Artificial intelligence-based industrial and commercial energy storage peak shaving method

Abstract

The application discloses an industrial and commercial energy storage peak shaving method based on artificial intelligence and relates to the technical field of industrial and commercial energy storage peak shaving, and comprises the following steps: in the discharging process of an energy storage system, the theoretical discharge power and the actual discharge power of the energy storage system are continuously collected to obtain energy storage system discharge data, and environmental temperature data are collected; and the standard energy storage discharge data and the standard temperature data are obtained through abnormal analysis and processing; efficiency calculation and analysis are performed according to the standard energy storage discharge data, and a discharge efficiency prediction model is constructed according to the standard temperature data; the discharge efficiency is predicted based on the discharge efficiency prediction model, and the target storage power of the energy storage system is adjusted; the application is used to solve the problem that the existing industrial and commercial energy storage peak shaving technology cannot predict the discharge efficiency of future dates according to the discharge efficiency data and the environmental temperature data of the energy storage system when configuring the power stored by the energy storage system, and ensure the accuracy of power configuration.

Description

Technical Field

[0001] This invention relates to the field of industrial and commercial energy storage peak shaving technology, specifically to an artificial intelligence-based industrial and commercial energy storage peak shaving method. Background Technology

[0002] Commercial and industrial energy storage peak-shaving technology refers to a user-side energy optimization management technology aimed at commercial and industrial users. Its core objectives are to smooth out fluctuations in their own electricity load, reduce peak electricity costs, and minimize the impact on the public power grid. This is achieved through the charging and discharging sequence control of the energy storage system, which enables energy storage during off-peak hours and discharge during peak hours. It is a core technology for commercial and industrial users to improve the economic efficiency of electricity use and ensure the stability of electricity supply.

[0003] Existing commercial and industrial energy storage peak-shaving technologies often first establish a predictive model to forecast future electricity demand when configuring the stored capacity of the energy storage system. The predicted capacity is then adjusted using a fixed discharge efficiency to ensure that future electricity needs are met. However, this approach often neglects the impact of ambient temperature on discharge efficiency. In reality, ambient temperature directly affects the discharge efficiency of the energy storage system. Increased ambient temperature increases the internal heat dissipation pressure of the battery, leading to lower actual discharge efficiency. Decreasing ambient temperature reduces discharge efficiency through both inhibiting electrochemical reactions and increasing internal resistance losses. Ignoring the impact of future ambient temperature changes on discharge efficiency introduces numerous problems to energy storage peak-shaving operations. The energy storage capacity configuration scheme calculated based on electrical efficiency can be affected by fluctuations in discharge efficiency caused by actual temperature changes. This can easily lead to a mismatch between the actual output of the energy storage system and the electricity demand, potentially causing power shortages during peak electricity consumption periods and necessitating the activation of emergency power supply mechanisms, which would increase power supply costs. The originally planned energy storage and release rhythm is disrupted, and some of the energy to be allocated cannot be integrated into the peak shaving process as expected, resulting in a decrease in the actual utilization efficiency of the energy storage system and making it difficult for energy storage peak shaving to achieve the design effect. Therefore, existing industrial and commercial energy storage peak shaving technologies cannot predict the discharge efficiency of future dates based on the discharge efficiency data of the energy storage system and ambient temperature data when configuring the energy stored in the energy storage system, thus failing to ensure the accuracy of the capacity configuration. Summary of the Invention

[0004] This invention aims to at least partially solve one of the technical problems in the prior art. It obtains energy storage system discharge data by continuously collecting theoretical and actual discharge data during the discharge process of the energy storage system, along with ambient temperature data. Anomaly analysis is performed on both the discharge data and ambient temperature data to obtain standard energy storage discharge data and standard temperature data. Efficiency calculation and analysis are performed based on the standard discharge data, and a discharge efficiency prediction model is constructed based on the standard temperature data. The discharge efficiency is predicted based on the discharge efficiency prediction model, and the target storage capacity of the energy storage system is adjusted. This addresses the problem that existing industrial and commercial energy storage peak-shaving technologies cannot predict the discharge efficiency of future dates based on the energy storage system's discharge efficiency data and ambient temperature data when configuring the stored capacity, thus failing to ensure the accuracy of capacity configuration.

[0005] To achieve the above objectives, this application provides an artificial intelligence-based method for industrial and commercial energy storage peak shaving, comprising the following steps:

[0006] During the discharge process of the energy storage system, the theoretical and actual discharge amounts of the energy storage system are continuously collected to obtain the discharge data of the energy storage system, and ambient temperature data is also collected.

[0007] Anomaly analysis was performed on the discharge data and ambient temperature data of the energy storage system to obtain standard energy storage discharge data and standard temperature data.

[0008] Efficiency calculation and analysis are performed based on standard energy storage discharge data, and a discharge efficiency prediction model is constructed based on standard temperature data.

[0009] The discharge efficiency is predicted based on the discharge efficiency prediction model, and the target storage capacity of the energy storage system is adjusted accordingly.

[0010] Furthermore, during the discharge process of the energy storage system, the theoretical and actual discharged power of the energy storage system are continuously collected to obtain the discharge data of the energy storage system. Collecting ambient temperature data includes the following sub-steps:

[0011] Let any one energy storage system be referred to as the first energy storage system. During the discharge process of the first energy storage system, all loads connected to the first energy storage system are referred to as energy storage discharge loads. The amount of electricity reduced by the first energy storage system is referred to as the theoretical amount of electricity released by the first energy storage system. The sum of the actual amount of electricity obtained by all energy storage discharge loads is referred to as the actual amount of electricity released by the first energy storage system.

[0012] The daily discharge period of the first energy storage system is obtained and recorded as the energy storage discharge period; the date of discharge of any first energy storage system is recorded as the first date.

[0013] Furthermore, during the discharge process of the energy storage system, continuously collecting the theoretical and actual discharged power of the energy storage system to obtain the energy storage system discharge data, and collecting ambient temperature data also includes the following sub-steps:

[0014] The first cycle size is set to t1. For the first date, during the discharge process of the first energy storage system, the theoretical and actual discharge amounts of the first energy storage system in each first cycle are periodically collected and the collection time is recorded. The data are arranged in chronological order and recorded as the theoretical discharge sequence and actual discharge sequence of the first date, respectively, and marked as the energy storage system discharge data of the first date.

[0015] Meanwhile, during the discharge process of the first energy storage system, the ambient temperature of the external environment where the first energy storage system is located is collected at the first time interval, and the collection time is recorded. The data is arranged in chronological order and recorded as the ambient temperature sequence of the first date, and marked as ambient temperature data.

[0016] The system repeatedly collects energy storage system discharge data and ambient temperature data for each day.

[0017] Furthermore, anomaly analysis is performed on the energy storage system discharge data and ambient temperature data to obtain standard energy storage discharge data and standard temperature data, including the following sub-steps:

[0018] Based on the theoretical discharge sequence and the actual discharge sequence of the first date, the discharge efficiency is recorded by dividing the actual discharge amount in each cycle by the theoretical discharge amount. After completion, the discharge efficiency sequence is obtained.

[0019] Calculate the coefficient of variation of the discharge efficiency sequence, denoted as CV0; denote any discharge efficiency in the discharge efficiency sequence as the first discharge efficiency PN(i), where i represents the position number, and set the initial window size to k1;

[0020] With PN(i) as the center of the initial window, this initial window is denoted as the initial neighborhood window of PN(i); calculate the coefficient of variation of the data within the initial neighborhood window of PN(i), and denot it as CV1;

[0021] If CV1 is not greater than CV0, then the initial neighborhood window of PN(i) is denoted as the reference window i of PN(i). If CV1 is greater than CV0, keep PN(i) as the center position of the initial window unchanged, expand the size of the initial window to k2*k1, and denot the expanded initial window as the reference window i of PN(i), where k2 is the set scaling factor.

[0022] Furthermore, the anomaly analysis and processing of the energy storage system discharge data and ambient temperature data to obtain standard energy storage discharge data and standard temperature data also includes the following sub-steps:

[0023] For reference window i, calculate the relative difference XP(i) corresponding to PN(i), where XP(i) = |PN(i) - PN(i-1)| / PN(i-1); repeat the calculation of the relative differences corresponding to all data in reference window i to obtain the relative difference set i, calculate the mean and standard deviation of the relative difference set i, and denote them as AP1 and AB1 in order;

[0024] If XP(i) is not located in [AP1-k3*AB1, AP1+k3*AB1], then mark PN(i) as a suspected point of trend abnormality; otherwise, do not mark it, and repeat the marking of all discharge efficiencies in the discharge efficiency sequence, where k3 is the set proportional coefficient;

[0025] Calculate the mean and standard deviation of all data in reference window i, and denote them as AP2 and AB2 in order. If PN(i) is not located in [AP1-k4*AB1, AP1+k4*AB1], then mark PN(i) as a candidate point of fluctuation anomaly, where k4 is the set proportional coefficient.

[0026] If PN(i) is a candidate point for fluctuation anomaly, then the density clustering algorithm is used to divide all data in the reference window i into multiple clusters, and the cluster with the largest amount of data is recorded as the normal cluster. If PN(i) is not located in the normal cluster, then PN(i) is marked as a suspected point for fluctuation anomaly; otherwise, it is not marked.

[0027] If PN(i) is a suspected point of fluctuation anomaly and a suspected point of trend anomaly, then mark PN(i) as abnormal data; otherwise, mark PN(i) as normal data. Repeat the marking of all discharge efficiencies in the discharge efficiency sequence. After completion, the marked efficiency sequence of the first date is obtained and recorded as the standard energy storage discharge data of the first date.

[0028] Furthermore, the anomaly analysis and processing of the energy storage system discharge data and ambient temperature data to obtain standard energy storage discharge data and standard temperature data also includes the following sub-steps:

[0029] For the ambient temperature sequence of the first date, any ambient temperature in the ambient temperature sequence is denoted as HT(j), where j represents the position number; if HT(j) is simultaneously greater than or simultaneously less than two adjacent ambient temperatures, then HT(j) is marked as a split point; all split points are obtained repeatedly, and the ambient temperature sequence is divided into multiple subsequences according to all split points; the subsequence where HT(j) is located is denoted as the first subsequence;

[0030] The first subsequence is subjected to nonlinear fitting to obtain the corresponding fitting function, which is denoted as the first fitting function. The fitting value of HT(j) is obtained according to the first fitting function and the position of HT(j).

[0031] If we take k5 ambient temperatures before and after HT(j) and record them as neighborhood reference temperatures, and obtain the corresponding fitted values, where k5 is the number of values ​​set;

[0032] If HT(j) is greater than the corresponding fitted value, and there are at least ⌈k6*k5⌉ neighboring reference temperatures that are less than the corresponding fitted value, then HT(j) is marked as outlier data; if HT(j) is less than the corresponding fitted value, and there are at least ⌈k6*k5⌉ neighboring reference temperatures that are greater than the corresponding fitted value, then HT(j) is marked as outlier data; otherwise, HT(j) is marked as normal data, where k6 is the set scaling factor.

[0033] Repeatedly label all ambient temperatures in the ambient temperature sequence to obtain the corresponding labeled temperature sequence, which is recorded as the standard temperature data for the first date.

[0034] Furthermore, the efficiency calculation and analysis based on standard energy storage discharge data, and the construction of a discharge efficiency prediction model based on standard temperature data, include the following sub-steps:

[0035] Set the first time length to t2, and divide the marked efficiency sequence and marked temperature sequence of the first date into multiple corresponding segments based on t2, which are respectively denoted as the segmented efficiency sequence and the segmented temperature sequence;

[0036] Let any segment with a duration of t2 in the efficiency sequence be denoted as the first efficiency segment, and let any discharge efficiency in the first efficiency segment be denoted as the first efficiency PG. Set the size of the first window to k7.

[0037] Taking PG as the starting position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the first reference standard deviation GB1 of PG; taking PG as the middle position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the second reference standard deviation GB2 of PG; then taking PG as the end position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the third reference standard deviation GB3 of PG.

[0038] Calculate the weighted standard deviation QG of PG, where QG = 1 / [(GB1 + GB2 + GB3) / 3]; repeatedly obtain the weighted standard deviation of all discharge efficiencies in the first efficiency segment, and sum them up, denoted as the weighted benchmark QW0;

[0039] Calculate the weighted efficiency QP corresponding to PG, where QP = (QG / QW0) * PG. Repeatedly obtain the weighted efficiency corresponding to all discharge efficiencies in the first efficiency segment and sum them to obtain the reference discharge efficiency of the first efficiency segment.

[0040] Repeatedly obtain all reference discharge efficiencies in the efficiency sequence and arrange them in the corresponding order, denoted as the reference discharge efficiency sequence for the first date.

[0041] Furthermore, the process of calculating and analyzing efficiency based on standard energy storage discharge data, and constructing a discharge efficiency prediction model based on standard temperature data, also includes the following sub-steps:

[0042] For dividing the temperature sequence, any segment with a duration of t2 in the divided temperature sequence is denoted as the first temperature segment. The first temperature segment is subjected to nonlinear fitting to obtain the corresponding fitting function. Then, the average temperature of the first temperature segment is calculated based on the corresponding fitting function. All average temperatures in the divided temperature sequence are obtained repeatedly and arranged in the corresponding order, and denoted as the average temperature sequence of the first date.

[0043] The reference discharge efficiency sequence and average temperature sequence for the first date are recorded as the temperature efficiency data for the first date; the temperature efficiency data for each day are repeatedly obtained and recorded as the temperature efficiency reference data.

[0044] Furthermore, the process of calculating and analyzing efficiency based on standard energy storage discharge data, and constructing a discharge efficiency prediction model based on standard temperature data, also includes the following sub-steps:

[0045] An original prediction model is constructed based on an LSTM model. The original prediction model includes an input layer, a hidden layer, and an output layer. The output of the original prediction model is set as a reference discharge efficiency sequence for future dates. The original prediction model is trained using temperature efficiency reference data, and the discharge efficiency prediction model is obtained after the training is completed.

[0046] Furthermore, predicting the discharge efficiency based on the discharge efficiency prediction model and adjusting the target storage capacity of the energy storage system includes the following sub-steps:

[0047] Obtain the average temperature sequence corresponding to future dates, use the discharge efficiency prediction model to predict the discharge efficiency of future dates, and obtain the reference discharge efficiency sequence for future dates;

[0048] Calculate the average discharge efficiency for future dates based on the reference discharge efficiency sequence for future dates; obtain the estimated storage capacity of the energy storage system for future dates, and record it as the target storage capacity of the energy storage system.

[0049] The target storage capacity of the energy storage system is corrected by using the average discharge efficiency of future dates, resulting in the corrected storage capacity of the energy storage system for future dates, and energy is stored based on the corrected storage capacity.

[0050] The beneficial effects of this invention are as follows: This invention continuously collects the theoretical and actual discharge capacity of the energy storage system during discharge, obtaining discharge data and ambient temperature data. Anomaly analysis is performed on both the discharge data and ambient temperature data to obtain standard discharge data and standard temperature data. Efficiency calculation and analysis are performed based on the standard discharge data, and a discharge efficiency prediction model is constructed based on the standard temperature data. The discharge efficiency is predicted based on the discharge efficiency prediction model, and the target storage capacity of the energy storage system is adjusted accordingly. When configuring the storage capacity of the energy storage system, a prediction model can be established based on the discharge efficiency data and ambient temperature data to predict the discharge efficiency for future dates, ensuring the accuracy of the power allocation.

[0051] This invention uses the coefficient of variation to determine whether to expand the window, allowing anomaly detection to adaptively adjust the reference window size based on fluctuations in surrounding data. This reduces computation in stable segments and expands the field of view in fluctuating segments for more robust judgments, reducing the false positive rate. It uses relative difference to detect abrupt changes in trend direction and density clustering to detect suspected points of fluctuation anomalies, distinguishing between abrupt anomalies and persistent jitter anomalies, which is more accurate than a single threshold method. For each discharge efficiency, the standard deviation of three location windows is calculated to obtain weighted standard deviations. Weighting is then performed according to these weighted standard deviations, assigning greater weight to more stable data points within a time period. This ensures that the reference discharge efficiency of a segment is primarily composed of stable and reliable data, avoiding bias from occasional jitters and guaranteeing that subsequent models can learn the correct feature relationships. Attached Figure Description

[0052] Figure 1 This is a flowchart illustrating the steps of the method of the present invention;

[0053] Figure 2 This is a flowchart of the trend anomaly suspected point marking process of the present invention;

[0054] Figure 3 This is a flowchart illustrating the calculation of the reference discharge efficiency of the present invention;

[0055] Figure 4 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0056] 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.

[0057] Example 1, please refer to Figure 1 As shown, this application provides an artificial intelligence-based method for industrial and commercial energy storage peak shaving, including the following steps:

[0058] Step S1 involves continuously collecting the theoretical and actual discharged power of the energy storage system during its discharge process to obtain discharge data, and also collecting ambient temperature data. Step S1 includes the following sub-steps:

[0059] Step S101: Any energy storage system is designated as the first energy storage system. During the discharge process of the first energy storage system, all loads connected to the first energy storage system are designated as energy storage discharge loads. The amount of electricity reduced by the first energy storage system is designated as the theoretical output amount of the first energy storage system, i.e., the amount of energy reduction seen from the energy storage system side, representing the energy that the system considers to be output. The sum of the actual electricity received by all energy storage discharge loads is designated as the actual output amount of the first energy storage system; i.e., the energy actually received by the load side, representing the energy that is truly transferred to the load.

[0060] Step S102: Obtain the daily discharge period of the first energy storage system and record it as the energy storage discharge period; record the date of discharge of any first energy storage system as the first date; the energy storage system does not necessarily discharge all day; the discharge period is obtained by sampling and analyzing only during the actual discharge period to avoid mixing idle or charging time into the discharge efficiency calculation, thereby improving the focus of the analysis.

[0061] Step S103: Set the first cycle size to t1. For the first date, during the discharge process of the first energy storage system, periodically collect the theoretical and actual discharge amounts of the first energy storage system in each first cycle, and record the collection time. Arrange them in chronological order and record them as the theoretical discharge sequence and actual discharge sequence of the first date, respectively, and mark them as the energy storage system discharge data of the first date. In this example, t1 = 1 second, which can be set flexibly.

[0062] In step S104, during the discharge process of the first energy storage system, the ambient temperature of the external environment where the first energy storage system is located is collected at a first time interval, and the collection time is recorded. The data is arranged in chronological order and recorded as the ambient temperature sequence of the first date, and marked as ambient temperature data. If the first energy storage system is indoors, the indoor ambient temperature needs to be collected. The ambient temperature of the future date used for subsequent prediction can be obtained from the weather forecast, while the indoor ambient temperature of the future date needs to be predicted by the system itself to meet subsequent needs.

[0063] Step S105: Repeatedly collect energy storage system discharge data and ambient temperature data for each day;

[0064] In practical implementation, for mainstream lithium battery energy storage systems, a decrease in ambient temperature reduces discharge efficiency by inhibiting electrochemical reactions and increasing internal resistance losses. An increase in ambient temperature may temporarily reduce electrolyte viscosity and accelerate ion migration, potentially leading to a slight increase in discharge efficiency. However, the increased ambient temperature also increases the internal heat dissipation pressure of the battery. Under prolonged discharge conditions, heat accumulation can easily lead to increased energy loss, causing the discharge efficiency to gradually fall below the normal level. Therefore, ambient temperature has a direct impact on the discharge efficiency of energy storage systems.

[0065] Step S2 involves performing anomaly analysis on the energy storage system discharge data and ambient temperature data to obtain standard energy storage discharge data and standard temperature data. Step S2 includes the following sub-steps:

[0066] Step S201: Based on the theoretical discharge sequence and actual discharge sequence of the first date, divide the actual discharged amount of each cycle by the theoretical discharged amount to obtain the discharge efficiency, and then obtain the discharge efficiency sequence.

[0067] For step S202, please refer to... Figure 2 As shown, the coefficient of variation of the discharge efficiency sequence is calculated and denoted as CV0; any discharge efficiency in the discharge efficiency sequence is denoted as the first discharge efficiency PN(i), where i represents the position number, and the initial window size is set to k1; in this embodiment, k1=20, which can be flexibly set according to the actual application scenario; CV reflects the relative fluctuation intensity of the data within the window; the larger the CV, the more violent the fluctuation.

[0068] Step S203: Using PN(i) as the center position of the initial window, the initial window at this time is denoted as the initial neighborhood window of PN(i); calculate the coefficient of variation of the data within the initial neighborhood window of PN(i), denoted as CV1;

[0069] Step S204: If CV1 is not greater than CV0, then the initial neighborhood window of PN(i) is recorded as the reference window i of PN(i). If CV1 is greater than CV0, PN(i) remains at the center of the initial window, and the size of the initial window is expanded to k2*k1. The expanded initial window is recorded as the reference window i of PN(i), where k2 is a set scaling factor; k2>1. In this embodiment, k2=2, which can be flexibly set. The actual discharge efficiency sequence often has local stable segments and local fluctuating segments. Using a fixed small window for the entire sequence may result in frequent misjudgments in the fluctuating region due to insufficient data, while a fixed large window may mask short-term mutations. Therefore, adaptively comparing the local coefficient of variation with the global coefficient of variation can dynamically determine whether to use a small window or a large window. When the data is already highly volatile, expanding the window can reduce the probability of marking normal high fluctuations as abnormal.

[0070] Step S205: For reference window i, calculate the relative difference XP(i) corresponding to PN(i), where XP(i) = |PN(i) - PN(i-1)| / PN(i-1); Repeat the calculation of the relative differences corresponding to all data in reference window i to obtain the relative difference set i, and calculate the mean and standard deviation of the relative difference set i, which are denoted as AP1 and AB1 in order; The relative difference directly quantifies the instantaneous slope or abrupt change magnitude. The relative difference reflects the significance of the change better than the absolute difference and can accurately identify trend anomalies;

[0071] Step S206: If XP(i) is not located in [AP1-k3*AB1, AP1+k3*AB1], then mark PN(i) as a suspected point of trend abnormality; otherwise, do not mark it, and repeat the marking of all discharge efficiencies in the discharge efficiency sequence, where k3 is a set proportional coefficient; in this embodiment, k3=3, which can be set flexibly.

[0072] Step S207: Calculate the mean and standard deviation of all data in reference window i, and denot them as AP2 and AB2 respectively. If PN(i) is not located in [AP1-k4*AB1, AP1+k4*AB1], then mark PN(i) as a candidate point for fluctuation anomaly, where k4 is the set proportional coefficient; in this embodiment, k4=2, which can be set flexibly; simply using the mean and threshold is not accurate enough and may misjudge real anomalies and normal fluctuations, so further judgment is needed on the candidate points for fluctuation anomalies.

[0073] Step S208: If PN(i) is a candidate point for fluctuation anomaly, then the density clustering algorithm is used to divide all data in the reference window i into multiple clusters, and the cluster with the largest amount of data is recorded as the normal cluster. If PN(i) is not located in the normal cluster, then PN(i) is marked as a suspected point for fluctuation anomaly; otherwise, it is not marked. The core feature of a real anomaly point, such as acquisition error or sensor failure, is that it exists in isolation in the local time series and has no similar efficiency value around it. However, normal changes caused by load fluctuation will form local clusters, and the surrounding efficiency values ​​change synchronously with the load, resulting in high density. Therefore, if PN(i) is not located in the normal cluster, then PN(i) is marked as a suspected point for fluctuation anomaly.

[0074] Step S209: If PN(i) is a suspected point of fluctuation abnormality and PN(i) is a suspected point of trend abnormality, then mark PN(i) as abnormal data; otherwise, mark PN(i) as normal data. Repeat the marking of all discharge efficiencies in the discharge efficiency sequence. After completion, the marked efficiency sequence of the first date is obtained and recorded as the standard energy storage discharge data of the first date. It is required that both types of abnormal conditions be met at the same time before it is finally judged as abnormal, which helps to reduce the misjudgment of real but rare discharge efficiency values.

[0075] Step S210: For the ambient temperature sequence of the first date, any ambient temperature in the ambient temperature sequence is denoted as HT(j), where j represents the position number; if HT(j) is simultaneously greater than or simultaneously less than two adjacent ambient temperatures, then HT(j) is marked as a split point; all split points are repeatedly obtained, and the ambient temperature sequence is divided into multiple subsequences according to all split points; the subsequence where HT(j) is located is denoted as the first subsequence; local extrema are detected as split points to cut the temperature sequence into several segments, making each segment closer to a single trend and improving the accuracy of fitting;

[0076] Step S211: Perform nonlinear fitting on the first subsequence to obtain the corresponding fitting function, denoted as the first fitting function. Obtain the fitting value corresponding to HT(j) based on the first fitting function and the position corresponding to HT(j).

[0077] Step S212: If k5 ambient temperatures before and after HT(j) are taken as neighborhood reference temperatures, and the corresponding fitted values ​​are obtained, where k5 is the set number; in this embodiment, k5=2, generally 2 or 3;

[0078] Step S213: If HT(j) is greater than the corresponding fitted value, and there are at least ⌈k6*k5⌉ neighboring reference temperatures less than the corresponding fitted value, then HT(j) is marked as abnormal data; if HT(j) is less than the corresponding fitted value, and there are at least ⌈k6*k5⌉ neighboring reference temperatures greater than the corresponding fitted value, then HT(j) is marked as abnormal data; otherwise, HT(j) is marked as normal data, where k6 is a set scaling factor; in this embodiment, k6=0.9, which can be flexibly set; nonlinear fitting gives the theoretical or expected temperature of the point in the current segment. In the natural environment, normal temperature fluctuations will cause multiple consecutive points to deviate slightly from the trend, for example, gusts of wind cause a short-term drop in temperature, which will then recover the trend. However, most acquisition errors are isolated deviations, that is, deviations from the neighborhood points. By comparing the actual measurement with the expected value and combining it with the neighborhood distribution, instantaneous disturbances and true abnormal readings can be distinguished.

[0079] Step S214: Repeat the marking of all ambient temperatures in the ambient temperature sequence. After completion, the corresponding marked temperature sequence is obtained and recorded as the standard temperature data for the first date.

[0080] In the actual implementation process, the theoretical discharge sequence and the actual discharge sequence may change suddenly due to the load connected to the energy storage system. It is impossible to distinguish between normal fluctuations and erroneous mutations. However, the discharge efficiency does not change suddenly and is relatively stable. Therefore, calculating the corresponding discharge efficiency sequence and analyzing the discharge efficiency sequence can reduce the difficulty of analysis and the amount of calculation.

[0081] Step S3 involves calculating and analyzing efficiency based on standard energy storage discharge data, and constructing a discharge efficiency prediction model based on standard temperature data. Step S3 includes the following sub-steps:

[0082] Step S301: Set the first time length to t2, and divide the labeled efficiency sequence and labeled temperature sequence of the first date into multiple corresponding segments according to t2, which are respectively denoted as the segmented efficiency sequence and the segmented temperature sequence; in this embodiment, t2 = 15 minutes, which is also the minimum time scale for subsequent prediction; the sampling frequency of the original data is very high, making direct prediction costly and difficult, and it is also easily dominated by noise; segmentation can transform the data into stable statistics, reducing the difficulty of prediction;

[0083] For step S302, please refer to... Figure 3 As shown, any segment with a duration of t2 in the efficiency sequence is denoted as the first efficiency segment, and any discharge efficiency in the first efficiency segment is denoted as the first efficiency PG. The size of the first window is set to k7; in this embodiment, k7=10, which can be flexibly set.

[0084] Step S303: Taking PG as the starting position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the first reference standard deviation GB1 of PG; taking PG as the middle position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the second reference standard deviation GB2 of PG; then taking PG as the end position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the third reference standard deviation GB3 of PG; taking the average of the three positions can more robustly reflect the local stability of the point under different relative positions.

[0085] Step S304: Calculate the weighted standard deviation QG of PG, where QG = 1 / [(GB1+GB2+GB3) / 3]; Repeatedly obtain the weighted standard deviation of all discharge efficiencies in the first efficiency segment, and sum them up, denoted as the weighted benchmark QW0;

[0086] Step S305: Calculate the weighted efficiency QP corresponding to PG, where QP = (QG / QW0) * PG. Repeatedly obtain the weighted efficiency corresponding to all discharge efficiencies in the first efficiency segment and sum them to obtain the reference discharge efficiency of the first efficiency segment. Directly using the average discharge efficiency in the segment will cause fluctuations to skew the result. Weighting by standard deviation allows more stable data points within the time period to be given greater weight, so that the reference discharge efficiency of the segment is mainly composed of stable data, rather than being skewed by occasional fluctuations. The reference discharge efficiency is more biased towards long-term stable contributors in the segment, thereby improving the robustness of the segment reference value.

[0087] Step S306: Repeatedly obtain all reference discharge efficiencies in the divided efficiency sequence, and arrange them in the corresponding order, recording them as the reference discharge efficiency sequence for the first date.

[0088] Step S307: For the segmented temperature sequence, any segment with a duration of t2 in the segmented temperature sequence is denoted as the first temperature segment. A nonlinear fitting is performed on the first temperature segment to obtain the corresponding fitting function. Then, the average temperature of the first temperature segment is calculated based on the corresponding fitting function. This process is repeated to obtain all average temperatures in the segmented temperature sequence and arranged in the corresponding order, denoted as the average temperature sequence of the first date. The first temperature segment may contain noise; averaging over the segment using the fitting function better reflects the intrinsic temperature level or trend of the segment and reduces the influence of noise. For example, the temperature may rise or fall slowly within the segment; the fitting curve can capture this trend and make the average temperature of the segment more physically meaningful.

[0089] Step S308: Record the reference discharge efficiency sequence and average temperature sequence of the first date as the temperature efficiency data of the first date; repeatedly obtain the temperature efficiency data of each day and record it as the temperature efficiency reference data.

[0090] Step S309: Construct an original prediction model based on the LSTM model. The original prediction model includes an input layer, a hidden layer, and an output layer. Set the output of the original prediction model to a reference discharge efficiency sequence for future dates. Train the original prediction model using temperature efficiency reference data to obtain the discharge efficiency prediction model. The relationship between temperature and discharge efficiency is often nonlinear, and LSTM can fit this complex nonlinear mapping.

[0091] In the specific implementation process, if PG is at the edge of the fragment and the first window cannot meet all the placement requirements, and there is only one or two reference standard deviations, then the average value is directly calculated based on the obtained standard deviations for subsequent processing, without needing to make up three.

[0092] Step S4 involves predicting the discharge efficiency based on the discharge efficiency prediction model and adjusting the target storage capacity of the energy storage system. Step S4 includes the following sub-steps:

[0093] Step S401: Obtain the average temperature sequence corresponding to future dates, and use the discharge efficiency prediction model to predict the discharge efficiency of future dates to obtain the reference discharge efficiency sequence for future dates.

[0094] Step S402: Calculate the average discharge efficiency for future dates based on the reference discharge efficiency sequence for future dates; obtain the estimated storage capacity of the energy storage system for future dates, and record it as the target storage capacity of the energy storage system, i.e., the electricity consumption for future dates predicted by the algorithm or model.

[0095] Step S403: Correct the target storage capacity of the energy storage system using the average discharge efficiency of future dates to obtain the corrected storage capacity of the energy storage system for future dates, and store energy according to the corrected storage capacity; for example, if the target storage capacity is 100kWh and the predicted average discharge efficiency for tomorrow is 90%, then in order to ensure the basic electricity demand for tomorrow, at least (100 / 0.9)kWh of energy must be stored.

[0096] In the actual implementation process, the average discharge efficiency for future dates can be weighted and averaged according to the importance of each time period or the possible power consumption. The specific correction method for the target storage power can also be set and adjusted according to the actual application scenario.

[0097] Example 2, please refer to Figure 4 As shown, Figure 4 A 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 similar to those in an AI-based industrial and commercial energy storage peak-shaving method to achieve the following functions: During the discharge process of the energy storage system, continuously collect the theoretical and actual discharged power of the energy storage system to obtain discharge data, and collect ambient temperature data; perform anomaly analysis on the discharge data and ambient temperature data to obtain standard discharge data and standard temperature data; perform efficiency calculation and analysis based on the standard discharge data, and construct a discharge efficiency prediction model based on the standard temperature data; predict the discharge efficiency based on the discharge efficiency prediction model, and adjust the target stored power of the energy storage system.

[0098] Furthermore, when the logical instructions in the aforementioned memory can be implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion 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.

[0099] Example 3: This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program performs steps such as those in the AI-based industrial and commercial energy storage peak-shaving method to achieve the following functions: During the discharge process of the energy storage system, continuously collect the theoretical and actual discharged power of the energy storage system to obtain energy storage system discharge data, and collect ambient temperature data; perform anomaly analysis processing on the energy storage system discharge data and ambient temperature data respectively to obtain standard energy storage discharge data and standard temperature data; perform efficiency calculation and analysis based on the standard energy storage discharge data, and construct a discharge efficiency prediction model based on the standard temperature data; predict the discharge efficiency based on the discharge efficiency prediction model, and adjust the target stored power of the energy storage system.

[0100] 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.

[0101] 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.

[0102] 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. An artificial intelligence-based method for industrial and commercial energy storage peak shaving, characterized in that, Includes the following steps: During the discharge process of the energy storage system, the theoretical and actual discharge amounts of the energy storage system are continuously collected to obtain the discharge data of the energy storage system, and ambient temperature data is also collected. Anomaly analysis was performed on the discharge data and ambient temperature data of the energy storage system to obtain standard energy storage discharge data and standard temperature data. Efficiency calculation and analysis are performed based on standard energy storage discharge data, and a discharge efficiency prediction model is constructed based on standard temperature data. The discharge efficiency is predicted based on the discharge efficiency prediction model, and the target storage capacity of the energy storage system is adjusted accordingly. The efficiency calculation and analysis based on standard energy storage discharge data, and the construction of a discharge efficiency prediction model based on standard temperature data, include the following sub-steps: Set the first time length to t2, and divide the marked efficiency sequence and marked temperature sequence of the first date into multiple corresponding segments based on t2, which are respectively denoted as the segmented efficiency sequence and the segmented temperature sequence; Let any segment with a duration of t2 in the efficiency sequence be denoted as the first efficiency segment, and let any discharge efficiency in the first efficiency segment be denoted as the first efficiency PG. Set the size of the first window to k7. Taking PG as the starting position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the first reference standard deviation GB1 of PG; taking PG as the middle position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the second reference standard deviation GB2 of PG; then taking PG as the end position of the first window, calculate the standard deviation of the charging efficiency within the first window at this time, and record it as the third reference standard deviation GB3 of PG. Calculate the weighted standard deviation QG of PG, where QG = 1 / [(GB1 + GB2 + GB3) / 3]; repeatedly obtain the weighted standard deviation of all discharge efficiencies in the first efficiency segment, and sum them up, denoted as the weighted benchmark QW0; Calculate the weighted efficiency QP corresponding to PG, where QP = (QG / QW0) * PG. Repeatedly obtain the weighted efficiency corresponding to all discharge efficiencies in the first efficiency segment and sum them to obtain the reference discharge efficiency of the first efficiency segment. Repeatedly obtain all reference discharge efficiencies in the efficiency sequence and arrange them in the corresponding order, which is recorded as the reference discharge efficiency sequence for the first date; For dividing the temperature sequence, any segment with a duration of t2 in the divided temperature sequence is denoted as the first temperature segment. The first temperature segment is subjected to nonlinear fitting to obtain the corresponding fitting function. Then, the average temperature of the first temperature segment is calculated based on the corresponding fitting function. All average temperatures in the divided temperature sequence are obtained repeatedly and arranged in the corresponding order, and denoted as the average temperature sequence of the first date. The reference discharge efficiency sequence and average temperature sequence for the first date are recorded as the temperature efficiency data for the first date; the temperature efficiency data for each day are repeatedly obtained and recorded as the temperature efficiency reference data. An original prediction model is constructed based on an LSTM model. The original prediction model includes an input layer, a hidden layer, and an output layer. The output of the original prediction model is set as a reference discharge efficiency sequence for future dates. The original prediction model is trained using temperature efficiency reference data, and the discharge efficiency prediction model is obtained after the training is completed.

2. The industrial and commercial energy storage peak-shaving method based on artificial intelligence according to claim 1, characterized in that, During the discharge process of the energy storage system, the theoretical and actual discharged power of the energy storage system are continuously collected to obtain the discharge data of the energy storage system. The collection of ambient temperature data includes the following sub-steps: Let any one energy storage system be referred to as the first energy storage system. During the discharge process of the first energy storage system, all loads connected to the first energy storage system are referred to as energy storage discharge loads. The amount of electricity reduced by the first energy storage system is referred to as the theoretical amount of electricity released by the first energy storage system. The sum of the actual amount of electricity obtained by all energy storage discharge loads is referred to as the actual amount of electricity released by the first energy storage system. The daily discharge period of the first energy storage system is obtained and recorded as the energy storage discharge period; the date of discharge of any first energy storage system is recorded as the first date.

3. The industrial and commercial energy storage peak-shaving method based on artificial intelligence according to claim 2, characterized in that, During the discharge process of the energy storage system, the theoretical and actual discharged power of the energy storage system are continuously collected to obtain the discharge data of the energy storage system. The collection of ambient temperature data also includes the following sub-steps: The first cycle size is set to t1. For the first date, during the discharge process of the first energy storage system, the theoretical and actual discharge amounts of the first energy storage system in each first cycle are periodically collected and the collection time is recorded. The data are arranged in chronological order and recorded as the theoretical discharge sequence and actual discharge sequence of the first date, respectively, and marked as the energy storage system discharge data of the first date. Meanwhile, during the discharge process of the first energy storage system, the ambient temperature of the external environment where the first energy storage system is located is collected at the first time interval, and the collection time is recorded. The data is arranged in chronological order and recorded as the ambient temperature sequence of the first date, and marked as ambient temperature data. The system repeatedly collects energy storage system discharge data and ambient temperature data for each day.

4. The industrial and commercial energy storage peak-shaving method based on artificial intelligence according to claim 3, characterized in that, Anomaly analysis and processing are performed on the discharge data and ambient temperature data of the energy storage system to obtain standard energy storage discharge data and standard temperature data, including the following sub-steps: Based on the theoretical discharge sequence and the actual discharge sequence of the first date, the discharge efficiency is recorded by dividing the actual discharge amount in each cycle by the theoretical discharge amount. After completion, the discharge efficiency sequence is obtained. Calculate the coefficient of variation of the discharge efficiency sequence, denoted as CV0; denote any discharge efficiency in the discharge efficiency sequence as the first discharge efficiency PN(i), where i represents the position number, and set the initial window size to k1; With PN(i) as the center of the initial window, this initial window is denoted as the initial neighborhood window of PN(i); calculate the coefficient of variation of the data within the initial neighborhood window of PN(i), and denot it as CV1; If CV1 is not greater than CV0, then the initial neighborhood window of PN(i) is denoted as the reference window i of PN(i). If CV1 is greater than CV0, keep PN(i) as the center position of the initial window unchanged, expand the size of the initial window to k2*k1, and denot the expanded initial window as the reference window i of PN(i), where k2 is the set scaling factor.

5. The industrial and commercial energy storage peak-shaving method based on artificial intelligence according to claim 4, characterized in that, The process of performing anomaly analysis on the energy storage system discharge data and ambient temperature data to obtain standard energy storage discharge data and standard temperature data also includes the following sub-steps: For reference window i, calculate the relative difference XP(i) corresponding to PN(i), where XP(i) = |PN(i) - PN(i-1)| / PN(i-1); repeat the calculation of the relative differences corresponding to all data in reference window i to obtain the relative difference set i, calculate the mean and standard deviation of the relative difference set i, and denote them as AP1 and AB1 in order; If XP(i) is not located in [AP1-k3*AB1, AP1+k3*AB1], then mark PN(i) as a suspected point of trend abnormality; otherwise, do not mark it, and repeat the marking of all discharge efficiencies in the discharge efficiency sequence, where k3 is the set proportional coefficient; Calculate the mean and standard deviation of all data in reference window i, and denote them as AP2 and AB2 in order. If PN(i) is not located in [AP1-k4*AB1, AP1+k4*AB1], then mark PN(i) as a candidate point of fluctuation anomaly, where k4 is the set proportional coefficient. If PN(i) is a candidate point for fluctuation anomaly, then the density clustering algorithm is used to divide all data in the reference window i into multiple clusters, and the cluster with the largest amount of data is recorded as the normal cluster. If PN(i) is not located in the normal cluster, then PN(i) is marked as a suspected point for fluctuation anomaly; otherwise, it is not marked. If PN(i) is a suspected point of fluctuation anomaly and a suspected point of trend anomaly, then mark PN(i) as abnormal data; otherwise, mark PN(i) as normal data. Repeat the marking of all discharge efficiencies in the discharge efficiency sequence. After completion, the marked efficiency sequence of the first date is obtained and recorded as the standard energy storage discharge data of the first date.

6. The industrial and commercial energy storage peak-shaving method based on artificial intelligence according to claim 5, characterized in that, The process of performing anomaly analysis on the energy storage system discharge data and ambient temperature data to obtain standard energy storage discharge data and standard temperature data also includes the following sub-steps: For the ambient temperature sequence of the first date, any ambient temperature in the ambient temperature sequence is denoted as HT(j), where j represents the position number; if HT(j) is simultaneously greater than or simultaneously less than two adjacent ambient temperatures, then HT(j) is marked as a split point; all split points are obtained repeatedly, and the ambient temperature sequence is divided into multiple subsequences according to all split points; the subsequence where HT(j) is located is denoted as the first subsequence; The first subsequence is subjected to nonlinear fitting to obtain the corresponding fitting function, which is denoted as the first fitting function. The fitting value of HT(j) is obtained according to the first fitting function and the position of HT(j). If we take k5 ambient temperatures before and after HT(j) and record them as neighborhood reference temperatures, and obtain the corresponding fitted values, where k5 is the number of values ​​set; If HT(j) is greater than the corresponding fitted value, and there are at least ⌈k6*k5⌉ neighboring reference temperatures that are less than the corresponding fitted value, then HT(j) is marked as outlier data; if HT(j) is less than the corresponding fitted value, and there are at least ⌈k6*k5⌉ neighboring reference temperatures that are greater than the corresponding fitted value, then HT(j) is marked as outlier data; otherwise, HT(j) is marked as normal data, where k6 is the set scaling factor. Repeatedly label all ambient temperatures in the ambient temperature sequence to obtain the corresponding labeled temperature sequence, which is recorded as the standard temperature data for the first date.

7. The industrial and commercial energy storage peak-shaving method based on artificial intelligence according to claim 6, characterized in that, Predicting discharge efficiency based on a discharge efficiency prediction model and adjusting the target storage capacity of the energy storage system includes the following sub-steps: Obtain the average temperature sequence corresponding to future dates, use the discharge efficiency prediction model to predict the discharge efficiency of future dates, and obtain the reference discharge efficiency sequence for future dates; Calculate the average discharge efficiency for future dates based on a reference discharge efficiency sequence for future dates; Obtain the estimated storage capacity of the energy storage system for a future date, and record it as the target storage capacity of the energy storage system; The target storage capacity of the energy storage system is corrected by using the average discharge efficiency of future dates, resulting in the corrected storage capacity of the energy storage system for future dates, and energy is stored based on the corrected storage capacity.

Images