Energy storage control method based on rolling type real-time electricity price prediction

Abstract

The invention discloses an energy storage control method based on rolling type real-time electricity price prediction, and relates to the technical field of energy storage control, and the method specifically comprises the following steps: screening out all rolling prediction periods with the number of times of prediction direction reversal exceeding a preset reversal number threshold value based on a formed prediction direction sequence, and calibrating the rolling prediction periods as trend repetition periods; performing real-time analysis on each trend repetition period, evaluating the prediction trend stability degree under the condition that the prediction direction inversion frequency exceeds a preset threshold value, and dividing the prediction trend stability degree into a high stability degree, a medium stability degree and a low stability degree; and respectively executing corresponding energy storage scheduling frequency control strategies according to evaluation results. According to the method, the problem that an existing energy storage control method cannot recognize and predict the trend oscillation state, and consequently frequent starting and stopping are caused is solved, energy storage scheduling frequency dynamic regulation and continuous optimization based on trend stability evaluation are achieved, and the response stability and operation efficiency of an energy storage system are effectively improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of energy storage control, and specifically relates to an energy storage control method based on rolling real-time electricity price prediction. BACKGROUND

[0002] Energy storage control refers to the intelligent management of energy storage devices (such as battery energy storage systems, pumped storage, electric vehicle energy storage, etc.) to optimize their charging and discharging processes, so that they can store or release energy at the right time, thereby improving energy utilization efficiency, reducing costs or enhancing grid stability. Traditional energy storage control is usually based on fixed rules or simple electricity price period division, which is difficult to adapt to the dynamic changes of electricity market prices, resulting in limited economic efficiency and scheduling flexibility of the energy storage system. Energy storage control based on rolling real-time electricity price prediction is a more intelligent and dynamic optimization method. This method continuously collects the latest electricity price data and combines historical trends to update and predict in real time, so that the energy storage system can predict future electricity price changes in advance and dynamically adjust the charging and discharging strategy accordingly, thereby achieving better economic efficiency and grid support capability. The core reason for using rolling prediction is that the electricity price of the electricity market is often affected by load demand, renewable energy output, market trading rules and other factors, and has strong volatility and uncertainty. Single static prediction or fixed scheduling strategy cannot accurately adapt to the rapidly changing electricity price environment. Therefore, by rolling the prediction data, the system can continuously correct the prediction error to ensure that the energy storage control decision is always based on the latest and most accurate market information, thereby actively charging at low electricity prices and reasonably discharging at high electricity prices to maximize economic benefits, reduce unnecessary energy loss, optimize grid load distribution, and improve the overall stability and sustainability of the energy system.

[0003] The existing energy storage control technology based on rolling real-time electricity price prediction mainly optimizes the charging and discharging strategy in real time by continuously updating the electricity price prediction model and combining the operation characteristics of the energy storage system to maximize economic benefits and grid stability. Specifically, this technology first uses a rolling time window to continuously obtain the latest key data such as electricity market prices, load demand, weather conditions, and renewable energy generation, and uses machine learning, time series analysis (such as ARIMA, LSTM), optimization algorithms (such as dynamic programming, reinforcement learning), and other methods to build a short-term electricity price prediction model based on these data to predict future electricity price fluctuations. Then, the system combines the predicted electricity price information with the current state of the energy storage device (such as battery remaining capacity, charging and discharging efficiency, life constraints, etc.) to adjust the charging and discharging plan of the energy storage system in real time through a rolling optimization scheduling algorithm, ensuring that it charges preferentially at low electricity prices and discharges reasonably at high electricity prices to reduce electricity costs or increase revenue. In addition, this technology continuously updates the prediction data during execution, recalculates the electricity price prediction value every certain time (such as every hour or every 15 minutes), and corrects the energy storage control strategy to ensure that the energy storage system is always dynamically adjusted based on the latest market information, avoiding decision-making errors due to prediction bias. At the same time, some advanced systems also combine demand response mechanisms to optimize energy storage scheduling based on user-side load conditions, further improving energy utilization efficiency. Overall, the energy storage control technology based on rolling real-time electricity price prediction realizes efficient and intelligent energy storage management through a dynamic closed-loop process of real-time data collection—prediction model update—charging and discharging optimization decision—rolling correction, effectively improving the economic efficiency and stability of the energy storage system.

[0004] The existing technology has the following shortcomings: In the energy storage scheduling operation process, if the electricity price prediction results in continuous multiple rolling prediction periods frequently appear trend reversal, for example, the direction switching phenomenon such as "up-down-up" in a short time, the energy storage control will be switched "charging-stop-discharging-stop" instructions in a very short time; this situation usually occurs when the market electricity price fluctuation frequency increases and the prediction model short-term output is unstable. Since each round of scheduling relies on the current single-point prediction result to generate the decision, without analyzing and modeling the continuity of the trend change of the prediction in the recent rounds, the system cannot identify whether the current prediction is in a stable trend or a direction shock state, so it will mistakenly regard this dramatic repeated trend as an executable scheduling signal, causing the scheduling frequency to abnormally increase. The existing energy storage control method based on rolling real-time electricity price prediction cannot dynamically control the energy storage scheduling frequency according to the prediction trend stability in the case of the above-mentioned frequent prediction direction reversal, which will cause the energy storage device to start and stop frequently in a short time, causing the battery service life to be shortened, the system scheduling efficiency to be reduced, and more seriously, the capacity may be exhausted in advance or the protection mechanism may be triggered frequently, missing the subsequent real high-value discharge or charging window, thereby reducing the overall regulation efficiency and economic return.

[0005] The above information disclosed in the BACKGROUND section is only for the purpose of enhancing the understanding of the background of the present disclosure, and therefore it can include information that does not constitute the prior art known to those of ordinary skill in the art. SUMMARY

[0006] The purpose of the present application is to provide an energy storage control method based on rolling real-time electricity price prediction to solve the problems in the above background.

[0007] In order to achieve the above-mentioned purpose, the present application provides the following technical solution: an energy storage control method based on rolling real-time electricity price prediction, specifically comprising the following steps: In the energy storage control period, the electricity price prediction values and corresponding energy storage scheduling records of all available rolling prediction periods in the current time period are obtained and processed, the electricity price prediction trend direction and prediction slope between each rolling prediction period are extracted, and the prediction direction sequence and trend slope sequence are formed; Based on the formed prediction direction sequence, all rolling prediction periods with the number of prediction direction reversals exceeding the preset reversal number threshold are screened out and marked as trend repetition periods; Real-time analysis is performed on each trend repetition period, the prediction trend stability in the case of the number of prediction direction reversals exceeding the preset threshold is evaluated, and it is divided into high stability, medium stability and low stability; According to the evaluation results, the corresponding energy storage scheduling frequency control strategy is executed; The execution result of the energy storage scheduling frequency control strategy and the evaluation result of the trend stability degree are recorded and fed back, which are used for subsequent evaluation optimization and dynamic adjustment of the control strategy, so as to realize continuous learning and adaptive optimization of the predicted trend fluctuation characteristics.

[0008] Preferably, based on the formed prediction direction sequence, all rolling prediction periods with a prediction direction reversal number exceeding a preset reversal number threshold are screened out and marked as trend repetition periods, specifically: the formed prediction direction sequence is traversed, the change of the prediction direction in each group of adjacent rolling prediction periods is counted, and if the prediction direction of the current period is different from that of the previous period, it is recorded as a direction reversal; the number of direction reversals in all rolling prediction periods is accumulated, and all rolling prediction period segments with an accumulated reversal number exceeding the preset reversal number threshold are marked as trend repetition periods.

[0009] Preferably, each trend repetition period is analyzed in real time, the prediction trend stability degree under the condition that the prediction direction reversal number exceeds the preset threshold is evaluated, and it is divided into high stability degree, medium stability degree and low stability degree, specifically including the following steps: Extract the trend disturbance feature information generated in each trend repetition period and pre-process after extraction; Extract key disturbance elements from the pre-processed trend disturbance feature information and analyze after extraction to generate a trend direction change index and a trend slope disturbance coefficient, respectively; A prediction trend evaluation model is constructed based on the generated trend direction change index and trend slope disturbance coefficient, and a prediction trend coefficient is generated by weighted summation; Determine a preset prediction trend coefficient threshold interval, and compare it with the generated prediction trend coefficient after determination, evaluate the prediction trend stability degree under the condition that the prediction direction reversal number exceeds the preset threshold according to the comparison result, and divide it into high stability degree, medium stability degree and low stability degree.

[0010] Preferably, the key disturbance elements are extracted from the pre-processed trend disturbance feature information, and the specific steps of generating the trend direction change index are as follows: Extract the prediction direction at different rolling prediction moments in each trend repetition period and mark it as , represents the prediction direction at the i-th rolling prediction moment in the j-th trend repetition period, if the prediction electricity value at this moment is higher than that at the previous moment, then , which indicates that the trend is rising; if the prediction electricity value at this moment is lower than that at the previous moment, then , which indicates that the trend is falling; wherein, ​​, , is a positive integer; Based on the prediction direction, determine whether any two adjacent prediction moments and direction change, if change, record the direction of the reverse flag , otherwise ; Calculate the direction disturbance factor , used to represent the current time prediction direction change relative to the previous time disturbance intensity, defined as ; Calculate the trend direction change index, the specific calculation formula is as follows: ; In the formula, is the trend direction change index.

[0011] Preferably, the key disturbance elements are extracted from the preprocessed trend disturbance feature information, and the specific steps of generating the trend slope disturbance coefficient are as follows: Extract the electricity price prediction slope at different rolling prediction moments in each trend repetition period, and mark it as , wherein represents the electricity price change slope at the th rolling prediction moment in the th trend repetition period, which is defined as the change rate obtained by dividing the prediction electricity price difference between the current time and the previous time by the time interval; wherein, , , is a positive integer; Based on the electricity price prediction slope, calculate the slope change amplitude between adjacent two rolling prediction moments, marked as , which is defined as ; Calculate the slope disturbance proportion coefficient , which is defined as , wherein represents the average of the prediction slope of all rolling prediction moments in the th trend repetition period; Calculate the trend slope disturbance coefficient, and the specific calculation formula is as follows: ; In the formula, is the trend slope disturbance coefficient.

[0012] Preferably, the generated trend direction change index and the trend slope disturbance coefficient A prediction trend evaluation model is constructed, and a prediction trend coefficient is generated by weighted summation, and the specific calculation formula is as follows:

[0013] In the formula, is a prediction trend coefficient, and is a non-zero weight coefficient of a trend direction change index and a trend slope disturbance coefficient , and .

[0014] Preferably, a preset prediction trend coefficient threshold interval is determined, and after being determined, the generated prediction trend coefficient is compared, according to the comparison result, the prediction trend stability degree in the case that the prediction direction reversal frequency exceeds the preset threshold is evaluated, and is divided into high stability degree, medium stability degree and low stability degree, and the specific comparison and analysis are as follows: If , the prediction trend stability degree in the case that the prediction direction reversal frequency exceeds the preset threshold is high stability degree; If , the prediction trend stability degree in the case that the prediction direction reversal frequency exceeds the preset threshold is medium stability degree; If , the prediction trend stability degree in the case that the prediction direction reversal frequency exceeds the preset threshold is low stability degree.

[0015] Preferably, according to the evaluation result, corresponding energy storage scheduling frequency control strategies are respectively executed, and specifically: When the evaluation result is high stability degree, the energy storage scheduling frequency control strategy executed is specifically: a low-frequency energy storage scheduling strategy is executed, the response cycle of energy storage control is prolonged, the instruction issuing frequency is reduced, so as to reduce the start-stop frequency of energy storage equipment; When the evaluation result is medium stability degree, the energy storage scheduling frequency control strategy executed is specifically: a medium-frequency energy storage scheduling strategy is executed, the response time interval is shortened, the control sensitivity is improved, and the energy storage equipment is kept in a schedulable state; When the evaluation result is low stability degree, the energy storage scheduling frequency control strategy executed is specifically: a frequency-limiting protection type energy storage scheduling strategy is executed, by prolonging the control cycle and suspending the frequent scheduling instruction, high-frequency switching behavior caused by trend shock is suppressed, and the situation that the energy storage equipment appears energy consumption rise and performance attenuation is avoided.

[0016] In the above technical solution, the technical effects and advantages provided by the present application are as follows: 1. The application realizes quantitative evaluation and classification discrimination of the stability degree of the prediction trend by constructing two parameters of "trend direction change index" and "trend slope disturbance coefficient", multi-dimensionally modeling the direction change characteristics and numerical disturbance amplitude of the electricity price prediction results in the rolling prediction period, and generating a prediction trend coefficient by weighted fusion of the two.

[0017] 2. On the basis of completing the trend stability level evaluation, the application proposes three types of energy storage dispatching frequency control strategies that can be correspondingly adapted, including low-frequency dispatching, medium-frequency dispatching and limited-frequency protection type strategy.

[0018] 3. The application further constructs a record and feedback mechanism based on the trend recognition result and the dispatching execution result, which can associate and store the trend evaluation label and the actual control behavior after each round of energy storage control period, and form a long-term sample set for subsequent evaluation model optimization and control strategy adjustment. BRIEF DESCRIPTION OF DRAWINGS

[0019] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art according to these drawings.

[0020] Figure 1 The flowchart of the energy storage control method based on the rolling real-time electricity price prediction of the present application. DETAILED DESCRIPTION

[0021] Example implementations are now described with reference to the drawings. Example implementations can, however, be implemented in many different forms and should not be construed as limited to the examples set forth herein; rather, these example implementations are provided so that this disclosure will be thorough and complete, and will fully convey the inventive gist to those skilled in the art.

[0022] The application provides a storage control method based on rolling real-time electricity price prediction as shown in the specification, which specifically comprises the following steps: Figure 1 The storage control method based on rolling real-time electricity price prediction as shown in the specification, which specifically comprises the following steps: In the storage control period, the electricity price prediction values and corresponding storage scheduling records of all available rolling prediction periods in the current time period are obtained and processed to extract the electricity price prediction trend direction and prediction slope between each rolling prediction period, forming a prediction direction sequence and a trend slope sequence. In the storage control period, the electricity price prediction results in the continuous rolling prediction periods (for example, nearly 1 hour, nearly 12 hours, etc.) in the current time period can be obtained by calling the electricity price prediction module deployed in the scheduling control system at a set time interval (for example, once every 5 minutes). These prediction results are usually returned in the form of a time sequence, including the prediction time point and the corresponding electricity value. At the same time, the system can also call the storage control log or database to obtain the storage scheduling instruction records issued and executed by the system in the corresponding time period, such as the charging, discharging or standby instructions and their power values in each time slice. This process can be automatically implemented through software methods such as data interface, database query or API calling, and finally the obtained electricity price prediction values and storage scheduling records are synchronized and sorted into a structured data table in chronological order as the input for subsequent trend extraction analysis.

[0023] After the sequence of electricity price prediction values is obtained, the system can compare the prediction electricity values of each two adjacent rolling prediction periods by writing software processing logic, and if the prediction electricity value at the later time point is higher than that at the earlier time point, it is marked as an upward trend, and vice versa, thereby generating a "prediction direction sequence" one by one, that is, a trend label sequence composed of "upward / downward". Then, the system calculates the change rate between each pair of adjacent prediction values, adopts simple difference or standard slope calculation (such as change value per unit time), and forms a set of "trend slope sequence", which can reflect the change amplitude of the electricity price prediction trend at the numerical level. The above processing logic can be realized through the mathematical operation library (such as the sliding window analysis function based on time sequence) embedded in the data processing module, realizing completely software-based and automated data conversion.

[0024] The reason for obtaining complete rolling prediction data and scheduling records within the energy storage control period, and further extracting prediction trend direction and slope information, is that if the energy storage control system is based only on the current prediction point to make scheduling decisions, it often cannot identify the continuity and stability of the prediction trend, and is prone to misjudge the market signal in the scene of sharp price fluctuations or frequent direction switching, leading to frequent start and stop of energy storage devices, causing problems such as shortened battery life and decreased system efficiency. By extracting the direction sequence and the slope sequence, not only can the time evolution structure of the prediction trend be constructed to identify its overall direction characteristics, but also the strength of the trend fluctuation can be quantified, thereby laying a data foundation for subsequent trend stability evaluation, control strategy classification and execution. This processing method is the starting point of the entire trend perception and dynamic control chain, ensuring that the system has structured understanding and forward-looking analysis capabilities for prediction inputs, and is a key link to achieve "from point to sequence, from static judgment to dynamic strategy".

[0025] Based on the formed prediction direction sequence, all rolling prediction periods with a prediction direction reversal number exceeding a preset reversal number threshold are screened out and are marked as trend reversal periods. In this embodiment, based on the formed prediction direction sequence, all rolling prediction periods with a prediction direction reversal number exceeding a preset reversal number threshold are screened out and are marked as trend reversal periods, specifically: the formed prediction direction sequence is traversed, the change of the prediction direction in each group of adjacent rolling prediction periods is counted, and if the prediction direction of the current period is different from that of the previous period, it is recorded as a direction reversal; the number of direction reversals in all rolling prediction periods is accumulated, and all rolling prediction period segments with an accumulated reversal number exceeding a preset reversal number threshold are marked as trend reversal periods.

[0026] To identify whether the electricity price prediction trend frequently reverses, the system can count the trend changes between adjacent rolling prediction periods by traversing the formed prediction direction sequence. The sequence is usually composed of two trend labels, "up" and "down". The system can use a loop traversal method to compare the values of each pair of adjacent elements in turn, that is, to judge whether the trend label of the current prediction period is the same as that of the previous period. If they are different, it means that the trend direction has changed, that is, a trend reversal. Each time a direction change is identified, the counter is incremented by one until the entire direction sequence is traversed. This process can be easily implemented through basic software logic structures such as Boolean judgment, conditional jump, and counter accumulation, and has the characteristics of repeatability, scalability, and automation. The core purpose of this is to convert the trend reversal from a qualitative judgment to a quantifiable parameter, so as to further compare it with the preset threshold to identify whether the trend change has the characteristics of abnormal fluctuations. In this way, the system not only identifies the degree of repeated prediction trends, but also provides a clear data basis for subsequent trend stability assessment and dynamic control strategy selection, enhancing the understanding and adaptability of the entire energy storage control method to the electricity price prediction trend.

[0027] To determine whether the electricity price prediction trend in a certain period of time is repeatedly fluctuating, the system can count the number of trend reversals recorded after completing the traversal and direction reversal identification of the prediction direction sequence. This accumulation process can be achieved by setting a counter variable. Each time a trend reversal is detected, the counter is incremented by one until the traversal is complete. Subsequently, the system compares the accumulated number of direction reversals with the pre-set number of reversals threshold. If the number of reversals is greater than the threshold, the rolling prediction period corresponding to the time period is marked as a trend reversal period. The entire judgment logic can be implemented through value comparison statements and conditional judgment statements in software, which has high automation and embeddability. The purpose of this is to convert the originally ambiguous "trend whether repeated" judgment standard into clear, quantifiable, and programmable logic judgment, so that the energy storage scheduling system can dynamically identify whether the electricity price prediction trend is in an unstable state in actual operation, and trigger trend stability assessment and control measures accordingly, thereby effectively avoiding frequent charging and discharging behaviors caused by dramatic prediction trend reversals, and improving the adaptability and operational stability of the energy storage control system to market fluctuations.

[0028] Real-time analysis of each trend reversal period to assess the prediction trend stability when the number of direction reversals exceeds the pre-set threshold, and to classify it into high stability, medium stability, and low stability. In this embodiment, real-time analysis is performed on each trend repeating cycle to evaluate the stability of the predicted trend when the number of direction reversals exceeds the preset threshold, and the stability is divided into high, medium and low, which includes the following steps: The trend disturbance feature information generated in each trend repeating cycle is extracted and preprocessed after extraction. The extraction of trend disturbance feature information can be based on the collected price prediction data and scheduling records in each trend repeating cycle. The system calls the data processing module to structure and extract multiple dimensional data in this cycle. This process can be automatically executed by a feature information construction function, which mainly includes: extracting the predicted direction, predicted slope, and slope jump amplitude (i.e. the difference between adjacent slopes) at the rolling time from the price prediction sequence; extracting the energy storage response state (charge / discharge / standby) and power change amplitude from the historical scheduling records. The extraction process converts the above raw data into a vector structure through data analysis and variable mapping. Each trend repeating cycle is packaged as a set of multi-dimensional feature vectors, and this vector set constitutes the trend disturbance feature information. The above processing logic can be automatically implemented through regular traversal of data field tags, time synchronization, vector packaging, etc., to ensure data dimension consistency, time correspondence and operability, providing uniform input for subsequent trend modeling.

[0029] The pre-processing of the extracted trend disturbance feature information is an important step before stability evaluation, aiming to improve the accuracy of subsequent parameter generation and the discriminability of the model. Since the original feature data may have different scales, abnormal fluctuations, noise interference, etc. between different trend repeating cycles, direct use for calculation will lead to distorted results or high volatility sensitivity. Therefore, the pre-processing process mainly includes three types of operations: first, normalization processing, which is used to unify various numerical dimensions to the same dimension range (such as the [0, 1] interval), common methods include maximum and minimum scaling or Z-score standardization; second, denoising processing, which is used to smooth the slope jump value or eliminate abnormal reversal records, and can use moving average or median filtering technology; third, missing value completion and boundary cleaning, to ensure that a complete analysis sequence can be formed even if there are missing values or data breakpoints at some prediction points. The above operations can be realized by the built-in feature preprocessing module of the software, which processes each feature vector dimension by dimension according to the set rules, making the disturbance feature information more stable and analyzable in mathematical structure, laying a reliable foundation for subsequent parameter extraction and evaluation model construction.

[0030] Key disturbance elements are extracted from the pre-processed trend disturbance feature information, and analyzed to generate trend direction change index and trend slope disturbance coefficient respectively. The process of extracting key disturbance elements from pre-processed trend disturbance feature information mainly relies on the combination of feature selection algorithms and disturbance behavior pattern recognition logic. First, the system will perform structural analysis on the pre-processed feature vector set, and according to the variability and distribution characteristics of each dimension data in multiple trend cycles, calculate the response strength of each type of feature in the fluctuation behavior, such as the frequency of direction switching, the amplitude of slope change, the frequency of direction continuity interruption, etc. Then, the system calls the built-in disturbance sensitivity evaluation function to score the disturbance contribution of all feature dimensions, and selects the feature dimensions with high sensitivity and high discrimination to trend stability changes as key disturbance elements. This process can be completed by combining the use of information gain score, variance analysis, feature embedding score based on disturbance label, etc. algorithm modules, and automatically output the feature index set with high disturbance correlation. Finally, the system extracts the numerical sequences corresponding to these features from the original feature vector, and constructs them into mathematical expression structures of key disturbance behavior, providing data basis for the generation of trend direction change index and trend slope disturbance coefficient. This process is highly automated and has the ability to adapt to different cycle feature structures, ensuring that the extracted disturbance elements are representative and valuable in mathematical statistics.

[0031] Construct a prediction trend evaluation model for the generated trend direction change index and trend slope disturbance coefficient, and generate a prediction trend coefficient by weighted summation; Determine the pre-set prediction trend coefficient threshold interval, and compare it with the generated prediction trend coefficient. According to the comparison result, evaluate the prediction trend stability when the prediction direction reversal frequency exceeds the pre-set threshold, and divide it into high stability, medium stability and low stability.

[0032] In the software implementation, the determination of the preset prediction trend coefficient threshold interval is usually automatically completed by combining historical data distribution analysis with model training results. Specifically, the system first calls the historical electricity price prediction and corresponding dispatch feedback database to extract prediction trend coefficient samples in multiple known trend repetition periods. These samples are obtained by weighted summation of the trend direction change index and the trend slope disturbance coefficient generated in the historical period. Subsequently, the system performs statistical analysis on these historical prediction trend coefficient samples to construct their probability distribution model. Common methods include kernel density estimation, histogram distribution fitting or Gaussian mixture modeling to identify the numerical distribution boundaries under different stability performances. On this basis, the system can automatically divide into three intervals according to the distribution characteristics, such as by setting two quantile points (e.g. 33% and 66%) or setting interval boundaries according to cluster centers, to generate prediction trend coefficient threshold intervals for judging high, medium and low stability levels. At the same time, the threshold interval can also be dynamically optimized through a dispatch simulation verification mechanism, i.e. the system continuously adjusts the interval boundaries in simulation to minimize the misjudgment rate or improve the stability classification accuracy. The entire process is implemented through a threshold self-adaptive generation module in the software, without the need for human intervention, and has scalability and self-learning ability, providing reliable and adjustable interval basis for subsequent stability level classification.

[0033] In this embodiment, the key disturbance elements are extracted from the preprocessed trend disturbance feature information, and the specific steps for generating the trend direction change index are as follows: The prediction direction at each rolling prediction moment in each trend repetition period is extracted and labeled as , represents the prediction direction at the th rolling prediction moment in the th trend repetition period. If the predicted electricity value at this moment is higher than the previous moment, then , indicating a rising trend; if the predicted electricity value at this moment is lower than the previous moment, then , indicating a falling trend; where , , is a positive integer; The prediction direction of each rolling prediction time in each trend repeating cycle is extracted, which is usually completed by time difference processing on the electricity value in the rolling prediction sequence. Specifically, after identifying the trend repeating cycle, the system calls the historical prediction data module to obtain the electricity price prediction value at all rolling prediction times in the cycle, and sorts them in time sequence. Then, by comparing the electricity price prediction value of the jth time with that of the previous time (i.e. the j−1th time) in each trend repeating cycle, a trend direction sequence is formed. Each element in this sequence corresponds to an actual prediction time and is uniquely determined by the direction of the price change between the time and the previous time. The entire process can be achieved through automatic traversal and difference calculation modules, which judge the difference of the prediction values and generate trend labels without relying on external input or human setting, and have high reusability and embeddability. The system finally assigns a unique identifier to the sequence , representing the prediction direction of the jth rolling prediction time in the ith trend repeating cycle, as an important input for subsequent direction reversal judgment and trend disturbance index calculation.

[0034] Variable The value logic is based on the most basic form of trend direction identification, which is the single-step comparison of electricity price prediction values, and belongs to the typical labeled direction classification method. In the ith trend repeating cycle, for the jth rolling prediction time, the system automatically retrieves the corresponding prediction electricity value and compares it with the prediction value of the previous time j−1. If the prediction electricity value of the jth time is greater than that of the j−1th time, it is considered that the electricity price trend of the current time is upward, and therefore is set; on the contrary, if the current prediction value is less than the previous time, it is considered that the current trend is downward, and is set. This judgment process can be automatically executed through the conditional logic judgment module in the software, and has high scalability. This method is essentially a symbolic processing, which converts continuous numerical sequences into discrete label sequences, making it easy to use Boolean judgment method to quickly identify direction reversal events. In addition, this value taking method has the advantages of high trend sensitivity, clear logic, and low computational overhead, and is a standard preprocessing method commonly used in building prediction trend change models. The trend direction label generated by this rule not only clearly expresses the trend structure, but also effectively avoids the noise interference caused by small price fluctuations, thereby improving the robustness and accuracy of the subsequent disturbance factor and trend index.

[0035] Based on the prediction direction, it is judged whether the direction of any two adjacent prediction times and changes, if it changes, the direction reversal label is recorded, otherwise ; Determining whether a change in trend direction occurs between any two adjacent prediction times is typically accomplished by performing Boolean conditional analysis on the trend direction label sequence using a logic judgment module. Specifically, the system completes the trend direction labeling... After generation, the predicted direction sequence within the same trend recurrence cycle will be traversed to extract the current time step by step. Compared to the previous moment The value of is taken and a sign comparison is performed. If the prediction direction at the current moment is different from that at the previous moment, that is... This indicates that the electricity price trend has reversed from rising to falling or from falling to rising, and the system then sets the direction reversal flag for that moment as [value missing]. This indicates a jump in trend direction; if the predicted directions at two different times are the same, the trend direction is considered to remain unchanged. This logical judgment process does not rely on the absolute value of the predicted electricity price, but focuses on the directional change itself, exhibiting good robustness in trend identification. The direction reversal marker sequence constructed in this way... This provides foundational structural feature data for subsequent trend disturbance modeling. The entire process can be automated within the software system through Boolean judgments, conditional branches, and logical flag function modules, making it suitable for the rapid processing and identification of large-scale trend sequences.

[0036] Calculate the directional perturbation factor , used to represent the disturbance strength of the predicted direction change at the current moment relative to the previous moment, is defined as follows: ; Directional perturbation factor The calculation aims to further quantify the strength of structural disturbances caused when trend direction changes, thereby introducing a weighting mechanism for subsequent trend direction change indices. The design of this factor is based on the value characteristics of the predicted direction, i.e., the predicted direction... The value of is only +1 (up) or -1 (down). Therefore, when the predicted direction changes between two adjacent moments (such as +1 and -1 or -1 and +1), their sum will be 0. Substituting into the formula... The post-obtain disturbance factor value is 1; when the two prediction directions are +1 in succession, the sum is +2, and the disturbance factor is |2-1|=1; and when they are -1 in succession, the sum is -2, and the disturbance factor is | -2-1|=3. Thus, the disturbance factor is a mathematically sensitive index of the "suddenness" of the change in the trend direction, which can not only reflect whether a reversal occurs, but also further reveal the degree of the change through the value size. The formula can be automatically realized in software through the arithmetic addition, constant offset and absolute value operation modules, and has simple processing logic but strong expression ability, and is suitable for nonlinear amplification processing of the change behavior of the prediction direction, especially in constructing a trend disturbance model with an exponential decay mechanism, by adjusting the influence weight of each direction switching in the overall trend change index, so that the model can more accurately capture the trend instability caused by violent shocks.

[0037] The trend direction change index is calculated, and the specific calculation formula is as follows: ; In the formula, is the trend direction change index.

[0038] The calculation method of the trend direction change index aims to quantify the frequency and disturbance degree of the prediction direction reversal in each trend repetition period, and to comprehensively evaluate in multiple period dimensions. The outer summation term averages the disturbance levels of all trend repetition periods, ensuring that the obtained index has period-to-period comparability and global representativeness; the inner summation is used to count the direction changes at each rolling prediction time in a single period, and through normalization processing, the disturbance density between periods of different lengths has a unified evaluation standard; and the direction reversal flag records whether the direction switching occurs at the jth time in the ith period, which is a direct measurement of the trend fluctuation behavior; and the index term introduces the exponential decay mechanism of the disturbance intensity factor, and as the disturbance factor increases, its weight decreases nonlinearly, effectively reducing the dominant effect of severe disturbance on the overall evaluation and improving the stability and anti-bias ability of trend identification. The overall formula realizes high-sensitivity modeling of the frequent switching of the prediction direction in the trend repetition period through nested normalization, weighting and exponential control, and expresses the disturbance behavior in the form of a numerical index with a unified dimension, which is convenient for use as a key judgment basis in subsequent trend stability level division.

[0039] The numerical size of the trend direction change index is significantly negatively correlated with the stability of the predicted trend under the condition of frequent reversal of the direction, that is, The larger the value is, the more times the prediction direction reverses in the trend repetition period (i.e. The value is frequently 1), and these reversal behaviors are accompanied by a small perturbation factor (i.e. Smaller (with a small exponential decay), characterized by frequent shifts in trend direction within a short period and concentrated fluctuations, resulting in a more disordered and unstable overall trend structure; conversely, if Smaller values ​​indicate that although a change in direction may have occurred, the disturbances between these changes are either strong or infrequent, reflecting a relatively gentle or sparse shift in trend direction, and a relatively stable trend. Therefore, This index serves as a key indicator for quantitatively predicting whether there is sustained oscillation or frequent disturbance in the trend direction under the premise that "the number of reversals has exceeded the threshold." A higher value indicates a more unstable trend, making it more likely to be classified as low-stability; a lower value indicates greater directional continuity, making it more likely to be classified as high-stability. This index provides accurate and objective numerical basis for assessing trend stability levels by combining the density of directional changes with the intensity of disturbances.

[0040] In this embodiment, the specific steps for extracting key disturbance elements from the preprocessed trend disturbance feature information, analyzing them, and generating trend slope disturbance coefficients are as follows: Extract the electricity price forecast slope at different rolling forecast moments within each trend recurrence cycle, and calibrate it as... ,in Indicates the first Within the first trend recurrence cycle The slope of electricity price change at each rolling forecast time point is defined as the rate of change obtained by dividing the difference in forecasted electricity price between the current time point and the previous time point by the time interval; where... , , It is a positive integer; Extracting the electricity price forecast slope at different rolling forecast moments within each trend recurrence cycle is typically based on time-series difference calculation and standardization of the predicted electricity price sequence. Specifically, the system first iteratively processes each identified trend recurrence cycle to obtain the electricity price forecast values ​​for all rolling forecast moments within that cycle, arranging them into an ordered numerical sequence in chronological order. Then, the slope calculation module is invoked to calculate the difference between the electricity price value at each j-th forecast moment and its previous moment, and a standard division operation is performed using the forecast window length or a constant rolling interval (e.g., in minutes or hours) to obtain the forecast slope corresponding to that forecast moment. During this process, the system automatically determines whether the prediction time intervals are uniform. If irregular sampling exists, it calls the timestamp alignment module to complete or perform weighted correction, ensuring the accuracy and consistency of the calculation results. All calculation results are uniformly labeled as follows: , used to represent the rate of change of electricity price at the j-th rolling forecast time point within the i-th trend recurrence cycle, constitutes a slope time series. This series not only mathematically reflects the intensity of local changes in the predicted electricity price, but also serves as an important input parameter for subsequently measuring the level of trend numerical disturbance. The entire process can be automatically implemented through difference functions, sliding window calculators, and normalization processing logic, possessing high efficiency, high accuracy, and high adaptability, suitable for the dynamic feature extraction needs of real-time forecast data streams.

[0041] Based on the electricity price forecast slope, the slope change between two adjacent rolling forecast times is calculated and calibrated as... Its definition is , used to measure the abrupt change in the predicted slope; Slope variation range The extraction of the slope is used to quantify the dramatic change in the electricity price forecast slope between adjacent rolling forecast times, and is an important indicator of the stability of the trend. Specifically, the system extracts the forecast slope and generates... After the sequence, the slope jump analysis module will be called to iterate through all predicted slope values ​​within the same trend recurrence cycle, and for each slope at time j... Compared to the previous moment Perform a numerical difference operation and calculate the absolute value of the difference, i.e. This calculation step aims to remove directional factors, retaining only the intensity of slope changes to reflect the abrupt changes in predicted values ​​over continuous time periods. Extracting this type of jump amplitude is highly sensitive to discontinuous fluctuations in the trend, such as "sharp rise—rapid fall" or "slow climb—sudden drop," effectively identifying key fluctuation points in the predicted trend that may cause misjudgments in energy storage regulation. The software system typically uses a sliding window mechanism, a loop pointer structure, and absolute value calculation logic to quickly extract all slope jumps throughout the entire trend's recurring cycle, ultimately outputting a set of values ​​arranged chronologically. The sequence serves as an important input for calculating the trend slope perturbation coefficient.

[0042] Calculate the slope disturbance specificity coefficient Its definition is ,in Indicates the first The average forecast slope of all rolling forecast moments within a trend recurrence cycle is used to normalize the jump amplitude to reflect its relative disturbance intensity within the cycle. Slope perturbation ratio The calculation of this coefficient measures the relative perturbation strength of the slope jump at a given rolling forecast time throughout the entire trend recurrence cycle, and is one of the key indicators for establishing a structural evaluation of trend fluctuations. This coefficient is derived from the slope change obtained in the previous step. Normalization is performed, and a square amplification mechanism is introduced to enhance the response capability to local high-perturbation behavior. Specifically, in the calculation process, the system first calculates the prediction slope for all rolling prediction times within the i-th trend recurrence period. The summation is then divided by the total number of time points h in the cycle to obtain the average predicted slope. Then, the jump amplitude at each moment. The ratio to this average value is calculated, and the square of the ratio is taken to form the slope perturbation weight coefficient. This calculation structure ensures that, within the same period, if the jump amplitude of a certain prediction point is much higher than the overall average slope level, its corresponding... The value will be significantly amplified, reflecting the strong disruptive effect of this point on the stability of trend fluctuations; conversely, if the jump amplitude is small or close to the mean, the point contributes less to the overall disturbance. This normalization and squaring operation can be automatically completed by the mean calculation module and exponential function module in the software, possessing characteristics such as clear structure, universal parameters, and stable performance. Through the dynamic adjustment of this coefficient, the system can achieve fine-grained identification of the predicted slope disturbance distribution, improving the accuracy of trend fluctuation modeling and the reliability of trend stability assessment.

[0043] The formula for calculating the trend slope perturbation coefficient is as follows: ; In the formula, This is the trend slope disturbance coefficient.

[0044] The calculation method for the trend slope perturbation coefficient aims to quantify the fluctuation intensity and cumulative perturbation characteristics of the electricity price forecast slope within the trend recurrence cycle from multiple dimensions, constructing a highly sensitive assessment model for trend numerical instability. The formula structure consists of two layers of summation. The outer layer achieves a global average of the perturbation levels across all trend recurrence cycles, ensuring consistency in assessment results across different cycles. The inner layer aggregates the slope perturbation performance at all rolling forecast moments within each cycle and uses a normalization term. To ensure fairness in comparisons of different period lengths. (The following is a list of components / mechanisms.) This indicates the magnitude of the jump in the predicted slope between adjacent time points, used to characterize the instantaneous fluctuation of the trend slope, while the disturbance weight coefficient... The relative perturbation contribution within the period is reflected by squared the ratio of each jump to the average slope of the period. Multiplying the two and taking... , forming a nonlinear amplification mechanism, so that the sharp jump higher than the average has a higher weight in the evaluation, thereby improving the ability to identify structural sharp fluctuation trends. The formula as a whole adopts a weighted superposition and square ratio normalization strategy, which can not only reflect instantaneous disturbance but also measure the degree of structural destruction within the cycle, and is an important modeling means for the stability of the trend value level.

[0045] Trend slope disturbance coefficient The numerical size of is negatively correlated with the stability of the predicted trend under the condition that the number of direction reversals exceeds the preset threshold, that is, The larger the value is, the more intense the fluctuation of the predicted slope within the trend repetition cycle is, the stronger the numerical change discontinuity is, and the more unstable the overall predicted trend is. Since the calculation of not only considers the instantaneous jump amplitude of the slope between adjacent rolling prediction moments but also combines the normalized disturbance proportion of the jump amplitude relative to the average slope of the cycle , and enhances the sensitivity to abnormal slope disturbance through the product and square root amplification mechanism, therefore when the trend prediction result is accompanied by strong numerical oscillation on the basis of frequent direction switching, it will rise significantly, indicating that the disturbance and uncontrollability of the trend are extremely high, and the stability is low; on the contrary, if the change amplitude of the predicted slope is gentle or relatively balanced, even if the direction reversal is frequent, it will still remain at a low level, indicating that although the trend changes frequently in direction, the structure is still relatively smooth, and the stability is relatively high. Therefore,

[0046] can be used as an important indicator reflecting the disturbance intensity of the predicted slope value level within the trend repetition cycle, combined with the direction change frequency index (such as TDDI), to support the grade division of the stability of the predicted trend together, and provide key basis for the decision of dynamic energy storage regulation strategy. In this embodiment, the generated trend direction change index and the trend slope disturbance coefficient are used to construct a predicted trend evaluation model, and a predicted trend coefficient is generated by weighted summation, and the specific calculation formula is as follows: ; In the formula, is the predicted trend coefficient, and are the non-zero weight coefficients of the trend direction change index and the trend slope disturbance coefficient , respectively.

[0047] The generation is achieved by weighting and summing the trend direction change index and the trend slope disturbance coefficient. Its core purpose is to comprehensively reflect the overall stability of the trend across both structural direction fluctuations and numerical slope disturbances. Weighting coefficients and Used for regulation and In the final assessment, both the impact proportions are real numbers greater than zero and satisfy the normalization condition. This ensures that the predicted trend coefficient remains within a comparable range and of the same dimensions as the original parameters. In practical applications, these two weights can be automatically learned by the system using error minimization methods based on training sample data under different scenarios. For example, the matching accuracy of historical scheduling behavior can be used as the loss function optimization objective, and a set of optimal and stable weight configurations can be obtained through linear regression or machine learning parametric regression algorithms for fitting and training. Alternatively, they can be manually set by users or engineers under specific rules (such as...). (Those that focus more on directional abrupt changes are given higher weights). Through this weighted model, the system can dynamically integrate the two types of trend disturbance information and output a unified predicted trend coefficient (PTC), thereby providing a more comprehensive and accurate evaluation basis for subsequent trend stability level classification and energy storage dispatch strategy matching.

[0048] In this embodiment, a pre-set threshold range for the predicted trend coefficient is determined. And after being determined, it is compared with the generated predictive trend coefficient. A comparison was conducted, and the stability of the predicted trend was assessed based on the comparison results when the number of predicted direction reversals exceeded a preset threshold. This stability was then categorized into high stability, medium stability, and low stability. The specific comparison analysis is as follows: like The stability of the predicted trend is considered high when the number of predicted direction reversals exceeds a preset threshold. This situation means that although the number of predicted direction reversals exceeds a set threshold within the trend's recurrence cycle, the overall trend structure remains highly stable. This typically occurs in scenarios where the predicted direction changes frequently, but the magnitude of these changes is small, the disturbance intensity is weak, and the slope fluctuations are relatively gentle, indicating a clear directional consistency or continuity in the trend. This state demonstrates that the prediction system has strong resilience to short-term disturbances and good trend controllability. In practical energy storage regulation, under this state, the energy storage control response cycle can be appropriately extended, and the control frequency reduced, thereby reducing frequent start-ups and shutdowns of energy storage equipment, extending equipment lifespan, and avoiding energy losses caused by excessive scheduling within the stable operating range.

[0049] like When the number of predicted direction reversals exceeds a preset threshold, the stability of the predicted trend is considered to be of medium stability. This situation indicates that the predicted direction reversal and slope disturbance in the current trend repetition cycle are at a medium level, and the trend stability has certain fluctuations but has not yet reached an extremely unstable state. This situation shows that the trend structure contains both several obvious direction switches and a certain amplitude of slope jump, and the system has certain uncertainty in judging the future short-term prediction results. At the energy storage control level, such a trend should be identified as a "warning interval", and the system needs to switch to a medium response strategy, such as shortening the control interval time, moderately increasing the energy storage operation sensitivity, and enabling the intermediate fault-tolerant mechanism when the system has prediction correction capability, to balance the contradiction between scheduling flexibility and device protection, and improve the adaptability of the system in the fluctuation interval.

[0050] If , the predicted trend stability degree in the case where the number of predicted direction reversals exceeds the preset threshold is low stability degree.

[0051] This situation shows that the predicted direction fluctuates violently and the slope disturbance intensity is extremely high in the current trend repetition cycle, and the trend as a whole is in a state of significant shock. This state means that the short-term output results of the prediction model have lost continuity and reliability, showing high-frequency repetition and unpredictable trend characteristics, which can easily lead to frequent triggering of charge-discharge switching of energy storage control and over-regulation phenomenon. Under this trend background, if the conventional control frequency is still used, it will significantly increase the load pressure of energy storage devices and increase the system energy consumption and battery aging risk. Therefore, when identified as low stability degree, the system should immediately enable the suppression type scheduling strategy to actively extend the control response window and reduce the scheduling frequency, or even trigger the energy storage locking mechanism to avoid wasting resources and reducing efficiency of the system due to misjudgment, and to gain a time window for further correction of the prediction model.

[0052] According to the evaluation results, the corresponding energy storage scheduling frequency control strategies are respectively executed; In this embodiment, according to the evaluation results, the corresponding energy storage scheduling frequency control strategies are respectively executed, specifically: When the evaluation result is high stability degree, the energy storage scheduling frequency control strategy executed is specifically: a low-frequency energy storage scheduling strategy is executed, the response cycle of energy storage control is extended, and the instruction issuing frequency is reduced to reduce the start-stop number of energy storage devices and optimize the long-term operation stability of energy storage resources; In the case of high stability, the software system can call the low-frequency scheduling strategy logic through the policy decision module to actively extend the scheduling period of the energy storage control. The specific way is: before the energy storage control period starts, first determine that the current trend stability label is "high", then set the energy storage scheduling instruction trigger interval to a larger time threshold, and limit the maximum number of charging and discharging instructions or maintaining the existing state within the period. At the same time, the system will record this state in the policy state cache area to avoid frequent activation of the scheduling process due to other external minor fluctuations or abnormal prediction values. The essence of this implementation is to extend the standby time of the energy storage system through the "time threshold adjustment + behavior suppression logic" dual mechanism to reduce the frequency of device start-stop and energy back-and-forth. The reason for doing so is that high trend stability indicates that the current market prediction signal has high credibility and low disturbance, and frequent scheduling will not bring additional benefits, but may lead to inefficient consumption of system resources and accelerated device aging. Therefore, the low-frequency strategy is a power-saving protection scheduling behavior based on "restoring by braking", which is suitable for running optimization in stable market conditions.

[0053] When the evaluation result is medium stability, the specific frequency control strategy for energy storage scheduling is: execute the medium-frequency energy storage scheduling strategy, shorten the response time interval, improve the control sensitivity, and keep the energy storage device in a schedulable state to enhance the system's ability to adapt to fluctuating trends. When the evaluation result is medium stability, the software system will enter the dynamic scheduling control strategy interval. The implementation includes: first, the control engine automatically calls out the medium-frequency strategy template, and divides the time window within the scheduling period into several sub-sections for more fine-grained monitoring of the prediction trend changes; then, according to the prediction trend, dynamically determine whether the local threshold (such as the profit threshold or the price slope mutation threshold) for triggering the charging and discharging operation is met in each sub-section; if it is met, execute a one-time scheduling operation immediately, if it is not met, go to the next sub-period for judgment. The system can also adjust the sensitivity parameter through software settings to improve the scheduling response speed in medium fluctuation conditions to enhance the system's trend tracking ability. The reason for adopting this medium-frequency strategy is that although the trend fluctuates, it has not reached the extreme shock level, and the system can still obtain certain arbitrage benefits or power balance effects through more active and flexible control responses, without causing overload pressure on the energy storage device, so the medium-frequency strategy is a scheduling method that balances benefits and steady-state balance.

[0054] When the evaluation result is low stability, the specific frequency control strategy for energy storage scheduling is: execute the frequency-limited protection type energy storage scheduling strategy, which prolongs the control period and suspends frequent scheduling instructions to suppress high-frequency switching behavior caused by trend shocks and avoid energy consumption and performance degradation of the energy storage device.

[0055] When the trend evaluation result is low stability, the software system needs to trigger the frequency limiting protection mechanism to maximize the suppression of the scheduling disorder caused by the trend of violent shock. The specific implementation is as follows: the system first marks the scheduling frequency as a controlled state, and locks the maximum upper limit of the energy storage scheduling times in the current period through the control logic, and sets the minimum interval time window; if the scheduling instruction is continuously triggered more than the preset limit value in a short time, the system will automatically suspend the current scheduling request and enter the “cooling delay” state. In this state, only safety maintenance operations (such as maintaining the current power level or the minimum protection discharge) are allowed, and the price prediction signal is no longer responded to. At the same time, the system can also record the abnormal fluctuation behavior into the trend disturbance log and feed back to the prediction correction module as the basis for subsequent parameter adjustment. The reason for adopting this strategy is that the low stability trend reflects that the prediction result is extremely unstable and uncontrollable in the short term. If the frequency limiting measure is not taken, it is likely to cause the energy storage device to start and stop frequently under the wrong signal driving, which not only reduces the economic benefit, but also accelerates the equipment wear and tear and induces the risk of systematic overload. Therefore, the frequency limiting protection strategy is a safety priority type response mechanism of “risk avoidance first, then correction”, which is a necessary means to ensure the reliability of the system and the safety of the assets under the extreme fluctuation trend.

[0056] The execution results of the energy storage scheduling frequency control strategy and the evaluation results of the trend stability degree are recorded and fed back for subsequent evaluation optimization and dynamic adjustment of the control strategy, so as to realize the continuous learning and adaptive optimization of the prediction trend fluctuation characteristics.

[0057] In order to realize the recording and feedback of the energy storage scheduling frequency control strategy and the trend stability evaluation results, the system needs to design a “control feedback module” and a “trend learning cache module” to form a data closed loop. Specifically, after each round of energy storage control period ends, the system will automatically collect the following two types of key data: first, the energy storage scheduling frequency control strategy finally executed in this period, including the strategy category (such as low frequency, medium frequency or frequency limiting), the scheduling times, the actual execution interval, the strategy triggering time point, etc.; the second is the prediction trend stability evaluation result label corresponding to this period, including the finally generated prediction trend coefficient (PTC) and its corresponding stability level (high, medium or low). The system will combine the above data into a “control-trend contrast data pair” and store it in the trend evolution learning cache area in chronological order to form a long-term time series sample set. In order to avoid storage redundancy and performance bottleneck, the system can use a sliding window mechanism to periodically clean up expired samples and improve query efficiency through a primary key index structure. This feedback data set can be used as the input basis for subsequent dynamic control models, and also provides a data basis for the bias correction between the strategy execution effect and the trend identification accuracy.

[0058] The core purpose of constructing the record and feedback mechanism is to improve the adaptive ability and model evolution ability of the energy storage control system in the complex trend fluctuation environment. In actual operation, the electricity price prediction model may be inaccurate for a short time due to external disturbances, parameter drift or market mutations, and the energy storage strategy may easily fail to execute or over-control if it relies on static evaluation rules for a long time. Through systematic recording of the execution of each round of energy storage strategy and the trend evaluation results, the software system can fine-tune the judgment boundary of the current evaluation model in subsequent operation by means of statistical models, rule optimization algorithms or lightweight learning models. For example, if the system detects that the current medium-frequency strategy frequently triggers high energy consumption or low returns in multiple moderately stable trend periods, it can automatically tighten the response threshold in the strategy triggering rule and migrate to the low-frequency strategy. For another example, the system can optimize the weight distribution method of the prediction trend coefficient based on the response results of the historical high-stability period, and improve the sensitivity and accuracy of trend identification. Finally, through the construction of the closed-loop path of “evaluation-strategy-feedback-correction”, the mechanism realizes the transformation from static regulation to dynamic optimization, and improves the long-term stability and economy of the system in the volatile power market environment.

[0059] The above formulas are dimensionless numerical calculations, and the formulas are obtained by software simulation of a large amount of data to obtain a formula of the latest real situation. The preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0060] The above embodiments can be realized wholly or partially by software, hardware, firmware or any other combination. When realized by software, the above embodiments can be realized wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another, for example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center through a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0061] It should be understood that the size of the sequence number of the above processes does not mean the order of execution in various embodiments of the present application, and the execution order of the processes should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0062] Those skilled in the art can appreciate that the units and algorithm steps of the examples described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0063] In several embodiments provided in the present application, it should be understood that the disclosed system and method can be implemented in other ways. For example, the above-described embodiments are only illustrative, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units or components shown or discussed can be indirect coupling or communication connection through some interfaces, devices or units, which can be electrical, mechanical or other forms.

[0064] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or they can be distributed on multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the present embodiment.

[0065] In addition, the functional units in each embodiment of the present application can be integrated in one processing unit, or each unit can exist physically, or two or more units can be integrated in one unit.

[0066] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered in the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A method for energy storage control based on rolling real-time price prediction, characterized in that, Specifically comprising the following steps: In the energy storage control cycle, obtain the electricity price prediction value and the corresponding energy storage scheduling record of all available rolling prediction periods in the current time period, and process them to extract the electricity price prediction trend direction and prediction slope between each rolling prediction period to form a prediction direction sequence and a trend slope sequence; Based on the formed prediction direction sequence, all rolling prediction periods with a prediction direction reversal number exceeding a preset reversal number threshold are screened out and marked as trend repeating periods; Real-time analysis is performed on each trend repeating period to evaluate the prediction trend stability in the case where the prediction direction reversal number exceeds the preset threshold, and the trend stability is divided into high stability, medium stability and low stability; According to the evaluation results, corresponding energy storage scheduling frequency control strategies are respectively executed; The execution results of the energy storage scheduling frequency control strategy and the evaluation results of the trend stability are recorded and fed back for subsequent evaluation optimization and dynamic adjustment of the control strategy to realize continuous learning and adaptive optimization of the prediction trend fluctuation characteristics.

2. The energy storage control method based on rolling real-time electricity price prediction of claim 1, wherein, Based on the formed prediction direction sequence, all rolling prediction periods with a prediction direction reversal number exceeding a preset reversal number threshold are screened out and marked as trend repeating periods, specifically: the formed prediction direction sequence is traversed, the change of the prediction direction in each group of adjacent rolling prediction periods is counted, and if the prediction direction of the current period is different from that of the previous period, it is recorded as a direction reversal; The number of direction reversals in all rolling prediction periods is accumulated, and all rolling prediction periods with an accumulated reversal number exceeding a preset reversal number threshold are marked as trend repeating periods.

3. The method of claim 2, wherein, Real-time analysis is performed on each trend repeating period to evaluate the prediction trend stability in the case where the prediction direction reversal number exceeds the preset threshold, and the trend stability is divided into high stability, medium stability and low stability, specifically comprising the following steps: Extract the trend disturbance characteristic information generated in each trend repeating period and pre-process it after extraction; From the pre-processed trend disturbance characteristic information, key disturbance elements are extracted and analyzed to generate a trend direction change index and a trend slope disturbance coefficient; A prediction trend evaluation model is constructed based on the generated trend direction change index and trend slope disturbance coefficient, and a prediction trend coefficient is generated by weighted summation; Determine the preset prediction trend coefficient threshold interval, and compare it with the generated prediction trend coefficient after determination, and evaluate the prediction trend stability in the case where the prediction direction reversal number exceeds the preset threshold according to the comparison result, and divide it into high stability, medium stability and low stability.

4. The method of claim 3, wherein, The specific steps of generating a trend direction change index by extracting key disturbance elements from pre-processed trend disturbance characteristic information and analyzing them are as follows: Extract the prediction direction at different rolling prediction moments in each trend repeating cycle, and mark it as , , represents the prediction direction at the th rolling prediction moment in the th trend repeating cycle. If the predicted electricity value at this moment is higher than the previous moment, then , it indicates a rising trend; if the predicted electricity value at this moment is lower than the previous moment, then , it indicates a falling trend; wherein , , is a positive integer; Based on the predicted direction, determine whether any two adjacent predicted time instants have a direction change, and if so, record a direction reversal flag ; otherwise , do nothing ; Computing directional perturbation factors , for indicating the perturbation intensity of the predicted direction change at the current moment relative to the previous moment, defined as ; The trend direction change index is calculated, and the specific calculation formula is as follows: ； In the formula, is the trend direction change index.

5. The method of claim 4, wherein, The specific steps of generating a trend slope disturbance coefficient by extracting key disturbance elements from pre-processed trend disturbance characteristic information and analyzing them are as follows: Extract the electricity price forecast slope at different rolling forecast moments within each trend recurrence cycle, and calibrate it as... ,in Indicates the first Within the first trend recurrence cycle The slope of electricity price change at each rolling forecast time point is defined as the rate of change obtained by dividing the difference in forecasted electricity price between the current time point and the previous time point by the time interval; where... , , It is a positive integer; Based on the electricity price prediction slope, the slope change amplitude between the adjacent two rolling prediction moments is calculated, which is marked as , which is defined as ; Computing slope perturbation weight coefficients defined as wherein denotes the average of the prediction slopes of all rolling prediction instants within the trend repetition period; The trend slope disturbance coefficient is calculated, and the specific calculation formula is as follows: ； In the formula, is the trend slope perturbation coefficient.

6. The method of claim 5, wherein, The generated trend direction change index The trend slope disturbance coefficient The prediction trend evaluation model is constructed, and the prediction trend coefficient is generated by weighted summation. The specific calculation formula is as follows: ； wherein is a trend coefficient to be predicted, and are trend direction change indices and trend slope disturbance coefficients with non-zero weight coefficients, and .

7. The method of claim 6, wherein, Determining a preset prediction trend coefficient threshold interval And after determining, comparing with the generated prediction trend coefficient According to the comparison result, the prediction trend stability degree when the prediction direction reversal times exceed the preset threshold is evaluated, and it is divided into high stability degree, medium stability degree and low stability degree. The specific comparison analysis is as follows: If , the prediction trend stability degree is a high stability degree when the number of prediction direction reversals exceeds a preset threshold. If , the prediction trend stability degree is the medium stability degree in the case that the number of prediction direction reversals exceeds the preset threshold value; If , the prediction trend stability degree in the case where the number of prediction direction reversals exceeds the preset threshold is a low stability degree.

8. The method of claim 7, wherein, According to the evaluation result, the corresponding energy storage scheduling frequency control strategy is executed, specifically: When the evaluation result is high stability, the energy storage scheduling frequency control strategy executed is specifically: execute low-frequency energy storage scheduling strategy, extend the response cycle of energy storage control, reduce the instruction issuing frequency, and reduce the start-stop frequency of energy storage equipment; When the evaluation result is medium stability, the energy storage scheduling frequency control strategy executed is specifically: execute medium-frequency energy storage scheduling strategy, shorten the response time interval, improve the control sensitivity, and keep the energy storage equipment in a schedulable state; When the evaluation result is low stability, the energy storage scheduling frequency control strategy executed is specifically: execute frequency-limited protection type energy storage scheduling strategy, by extending the control cycle and suspending the frequent scheduling instructions, suppress the high-frequency switching behavior caused by trend oscillation, and avoid the energy consumption rise and performance degradation of energy storage equipment.

Images