Energy storage system capacity configuration optimization method and device and storage medium

Abstract

The invention discloses an energy storage system capacity configuration optimization method and device and a storage medium, and relates to the technical field of energy storage system capacity configuration.The energy storage system capacity configuration optimization method comprises the following steps that daily total electricity utilization power information of a factory is continuously collected, corresponding weather information is obtained, and factory power weather data is obtained; denoising processing is carried out, an electricity consumption power prediction model is constructed, and the daily total electricity consumption power of the factory is predicted; calculating a prediction error, carrying out error rule analysis, and correcting the predicted total power consumption power to obtain corrected power prediction data; calculating predicted electricity consumption, and performing configuration optimization on the capacity of the energy storage system of the factory according to the predicted electricity consumption; the method is used for solving the problems that when an existing energy storage system capacity configuration technology configures the capacity of an energy storage system of a factory through a power consumption prediction model, the rule of prediction deviation of the model cannot be analyzed, meanwhile, a prediction result cannot be corrected, and the accuracy of capacity configuration of the energy storage system cannot be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage system capacity configuration, in particular to an energy storage system capacity configuration optimization method, device and storage medium. BACKGROUND

[0002] The energy storage system capacity configuration technology refers to a systematic technical design process of determining the energy storage capacity, charging and discharging power capacity and standby redundancy of the energy storage system to meet the safe, reliable, economic and efficient operation target and realize precise matching with the demand through technical analysis and quantitative calculation under the constraint of a specific application scenario or the whole life cycle.

[0003] The existing energy storage system capacity configuration technology often establishes a prediction model for predicting the power consumption of the factory on a future date when configuring the capacity of the energy storage system of the factory, so as to meet the demand for precise configuration of the capacity of the energy storage system. The prediction model established has high prediction accuracy when it is first used, can accurately predict the power consumption of the factory on a future date, and thus the capacity configuration of the energy storage system is also very accurate. In the peak-valley arbitrage scenario, the power consumption cost of the factory can be greatly reduced through the charging and discharging price difference. However, as the use time of the prediction model is prolonged, the prediction accuracy will gradually decrease due to production adjustment, weather change and temporary failure and other factors, and the prediction deviation will gradually increase. The technical personnel often find problems through regular review, which has obvious lag, and the prediction deviation will be discovered and adjusted by the relevant technical personnel only when it reaches a certain degree. Before being discovered, there will be a long period of time with large prediction deviation, which will inevitably lead to inaccurate capacity configuration of the energy storage system and affect the benefits of the peak-valley arbitrage of the factory. Therefore, when the existing energy storage system capacity configuration technology configures the capacity of the energy storage system of the factory through the power consumption prediction model, it cannot analyze the law of the prediction deviation of the model and correct the prediction result, so as to improve the accuracy of the capacity configuration of the energy storage system. SUMMARY

[0004] The present application aims to at least solve one of the technical problems in the prior art, by continuously collecting the total power consumption information of the factory every day, obtaining the corresponding weather information, obtaining the factory power weather data; the factory power weather data is denoised, and an electricity power prediction model is constructed to predict the total power consumption of the factory every day, and obtain the power prediction data; the prediction error is calculated according to the power prediction data, and the error law is analyzed, and the predicted total power consumption is corrected to obtain the corrected power prediction data; the predicted power consumption is calculated according to the corrected power prediction data, and the capacity of the energy storage system of the factory is configured and optimized according to the predicted power consumption, so as to solve the problem that the existing energy storage system capacity configuration technology cannot analyze the prediction deviation law of the model when configuring the capacity of the energy storage system of the factory through the power prediction model, and the prediction result is corrected, and the accuracy of the capacity configuration of the energy storage system is improved.

[0005] To achieve the above-mentioned purpose, the present application provides a kind of energy storage system capacity configuration optimization method comprising the following steps: Continuously collect the total power consumption information of the factory every day, and obtain the corresponding weather information, obtain the factory power weather data; The factory power weather data is denoised, and an electricity power prediction model is constructed to predict the total power consumption of the factory every day, and obtain the power prediction data; The prediction error is calculated according to the power prediction data, and the error law is analyzed, and the predicted total power consumption is corrected to obtain the corrected power prediction data; The predicted power consumption is calculated according to the corrected power prediction data, and the capacity of the energy storage system of the factory is configured and optimized according to the predicted power consumption.

[0006] Further, continuously collect the total power consumption information of the factory every day, and obtain the corresponding weather information, obtain the factory power weather data comprising the following sub-steps: Divide each day into k1 time periods, in turn, as time period 1-time period k1;And any one time period is recorded as the first time period;Select weather environment parameters related to the power consumption of the factory, recorded as weather parameters, and any one weather parameter is recorded as the first parameter;Any one date is recorded as the first date; The total power consumption of the factory is collected at the first time interval on the first date, and the collection time is recorded, arranged in time sequence, recorded as the total power consumption sequence of the first date;Wherein, the first time interval is t1; The average value of the first parameter in the corresponding time period is collected in each time period of the first date, and is arranged in the order of time period and recorded as the parameter change sequence of the first parameter of the first date;Repeat to obtain the parameter change sequence of all weather parameters of the first date; The total power consumption sequence for the first date and the parameter change sequence of all weather parameters are recorded as the factory power weather information for the first date; Repeatedly collect factory power weather information for each date, and record it as factory power weather data.

[0007] Furthermore, the factory's power weather data is denoised, and a power consumption prediction model is constructed to predict the factory's total daily power consumption. The resulting power consumption prediction data includes the following sub-steps: The total power consumption sequence for the first date is denoted as the first power sequence; the first power sequence is evenly divided into multiple subsequences of length k1, and any one of the subsequences is denoted as the first subsequence, where k1 is the set length; Calculate the standard deviation of the first subsequence, denoted as the fluctuation intensity AB of the first subsequence, and calculate the absolute difference between any two adjacent data in the first subsequence, denoted as the adjacent absolute difference, and denote the maximum adjacent absolute difference as the peak value of the change of the first subsequence AC; Repeatedly obtain the fluctuation intensity and peak value of all subsequences of the first power sequence, and arrange them in ascending order to be denoted as the fluctuation intensity sequence and peak value sequence of the first power sequence; The fluctuation intensity sequence is divided into two clusters using density clustering algorithm, and the maximum value in the cluster with the smaller value is taken as the fluctuation threshold; the change peak sequence is also divided into two clusters using density clustering algorithm, and the maximum value in the cluster with the smaller value is taken as the change threshold. Obtain the position number of the fluctuation threshold in the fluctuation intensity sequence, denoted as BM0, and obtain the position number of the change threshold in the change peak sequence, denoted as CM0. Then, BM0+CM0 is denoted as the judgment threshold AM0.

[0008] Furthermore, the factory's power weather data is denoised, and a power consumption prediction model is constructed to predict the factory's total daily power consumption. The resulting power consumption prediction data also includes the following sub-steps: Obtain the position indices of the fluctuation intensity and change peak of the first subsequence in the corresponding fluctuation intensity sequence and change peak sequence, and sum them up, denoted as the index and DM of the first subsequence; If DM is not greater than AM0, the first subsequence is marked as a stable subsequence; if DM is greater than AM0, the first subsequence is marked as a mutant subsequence. The marking of all subsequences in the first power sequence is repeated, and adjacent mutant subsequences or adjacent stable subsequences are merged and recorded as mutant segments and stable segments, respectively. Denote any stable segment in the first power sequence as the first stable segment; set the size of the second sliding window to k2 and the sliding step size to 1, and slide it sequentially from the starting position of the first stable segment. For each sliding window, sort the data in the window by size, obtain the median, and then replace the data at the center position of the window with the median; repeat the operation to obtain the corresponding denoised stable segment; repeat to obtain the denoised stable segments of all stable segments in the first power sequence; For each abrupt change segment in the first power sequence, wavelet transform is used for denoising to obtain the corresponding denoised abrupt change segment; and it is then merged with all denoised stable segments to obtain the denoised power sequence for the first date. Linear fitting is performed on the total power consumption in the first time period of the denoised power sequence, and then the average total power consumption in the first time period is obtained. The average total power consumption of all time periods in the denoised power sequence is obtained repeatedly and arranged in chronological order, and recorded as the average power sequence of the first date. The average power sequence for each date in the factory power weather data is repeatedly obtained and merged with the factory power weather information. The data is then sorted by date and recorded as the denoised power weather data.

[0009] Furthermore, the factory's power weather data is denoised, and a power consumption prediction model is constructed to predict the factory's total daily power consumption. The resulting power consumption prediction data also includes the following sub-steps: An initial model is built based on an LSTM model, which includes an input layer, a hidden layer, and an output layer. The output of the initial model is set to the average total power consumption for each future time period. The initial model is trained using denoised power weather data to obtain a power prediction model. Based on historical denoised power weather data, input the power prediction model to obtain the average total power consumption for each time period of future dates, which is recorded as the predicted power sequence for the corresponding dates. The predicted power sequence for each date is collected repeatedly and recorded as the electric power prediction data.

[0010] Furthermore, the prediction error is calculated based on the power prediction data, and the error pattern is analyzed. The predicted total power consumption is then corrected to obtain the corrected power prediction data, which includes the following sub-steps: Based on the average power sequence of the predicted power sequence for the first date, obtain the predicted average total power consumption and the actual average total power consumption for the first time period, respectively; denoted as YP and SP in sequence; and (SP-YP) is recorded as the prediction deviation for the first time period. Repeat the acquisition of prediction deviations for all time periods of the first date, and arrange them in chronological order, recording them as the prediction deviation sequence for the first date; repeat the acquisition of prediction deviation sequences for each date. When it is necessary to configure the capacity of the energy storage system, the predicted power sequence for future dates is obtained by using the power prediction model, and it is denoted as the future power sequence. Obtain the most recent k3 dates, denoted as the most recent reference dates. Based on the prediction deviation sequence of each most recent reference date, obtain the prediction deviation of each most recent reference date in the first time period, and sort them in chronological order from oldest to most recent, denoted as the reference deviation sequence for the first time period; where k3 is the set number of dates. The reference deviation sequence of the first time period is evenly divided into segments of length k4, which are denoted as segment 1 to segment k4. Any two segments are combined and denoted as segment combination. The two segments of any segment combination are denoted as segment i and segment j, i∈[1,k4], j∈[1,k4]. Calculate the absolute difference between the data at the same position in segments i and j, sum all the absolute differences, and denot this sum as the absolute difference sum DY(ij) between segments i and j; calculate the average values ​​of segments i and j respectively, sum them, and denot this sum as the average value sum DP(ij) between segments i and j; and calculate the average deviation ratio WB(ij) of the corresponding segment combination, where WB(ij) = DY(ij) / DP(ij); Repeatedly obtain the average deviation ratio of all segment combinations, and calculate the similarity combination for segment combinations with an average deviation ratio not greater than k5. Count the proportion of the number of similar combinations to the total number of segment combinations, and record it as the similarity combination ratio. If the similarity combination ratio is not less than k6, mark the first time period as the deviation regularity period; otherwise, mark the first time period as the deviation randomness period. Here, k5 and k6 are the set threshold ratios.

[0011] Furthermore, the process of calculating the prediction error based on the power prediction data, analyzing the error pattern, and correcting the predicted total power consumption to obtain the corrected power prediction data also includes the following sub-steps: If the first time period is a period of deviation pattern, then any predicted deviation in the reference deviation sequence of the first time period is denoted as RC(m), where m represents the position number; and the sum of all position numbers in the reference deviation sequence of the first time period is calculated and denoted as N0. Calculate the weighted deviation QC(m) of RC(m), where QC(m) = (m / N0) * RC(m); repeatedly calculate the weighted deviation of all predicted deviations in the reference deviation sequence of the first time period, and sum them up, denoted as the corrected deviation XW of the first time period; repeatedly obtain the corrected deviation for all deviation pattern periods; The average total power consumption for the first period in the future power sequence is denoted as YW; the predicted average total power consumption for the first period is replaced by (YW+XW); The average total power consumption for all deviation periods in the future power sequence is repeatedly replaced to obtain the future corrected power sequence, which is denoted as the corrected power prediction data.

[0012] Furthermore, the process of calculating the predicted electricity consumption based on the corrected power forecast data and optimizing the configuration of the factory's energy storage system capacity based on the predicted electricity consumption includes the following sub-steps: Record the date corresponding to the future revised power sequence as the future date, obtain the peak, normal and valley periods of the electricity price on the future date; and calculate the electricity consumption during the peak, normal and valley periods on the future date according to the future revised power sequence; record them in order as peak predicted electricity consumption, normal predicted electricity consumption and valley predicted electricity consumption. The energy storage capacity of the factory's energy storage system is adjusted based on the peak and flat electricity consumption forecasts.

[0013] Secondly, this application provides an electronic device including a processor and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the steps in the method described above are performed.

[0014] Thirdly, this application provides a storage medium on which a computer program is stored, which, when executed by a processor, performs the steps of the method described above.

[0015] The beneficial effects of this invention are as follows: This invention continuously collects the total daily power consumption information of a factory and obtains corresponding weather information to obtain factory power-weather data; it performs noise reduction processing on the factory power-weather data and constructs a power consumption prediction model to predict the factory's total daily power consumption, obtaining power prediction data; it calculates the prediction error based on the power prediction data, analyzes the error law, and corrects the predicted total power consumption to obtain corrected power prediction data; it calculates the predicted electricity consumption based on the corrected power prediction data and optimizes the configuration of the factory's energy storage system capacity based on the predicted electricity consumption; when configuring the capacity of the factory's energy storage system through the power consumption prediction model, it can analyze the law of prediction deviation of the model and correct the prediction results, thereby improving the accuracy of the energy storage system capacity configuration and improving the factory's peak-valley arbitrage benefits; This invention divides the daily total power consumption sequence into subsequences and uses fluctuation intensity and peak value combined with density clustering to identify stable and abrupt segments for segmentation. For stable segments, it effectively removes random noise and isolated outliers while preserving long-term trends and periodic characteristics. For abrupt segments, it better preserves the edges of abrupt changes, reducing data noise without losing the characteristic information of the abrupt changes and avoiding data distortion caused by excessive smoothing. The prediction deviations for each period of the most recent k3 days are arranged in chronological order to form a reference deviation sequence, and the existence of deviation patterns is determined by segmentation and combination to avoid overcorrection of inherently random errors and prevent the introduction of new deviations. The deviations are weighted and summed to avoid treating historical deviations equally and to emphasize recent information, thereby improving the correction effect. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of the method of the present invention; Figure 2 This is a flowchart of the stable subsequence determination process of the present invention; Figure 3 This is a flowchart of the process for obtaining the correction deviation of the present invention; Figure 4 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0017] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0018] Example 1, please refer to Figure 1 As shown, this application provides a method for optimizing the capacity configuration of an energy storage system, including the following steps: Step S1 involves continuously collecting the factory's daily total power consumption information and obtaining corresponding weather information to obtain the factory's power-weather data. Step S1 includes the following sub-steps: Step S101: Divide each day into k1 time periods, denoted sequentially as period 1 to period k1; and designate any one of these time periods as the first period; select weather environmental parameters related to the factory's power consumption, denoted as weather parameters, and designate any one of these weather parameters as the first parameter; designate any one of these dates as the first date; the weather environmental parameters related to the factory's power consumption can be selected according to the actual application scenario, such as temperature, humidity, wind speed, and rainfall; k1 can be flexibly set according to the actual application scenario. In this embodiment, k1=24, that is, one hour is one time period; the energy storage capacity is usually determined based on peak, flat, and valley periods or several business periods. The segmentation can directly correspond to these decision moments, so that the subsequent prediction output can be directly used for capacity calculation; Step S102: Collect the total power consumption of the factory at a first time interval on the first date and record the collection time. Arrange the data in chronological order and record it as the total power consumption sequence of the first date. The first time interval is t1. In this embodiment, t1 = 0.5s, which can be flexibly set. Step S103: Collect the average value of the first parameter in each time period of the first date, such as average temperature, average wind speed and average rainfall, and arrange them in the order of time periods to record the parameter change sequence of the first parameter on the first date; repeat to obtain the parameter change sequence of all weather parameters on the first date. Step S104: Record the total power consumption sequence of the first date and the parameter change sequence of all weather parameters as the factory power weather information for the first date; Step S105: Repeatedly collect the factory power weather information for each day and record it as factory power weather data; use it as samples for subsequent model training. In practice, weather often affects industrial loads; for example, temperature directly affects the cooling and heating loads; humidity affects dehumidification loads and process conditions; and wind speed may indirectly affect external equipment or cooling efficiency. Using these weather parameters as inputs can improve the accuracy of later forecasts.

[0019] Step S2 involves denoising the factory's power weather data and constructing a power consumption prediction model to predict the factory's total daily power consumption, thus obtaining power consumption prediction data. Step S2 includes the following sub-steps: For step S201, please refer to... Figure 2 As shown, the total power consumption sequence for the first date is denoted as the first power sequence. The first power sequence is evenly divided into multiple subsequences of length k1, and any one of the subsequences is denoted as the first subsequence, where k1 is the set length. In this embodiment, k1 = 120 data points. Compared with the overall unified processing, short-segment statistics can reflect the behavior of the load at different time scales within the day, which is convenient for local adaptive noise reduction. Step S202: Calculate the standard deviation of the first subsequence, denoted as the fluctuation intensity AB of the first subsequence, and calculate the absolute difference between any two adjacent data in the first subsequence, denoted as the adjacent absolute difference. The largest adjacent absolute difference is denoted as the peak value of the change in the first subsequence AC. The standard deviation reflects the overall noise level within the subsequence, while the largest adjacent difference reflects the abrupt change characteristics within the subsequence. The combination of the two helps to distinguish between the true abrupt change segment and the stable segment. Step S203: Repeatedly obtain the fluctuation intensity and change peak value of all subsequences of the first power sequence, and arrange them in ascending order as the fluctuation intensity sequence and change peak value sequence of the first power sequence; Step S204: Divide the fluctuation intensity sequence into two clusters using a density clustering algorithm, and obtain the maximum value in the cluster with the smaller value, which is recorded as the fluctuation threshold; and divide the change peak sequence into two clusters using the density clustering algorithm, and obtain the maximum value in the cluster with the smaller value, which is recorded as the change threshold; the fluctuation threshold and change threshold are automatically determined by the distribution of the data itself, which can avoid the problem of manually set thresholds being prone to failure; the cluster with the smaller value represents a small value cluster that is mostly stable and normal, and the maximum value is the upper limit of this normal state. Exceeding this value indicates entering the range of high fluctuation and sudden change.

[0020] Step S205: Obtain the position number of the fluctuation threshold in the fluctuation intensity sequence, denoted as BM0, and obtain the position number of the change threshold in the change peak sequence, denoted as CM0. BM0+CM0 is denoted as the judgment threshold AM0. By merging the position numbers of the two thresholds into AM0, the division criteria of the stable segment and the abrupt segment can be quantified into a judgment threshold, which is convenient for subsequent division of each subsequence using the same standard. Step S206: Obtain the position numbers of the fluctuation intensity and change peak of the first subsequence in the corresponding fluctuation intensity sequence and change peak sequence, and sum them up, denoted as the first subsequence number and DM; DM combines the ranking information of the two indicators into a single score to avoid misjudgment by a single indicator. Step S207: If DM is not greater than AM0, the first subsequence is marked as a stable subsequence; if DM is greater than AM0, the first subsequence is marked as a mutant subsequence. Repeat the marking of all subsequences in the first power sequence and merge adjacent mutant subsequences or adjacent stable subsequences, and record them as mutant segments and stable segments respectively. For example, if several adjacent subsequences are marked as mutant segments, the whole should be regarded as a mutant segment.

[0021] Step S208: Denote any stable segment in the first power sequence as the first stable segment; set the size of the second sliding window to k2 and the sliding step size to 1, and slide it sequentially from the starting position of the first stable segment. For each sliding window, sort the data in the window by size, obtain the median, and then replace the data at the center position of the window with the median; repeat the operation to obtain the corresponding denoised stable segment; repeatedly obtain the denoised stable segments of all stable segments in the first power sequence; in this embodiment, k2=5, which can be flexibly set; the stable segment is random noise with a flat data trend. Median filtering is highly robust to isolated abrupt changes and will not be skewed by extreme values ​​like mean filtering, making it suitable for stable segments with stable but random spikes; Step S209: For each abrupt change segment in the first power sequence, wavelet transform is used for denoising to obtain the corresponding denoised abrupt change segment; and it is merged with all denoised stable segments to obtain the denoised power sequence of the first date; the noise of the abrupt change segment is superimposed on the information data, and the abrupt change characteristics of the data need to be preserved during denoising. Wavelet transform can remove noise while preserving abrupt change characteristics; the selection of wavelet basis and the number of decomposition layers of wavelet transform need to be matched with the sampling frequency and signal bandwidth, and can be flexibly set according to the actual application scenario; Step S210: Linearly fit the total power consumption in the first time period of the denoised power sequence, and then obtain the average total power consumption in the first time period. Repeat this process to obtain the average total power consumption of all time periods in the denoised power sequence, and arrange them in chronological order, denoised as the average power sequence of the first date. The collected weather data is the average value of the time period, and the subsequent modeling uses the average power of each time period as the output target. Therefore, the denoised time series needs to be aggregated to the divided time period scale. Linear fitting can remove weak trends or short-term drifts within the time period, making the obtained average value more accurate. Step S211: Repeatedly obtain the average power sequence for each date in the factory power weather data, merge it with the factory power weather information, sort it by date, and record it as the denoised power weather data; weather parameters usually have multiple accurate acquisition channels that can be mutually verified, and the obtained weather parameters are usually very accurate, so there is no need to perform denoising processing.

[0022] Step S212: Construct an initial model based on the LSTM model. The initial model includes an input layer, a hidden layer, and an output layer. Set the output of the initial model to the average total power consumption for each future time period. Train the initial model using denoised power weather data to obtain the power prediction model. The denoised data reduces model overfitting and gradient fluctuations caused by noise, making the training more robust. Step S213: Based on historical denoised power weather data, input the power prediction model to obtain the average total power consumption for each time period of future dates, which is recorded as the predicted power sequence for the corresponding date; the input may include the average power of the denoised time period of the past several days and the weather parameters of the same time period, as well as the weather parameters of the future time period; Step S214: Repeatedly collect the predicted power sequence for each date, and record it as power prediction data; In practical implementation, traditional global single denoising algorithms, such as full-sequence FIR filtering, are prone to incomplete noise removal in stable segments or excessive smoothing of features in abrupt segments. By dividing the sequence into segments based on its own features, matching suitable denoising algorithms to the two types of segments, and finally merging them, it is possible to remove noise while preserving features as much as possible.

[0023] Step S3 involves calculating the prediction error based on the predicted power data, analyzing the error pattern, and correcting the predicted total power consumption to obtain corrected power prediction data. Step S3 includes the following sub-steps: Step S301: Based on the average power sequence of the predicted power sequence for the first date, obtain the predicted average total power consumption and the actual average total power consumption for the first time period, respectively; denoted as YP and SP in sequence; and (SP-YP) is recorded as the prediction deviation for the first time period. Repeatedly obtain the prediction deviations for all time periods of the first date, and arrange them in chronological order, recording them as the prediction deviation sequence for the first date; repeat the prediction deviation sequence for each date; as the raw data for subsequent deviation pattern analysis. Step S302: When it is necessary to configure the capacity of the energy storage system, the predicted power sequence for future dates is obtained by using the power prediction model, and recorded as the future power sequence. For step S303, please refer to... Figure 3 As shown, the most recent k3 dates are obtained and denoted as the most recent reference dates. Based on the prediction deviation sequence of each most recent reference date, the prediction deviation of each most recent reference date in the first time period is obtained and sorted in chronological order from oldest to newest, and denoted as the reference deviation sequence of the first time period. Here, k3 is the number of dates set. In this embodiment, k3=30, which can be adjusted flexibly; however, it should not be too large, as a large number may contain outdated information. Step S304: The reference deviation sequence of the first time period is evenly divided into segments of length k4, which are denoted as segment 1 to segment k4. Any two segments are combined, which is called a segment combination. The two segments of any segment combination are denoted as segment i and segment j, i∈[1,k4], j∈[1,k4]. The purpose of segmenting and combining the sequence is to detect whether the deviation has a stable shape or proportional relationship, rather than judging it based on single-point similarity. If many segment combinations are statistically similar, it indicates that there is a repeatable pattern inside the deviation sequence. In this embodiment, k4=6 data points, which can be flexibly adjusted. Step S305: Calculate the absolute difference between the data at the same position in segments i and j, sum all the absolute differences, and denot this as the sum of the corresponding absolute differences DY(ij) for segments i and j; calculate the average values ​​for segments i and j respectively, sum them, and denot this as the sum of the average values ​​DP(ij) for segments i and j; and calculate the average deviation ratio WB(ij) for the corresponding segment combination, where WB(ij) = DY(ij) / DP(ij); DY is the accumulation of absolute error, and DP incorporates the scale into the denominator, making WB a relative deviation measure; this allows for the comparison of small-amplitude but relatively consistent sequences and large-amplitude consistent sequences under the same dimension. Step S306: Repeatedly obtain the average deviation ratio of all segment combinations, and calculate the similarity combination for segment combinations with an average deviation ratio not greater than k5. Count the proportion of similar combinations to the total number of segment combinations, and record it as the similarity combination ratio. If the similarity combination ratio is not less than k6, mark the first time period as the deviation regularity period; otherwise, mark the first time period as the deviation randomness period. Here, k5 and k6 are set threshold ratios. A small WB indicates that the two segments are similar in shape and scale, and a large WB indicates that they are not similar. By statistically analyzing the WB distribution of all combinations, it can be determined whether there is a repetitive structure in the deviation sequence. In this embodiment, k5=0.25 and k6=0.6, which can be adjusted according to the actual application scenario.

[0024] Step S307: If the first time period is a deviation pattern period, then any predicted deviation in the reference deviation sequence of the first time period is denoted as RC(m), where m represents the position number; and the sum of all position numbers in the reference deviation sequence of the first time period is calculated and denoted as N0; as the weight benchmark. Step S308: Calculate the weighted deviation QC(m) of RC(m), where QC(m) = (m / N0) * RC(m); Repeatedly calculate the weighted deviation of all predicted deviations in the reference deviation sequence of the first time period, and sum them up, denoted as the corrected deviation XW of the first time period; Repeatedly obtain the corrected deviation of all deviation pattern periods; The larger m is, the closer it is to the historical day of the current day, and the higher the weight; Give more weight to recent dates to reflect that the recent deviation is more representative of the upcoming deviation trend; Step S309: For the average total power consumption of the first time period in the future power sequence, denoted as YW; replace the predicted average total power consumption of the first time period with (YW+XW); Step S310: Repeat the replacement of the average total power consumption for all deviation periods in the future power sequence. After completion, the future corrected power sequence is obtained, which is recorded as the corrected power prediction data. The deviation in the random deviation period is random and can be without pattern, so it is not processed. If it is forcibly processed, it may introduce new deviations and cause errors in the results. In practice, directly adding the weighted sum of historical systematic deviations to the prediction can compensate for the model's persistent deviations during that period, thereby making the peak, flat, and valley energy calculated by time period more accurate, and thus making energy storage configuration more economical and reliable.

[0025] Step S4 involves calculating the predicted electricity consumption based on the corrected power prediction data, and optimizing the configuration of the factory's energy storage system capacity based on the predicted electricity consumption. Step S4 includes the following sub-steps: Step S401: Record the date corresponding to the future corrected power sequence as the future date, obtain the peak period, normal period and valley period of the electricity price on the future date; and calculate the electricity consumption during the peak period, normal period and valley period of the future date according to the future corrected power sequence; and record them in order as peak predicted electricity consumption, normal predicted electricity consumption and valley predicted electricity consumption. Step S402, and adjust the energy storage capacity of the factory's energy storage system according to the peak and flat electricity consumption forecasts; so that the energy stored in the energy storage system can meet the electricity consumption during peak periods. In the specific implementation process, other algorithms can also be used to construct a mathematical model and capacity model of the energy storage system, and then the optimal capacity of the energy storage system can be determined based on the predicted electricity consumption, so as to achieve a balance between economy, technology and safety throughout the entire life cycle and improve the benefits of peak-valley arbitrage.

[0026] Example 2, please refer to Figure 4 As shown, Figure 4A schematic diagram of an electronic device is provided, which may include a processor, a communication interface, a memory, and a communication bus. The processor, communication interface, and memory communicate with each other via the communication bus. The memory stores computer-readable instructions, and the processor can call these instructions. When the processor executes a computer-readable instruction, it performs steps as described in an energy storage system capacity configuration optimization method to achieve the following functions: continuously collecting daily total power consumption information of the factory and obtaining corresponding weather information to obtain factory power-weather data; denoising the factory power-weather data and constructing a power consumption prediction model to predict the factory's daily total power consumption, obtaining power prediction data; calculating the prediction error based on the power prediction data, analyzing the error pattern, and correcting the predicted total power consumption to obtain corrected power prediction data; calculating the predicted electricity consumption based on the corrected power prediction data, and optimizing the configuration of the factory's energy storage system capacity based on the predicted electricity consumption.

[0027] Furthermore, the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, and can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0028] Example 3: This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs the steps of a method for optimizing the capacity configuration of an energy storage system to achieve the following functions: continuously collecting the total daily power consumption information of a factory and obtaining corresponding weather information to obtain factory power weather data; denoising the factory power weather data and constructing a power consumption prediction model to predict the total daily power consumption of the factory to obtain power prediction data; calculating the prediction error based on the power prediction data, analyzing the error pattern, and correcting the predicted total power consumption to obtain corrected power prediction data; calculating the predicted electricity consumption based on the corrected power prediction data, and optimizing the capacity configuration of the factory's energy storage system based on the predicted electricity consumption.

[0029] Based on the above description of the embodiments, the embodiments of the present invention can be provided as methods, systems, or computer program products. Based on this understanding, the above technical solutions, in essence or in terms of their contribution to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or certain parts of the embodiments.

[0030] In the embodiments provided in this application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely illustrative. For example, the division of modules or units is only a logical functional division, and there may be other division methods in actual implementation. Furthermore, multiple modules or units may be combined or integrated into another system, or some features may be ignored or not executed. Additionally, the coupling or direct coupling or communication connection shown or discussed may be through some communication interfaces. The indirect coupling or communication connection between systems, modules, and units may be electrical, mechanical, or other forms.

[0031] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application.

Claims

1. A method for optimizing the capacity configuration of an energy storage system, characterized in that, Includes the following steps: Continuously collect the factory's total daily power consumption information and obtain the corresponding weather information to obtain the factory's power weather data; The factory's power weather data is denoised, and a power consumption prediction model is constructed to predict the factory's total daily power consumption, thus obtaining power consumption prediction data. The prediction error is calculated based on the power prediction data, the error law is analyzed, and the predicted total power consumption is corrected to obtain the corrected power prediction data. The predicted electricity consumption is calculated based on the corrected power forecast data, and the capacity of the factory's energy storage system is configured and optimized based on the predicted electricity consumption.

2. The energy storage system capacity configuration optimization method according to claim 1, characterized in that, Continuously collect the factory's daily total power consumption information and obtain corresponding weather information to obtain the factory's power-weather data, which includes the following sub-steps: Divide each day into k1 time periods, which are sequentially denoted as time period 1 to time period k1; and denote any one of these time periods as the first time period; select weather and environmental parameters related to the factory's power consumption, and denote them as weather parameters, and denote any one of these weather parameters as the first parameter; and denote any one of these dates as the first date; The total power consumption of the factory is collected at a first time interval on the first date, and the collection time is recorded. The data is arranged in chronological order and recorded as the total power consumption sequence for the first date; where the first time interval is t1. The average value of the first parameter is collected for each time period on the first date, and the values ​​are arranged in the order of the time periods to form the parameter change sequence of the first parameter on the first date; the parameter change sequences of all weather parameters on the first date are collected repeatedly. The total power consumption sequence for the first date and the parameter change sequence of all weather parameters are recorded as the factory power weather information for the first date; Repeatedly collect factory power weather information for each date, and record it as factory power weather data.

3. The energy storage system capacity configuration optimization method according to claim 2, characterized in that, The process involves denoising the factory's power weather data and constructing a power consumption prediction model to predict the factory's total daily power consumption. The resulting power consumption prediction data includes the following sub-steps: The total power consumption sequence for the first date is denoted as the first power sequence; the first power sequence is evenly divided into multiple subsequences of length k1, and any one of the subsequences is denoted as the first subsequence, where k1 is the set length; Calculate the standard deviation of the first subsequence, denoted as the fluctuation intensity AB of the first subsequence, and calculate the absolute difference between any two adjacent data in the first subsequence, denoted as the adjacent absolute difference, and denote the maximum adjacent absolute difference as the peak value of the change of the first subsequence AC; Repeatedly obtain the fluctuation intensity and peak value of all subsequences of the first power sequence, and arrange them in ascending order to be denoted as the fluctuation intensity sequence and peak value sequence of the first power sequence; The fluctuation intensity sequence is divided into two clusters using density clustering algorithm, and the maximum value in the cluster with the smaller value is taken as the fluctuation threshold; the change peak sequence is also divided into two clusters using density clustering algorithm, and the maximum value in the cluster with the smaller value is taken as the change threshold. Obtain the position number of the fluctuation threshold in the fluctuation intensity sequence, denoted as BM0, and obtain the position number of the change threshold in the change peak sequence, denoted as CM0. Then, BM0+CM0 is denoted as the judgment threshold AM0.

4. The energy storage system capacity configuration optimization method according to claim 3, characterized in that, The process of denoising the factory's power weather data and constructing a power consumption prediction model to predict the factory's total daily power consumption, resulting in power consumption prediction data, also includes the following sub-steps: Obtain the position indices of the fluctuation intensity and change peak of the first subsequence in the corresponding fluctuation intensity sequence and change peak sequence, and sum them up, denoted as the index and DM of the first subsequence; If DM is not greater than AM0, the first subsequence is marked as a stable subsequence; if DM is greater than AM0, the first subsequence is marked as a mutant subsequence. The marking of all subsequences in the first power sequence is repeated, and adjacent mutant subsequences or adjacent stable subsequences are merged and recorded as mutant segments and stable segments, respectively. Denote any stable segment in the first power sequence as the first stable segment; set the size of the second sliding window to k2 and the sliding step size to 1, and slide it sequentially from the starting position of the first stable segment. For each sliding window, sort the data in the window by size, obtain the median, and then replace the data at the center position of the window with the median; repeat the operation to obtain the corresponding denoised stable segment; repeat to obtain the denoised stable segments of all stable segments in the first power sequence; For each abrupt change segment in the first power sequence, wavelet transform is used for denoising to obtain the corresponding denoised abrupt change segment; and it is then merged with all denoised stable segments to obtain the denoised power sequence for the first date. Linear fitting is performed on the total power consumption in the first time period of the denoised power sequence, and then the average total power consumption in the first time period is obtained. The average total power consumption of all time periods in the denoised power sequence is obtained repeatedly and arranged in chronological order, and recorded as the average power sequence of the first date. The average power sequence for each date in the factory power weather data is repeatedly obtained and merged with the factory power weather information. The data is then sorted by date and recorded as the denoised power weather data.

5. The energy storage system capacity configuration optimization method according to claim 4, characterized in that, The process of denoising the factory's power weather data and constructing a power consumption prediction model to predict the factory's total daily power consumption, resulting in power consumption prediction data, also includes the following sub-steps: An initial model is built based on an LSTM model, which includes an input layer, a hidden layer, and an output layer. The output of the initial model is set to the average total power consumption for each future time period. The initial model is trained using denoised power weather data to obtain a power prediction model. Based on historical denoised power weather data, input the power prediction model to obtain the average total power consumption for each time period of future dates, which is recorded as the predicted power sequence for the corresponding dates. The predicted power sequence for each date is collected repeatedly and recorded as the electric power prediction data.

6. The energy storage system capacity configuration optimization method according to claim 5, characterized in that, The prediction error is calculated based on the power prediction data, and the error pattern is analyzed. The predicted total power consumption is then corrected to obtain the corrected power prediction data, which includes the following sub-steps: Based on the average power sequence of the predicted power sequence for the first date, obtain the predicted average total power consumption and the actual average total power consumption for the first time period, respectively; denoted as YP and SP in sequence; and (SP-YP) is denoted as the prediction deviation for the first time period. Repeat the acquisition of the prediction deviation for all time periods of the first date, and arrange them in chronological order, denoted as the prediction deviation sequence for the first date; repeat the acquisition of the prediction deviation sequence for each date. When it is necessary to configure the capacity of the energy storage system, the predicted power sequence for future dates is obtained by using the power prediction model, and it is denoted as the future power sequence. Obtain the most recent k3 dates, denoted as the most recent reference dates. Based on the prediction deviation sequence of each most recent reference date, obtain the prediction deviation of each most recent reference date in the first time period, and sort them in chronological order from oldest to most recent, denoted as the reference deviation sequence for the first time period; where k3 is the set number of dates. The reference deviation sequence of the first time period is evenly divided into segments of length k4, which are denoted as segment 1 to segment k4. Any two segments are combined and denoted as segment combination. The two segments of any segment combination are denoted as segment i and segment j, i∈[1,k4], j∈[1,k4]. Calculate the absolute difference between the data at the same position in segments i and j, sum all the absolute differences, and denot this sum as the absolute difference sum DY(ij) between segments i and j; calculate the average values ​​of segments i and j respectively, sum them, and denot this sum as the average value sum DP(ij) between segments i and j; and calculate the average deviation ratio WB(ij) of the corresponding segment combination, where WB(ij) = DY(ij) / DP(ij); Repeatedly obtain the average deviation ratio of all segment combinations, and calculate the similarity combination for segment combinations with an average deviation ratio not greater than k5. Count the proportion of the number of similar combinations to the total number of segment combinations, and record it as the similarity combination ratio. If the similarity combination ratio is not less than k6, mark the first time period as the deviation regularity period; otherwise, mark the first time period as the deviation randomness period. Here, k5 and k6 are the set threshold ratios.

7. The energy storage system capacity configuration optimization method according to claim 6, characterized in that, The process of calculating the prediction error based on the predicted power data, analyzing the error pattern, and correcting the predicted total power consumption to obtain the corrected power prediction data also includes the following sub-steps: If the first time period is a period of deviation pattern, then any predicted deviation in the reference deviation sequence of the first time period is denoted as RC(m), where m represents the position number; and the sum of all position numbers in the reference deviation sequence of the first time period is calculated and denoted as N0. Calculate the weighted deviation QC(m) of RC(m), where QC(m) = (m / N0) * RC(m); repeatedly calculate the weighted deviation of all predicted deviations in the reference deviation sequence of the first time period, and sum them up, denoted as the corrected deviation XW of the first time period; repeatedly obtain the corrected deviation for all deviation pattern periods; The average total power consumption for the first period in the future power sequence is denoted as YW; the predicted average total power consumption for the first period is replaced by (YW+XW); The average total power consumption for all deviation periods in the future power sequence is repeatedly replaced to obtain the future corrected power sequence, which is denoted as the corrected power prediction data.

8. The energy storage system capacity configuration optimization method according to claim 7, characterized in that, Calculating the predicted electricity consumption based on the corrected power forecast data, and optimizing the configuration of the factory's energy storage system capacity based on the predicted electricity consumption, includes the following sub-steps: Record the date corresponding to the future revised power sequence as the future date, obtain the peak, normal and valley periods of the electricity price on the future date; and calculate the electricity consumption during the peak, normal and valley periods on the future date according to the future revised power sequence; record them in order as peak predicted electricity consumption, normal predicted electricity consumption and valley predicted electricity consumption. The energy storage capacity of the factory's energy storage system is adjusted based on the peak and flat electricity consumption forecasts.

9. An electronic device, characterized in that, It includes a processor and a memory, the memory storing computer-readable instructions that, when executed by the processor, perform the steps of the method as described in any one of claims 1-8.

10. A storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it performs the steps of the method as described in any one of claims 1-8.

Images