Dense area identification method based on fund net flow direction multi-time aggregation confirmation

Abstract

This invention relates to the field of on-chain behavior recognition, specifically to a method for identifying dense areas based on multiple aggregations and confirmations of net capital flows. The method includes: S1: Obtaining capital direction signals through asset grading; S2: Adaptive threshold adjustment based on market volatility; S3: Immediate generation of candidate markers, without reading final prices at this stage; S4: Final price completion and validity filtering. At the next hour, all candidate markers generated in the previous period are queried, and the confirmed opening price (0), highest price (H), lowest price (L), and closing price (C) of the corresponding period are obtained. The price field of the candidate markers is completed, and validity filtering is performed to delete markers with abnormal price data; S5: Multiple aggregation confirmation and marker density scoring; S6: Confirmation of valid dense areas using a two-way threshold; S7: Interval calculation, deduplication, and output. Steps S1-S7 can output accurate, stable, and practically instructive price dense areas.

Description

This invention relates to the field of on-chain behavior recognition, and more particularly to a method for identifying dense areas based on multiple aggregations of net fund flows. Background: The cryptocurrency market contains a large number of trading instruments. Automatically identifying price dense areas (i.e., price ranges that act as support or resistance in trading) for each instrument is a key technical problem in quantitative analysis, market data labeling, and trading signal generation. An existing method for identifying dense areas based on historical price peaks and troughs or trading volume distribution identifies candidate support and resistance levels by scanning local highs and lows in historical candlestick charts or statistically analyzing the trading volume density within specific price bands. Specifically, it involves backtracking and analyzing historical price sequences and trading volume distributions within a sliding time window, and then outputting several candidate price points or price bands. Representative implementations include technical indicators such as Bollinger Bands and the Relative Strength Index (RSI), which identify dense areas through price channels or overbought / oversold regions. The shortcomings of the existing technologies mentioned above are as follows: 1. These methods rely on price patterns or total trading volume for judgment, failing to distinguish between support candidates formed by active buying and resistance candidates formed by active selling, resulting in insufficient directional identification capabilities. 2. These methods directly output results based on single peaks and troughs or single-segment trading density phenomena, lacking a mechanism for aggregation and confirmation of multiple historical active events in the same direction, leading to a high probability of misjudging occasional price fluctuations as stable dense areas. 3. These methods output discrete price levels or static price bands, without considering the size of capital inflows, failing to distinguish between ordinary trading volume fluctuations and genuine capital accumulation behavior, resulting in a high false alarm rate. Summary of the Invention The purpose of this invention is to provide a dense area identification method based on multiple aggregation and confirmation of net capital flows, to solve the technical problems of insufficient directional identification capabilities, high misjudgment rates, and high false alarm rates in the background technologies. To achieve the above objectives, the present invention provides the following technical solution: a method for identifying dense areas based on multiple aggregation confirmations of net capital flows, comprising: S1: obtaining capital direction signals by classifying assets; S2: adaptive threshold adjustment based on market volatility; S3: instant generation of candidate markers, without reading the final price at this time; S4: final price completion and validity filtering, at the next hour, querying all candidate markers generated in the previous period, obtaining the confirmed opening price (0), highest price (H), lowest price (L), and closing price (C) final values ​​of the corresponding period, completing the price field of the candidate markers, and performing validity filtering after completion to delete markers with abnormal price data; S5: multiple aggregation confirmations and marker density scoring; S6: two-way threshold confirmation of valid dense areas; S7: interval calculation, deduplication, and output.By adopting the above technical solutions, step S1 obtains fund direction signals through asset grading, clearly identifying support candidates dominated by active buying and resistance candidates dominated by active selling, thus compensating for the shortcomings in direction identification. Step S5's multi-aggregation confirmation mechanism performs multiple rounds of aggregation analysis on historical active events in the same direction, significantly reducing the probability of misjudging occasional price fluctuations as stable consolidation zones. Simultaneously, the entire methodology fully considers the magnitude of fund inflows, combining the adaptive threshold adjustment in S2 and the effectiveness filtering in S4 to accurately distinguish between ordinary trading volume fluctuations and genuine fund accumulation behavior, effectively reducing the false alarm rate. Finally, through the interval calculation, deduplication, and output steps in S7, accurate, stable, and practically instructive price consolidation zones can be obtained, providing reliable technical support for quantitative analysis, market data labeling, and trading signal generation in the cryptocurrency market. Furthermore, the asset grading in step S1 includes the system dividing the assets to be analyzed into multiple grades according to market position and liquidity. Each grade is independently configured with the following trigger parameters: capital inflow threshold, confirmation time window type, and price stability upper limit. The window type includes short window or long window. Furthermore, the design principles of the asset tiering are as follows: Top-tier large-cap assets: Highest market depth, large daily trading volume, set with a higher capital inflow threshold, and a longer confirmation time window (H2, covering the first two observation periods) to reduce the interference of abnormal fluctuations in a single period on the results; Quasi-top-tier assets: Capital sensitivity between top-tier and mainstream assets, with both capital inflow threshold and time window one level lower than top-tier assets; Mainstream assets: Good overall liquidity, using a long window (H2), and a capital inflow threshold set at a medium level; Selected assets: Highest requirements for price stability, lowest capital inflow threshold, using a short time window (H1) to enable rapid system response; Mid-cap assets: Using a short time window (H1), with a capital inflow threshold at a low to medium level; Small-cap assets: Most sensitive to capital inflows, lowest capital inflow threshold, using a short time window (H1); The tiering parameters are fully configurable, the system does not hardcode the number of tiers, and tiers can be added or merged according to changes in market structure. Furthermore, the acquisition of the capital direction signal includes, for the asset to be analyzed, based on the transaction flow data within the corresponding time window, separately calculating the active buying volume Fb and the active selling volume Fs, and calculating the net active trading volume: Fn = Fb - Fs, where: Fn is the net active trading volume, a positive value indicates buyer dominance, and a negative value indicates seller dominance; Fb is the active buying volume within the observation time window; and Fs is the active selling volume within the observation time window. Furthermore, the benchmark capital inflow threshold for each level is statically configured.Based on this, the system dynamically adjusts the threshold according to the market volatility over the past 2-4 months to obtain the actual trigger threshold: Thresho Id(t) = Thresho IdBase X Ka, where the adjustment coefficient Ka is calculated according to the following formula: Ka = 1 + a X (cr(t) - a base) I a base, where: a(t) is the standard deviation of the historical fund flows of this asset class within the observation window of the past 2-4 months; a base is the historical standard deviation within the benchmark observation window; and Q is the adjustment coefficient. The calculation result of Ka is limited to the range [KaMin, KaMax] (KaMin > 0) to prevent Ka from having a non-positive value when the market volatility approaches zero during extremely calm periods. The final result of ThreshoId(t) is further limited to the range [ThreshoIdMin, ThresholdMax] to prevent over-adjustment. The design intention of adaptive adjustment is as follows: when the market is more active than the historical benchmark in the past 2-4 months (a(t) > abase), Ka > 1, and the actual threshold automatically increases to avoid misjudging daily high-frequency capital flows as candidates for dense areas; when the market is calmer than the historical benchmark in the past 2-4 months (a(t) < abase), Ka < 1, and the actual threshold automatically decreases to improve the identification sensitivity and avoid missing real entry behavior during calm periods. For newly launched assets with insufficient data accumulation, i.e., insufficient historical data for the benchmark observation window length, Ka = 1, degenerating to use the benchmark threshold ThresholdBase. To ensure stability during the cold start period, the method for real-time generation of candidate markers includes comparing Fn with the actual trigger threshold Thresho Id(t) at each hourly trigger point: if Fn ~ Thresho Id(t), a support candidate marker is generated; if Fn ~ - Thresho Id(t), a resistance candidate marker is generated; otherwise, no marker is generated. Each candidate marker record includes: asset identifier, level label, direction label, Fn value, creation timestamp, and the open, high, low, and close price fields to be completed. Furthermore, in step S4, "real-time triggering" and "final price landing" are separated on the timeline. Historical aggregation must be established after the final price is completed, because only when the price field is stable can subsequent dense area confirmation and interval calculation have a consistent price benchmark.Further, in step S5, the process of multiple aggregation confirmation includes: after completing the price 5 HK 20137912 A specification, grouping by asset identifier and direction label, extracting all valid markers in the same direction within a historical aggregation window of duration W, forming a candidate marker set S; performing the following judgment on set S: if Isl < Nmin, then the dense area is not confirmed in this round, the process terminates, and waits for the next cycle to re-determine; if Isl ~ Nmin, then the latest Nrecent markers are taken from set S to form a marker subset Srecent for the past 2-4 months, and subsequent dense area confirmation and interval calculation are based on Srecent; the significance of setting a minimum number of times Nmin is that: a single occasional capital fluctuation should not directly trigger the output of a dense area, only when the same direction of capital activity events repeatedly occur within the historical aggregation window can there be a basis for confirming a stable dense area; the method of marker density scoring is: for the marker subset Srecent that meets the number condition for the past 2-4 months, calculate the marker density score: Cs = Nrecent / Tspan; Where: Cs is the marker density score (unit: times / hour); Nrecent is the number of markers in the past 2-4 months; Tspan is the time span between the earliest and latest markers in Srecent, with a minimum value of 1 hour; the higher the Cs value, the more concentrated the active events of funds in the same direction appear in a shorter period of time, and the higher the confidence of the dense area. The Cs value is output as an auxiliary field of the dense area message, which is used by the downstream system to classify the signal according to the confidence level. Further, in step S6, for the time window covered by the marker subset Srecent of the past 2-4 months, the highest price maximum value PMax and the lowest price minimum value PMin of each marker period within the window are statistically analyzed, and the price fluctuation range is calculated as: Dp = (PMax - PMin) / PMin X 100%. Where: Dp is the price fluctuation range; PMax is the maximum value of the highest price in the subset of markers over the past 2-4 months; PMin is the minimum value of the lowest price in the subset of markers over the past 2-4 months. A lower threshold (DpMin) and an upper threshold (DpMax) are applied to Dp simultaneously: if Dp < DpMin, it is considered a single-point sporadic case of excessively concentrated prices, and a dense area is not confirmed; if Dp > DpMax, it is considered a discrete sporadic case of excessively dispersed prices, and a dense area is not confirmed; only when DpMin ~ Dp ~ DpMax is a valid dense area confirmed in that direction. Further, in step S7, the intermediate price range is calculated.For the recent 2-4 month subset of markers confirmed by the two-way threshold, calculate the median price for each marker's corresponding period: PMid = (H + L) / 2. Where: PMid is the median price; H is the highest price for that marker period; L is the lowest price for that marker period. Using the minimum of all median prices as the lower bound of the price range Pricelow, and the maximum of all median prices as the upper bound of the price range PriceHigh, we obtain the candidate support or resistance range [Pricelow, PriceHigh]. The reason for using the median price instead of the opening or closing price or extreme price is that the median price (H + L) / 2 simultaneously considers the highest and lowest points reached within the period, which is more representative of the price center of gravity within the entire period than the closing price on one side. This effectively avoids the selective bias caused by long upper or lower shadows (see the HK 20137912 A instruction manual). The interval coverage ratio is deduplicated by comparing the current candidate interval [Pricelow, PriceHig] with the most recent output interval for the same asset and in the same direction, and calculating the interval coverage ratio: R = LOverlap / min(LCurrent, LPrev) X 100% Where: R is the interval coverage ratio; LOverlap is the overlap length between the current interval and the previous output interval; LCurrent = PriceHig - Pricelow is the length of the current interval; LPrev is the length of the previous output interval. If R is higher than the deduplication threshold, the current output is suppressed, and it is considered a repeated trigger of the same dense area; if R is not higher than the deduplication threshold, or there is no previously output interval, the current interval result is output; the output dense area result includes: asset identifier, level label, direction label, confirmation time range, lower bound of price range (Pricelow), upper bound of price range (PriceHigh), and dense area confidence (Cs). The direction label refers to the support or resistance direction, and the confirmation time direction refers to the earliest marking time to the latest marking time. In summary, the present invention has the following beneficial effects compared with the prior art: 1. By constructing a net capital direction signal (Fn = Fb - Fs) by the difference between active buying and active selling transaction amounts, the two directions of buyer-led and seller-led are clearly distinguished, solving the problem that the prior art relies on total transaction volume and cannot distinguish the capital direction.2. By classifying assets into multiple independent tiers based on their market position and assigning each tier a minimum threshold and confirmation time window for fund inflows, assets of different sizes and liquidity can be identified with sensitivity matching their characteristics, solving the problem that a unified threshold cannot accommodate the differences among multiple asset tiers. 3. Through a "mark first, then complete" time separation mechanism, a directional marker is written immediately at the trigger moment without reading unconfirmed price data, and the final price is completed in the next cycle, fundamentally solving the coupling problem between real-time triggering and final price, ensuring the consistency and reproducibility of interval coordinates. 4. By requiring the number of markers in the same direction to reach a minimum within the historical aggregation window before entering the dense area confirmation process, and quantifying confidence using a marker density score (Cs = Nrecent / Tspan), misjudgments triggered by single, occasional fund movement are avoided, achieving a granular improvement in identification from "binary presence or absence" to "continuous confidence rating." 5. By applying a two-way threshold (DpMin~Dp~DpMax) to the price fluctuation range of the past 2-4 months, effective dense areas are identified. The minimum and maximum values ​​of the median price of each marking period (PMid = (H + L) / 2) are used as the interval boundaries to exclude two types of interference: single-point sporadic events and excessive dispersion, outputting dense areas with stable price centers. 6. The interval coverage rate (R = LOver Iap / min (LCurrent, LPrev) X 100%) is used to suppress duplicate outputs of the same dense area. Relative overlap ratios rather than absolute distances are used as the deduplication basis, automatically adapting to assets of different price levels and reducing downstream signal noise. 7. By using an adaptive threshold adjustment based on market volatility over the past 2-4 months (Threshold(t) = ThresholdBase X Ka, Ka = 1 + a X (cr(t) - a base) / a base), the inflow threshold automatically rises and falls with market activity. This addresses the issue of unstable sensitivity of the static threshold under different market cycles, and ensures stability in extreme situations through a lower bound constraint on Ka and a cold start benchmark. Figure 1 is the overall flowchart of the embodiment; Figure 2 is a schematic diagram of the adaptive threshold adjustment of the embodiment; Figure 3 is a multiple aggregation confirmation density scoring chart of the embodiment; Figure 4 is a confirmation chart of the two-way threshold and mid-price range in the embodiment.Detailed Description of Embodiments 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, 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. Embodiment 1' Referring to Figures 1-4, a method for identifying dense areas based on multiple aggregation confirmations of net capital flows is described. Trigger parameters are configured according to asset market position, and the trigger sensitivity is automatically adjusted with market volatility using an adaptive threshold algorithm. Then, a direction signal is constructed using net capital flows, and a direction mark is written at the trigger time, with the final price completed in the next cycle. The confidence of the dense area is quantified by multiple aggregation confirmations and mark density scoring within the historical aggregation window. Finally, the effective dense area is confirmed through a two-way price fluctuation threshold, and the intermediate price range is calculated and output with confidence scores after deduplication and suppression. Specifically, in this embodiment, the method for identifying dense areas based on multiple aggregation confirmations of net capital flows includes: 10 HK 20137912 A Specification S1: Obtaining capital direction signals through asset grading. S1.1, Asset Classification: The system classifies the assets to be analyzed into multiple levels according to their market position and liquidity. Each level is independently configured with the following trigger parameters: capital inflow threshold, confirmation time window type (short window H1 or long window H2), and price stability limit (DpMax). The design principles for the tiered structure are as follows: **Top-tier large-cap assets:** These have the highest market depth and high daily trading volume, resulting in a higher capital inflow threshold and a longer confirmation window (H2, covering the first two observation periods) to reduce the interference of abnormal fluctuations in a single period. **Near-top-tier assets:** Capital sensitivity is between that of top-tier and mainstream assets, with both capital inflow threshold and time window lower than top-tier assets. **Mainstream assets:** These have relatively good overall liquidity, using a long window (H2) and a medium-level capital inflow threshold. **Selected assets:** These have the highest requirements for price stability (lowest DpMax), the lowest capital inflow threshold, and a short time window (H1) for rapid system response. **Mid-cap assets:** These use a short time window (H1) with a low to medium level capital inflow threshold. **Small-cap assets:** These are the most sensitive to capital inflows, with the lowest capital inflow threshold, and a short time window (H1). The tiering parameters are fully configurable; the system does not hardcode the number of tiers and can add or merge tiers based on changes in market structure.S1.2 Acquisition of Fund Direction Signals: The 11 HK 20137912 A prospectus, for the asset to be analyzed, calculates the net active trading volume (Fn) based on the transaction flow data within the corresponding time window, specifically calculating the active buying volume (Fb) and active selling volume (Fs): Fn = Fb - Fs Where: Fn is the net active trading volume, a positive value indicates buyer dominance, and a negative value indicates seller dominance; Fb is the active buying volume within the observation time window; Fs is the active selling volume within the observation time window. S2: Adaptive Threshold Adjustment Based on Market Volatility: The benchmark fund inflow threshold (ThresholdBase) for each level is statically configured. Based on this, the system dynamically adjusts the threshold value according to the recent market volatility to obtain the actual trigger threshold value. In this embodiment, the recent period is 2-4 months, specifically 3 months: Thresho Id(t) = Thresho IdBase X Ka, where the adjustment coefficient Ka is calculated according to the following formula: Ka = 1 + a X (cr(t) - crbase) / abase, where: a(t) is the standard deviation of the historical capital flow of this asset class within the observation window of the past 2-4 months; abase is the historical standard deviation within the baseline observation window; a is the adjustment coefficient. The calculation result of Ka is limited to the range of [KaMin, KaMax] (KaMin > 0) to prevent Ka from having a non-positive value when the market volatility approaches zero during extreme calm periods; the final result of Thresho Id(t) is further limited to the range of [Thresho IdMin, ThresholdMax] to prevent over-adjustment. The design intent of adaptive adjustment is as follows: When the market has been more active than the historical benchmark in the past 2-4 months (a(t) > crbase), Ka > 1, and the actual threshold automatically increases to avoid misjudging daily high-frequency capital flows as candidates for dense areas; when the market has been calmer than the historical benchmark in the past 2-4 months (a(t) < abase), Ka < 1, and the actual threshold automatically decreases to improve recognition sensitivity and avoid missing real entry behavior during calm periods. For newly launched assets with insufficient data accumulation (historical data is insufficient for the benchmark observation window length), Ka = 1, degenerating to use the benchmark threshold ThresholdBase to ensure stability during the cold start period.S3: Instant Generation of Candidate Markers (No Reading of Final Price) At each hourly trigger, Fn is compared with the corresponding level's actual trigger threshold Thresho Id(t): If Fn ~ Thresho Id(t), a support candidate mark is generated; if Fn ~ - Thresho Id(t) (negative boundary), a resistance candidate mark is generated; otherwise, no mark is generated. Each candidate mark record includes: asset identifier, level label, direction label (support / resistance), Fn value, creation timestamp, and the open, high, low, and close price fields to be completed (set to null). This step only completes the instant marking of the funding event and does not read the unconfirmed final price at the same time. The reason is that the highest price, lowest price, and closing price of the period corresponding to the trigger time are in an unconfirmed state, and what is read is an intermediate state value. Before the end of the observation period, it will be continuously refreshed by subsequent transactions, causing the price of the candidate mark record to be inconsistent with the final K-line value. S4: Final Price Completion and Validity Filtering. At the next full hour, query all candidate tags generated in the previous period, obtain the confirmed opening price (0), highest price (H), lowest price (L), and closing price (C) of the corresponding period, and complete the price field of the candidate tags. After completion, perform validity filtering and delete tags with abnormal price data. This step separates "real-time triggering" and "final price landing" on the timeline. Historical aggregation must be established after the final price is completed, because only when the price field is stable can the subsequent dense area confirmation and interval calculation have a consistent price benchmark. S5: Multiple aggregation confirmation and tag density scoring S5.1 After the price completion is completed by multiple aggregation confirmation, group by asset identifier and direction label, extract all valid tags in the same direction within the historical aggregation window (duration is W), and form a candidate tag set S. Perform the following judgment on set S: If Isl < Nmin (lower limit of the number of times), then the dense area is not confirmed in this round, the process is terminated, and waits for the next period to re-determine. If Isl~Nmin', then the most recent Nrecent entries (Nrecent~Nmin) are taken from set S to form a subset Srecent, representing the data for the past 2-4 months. Subsequent dense area confirmation and interval calculation are based on Srecent. The significance of setting a lower limit Nmin is that a single, occasional fund movement should not directly trigger the output of a dense area. Only when similar fund activity events occur repeatedly within the historical aggregation window can a stable dense area be confirmed. This directly distinguishes it from the existing "single trigger, output" approach.S5.2 Label Density Score: For the label subset Srecent that meets the frequency condition within the past 2-4 months, calculate the label density score: 14 HK 20137912 A Specification Cs = Nrecent / Tspan. Where: Cs is the label density score (unit: times / hour); Nrecent is the number of labels within the past 2-4 months; Tspan is the time span (in hours) between the earliest and latest label in Srecent, with a minimum value of 1. A higher Cs value indicates that active capital events in the same direction occur more concentratedly within a shorter time, and the confidence level of the dense area is higher. The Cs value is output as an auxiliary field of the dense area message for downstream systems to use for signal classification based on confidence level. S6: Two-way threshold confirms valid dense areas. For the time window covered by the recent 2-4 month subset Srecent, the highest price (PMax) and lowest price (PMin) for each period within the window are statistically analyzed to calculate the price fluctuation range: Dp = (PMax - PMin) / PMin x 100%. Where: Dp is the price fluctuation range; PMax is the highest price within the recent 2-4 month subset; PMin is the lowest price within the recent 2-4 month subset. A lower threshold (DpMin) and an upper threshold (DpMax) are applied to Dp: if Dp < DpMin, it is considered a single-point, occasional case of excessive price concentration, and a dense area is not confirmed; if Dp > DpMax, it is considered a discrete, occasional case of excessive price dispersion, and a dense area is not confirmed; only when DpMin ~ Dp ~ DpMax is a valid dense area confirmed in that direction. The significance of setting a lower threshold: to exclude situations where prices barely expand and only occasionally accumulate within a very narrow range—while such situations may result in multiple markers, the price distribution is too concentrated, approaching a single-point trigger rather than an effective range. The significance of setting an upper threshold: to exclude discrete historical events that span too wide a range, avoiding the forced merging of price fluctuations that are far apart into the same dense area. S7: Range calculation, deduplication, and output. S7.1 Calculation of the median price range. For the marker subset Srecent confirmed through the two-way threshold over the past 2-4 months, calculate the median price for each marker's corresponding period: PMid = (H + L) / 2. Where: PMid is the median price; H is the highest price in the marker period; L is the lowest price in the marker period.The minimum of all median prices is used as the lower bound of the price range, Pricelow, and the maximum of all median prices is used as the upper bound, PriceHigh, to obtain the candidate support or resistance range [Pricelow, PriceHigh]. The reason for using median prices instead of opening / closing prices or extreme prices is that the median price (H + L) / 2 simultaneously considers both the highest and lowest prices reached within the period, representing the price center of gravity more effectively than the closing price on one side, thus avoiding the selective bias caused by long upper or lower shadows. S7.2 Range coverage deduplication. The candidate interval [Pricelow, PriceHig] is compared with the most recent output interval under the same asset and direction to calculate the interval coverage rate: R = LOverlap / min(LCurrent, LPrev) X 100% Where: R is the interval coverage rate; LOverlap is the overlap length between the current interval and the previous output interval; LCurrent = PriceHig - Pricelow is the length of the current interval; LPrev is the length of the previous output interval. If R is higher than the deduplication threshold, the current output is suppressed (considered as a repeated trigger of the same dense area); if R is not higher than the deduplication threshold, or there is no previous output interval, the current interval result is output. S7.3 Output Content: The output dense area results include: asset identifier, level label, direction label (support / resistance), confirmation time range (earliest marked time to latest marked time), lower bound of price range (Pricelow), upper bound of price range (PriceHigh), and dense area confidence level (Cs). In another implementation, a combination of statistical filtering and continuity verification equivalent to the main scheme can be used. That is, while keeping the core filtering link unchanged, trend and continuity judgments are completed using other statistical benchmarks, which can also achieve the effect of identifying real capital entry behavior. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms "a," "described," and "the" used in this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" as used herein refers to and includes any or all possible combinations of one or more associated listed items. It should be understood that although the invention may use terms such as first, second, third, etc. to describe various information, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from one another.For example, without departing from the scope of the invention, the first information in specification 17 HK 20137912 A may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" as used herein may be interpreted as "when," "when," or "in response to a determination." Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents. 18 HK 20137912 A Claim 1: A method for identifying dense areas based on multiple aggregation confirmations of net capital flows, characterized by comprising: S1: obtaining capital direction signals by asset classification; S2: adaptive threshold adjustment based on market volatility; S3: instant generation of candidate markers, without reading the final price at this time; S4: final price completion and validity filtering, at the next hour, querying all candidate markers generated in the previous period, obtaining the confirmed opening price (0), highest price (H), lowest price (L), and closing price (C) final values ​​of the corresponding period, completing the price field of the candidate markers, and performing validity filtering after completion to delete markers with abnormal price data; S5: multiple aggregation confirmations and marker density scoring; S6: two-way threshold confirmation of valid dense areas; S7: interval calculation, deduplication, and output. 2. The method for identifying dense areas based on multiple aggregation confirmations of net capital flows according to claim 1, characterized in that the asset classification in step S1 includes the system dividing the assets to be analyzed into multiple levels according to market position and liquidity, and each level is independently configured with the following trigger parameters: capital inflow threshold, confirmation time window type and price stability upper limit, wherein the window type includes short window or long window.3. The method for identifying dense areas based on multiple aggregations and confirmations of net capital flow as described in claim 2, characterized in that the design principle of the asset classification is as follows: Top-tier large-cap assets: highest market depth, large daily trading volume, set with a higher capital inflow threshold, and a longer confirmation time window (H2, covering the first two observation periods) to reduce the interference of single-period abnormal fluctuations on the results; Quasi-top-tier assets: capital sensitivity between top and mainstream assets, with capital inflow threshold and time window both lower than top-tier assets; Mainstream assets: good overall liquidity, using a long window (H2), and a capital inflow threshold set at a medium level; Selected assets: highest requirements for price stability, lowest capital inflow threshold, using a short time window (H1) to enable rapid system response; Mid-cap assets: using a short time window (H1), with a capital inflow threshold at a low to medium level; Small-cap assets: most sensitive to capital inflows, lowest capital inflow threshold, using a short time window (H1); The hierarchical parameters are fully configurable, and the system does not hardcode the number of levels. Levels can be added or merged according to changes in market structure. 4. A method for identifying dense areas based on multiple aggregations of net capital flows according to claim 1, characterized in that: the acquisition of the capital direction signal includes, for the asset to be analyzed, based on the transaction flow data within the corresponding time window, separately calculating the active buying transaction amount Fb and the active selling transaction amount Fs, and calculating the net active transaction amount: Fn = Fb - Fs Where: Fn is the net active transaction amount, a positive value indicates that the buyer is dominant, and a negative value indicates that the seller is dominant; Fb is the active buying transaction amount within the observation time window; Fs is the active selling transaction amount within the observation time window. 5. A method for identifying dense areas based on multiple aggregations of net capital flows according to claim 1, characterized in that: the benchmark capital inflow threshold for each level is statically configured. Based on this, the system dynamically adjusts the threshold according to the market volatility over the past 2-4 months to obtain the actual trigger threshold: Thresho Id(t) = Thresho IdBase X Ka, where the adjustment coefficient Ka is calculated according to the following formula: Ka = 1 + a X (a(t) - abase) / abase, where: a(t) is the standard deviation of the historical cash flow of this asset class within the observation window of the past 2-4 months; abase is the historical standard deviation within the benchmark observation window; and a is the adjustment coefficient.The calculation result of Ka is limited to the range of [KaMin, KaMax] (KaMin > 0) to prevent Ka from becoming non-positive when the market volatility approaches zero during periods of extreme calm. The final result of ThreshoId(t) is further limited to the range of [ThreshoIdMin, ThresholdMax] to prevent over-adjustment. The design intent of adaptive adjustment is as follows: when the market has been more active than the historical benchmark in the past 2-4 months (a(t) > abase), Ka > 1, and the actual threshold automatically increases to avoid misjudging daily high-frequency capital flows as candidates for dense areas; when the market has been calmer than the historical benchmark in the past 2-4 months (a(t) < abase), Ka < 1, and the actual threshold automatically decreases to improve the identification sensitivity and avoid missing real entry behavior during calm periods. For newly launched assets with insufficient data accumulation, i.e., insufficient historical data for the benchmark observation window length, Ka = 1, degenerating to use the benchmark threshold ThresholdBase to ensure stability during the cold start period. 6. A method for identifying dense areas based on multiple aggregations of net capital flows according to claim 1, characterized in that the method for generating candidate markers in real time includes comparing Fn with the actual trigger threshold Thresho Id(t) of the corresponding level at each hourly trigger time: if Fn ~ Thresho Id(t), then a support candidate marker is generated; if Fn ~ - Thresho Id(t), then a resistance candidate marker is generated; otherwise, no marker is generated; each candidate marker record includes: asset identifier, level label, direction label, Fn value, creation timestamp, and the opening, high, low, and closing price fields to be completed. 7. A method for identifying dense areas based on multiple aggregations of net capital flows according to claim 1, characterized in that in step S4, "real-time triggering" and "final price landing" are separated on the timeline. Historical aggregation must be established after the final price is completed, because only when the price field is stable can subsequent dense area confirmation and interval calculation have a consistent price benchmark.8. A method for identifying dense areas based on multiple aggregation confirmations of net capital flow according to claim 1, characterized in that, in step S5, the process of multiple aggregation confirmation includes, after completing price completion, grouping by asset identifier and direction label, extracting all valid markers in the same direction within a historical aggregation window of duration W, forming a candidate marker set S; performing the following judgment on set S: if Isl < Nmin, then no dense area is confirmed in this round, the process terminates, and waits for the next cycle to re-determine; if Isl ~ Nmin, then the latest Nrecent markers are taken from set S to form a marker subset Srecent for the past 2-4 months, and subsequent dense area confirmation and interval calculation are based on Srecent, wherein Nrecent is less than or equal to Nmin; the significance of setting a lower limit of Nmin is that: a single occasional capital fluctuation should not directly trigger the output of dense areas, only when the same direction of active capital events repeatedly occur within the historical aggregation window can a stable dense area be confirmed; The method for label density scoring is as follows: For a subset of labels Srecent that meets the frequency condition within the past 2-4 months, calculate the label density score: Cs = Nrecent / Tspan; where: Cs is the label density score (unit: times / hour); Nrecent is the number of labels within the past 2-4 months; Tspan is the time span between the earliest and latest labels in Srecent, with a minimum value of 1 hour; the higher the Cs value, the more concentrated the same-direction capital activity events appear within a shorter time, and the higher the confidence of the dense area. The Cs value is output as an auxiliary field of the dense area message for downstream systems to use for signal classification according to confidence. 9. A method for identifying dense areas based on multiple aggregations of net capital flows according to claim 1, characterized in that, in step S6, for the time window covered by the labeled subset Srecent over the past 2-4 months, the maximum highest price PMax and the minimum lowest price PMin of each labeled period within the window are statistically analyzed, and the price fluctuation range is calculated as: Dp = (PMax - PMin) / PMin X 100%. Where: Dp is the price fluctuation range; PMax is the maximum highest price within the labeled subset over the past 2-4 months; PMin is the minimum lowest price within the labeled subset over the past 2-4 months.Simultaneously apply a lower threshold (DpMin) and an upper threshold (DpMax) to Dp: If Dp < DpMin, it is considered a single-point occasional case where the price is too concentrated, and a dense area is not confirmed; if Dp > DpMax, it is considered a discrete occasional case where the price is too dispersed, and a dense area is not confirmed; only when DpMin ~ Dp ~ DpMax is a valid dense area confirmed to exist in that direction.10. A method for identifying dense areas based on multiple aggregations of net capital flows according to claim 1, characterized in that, in step S7, the mid-price interval calculation uses a subset of markers Srecent confirmed by a two-way threshold over the past 2-4 months, and calculates the mid-price for each marker corresponding to a period: PMid = (H + L) / 2 where: PMid is the mid-price; H is the highest price of the marker period; L is the lowest price of the marker period; the minimum value of all mid-prices is used as the lower bound of the price interval Pricelow, and the maximum value of all mid-prices is used as the upper bound of the price interval PriceHigh, to obtain the current support interval candidate or resistance interval candidate [Pricelow, PriceHigh]; the reason for using the mid-price instead of the opening and closing prices or extreme prices is that the mid-price (H + L) / 2 simultaneously considers the highest and lowest points reached in the period, which is more representative of the price center of gravity in the entire period than the closing price on one side, and can effectively avoid the selective bias caused by long upper shadows or long lower shadows; the interval inclusion rate deduplication: the current candidate interval [Pricelow, PriceHigh] is deduplicated. The interval coverage ratio is calculated by comparing the current interval with the most recent output interval under the same asset and direction: R = LOverlap / min(LCurrent, LPrev) X 100% Where: R is the interval coverage ratio; LOverlap is the overlap length between the current interval and the previous output interval; LCurrent = PriceHigh - Pricelow is the length of the current interval; LPrev is the length of the previous output interval. If R is higher than the deduplication threshold, the current output is suppressed, and it is considered a repeated trigger of the same dense area. If R is not higher than the deduplication threshold, or there is no previous output interval, the current interval result is output. The output dense area result includes: asset identifier, grade label, direction label, confirmation time range, lower bound of price interval (Pricelow), upper bound of price interval (PriceHigh), and dense area confidence (Cs). The direction label refers to the support direction or resistance direction, and the confirmed time orientation refers to the period from the earliest marked time to the latest marked time. 7 HK 20137912 A Instruction Manual Appendix 1 Figure 1 HK 20137912 A Instruction Manual Appendix 2 Figure 2 HK 20137912 A Instruction Manual Appendix 3 Figure 3 HK 20137912 A Instruction Manual Appendix 4 Figure 4 HK 20137912 A.

Images