value at the current moment, h t-1 Indicates the hidden state value of the previous moment, which is the hiddenSize size vector, b i Indicates that the bias of the input gate is a hiddenSize size vector, W i Represents the weight matrix corresponding to the sample input value in the input gate, with a size of [inputSize×hiddenSize], U i Represents the weight matrix corresponding to the output state value of the input gate at the previous moment, with a size of [hiddenSize×hiddenSize].

[0109] 2.3) The unit gate is recorded as: Among them, net c represents the unit gate network, b cRepresents the bias in the unit gate, which is the hiddenSize size vector, W C Represents the weight matrix corresponding to the sample input value in the unit gate, with a size of [inputSize×hiddenSize], U C Represents the weight matrix corresponding to the hidden state value of the previous moment in the unit gate, with a size of [hiddenSize×hiddenSize].

[0110] 2.4) The forget gate is recorded as: Among them, net f represents the forget gate network, b f Represents the bias of the forget gate, which is the hiddenSize size vector, W f Represents the weight matrix corresponding to the sample input value in the forget gate, with a size of [inputSize×hiddenSize], U f Represents the weight matrix corresponding to the hidden state value at the previous moment in the forget gate, with a size of [hiddenSize×hiddenSize].

[0111] 2.5) The current state is recorded as: Among them, C t-1 Indicates the historical state of the previous moment. Represents the multiplication of corresponding elements.

[0112] 2.6) The output gate is recorded as: Among them, net o represents the output gate network, b o Represents the bias of the output gate, which is the hiddenSize size vector, W o Represents the weight matrix corresponding to the sample input value in the output gate, with a size of [inputSize×hiddenSize], U o Represents the weight matrix corresponding to the hidden state value of the previous moment in the output gate, with a size of [hiddenSize×hiddenSize].

[0113] 2.7) The current hidden state is recorded as: The size is hiddenSize.

[0114] 2.8) The final output is recorded as: Among them, net y represents the output layer network, W out is the output layer weight matrix, size is [hiddenSize×outSize], b out Represents the bias of the output layer, with a size of outSize.

[0115] 2.9) The loss function is recorded as: Where: y tag Represents the label value of the training sample.

[0116] 2.10) Update weights and biases:

[0117] Output layer weights and bias updates:

[0118]

[0119] in:

[0120]

[0121] Among them, η is the learning efficiency, Is to find its partial derivative, the current hidden state is recorded as:

[0122]

[0123] In the formula, y tag Represents the label value of the training sample.

[0124] Output gate weights and bias updates:

[0125]

[0126] Where W and U matrices constitute the weight matrix, subscript O is the output gate, W processes the current data and U is used to process the data at the previous moment, that is, [W0, U0] is the output gate weight matrix, b0 is the bias matrix, Δ represents the directional derivative, and the standard update direction.

[0127] in:

[0128]

[0129] Input gate weights and bias updates:

[0130]

[0131] Where the subscript i represents the input gate.

[0132] in:

[0133]

[0134] Forget gate weights and bias updates:

[0135]

[0136] Wherein, the subscript f represents the forget gate.

[0137] in:

[0138]

[0139] Unit gate weights and bias updates:

[0140]

[0141] in:

[0142]

[0143] 3) x t represents the input sequence, h t Represents the state value of the LSTM hidden layer node, a ti is the LSTM hidden layer state feature h t The attention weight assigned to the current input, c t The state value of the last hidden layer is used as the input of the attention mechanism to analyze the temporal correlation between the historical state and the current state and assign different weights, s t Represents the state value of the last hidden layer node output by the attention mechanism. The calculation formula of the attention mechanism is as follows:

[0144]

[0145] s t =tanh(W c [C t ,h t ])

[0146] In the formula, score(h t ,h i ) represents the hidden state h used to measure the current moment t and the hidden state h at the previous moment i The correlation between them.

[0147] The load value of the next 24 hours is predicted based on the processed historical data. The daily load curve consists of 96 load data. The load information of the 60 days before 0:00 on the day to be predicted is all known. In order to make the most of the historical data and enable the model to be fully learned and trained, the iterative rolling prediction method is adopted, with the historical data of the previous day as input and the historical data of the next day as output for training the model. The model is trained to obtain the preferred prediction model, and the model with an accuracy of more than 95% in both the training set and the test set is selected as the prediction model.

[0148] In this embodiment, as described in step S400 above, for users who have installed energy storage, a monthly energy storage optimization model is constructed based on the load forecast data provided in step S300 to determine the maximum demand value to be reported for that month. With the maximum demand as a constraint, a day-ahead energy storage optimization model is constructed, with 31 days as the optimization cycle, and the cplex solver is used to optimize the charge and discharge power of the energy storage every day, which is repeated for 31 days. The maximum demand of the user for that month and the charge and discharge instructions for each day of the month are output.

[0149] The mathematical model established in the above steps. Taking the lowest monthly electricity fee paid by large industrial users as the objective function, the maximum demand value reported is determined. The electricity fee that users need to pay includes the electricity fee and the basic electricity fee, which can be specifically expressed as:

[0150] min F1=f1+f2

[0151] Where F1 is the total electricity fee for the month; f1 is the electricity cost for the month; f2 is the basic electricity cost for the month. With 15 minutes as the measurement interval, the user's electricity fee calculation formula is as follows:

[0152]

[0153] In the formula, e t is the time-of-use electricity price; T is the number of days per month; P L,i,t is the user load power at the tth moment on the i-th day; P d,i,t is the energy storage discharge power at the tth moment on the i-th day; P c,i,t is the energy storage charging power at the tth moment on the i-th day.

[0154] According to the demand billing rules, the basic monthly electricity charges are as follows:

[0155]

[0156] In the formula, c d is the unit price of the maximum demand; P NM is the demand value, i.e. the maximum demand value reported by the user; P act The actual demand.

[0157] The energy storage system needs to meet the following constraints:

[0158] 1) Energy storage state of charge (SOC) constraints

[0159] S min ≤S i,t ≤S max

[0160] In the formula, S i,t is the energy storage charge state at the tth moment on the i-th day, S min and Smax are the lower and upper limits of the state of charge, respectively.

[0161] 2) Energy storage charging and discharging state constraints

[0162] B d,i,t +B c,i,t ≤1

[0163] Where B d,i,t and B c,i,t is a variable with a value of 0 or 1, representing the discharge and charge state of the energy storage at the tth moment on the i-th day, respectively. d,i,t 1 means that the energy storage is in the discharge state at the tth moment of the i-th day, B c,i,t A value of 1 indicates that the energy storage is in a charging state at the tth moment on the i-th day.

[0164] 3) Energy storage power constraints

[0165]

[0166] Where: P dmax and P cmax are the maximum discharge and charge power of energy storage, P d,i,t is the energy storage discharge power at the tth moment on the i-th day.

[0167] 4) Energy storage charge state continuity constraints

[0168]

[0169] Where: S i,t+1 is the energy storage charge state at the tth moment on the i-th day; Δt is the charge and discharge time, which is 15 minutes; η c is the energy storage charging efficiency, η d is the energy storage discharge efficiency; E = βP, β is the battery energy rate, and P is the energy storage rated power.

[0170] 5) Peak and valley constraints

[0171] In order to prevent energy storage from being guided by peak and valley electricity prices and forming new load peaks, the peak and valley of user load are constrained as follows:

[0172] L i,min <P L,i,t -P d,i,t +P c,i,t <L i,max

[0173] Where L i,max and L i,min are the maximum and minimum values ​​of the user's power load on the i-th day; P L,i,t is the user load power at the tth moment on the i-th day, P d,i,tis the energy storage discharge power at the tth moment on the i-th day, P c,i,t is the energy storage charging power at the tth moment on the i-th day.

[0174] In this embodiment, as described in step S500, in the intraday scheduling model, maximizing user benefits is used as the objective function, as shown in the following formula:

[0175]

[0176] Where F2 is the optimized intraday user benefit; P L,t Optimize the user load power at time t within the day; P d,t To optimize the energy storage discharge power at the tth moment in the day; P c,t To optimize the energy storage charging power at time t within the day.

[0177] The constraints of this model are similar to those of the previous model, satisfying the energy storage state of charge constraint, charge and discharge limit constraint, energy storage state of charge continuity constraint, and charge and discharge state constraint. The difference is that the time subscript changes from (i, t) to t. In addition, in order to prevent the charging and discharging power of energy storage from exceeding the required charging standard and causing penalties, the following constraints must also be met:

[0178] 0≤P L,t -P d,t +P c,t ≤1.05P NM

[0179] In the scheme of this embodiment, for users with load peaks and large peak-valley price differences, under the optimized energy storage charging and discharging guidance strategy, the energy storage system automatically selects charging during the valley period and discharging during the peak period according to the peak-valley price difference, and at the same time controls the demand at the set value by reducing the load peak. Peak-valley arbitrage and demand control are achieved to ensure that the power curve is smooth and the random fluctuations are small during the static charging and discharging process. This demand control method based on the energy storage system has the following advantages: by optimizing demand and scheduling strategies, electricity expenditures can be effectively reduced; real-time scheduling and rolling optimization ensure the stable operation of the system and reduce the risks caused by load fluctuations; the method can dynamically adjust the strategy according to real-time data to adapt to different electricity usage environments and needs; the method can be applied to users of different sizes, from home users to large industrial users.

[0180] According to another aspect of the embodiment of the present application, there is also provided a demand control system based on an energy storage system, comprising the following modules:

[0181] A data collection module is used to collect data on power load characteristics and related influencing factors;

[0182] A data analysis module is used to analyze the collected data on power load characteristics and related influencing factors using a principal component-weighted grey correlation analysis method to extract key input variables for load forecasting;

[0183] The load forecasting module, based on the long short-term memory network of the Attention mechanism, is used to receive key input variables and output load forecast data for a period of time in the future;

[0184] The energy storage optimization module is used to build a month-ahead energy storage optimization model based on load forecast data to determine the predicted maximum demand value;

[0185] The dispatch optimization module is used to optimize the daily dispatch plan according to the maximum demand value, so as to achieve the optimal charging and discharging operation of the energy storage system.

[0186] According to another aspect of an embodiment of the present application, there is further provided an electronic device, including a processor and a memory, wherein the processor is configured to implement the steps of the method when executing a computer program stored in the memory.

[0187] In the above embodiments of the present invention, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0188] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. Among them, the device embodiments described above are only schematic. For example, the division of the units can be a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of units or modules, which can be electrical or other forms.

[0189] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0190] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of 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, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), mobile hard disk, magnetic disk or optical disk and other media that can store program codes.

[0191] The above is only a preferred embodiment of the present invention. It should be pointed out that for ordinary technicians in this technical field, several improvements and modifications can be made without departing from the principle of the present invention. These improvements and modifications should also be regarded as the scope of protection of the present invention.

Claims

1. A demand control method based on an energy storage system, characterized in that: The steps include: Collect data on power load characteristics and related influencing factors; The collected data on power load characteristics and their related influencing factors are analyzed using principal component-weighted grey relational analysis to extract key input variables for load forecasting; Inputting the key input variables into a long short-term memory load forecasting model based on the Attention mechanism, and outputting load forecasting data for a period of time in the future by training the long short-term memory load forecasting model; Based on the load forecast data obtained, a month-ahead energy storage optimization model is constructed to determine the predicted maximum demand value; According to the maximum demand value, the scheduling plan within the day is optimized on a rolling basis to perform optimal charging and discharging operations of energy storage.

2. The demand control method based on the energy storage system according to claim 1, characterized in that: Before using the principal component-weighted grey correlation analysis method to analyze the collected data on power load characteristics and related influencing factors, it also includes preprocessing the collected data on power load characteristics and related influencing factors.

3. The demand control method based on the energy storage system according to claim 1, characterized in that: The method of using principal component-weighted grey correlation analysis to analyze the collected data on power load characteristics and related influencing factors to extract key input variables for load forecasting includes: The collected data on power load characteristics and related influencing factors are averaged, and the average values ​​of each indicator are calculated to form an averaged sequence; Calculate the standard deviation sequence based on the mean value sequence to obtain the standard deviation of each indicator; The data were standardized using the Z-score standardization method to generate a standardized sequence; Calculate the correlation coefficient sequence based on the standardized sequence to determine the correlation between the data; Solve the characteristic equation for the correlation coefficient sequence, calculate and sort the eigenvalues ​​and corresponding eigenvectors; Sort by eigenvalues, calculate the contribution rate and cumulative contribution rate of each eigenvalue, and select the number of principal components whose cumulative contribution rate reaches a predetermined threshold; Using the first several selected principal components, the score of each data sample on the principal component is calculated according to its eigenvector to form a new data sequence; Grey relational analysis is performed on the new data series, and the time-weighted relational degree is calculated to extract the key input variables for load forecasting.

4. The demand control method based on the energy storage system according to claim 3, characterized in that: Methods for performing grey relational analysis on new data series include: The new data sequence and the load forecast target variable constitute a reference sequence and a comparison sequence; Calculate the absolute value of the difference between the reference sequence and each comparison sequence to form a difference matrix; Determine the maximum and minimum differences in the difference matrix; According to the maximum difference and the minimum difference in the determined difference matrix, the correlation coefficient of each time point is calculated using the grey correlation coefficient formula, which is: Among them, W is the time-weighted correlation, ξ is the correlation coefficient sequence, T is the historical time length of the data taken, t is the historical moment, and w is the weight value; The correlation coefficients at all time points are weighted and summed to calculate the time-weighted grey correlation degree of each input variable; The input variables are sorted according to the calculated time-weighted grey correlation degree, and the variables with higher correlation are extracted as the key input variables for load forecasting.

5. The demand control method based on the energy storage system according to claim 1, characterized in that: The step of inputting the key input variables into the long short-term memory load prediction model based on the Attention mechanism includes: Construct a long short-term memory load prediction model architecture based on the Attention mechanism, which includes the input layer, LSTM layer, Attention layer and output layer; The extracted key input variables are organized into time series data and divided into training sets, validation sets, and test sets; The training set is input into the long short-term memory load prediction model, the model parameters are adjusted using the back propagation algorithm, and the loss function is optimized through multiple rounds of iterations until convergence; The Attention mechanism is applied after each LSTM layer output to calculate the weight of each time step on the final prediction result; The time step output adjusted by the Attention mechanism is passed to the fully connected layer to generate the final load forecast data; Using the validation set to verify the prediction accuracy of the long short-term memory load prediction model, and adjusting the model parameters as needed to obtain the trained long short-term memory load prediction model; The test set is input into the trained long short-term memory load forecasting model to output load forecast data for a period of time in the future.

6. The demand control method based on the energy storage system according to claim 1, characterized in that: The steps of constructing a month-ahead energy storage optimization model based on the obtained load forecast data and determining the maximum demand value include: Establish a month-ahead energy storage optimization model, which takes minimizing the peak demand as the objective function and includes the following constraints: energy storage system capacity, charge and discharge power limit, charge and discharge efficiency, and operating cost of energy storage equipment; Using the optimization solution algorithm, the maximum demand peak is calculated based on the predicted load data and the parameters of the energy storage optimization model in the previous month; By adjusting the charging and discharging strategy of the pre-month energy storage optimization model, the pre-month energy storage optimization model is used to reduce the predicted maximum demand peak and determine the final maximum demand reporting value.

7. The demand control method based on the energy storage system according to claim 1, characterized in that: According to the maximum demand value, the daily dispatch plan is optimized to perform the optimal charging and discharging operation of energy storage. The specific steps include: Based on the determined maximum demand value, an intraday dispatch model is established, which includes the time period electricity price in the power market and the operation restrictions of the energy storage system; Use a rolling optimization strategy to update input data daily, including the latest forecast load data and electricity price information; Through the optimization algorithm, the charging and discharging operations of the energy storage system are planned at a given time according to the power market conditions and the status of the energy storage system on that day; Output daily charging and discharging instructions to maximize user economic benefits.

8. A demand control system based on an energy storage system, characterized in that: Includes the following modules: A data collection module is used to collect data on power load characteristics and related influencing factors; A data analysis module is used to analyze the collected data on power load characteristics and related influencing factors using a principal component-weighted grey correlation analysis method to extract key input variables for load forecasting; The load forecasting module, based on the long short-term memory network of the Attention mechanism, is used to receive key input variables and output load forecast data for a period of time in the future; The energy storage optimization module is used to build a month-ahead energy storage optimization model based on load forecast data to determine the predicted maximum demand value; The dispatch optimization module is used to optimize the daily dispatch plan according to the maximum demand value, so as to achieve the optimal charging and discharging operation of the energy storage system.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the processor is used to implement the steps of the demand control method based on the energy storage system as claimed in any one of claims 1 to 7 when executing the computer program stored in the memory.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the demand control method based on the energy storage system according to any one of claims 1 to 7 are executed.

Images