Dynamic correction and prediction method of seawater optical chlorophyll sensor profile data

Abstract

The application belongs to the technical field of marine sensor data processing, and particularly relates to a seawater optical chlorophyll sensor profile data dynamic correction and prediction method, which solves the problem that chlorophyll fluorescence reading lag deviation and real state are mixed caused by temperature, pressure and lifting motion in the full-depth continuous observation of a mobile platform, and leads to unclear correction and prediction basis. The method obtains chlorophyll original fluorescence reading, temperature, pressure and platform lifting motion state to form a full-depth observation sequence; generates a temperature and pressure motion lag response feature group according to the temperature and pressure motion driving relationship and lag change; inputs a DAG channel correlation discovery unit to calculate a deviation attribution amount, and directs channel edge weight to a deviation attribution edge and a real attribution edge; updates an environmental induction deviation state and a chlorophyll real state respectively to form a structured chlorophyll state; and dynamically corrects original fluorescence reading and generates a chlorophyll prediction result. The method is used for full-depth chlorophyll sensor data processing.

Description

Technical Field

[0001] This invention relates to the field of marine sensor data processing technology, and discloses a method for dynamic correction and prediction of profile data from a seawater optical chlorophyll sensor. Background Technology

[0002] Chlorophyll a is an important parameter reflecting changes in marine phytoplankton biomass, water nutrient status, and ecological processes. In marine field observations, chlorophyll fluorescence sensors are commonly used to emit excitation light of specific wavelengths and receive the red fluorescence signal produced by chlorophyll molecules, thereby obtaining continuous readings related to chlorophyll content. Compared to post-sampling laboratory extraction analysis, in-situ fluorescence observation can be installed on buoys, stations, submersibles, gliders, profiling buoys, and other mobile platforms to continuously collect water profile information according to time and depth. This is suitable for long-term monitoring of primary marine production processes, abnormal algal changes, and vertical water mass structure. For full-ocean-depth observations on mobile platforms, the platform experiences temperature gradients, pressure changes, motion state changes, and sampling interval changes simultaneously during ascent and descent. The sensor response also continuously changes with the water environment and platform movement. Therefore, raw chlorophyll fluorescence readings typically contain both true chlorophyll state information and state disturbances introduced by the observation environment and sensor response process. Subsequent data processing requires understanding the temporal relationships between various variables in the context of continuous full-ocean-depth observation.

[0003] Existing technologies use temperature, turbidity, depth, historical fluorescence, or multiple variable channels as unified correction or prediction inputs. This approach treats multiple variables as unified inputs, which easily preserves mixed state chains and fails to form a structured separation between environmentally induced bias states and the true state of chlorophyll, resulting in inaccurate results. Summary of the Invention

[0004] The core technical problem this application aims to solve is: during continuous full-ocean-depth observations on a mobile platform, given the lag in the response of raw chlorophyll fluorescence readings relative to the combined driving forces of temperature, pressure, and platform elevation and elliptical motion, how to utilize synchronously acquired observation sequences to identify the portion of the correlation between environmental variable channels and the target chlorophyll state attributable to environmental induced bias during DAG channel correlation detection, and how to separate this portion from the correlation of the true chlorophyll state, so that the dynamic correction and prediction of chlorophyll sensor data are based on a structured state that distinguishes between the environmental induced bias state and the true chlorophyll state. The technical solution is as follows:

[0005] A method for dynamic correction and prediction of profile data from a seawater optical chlorophyll sensor includes the following steps: S1. Acquire raw chlorophyll fluorescence readings, temperature, pressure, and platform elevation and descent status during continuous full-ocean-depth observations of the mobile platform to form a full-ocean-depth observation sequence; S2. Determine the temperature and pressure motion driving relationship based on the full ocean depth observation sequence, and generate a temperature and pressure motion hysteresis response characteristic group by combining the hysteresis change of the original chlorophyll fluorescence readings relative to the temperature and pressure motion driving relationship. S3. Input the temperature and pressure motion hysteresis response feature group into the channel correlation discovery unit of the DAG to calculate the deviation assignment between the environmental variable channel and the chlorophyll target state; S4. Based on the deviation assignment, the state of the channel edge weight in the channel correlation discovery unit of the DAG is redirected to form the deviation assignment edge pointing to the environmentally induced deviation state and the true assignment edge pointing to the true chlorophyll state. S5. Update the environmentally induced deviation state based on the deviation-assigned edge, and update the true chlorophyll state based on the true-assigned edge to form a structured chlorophyll state. S6. Dynamically correct the original chlorophyll fluorescence readings based on the structured chlorophyll state to obtain the corrected chlorophyll state. S7. Generate chlorophyll prediction results based on the corrected chlorophyll state, and output sensor data results including the corrected chlorophyll state and chlorophyll prediction results.

[0006] Preferably, in S1, raw chlorophyll fluorescence readings, temperature, pressure, and platform lifting and lowering motion status are collected during continuous full-ocean-depth observations of the mobile platform, and raw observation data are generated based on the collection time. The full ocean depth continuous observation boundary was determined based on the pressure changes in the raw observation data, and the raw chlorophyll fluorescence readings, temperature, pressure, and platform rise and fall motion status within the full ocean depth continuous observation boundary were used to form the boundary observation data. Based on the acquisition time in the observation data within the boundary, the original chlorophyll fluorescence readings, temperature, pressure, and platform lifting and lowering motion status are correlated over time to form time-corresponding observation data; A full-ocean-depth observation sequence is generated based on the pressure order and acquisition time order in the time-corresponding observation data.

[0007] Preferably, in S2, the temperature and pressure motion change sequence within the same observation window is determined based on the temperature, pressure, and platform rise and fall motion status in the full ocean depth observation sequence; Based on the temperature and pressure motion change sequence, a temperature and pressure motion driving relationship is constructed to characterize the combined effects of temperature, pressure, and platform lifting and lowering motion. Based on the temperature and pressure motion driving relationship, hysteresis matching was performed on the original chlorophyll fluorescence readings to obtain the hysteresis change of the original chlorophyll fluorescence readings relative to the temperature and pressure motion driving relationship. Based on the hysteresis change, determine whether there is a corresponding relationship between the original chlorophyll fluorescence reading and the initial temperature and pressure motion driving the response, and form a hysteresis response determination result; Based on the temperature and pressure motion driving relationship and the hysteresis response determination results, a temperature and pressure motion hysteresis response feature set is generated for the channel correlation discovery unit used as input to the DAG.

[0008] Preferably, in S3, the temperature and pressure motion hysteresis response feature group is input into the channel correlation discovery unit of the DAG, and the temperature, pressure and platform lifting motion state are channelized and mapped according to the temperature and pressure motion hysteresis response feature group to form environmental variable channels; Based on the correspondence between environmental variable channels and chlorophyll target states within the same observation window, calculate the channel-related basic quantities between environmental variable channels and chlorophyll target states; Based on the characteristic set of temperature and pressure motion hysteresis response, the source of deviation is determined for the relevant basic quantities of the channel, and the result of the deviation source determination is formed. Based on the determination of the source of deviation, the portion of the channel-related basic quantities caused by the temperature and pressure motion driving relationship is assigned and calculated to form the deviation correlation quantity; By mapping the deviation correlation quantity to the channel-related basic quantity, the deviation attribution quantity between the environmental variable channel and the chlorophyll target state is obtained; The deviation assignment is used as the basis for determining the state redirection of the channel edge weight in the channel correlation discovery unit of the DAG.

[0009] Preferably, for each environmental variable channel The deviation attribution is determined using the following formula: ; In the formula, Represents environment variable channels The corresponding deviation attribution; This indicates the temperature channel, pressure channel, and platform lifting / lowering motion status channel; Indicates that the environment points to the basic quantity. Corresponding environment variable channels The basic quantities related to the channel; This means that the value inside the parentheses is limited to the range of 0 to 1; This indicates that the value is 1 if the condition inside the parentheses is true and 0 if it is false. Indicates the sampling index; This indicates the number of sampling times within the same observation window; Indicates sampling index The result of the delayed response determination at the location; Indicates sampling index Environment variable channel Effective driving markers in thermo-pressure motion driving relationships; Indicates the source support threshold; Indicates sampling index Lagging changes at the location; This represents a positive constant to prevent the denominator from being zero; it is fixed in the example configuration. ; This represents the absolute value operation; Will , and The deviation attribution is obtained by combining the channels of temperature, pressure, and platform lifting motion in sequence. , , and These correspond to the temperature channel, pressure channel, and platform lifting motion status channel, respectively.

[0010] Preferably, in S4, based on the deviation assignment, the channel edge weights to be redirected between the environmental variable channel and the chlorophyll target state are determined in the channel correlation discovery unit of the DAG; Based on the edge weights of the channel to be redirected and the deviation attribution quantity, determine the edge weight portion of the edge weights of the channel to be redirected that belongs to the environmentally induced deviation state, and form the basic quantity of deviation attribution edge weights. Based on the basic quantity of the deviation attribution edge, the target state of the channel edge weight to be redirected is redirected from the chlorophyll target state to the environmentally induced deviation state, forming a deviation attribution edge pointing to the environmentally induced deviation state. Based on the portion of the edge weights in the channel to be redirected that do not belong to the environmentally induced deviation state, a true attribution edge base quantity pointing to the true state of chlorophyll is formed. The bias-attributed edge and the true-attributed edge are used as the channel edge weights in the DAG to update the environmentally induced bias state and the true chlorophyll state.

[0011] Preferably, to ensure that the edge weights of the channel to be redirected maintain the same numerical scale before and after redirection, and that both the baseline values ​​of the biased attributed edges and the baseline values ​​of the true attributed edges are generated from directly usable edge weight values, the following rules are adopted to determine them. and : ; In the formula, Represents environment variable channels The corresponding deviation belongs to the basic quantity of the edge; Represents environment variable channels The corresponding actual number of belonging edges; This indicates that the value is 1 if the condition inside the parentheses is true and 0 if it is false. Indicates the edge weight of the channel to be redirected The value; Indicates the effective threshold of edge weights. and They have the same numerical scale; Represents a symbolic function; This represents the absolute value operation; Represents environment variable channels The corresponding deviation attribution; This means that the value inside the parentheses is limited to the range of 0 to 1; Indicates that the environment points to the basic quantity. Corresponding environment variable channels The basic quantities related to the channel; The channel index indicates the temperature channel, pressure channel, and platform lifting motion status channel.

[0012] Preferably, in step S5, the deviation-assigned edge and the true-assigned edge are used as the channel edge weight results in the DAG. The deviation-assigned edge is used to update the environmentally induced deviation state, and the true-assigned edge is used to update the chlorophyll true state. A structured chlorophyll state is formed based on the updated environmentally induced deviation state and the updated chlorophyll true state. Let the environmental induced deviation state be denoted as The true state of chlorophyll is recorded as For sampling index Using the same sampling index, the environmental variable channel time correlation representation, the chlorophyll target state channel time correlation representation, the bias-attributed edge value, and the true-attributed edge value, the state update rules are formed according to the following rules. and : ; In the formula, Indicates sampling index Channels represent dimensional indexes Update results of environmentally induced bias state at the location; Indicates sampling index Channels represent dimensional indexes Updated results of chlorophyll true state at the location; Indicates the sampling index; Channels represent dimension indexes; The state continuity coefficient represents the state of environmentally induced deviation. The state continuity coefficient represents the true state of chlorophyll; This indicates the environmental induced deviation state value at the previous sampling index; This represents the true chlorophyll state value at the previous sampling index; Indicates the channel index of the environment variable; Indicates the temperature channel; Indicates a pressure channel; This indicates the platform's lifting and lowering motion status channel; Represents environment variable channels The edge weight value corresponding to the edge to which the deviation belongs; Represents environment variable channels The edge weight corresponding to the actual belonging edge; Represents environment variable channels In the sampling index Channels represent dimensional indexes The time-related numerical value at that location; Indicates the chlorophyll target state channel at the sampling index Channels represent dimensional indexes The time-related numerical value at that location; This represents the summation of the temperature channel, pressure channel, and platform lifting motion status channel; This represents the absolute value operation; Represents a positive constant to prevent the denominator from being zero; Channel representing chlorophyll target state Unit injection coefficient in chlorophyll true state update.

[0013] Preferably, S6 determines the dynamic correction amount of the environmentally induced deviation state relative to the original chlorophyll fluorescence reading based on the structured chlorophyll state. Based on the dynamic correction amount, the original chlorophyll fluorescence readings are dynamically corrected to form the initial corrected chlorophyll state; Based on the actual chlorophyll state in the structured chlorophyll state, the initial corrected chlorophyll state is judged for state consistency, and a state consistency judgment result is formed. Based on the state consistency determination results, the initial corrected chlorophyll state is adjusted to obtain the corrected chlorophyll state.

[0014] Preferably, S7 determines the chlorophyll state input used to generate the chlorophyll prediction result based on the corrected chlorophyll state; The chlorophyll state input is fed into the prediction output process of the DAG to generate chlorophyll prediction results; Based on the corrected chlorophyll state and chlorophyll prediction results, sensor data results are generated; The output includes sensor data results that correct for chlorophyll status and chlorophyll prediction.

[0015] Compared with the prior art, the beneficial effects of this application are as follows:

[0016] This invention proposes an improved DAG channel correlation discovery method. The key improvement lies not in incorporating temperature, pressure, and platform motion states as ordinary external features into the chlorophyll prediction model, but in introducing a deviation attribution quantity within the DAG's channel correlation discovery unit. (This is a direct improvement to the underlying computational stage of the main algorithm.) Conventional DAG or multivariate time-series models, even when detecting a relationship between environmental variable channels and target channels, typically use the relationship strength as the basis for prediction, without specifying whether the relationship should be attributed to sensor response deviation or actual chlorophyll change. This approach embeds the deviation attribution quantity within the DAG's channel correlation discovery unit, allowing the scene mechanism formed by the temperature and pressure motion hysteresis response feature group to participate in the channel correlation calculation itself. Therefore, the deviation attribution quantity is not an external correction coefficient, nor a post-prediction patch, but an internal determination quantity generated before edge weight formation and target state determination, enabling… This process involves determining which subsequent state object the relevant information of the environmental variable channel will be assigned to, and then using this deviation assignment to redirect the destination state of the channel edge weights pointing to the chlorophyll target state (this reflects a structural change distinct from existing technologies. Existing DAG-based methods typically focus on discovering whether there are directed dependencies between variables, with the destination state of the edge weights still revolving around the prediction of the target variable; this scheme changes the state object to which the channel edge weights point, splitting the environmental variable channels that might have previously uniformly pointed to the chlorophyll target state into environmentally induced deviation states and chlorophyll true states. In other words, the improvement is not about adding a more complex prediction layer, but about changing the direction of information injection within the same DAG main solution chain, allowing environmental response offsets to be independently received, while chlorophyll true changes can be retained in the true assignment chain. This processing provides an interpretable state source for subsequent structured chlorophyll states). Through this processing, the environmental channel-related edges that originally uniformly pointed to the chlorophyll target state are assigned as deviation assignment edges pointing to environmentally induced deviation states and true assignment edges pointing to the chlorophyll true state. Compared to conventional multivariate correction or prediction algorithms that compress the relationship between environmental variables and chlorophyll readings into a single input mapping, this invention structurally decomposes the target state of the correlation at the DAG edge weight level, so that the environmentally induced sensor response deviation no longer shares the same state link with the true state of chlorophyll, thereby providing a state basis with source attribution for dynamic correction.

[0017] This invention proposes a method for constraining the hysteresis response of temperature and pressure motion during full-ocean-depth mobile observation. This method constructs a temperature and pressure motion driving relationship based on temperature, pressure, and platform elevation / recession motion states in the full-ocean-depth observation sequence. It then combines this relationship with the hysteresis changes in raw chlorophyll fluorescence readings relative to this driving relationship to generate a set of temperature and pressure motion hysteresis response features. (This object transforms the physical response process in the full-ocean-depth mobile platform scenario into a computational basis usable within the DAG. Existing solutions often use temperature, pressure, or motion state as synchronous exogenous variable inputs, assuming that correlations within the same window can directly serve chlorophyll prediction. This solution emphasizes a hysteresis path where the temperature and pressure motion driving relationship occurs first, followed by the response from raw chlorophyll fluorescence readings. This path describes the formation process of sensor response offset. Thus, the relationship between environmental variable channels and the chlorophyll target state is no longer solely determined by the strength of numerical correlation but is constrained by the source of the hysteresis response. This feature set gives the bias attribution a scenario source, rather than relying on the model's black-box weights to absorb environmental disturbances.) This feature set is not used as a general enhancement input for the chlorophyll target state, but rather as a source constraint for the basic quantities related to the DAG channels (here, the hysteresis response features are not simply concatenated to the chlorophyll target state input, but rather the hysteresis response is used as a source constraint when the basic quantities related to the DAG channels are formed. Existing technologies tend to interpret the synchronous correlation between temperature, pressure, and platform rise and fall motion states and the original chlorophyll fluorescence readings as target state-driven; this scheme requires that the basic quantities related to the channels must simultaneously meet the source conditions of temperature and pressure change first and fluorescence reading change later before entering the deviation attribution calculation. This constraint changes the judgment criteria of the DAG channel correlation discovery unit, avoiding environmentally induced bias being written into the chlorophyll true state link from the beginning), used to determine whether the correlation between environmental variable channels and the chlorophyll target state has the source characteristic of temperature and pressure change occurring first and fluorescence reading responding later. Thus, the channel correlation discovery process can combine the actual observation sequence and sensor response process during profile rise and fall to determine the deviation attribution, avoiding compensation based solely on the synchronous correlation of variable values ​​within the same window.

[0018] This invention proposes a dynamic correction and prediction method based on structured chlorophyll states (which are the core receiving objects for continuing the aforementioned edge weight redirection results. Ordinary correction methods often only obtain a corrected reading, and prediction models often only receive a mixed target state; here, both the environmentally induced bias state and the true chlorophyll state are retained in the structured chlorophyll state, enabling dynamic correction to read the source of the bias and the prediction stage to read the correction result constrained by the true state. This object proof scheme is not a loosely connected "correct first, then predict" approach, but rather connects the bias attribution edge, the true attribution edge, the state update, dynamic correction, and prediction output into the same main algorithm link). This method uses the bias attribution edge to update the environmentally induced bias state and the true attribution edge to update the true chlorophyll state, combining the two to form a structured chlorophyll state. Based on this, the dynamic correction amount of the original chlorophyll fluorescence reading is determined according to the environmentally induced bias state, and the initial correction result is judged for state consistency according to the true chlorophyll state. Finally, the prediction result is generated using the corrected chlorophyll state. Therefore, the correction and prediction processes use the same structured and separated state objects, reducing the possibility of environmentally induced biases being passed on to the prediction output as actual changes in chlorophyll. This enables the full-ocean-depth chlorophyll sensor data processing to form a continuous processing chain of lag source identification, edge weight redirection, state updates, and post-correction prediction. Attached Figure Description

[0019] Figure 1 This is a flowchart of the application process; Figure 2 The following are basic feature maps of the continuous observation profile across the entire ocean depth: (a) is a temperature profile, (b) is a pressure profile, and (c) is a chlorophyll reading profile. Figure 3 A continuous comparison chart of chlorophyll readings and correction status; Figure 4 A graph showing the relationship between the hysteresis response and the deviation attribution in thermo-pressure motion; Figure 5 This is a diagram showing the redirection results of the biased assigned edges and the true assigned edges; Figure 6 A comparison chart of correction and prediction errors for different ablation schemes; Figure 7 Heatmaps of ablation error at different depth ranges; Figure 8 A three-dimensional calibration surface plot of chlorophyll state under the constraint of temperature and pressure motion deviation. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, 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.

[0021] A method for dynamic correction and prediction of profile data from a seawater optical chlorophyll sensor includes the following steps: S1. Acquire raw chlorophyll fluorescence readings, temperature, pressure, and platform elevation / recession status from continuous full-ocean-depth observations using the mobile platform, forming a full-ocean-depth observation sequence. This includes: The system collects raw chlorophyll fluorescence readings, temperature, pressure, and platform elevation and descent status during continuous full-ocean-depth observations using a mobile platform, and generates raw observation data based on the collection time. The full ocean depth continuous observation boundary was determined based on the pressure changes in the raw observation data, and the raw chlorophyll fluorescence readings, temperature, pressure, and platform rise and fall motion status within the full ocean depth continuous observation boundary were used to form the boundary observation data. Based on the acquisition time in the observation data within the boundary, the original chlorophyll fluorescence readings, temperature, pressure, and platform lifting and lowering motion status are correlated over time to form time-corresponding observation data; A full-ocean-depth observation sequence is generated based on the pressure order and acquisition time order in the time-corresponding observation data.

[0022] In this embodiment, step S1 specifically includes:

[0023] Raw chlorophyll fluorescence readings, temperature, pressure, and platform elevation and descent status were collected during continuous full-ocean-depth observations by the mobile platform, forming time-stamped sampling sequences. , , and . This represents the original fluorescence reading of chlorophyll. Indicates temperature. Indicates pressure, This indicates the platform's vertical movement. The four types of timestamped sampling sequences are written into the same data collection set in ascending order of acquisition time to form the raw observation data. Raw observation data Only the collection time, variable identifier, and variable value are retained. The original chlorophyll fluorescence readings are not corrected, and the temperature, pressure, and platform lifting and lowering motion are not converted into the target chlorophyll state.

[0024] Based on the original observation data Pressure changes within the ocean determine the boundary for continuous observation across the entire ocean depth. The observation mission assumes a pre-defined surface pressure benchmark. Full ocean depth pressure benchmark Surface pressure tolerance Continuous time interval threshold and pressure jump threshold Pressure sequence Arrange the data in ascending order of acquisition time, and search for continuous pressure subsequences within the pressure sequence. The continuous pressure subsequence satisfies that the interval between adjacent pressure acquisition times is no greater than [a certain value]. The variation of adjacent pressure values ​​does not exceed The pressure changes in the same direction of rise and fall, and the pressure coverage range is from... The corresponding range has been reached. Corresponding range. The starting acquisition time of the continuous pressure subsequence that meets the above conditions is denoted as... The end time of data collection is recorded as Forming a continuous observation boundary across the entire ocean depth When multiple continuous pressure subsequences meet the conditions, the continuous pressure subsequence with the largest pressure coverage and the longest continuous length at the acquisition time is selected as the boundary for continuous observation at full ocean depth.

[0025] Based on the full ocean depth continuous observation boundary From raw observation data Extract the acquisition time located at to The raw chlorophyll fluorescence readings, temperature, pressure, and platform up-and-down motion within the boundary form the observation data. Observational data within the boundary The original acquisition times for the four types of variables are retained for subsequent time mapping. A unified time axis is constructed using the acquisition times of pressure within the boundary. In a unified timeline Sort by collection time in ascending order. For each... Nearest neighbor time matching is performed within the sampling sequences that are within the boundaries of the raw chlorophyll fluorescence readings, temperature, and the plateau's rise and fall motion states. The matching index is written as... , .when At that time, the corresponding variable value will be bound to When any variable does not meet the time tolerance At that time, the generation will not be performed. The corresponding time corresponds to the observation record. Pressure values ​​are obtained using... The value of the pressure within the boundary at that point.

[0026] In this application, x represents the variable index for performing nearest neighbor time matching, taking the original chlorophyll fluorescence reading, temperature, and platform rise and fall motion state, i.e., x∈{F,T,U}; pressure is used as a unified time axis source, and the value of the pressure within the boundary at τq is directly adopted, so it is no longer included in the value range of x.

[0027] After completing the time mapping, time-corresponding observation data is generated. ,in The pressure sequence was determined based on the direction of pressure changes within the continuous ocean depth observation boundary, with the descending phase corresponding to the increasing pressure sequence and the ascending phase corresponding to the decreasing pressure sequence. The time-corresponding observation data were first arranged according to the order of acquisition time. Then, the continuity of the profiles of adjacent records is verified according to the pressure sequence; when the pressure change direction of adjacent records is inconsistent with the pressure change direction of the full-ocean-depth continuous observation boundary and the pressure change amplitude exceeds... If the record is not found in the full ocean depth observation sequence, it will not be written into the full ocean depth observation sequence. The verified records will be used to generate the full ocean depth observation sequence in chronological order of acquisition time. , , In a feasible fixed configuration, Full-ocean-depth observation sequence for The dimension is used as the input object for generating the thermo-baric motion hysteresis response feature set by S2.

[0028] S2. Determine the temperature and pressure motion driving relationship based on the full ocean depth observation sequence, and generate a temperature and pressure motion hysteresis response characteristic group by combining the hysteresis change of the original chlorophyll fluorescence readings relative to the temperature and pressure motion driving relationship.

[0029] Based on the temperature, pressure, and platform rise and fall motion status in the full ocean depth observation sequence, determine the temperature and pressure motion change sequence within the same observation window;

[0030] Based on the temperature and pressure motion change sequence, a temperature and pressure motion driving relationship is constructed to characterize the combined effects of temperature, pressure, and platform lifting and lowering motion.

[0031] Based on the temperature and pressure motion driving relationship, hysteresis matching was performed on the original chlorophyll fluorescence readings to obtain the hysteresis change of the original chlorophyll fluorescence readings relative to the temperature and pressure motion driving relationship.

[0032] Based on the hysteresis change, determine whether there is a corresponding relationship between the original chlorophyll fluorescence reading and the initial temperature and pressure motion driving the response, and form a hysteresis response determination result;

[0033] Based on the temperature and pressure motion driving relationship and the hysteresis response determination results, a temperature and pressure motion hysteresis response feature set is generated for the channel correlation discovery unit used as input to the DAG.

[0034] A key point of S2:

[0035] After the thermo-baric motion hysteresis response feature set is input into the channel correlation discovery unit of the DAG, source constraints are applied to the channel correlation baseline quantities between environmental variable channels and chlorophyll target states based on the thermo-baric motion hysteresis response feature set. The channel correlation baseline quantities that satisfy the source constraints are then used to calculate the deviation attribution quantity. This constraint embeds the thermo-baric motion hysteresis response feature set formed in S2 into the channel correlation calculation process in S3. Existing DAG channel correlation discovery typically only establishes connections based on the relationship values ​​between variable channels, making it difficult to identify whether the relationship values ​​originate from actual chlorophyll changes or from the sensor response process. This scheme first applies source constraints at the channel correlation baseline quantity level, and then uses the channel correlation baseline quantities that satisfy the constraints to calculate the deviation attribution quantity. In this way, the calculation of the deviation attribution quantity does not deviate from the existing channel correlation links in the DAG, nor does it form an independent correction module. Instead, it reconstructs the judgment criteria at the same key point of action in the main algorithm.

[0036] Step S2 is as follows:

[0037] Full-ocean-depth observation sequence obtained based on S1 ,in , In the fixed example configuration, The current 64 sampling times are used as the same observation window. In online continuous processing, length is used. Observation window with a step size of 1 and along the full ocean depth observation sequence The data collection time sequence and pressure sequence are processed within the window.

[0038] In each observation window Internally, the temperature and pressure change sequence is determined based on the temperature, pressure, and platform lifting and lowering motion. The temperature and pressure changes corresponding to the first sampling index within the window are set to zero. The temperature and pressure changes corresponding to subsequent sampling indices within the window are then determined as the changes in value between adjacent sampling times. The platform's lifting motion state is directly written into the motion state field of the same sampling index. This forms... .in, Indicates the amount of temperature change. This indicates the change in pressure. This indicates the platform's lifting and lowering motion state. It also includes a sequence of temperature and pressure changes. No change to the original chlorophyll fluorescence reading It is only used to express the sensor response driving conditions within the same observation window for temperature, pressure, and platform lifting motion.

[0039] Based on the temperature and pressure motion change sequence Constructing the temperature and pressure motion driving relationship Set a temperature change threshold. Pressure change threshold and motion effective marker threshold .when Not less than When, a valid temperature marker is generated; when Not less than When the direction of pressure change is consistent with the direction of rise and fall corresponding to the boundary of continuous full-ocean-depth observation, a valid pressure marker is generated; when the platform's rise and fall motion state When the platform is in a non-static state, and the direction of its lifting motion is consistent with the direction of pressure change, a valid motion marker is generated. The valid temperature marker, valid pressure marker, and valid motion marker are encoded in the channel order of temperature, pressure, and platform lifting motion state. .when When all three channels meet the valid driving conditions, the sampling index will be... Record in the driver event set Temperature and pressure motion driving relationship This is used to characterize the common driving conditions formed by temperature, pressure, and platform lifting motion state in relation to the sensor response process.

[0040] Based on the driving relationship of temperature and pressure motion Hysteresis matching was performed on the raw chlorophyll fluorescence readings. A threshold for chlorophyll change was set. and the lag candidate set In the fixed example configuration Based on the changes in chlorophyll fluorescence readings between adjacent sampling times, a set of chlorophyll change events is formed. For each ,exist Searching for the content that satisfies The subsequent chlorophyll change events, and adopted Determine the lag-matching sampling index. If no lag candidate set exists... Subsequent chlorophyll change events will be sampled using an index. The corresponding hysteresis change is set to zero.

[0041] Based on the hysteresis matching sampling index The hysteretic change in chlorophyll-generating raw fluorescence readings relative to the temperature-pressure motion-driven relationship. .when When it exists, Compared to The change in the value of is written as ;when If it does not exist, Set to zero. Further generate the hysteresis response determination result. ,in When sampling index Belongs to the set of driving events And there exists a candidate set that satisfies the lag condition. of ,and Not less than the chlorophyll change threshold At that time, Set to 1; otherwise, Set to 0. Hysteresis response determination result. This is used to indicate whether there is a correspondence where a thermo-baric motion-driven relationship occurs first, followed by a response from the original chlorophyll fluorescence reading.

[0042] Based on the driving relationship of temperature and pressure motion Lag changes and delayed response determination results Generate temperature and pressure motion hysteresis response characteristic set ,in In a fixed example configuration, the thermobaric motion hysteresis response characteristic group for Dimension. Characteristics of thermo-baric motion hysteresis response group The temperature and pressure motion hysteresis response feature set is used as a constraint input to the channel correlation discovery unit of the DAG, rather than as a general enhancement input to the chlorophyll target state. This statement defines the boundary between embedded reinforcement and ordinary feature stacking. If the temperature and pressure motion hysteresis response feature set is only used as an enhancement input to the chlorophyll target state, the DAG may still treat the hysteresis response as a target prediction feature and absorb it into the true chlorophyll state; this scheme places the temperature and pressure motion hysteresis response feature set at the constraint position of the channel correlation discovery unit, so that the DAG first determines the source of correlation of the environmental variable channel, and then calculates the bias attribution. This design makes the scene mechanism act on the channel correlation base quantity, rather than adding a general feature that can be absorbed by the black box to the target channel. The channel correlation discovery unit of the DAG receives the temperature and pressure motion hysteresis response feature set. Then, through hysteresis response mapping form The delayed response indicates that, This represents the hysteresis response mapping parameters. The channel correlation discovery unit of the DAG performs source constraints on the channel correlation fundamental quantities between environmental variable channels and the chlorophyll target state based on the hysteresis response representation; only the corresponding... ,and The channel-related basic quantities, which are jointly driven by temperature, pressure, and the platform's lifting and lowering motion, serve as the input basis for subsequent calculations of deviation attribution quantities.

[0043] S3. Input the temperature and pressure motion hysteresis response feature set into the channel correlation discovery unit of the DAG to calculate the deviation assignment between the environmental variable channels and the chlorophyll target state; including: The temperature and pressure motion hysteresis response feature set is input into the channel correlation discovery unit of the DAG, and the temperature, pressure and platform lifting motion state are channelized and mapped according to the temperature and pressure motion hysteresis response feature set to form environmental variable channels. Based on the correspondence between environmental variable channels and chlorophyll target states within the same observation window, calculate the channel-related basic quantities between environmental variable channels and chlorophyll target states; Based on the characteristic set of temperature and pressure motion hysteresis response, the source of deviation is determined for the relevant basic quantities of the channel, and the result of the deviation source determination is formed. Based on the determination of the source of deviation, the portion of the channel-related basic quantities caused by the temperature and pressure motion driving relationship is assigned and calculated to form the deviation correlation quantity; By mapping the deviation correlation quantity to the channel-related basic quantity, the deviation attribution quantity between the environmental variable channel and the chlorophyll target state is obtained; The deviation assignment is used as the criterion for retargeting the state of channel edge weights in the channel correlation detection unit of the DAG. This section explains the location of the deviation assignment. Unlike using the deviation assignment as an output correction ratio, this scheme uses it as the criterion for retargeting the state of channel edge weights, directly determining whether the information of the environmental variable channel enters an environmentally induced deviation state or a true chlorophyll state. This point of application is located inside the DAG channel correlation detection unit, allowing the scenario judgment generated by the temperature and pressure motion hysteresis response feature group to continue influencing the state of edge weights, rather than remaining at the external feature layer. Therefore, the subsequently formed deviation-assigned edges and true-assigned edges have a clear origin.

[0044] Step S3 is as follows: With full ocean depth observation sequence and temperature and pressure motion hysteresis response characteristics group For the input object, , , In the fixed example configuration, Full-ocean-depth observation sequence for Dimension, thermo-baric motion hysteresis response characteristics group for The channel correlation discovery unit of the DAG receives the thermobaric motion hysteresis response feature set. and the thermo-pressure motion hysteresis response characteristics group As the source constraint input in the channel-related calculation process.

[0045] Full ocean depth observation sequence When performing channelization mapping, the temperature channel is denoted as The pressure channel is denoted as The platform lifting motion state channel is recorded as And form a set of environmental variable channels. The channel corresponding to the original chlorophyll fluorescence reading is denoted as... This serves as the chlorophyll target state channel. The input mapping layer receives... , , and The 1D input at each sampling time is mapped through a fully connected network of 32 neurons to form the corresponding variable channel. 3D channel representation. The time-dependent computation layer performs calculations on each variable channel. 3D channel representation models continuous changes within the sampling time, resulting in , , and . , and Bind to environment variable channel sets respectively , Binding to chlorophyll target state channel .

[0046] The thermo-baric motion hysteresis response characteristic group After inputting the channel correlation discovery unit of the DAG, a hysteresis response mapping is used. Generate hysteresis response representation , for dimension, This represents the hysteresis response mapping parameters. Hysteresis response representation. and , , and In the same sampling index This approach embeds the temperature and pressure motion driving relationship, hysteresis changes, and hysteresis response determination results into the channel correlation detection unit of the DAG, rather than treating them as chlorophyll target state channels. The emphasis here is on the embedding position of the reinforcement action. Existing time-series prediction typically feeds exogenous variables, hysteresis quantities, or derived features into the target prediction channel, allowing the model to learn the weights automatically. This scheme places the temperature and pressure motion driving relationship, hysteresis changes, and hysteresis response determination results into the channel correlation detection unit of the DAG, enabling the source of sensor response offset to be identified during the correlation detection stage. This processing embeds scenario-based judgment into the existing channel correlation link of the DAG, directly changing the basis for forming channel relationships, rather than setting up a separate corrector after the DAG output. Normal input.

[0047] In the channel correlation detection unit of the DAG, 16 channel correlation neurons are configured. Each channel correlation neuron connects to two variable channels. The system represents the time-related correlation and outputs a 1D correlation value. Sixteen channel-related neurons collectively form the basic channel-related quantity. In terms of channel-related basic quantities In the process, extract the set of environment variable channels. Pointing to the chlorophyll target state channel The relational value forms the environmental pointer to the basic quantity. . , and The relationships between the temperature channel, pressure channel, platform lifting motion state channel, and chlorophyll target state channel are respectively represented.

[0048] Based on the characteristics of thermo-pressure motion hysteresis response group Environmentally oriented basic quantities The process involves determining the source of bias. This determination is a crucial step between identifying hysteresis response characteristics and identifying the bias attribution. Ordinary correlation analysis can only indicate that environmental variable channels and chlorophyll target states rise and fall together within a window or have a statistical relationship, but it cannot determine whether this relationship originates from a sensor process where temperature and pressure movements occur first and the readings follow. This approach requires that a source support threshold, a valid driving marker, and a corresponding relationship within the same observation window be met before classifying the corresponding environmental variable channel as a valid source. This determination incorporates the observation sequence during the full-depth profile rise and fall process into the DAG channel correlation calculation, reducing the risk of misclassifying ordinary covariance relationships as driving forces of the true chlorophyll state. A source support threshold is set. In the fixed example configuration Each sampling index. For , and Perform source constraint checks separately. For any environment variable channel, when there are at least [number missing] sources within the same observation window... Each sampling index satisfies ,and Includes valid driver tags for the corresponding environment variable channels, while also hysteresis changes. If the corresponding baseline quantity belongs to the same observation window, the deviation source determination result for the corresponding environmental variable channel is set to valid; otherwise, the deviation source determination result for the corresponding environmental variable channel is set to invalid. This forms the deviation source determination result. .

[0049] Based on the results of the deviation source determination Basic quantities related to the channel The portion of the data driven by temperature, pressure, and motion is assigned a channel. The channel correlation detection unit of the DAG is configured with three assigning neurons, each corresponding to an environmental variable channel formed by temperature, pressure, and the platform's vertical motion. Each assigning neuron simultaneously receives hysteresis response representations. and environment-oriented basic quantities The corresponding values ​​in the table are then processed by weighting, biasing, and compressing nonlinear units to output the corresponding environmental variable channel's attribution value. To ensure that the deviation correlation quantity corresponds to the channel-related basic quantity under the same source constraint, for each environmental variable channel... The deviation attribution is determined using the following formula: ; In the formula, Represents environment variable channels The corresponding deviation attribution; This indicates the temperature channel, pressure channel, and platform lifting / lowering motion status channel; Indicates that the environment points to the basic quantity. Corresponding environment variable channels The basic quantities related to the channel; This means that the value inside the parentheses is limited to the range of 0 to 1; This indicates that the value is 1 if the condition inside the parentheses is true and 0 if it is false. Indicates the sampling index; This indicates the number of sampling times within the same observation window; Indicates sampling index The result of the delayed response determination at the location; Indicates sampling index Environment variable channel Effective driving markers in thermo-pressure motion driving relationships; Indicates the source support threshold; Indicates sampling index Lagging changes at the location; This represents a positive constant to prevent the denominator from being zero; it is fixed in the example configuration. ; This represents absolute value operations.

[0050] Will , and The deviation attribution is obtained by combining the channels of temperature, pressure, and platform lifting motion in sequence. . This indicates the degree to which the correlation between environmental variable channels and the chlorophyll target state should be attributed to an environmentally induced bias state. This interpretation defines the true meaning of the bias attribution quantity. In existing models, edge weights typically represent the strength of a variable's contribution to the target. In this scheme, the bias attribution quantity represents the proportion or degree to which the same correlation should transition to an environmentally induced bias state, pointing to the attribution relationship rather than the predicted contribution. Because of this difference in meaning, the bias attribution quantity can serve as the basis for redirecting the target state of channel edge weights: the more the value supports the source of environmentally induced bias, the less the corresponding channel information should be directly entered into the chlorophyll true state link. This changes the way DAG edge weights are used from simple prediction enhancement to state allocation under the constraint of scenario failure mechanisms. The channel correlation detection unit of the DAG will use the bias attribution quantity... The output to the subsequent channel edge weight destination state redirection process makes the deviation assignment quantity... This serves as the basis for determining the edge weight of the channel to be redirected.

[0051] S4. Based on the deviation assignment, redirect the state of the channel edge weights in the channel correlation discovery unit of the DAG to form deviation assignment edges pointing to the environmentally induced deviation state and true assignment edges pointing to the true chlorophyll state; including: Based on the deviation assignment, the channel edge weights to be redirected between the environmental variable channel and the chlorophyll target state are determined in the channel correlation discovery unit of the DAG. Based on the edge weights of the channel to be redirected and the deviation attribution quantity, determine the edge weight portion of the edge weights of the channel to be redirected that belongs to the environmentally induced deviation state, and form the basic quantity of deviation attribution edge weights. Based on the baseline quantity of the deviation attribution edge, the target state of the channel edge weights to be redirected is redirected from the chlorophyll target state to the environment-induced deviation state, forming a deviation attribution edge pointing to the environment-induced deviation state. This is the direct focus of the core edge weight improvement. Existing multivariate prediction allows temperature, pressure, and platform lifting / lowering motion states to be injected into the chlorophyll target state along relevant edges, causing sensor response offsets to be interpreted as true chlorophyll changes. This scheme, when the baseline quantity of the deviation attribution edge is valid, changes the target state of the channel edge weights to be redirected to the environment-induced deviation state, preventing this information from contaminating the true chlorophyll state. This processing retains the environmental variable channel as the source and only changes the target state, indicating that the improvement occurs within the DAG edge weight pointing mechanism itself, rather than initiating a separate parallel model or post-processing step.

[0052] Based on the edge weights in the channel to be redirected that do not belong to the environmentally induced bias state, a true attribution edge pointing to the true state of chlorophyll is formed. The bias-attributed edge and the true-attributed edge are used as the channel edge weights in the DAG to update the environmentally induced bias state and the true chlorophyll state.

[0053] A key point of S4:

[0054] The bias assignment factor is applied to the edge weights of the channels to be redirected in the channel correlation discovery unit of the DAG, and the edge weights of the channels to be redirected are assigned as bias assignment edges pointing to the environmentally induced bias state and true assignment edges pointing to the true chlorophyll state, based on the bias assignment factor. This split allows the relevant information of the same environmental variable channel to obtain two different state destinations. The conventional approach usually only judges whether the environmental variable contributes to the chlorophyll target state, and the stronger the contribution, the more likely it is to be absorbed into the prediction channel by the model; this scheme assigns the edge weights of the channels to be redirected as bias assignment edges and true assignment edges based on the bias assignment factor, respectively receiving sensor response offset and environmentally relevant information that can still serve the update of the true state. This structure avoids the problem of simply deleting environmental variables or completely excluding environmental information, and also differs from the static correction method that directly uses all environmental variables as compensation parameters.

[0055] Step S4 is as follows: Deviation attribution based on S3 output , , and These correspond to the temperature channel, pressure channel, and platform lifting / lowering motion status channel, respectively. The DAG channel-related discovery unit has already formed a set of environmental variable channels. Chlorophyll target state channel Environmental orientation basic quantity and time-related representations of each environmental variable channel. , and S4 does not recalculate the characteristic set of temperature and pressure motion hysteresis response, nor does it change the source constraint results of the channel-related basic quantities. Instead, it assigns the deviation to the quantity. In the channel correlation discovery unit acting on the DAG, the environmental variable channel points to the chlorophyll target state channel. The constraint directly applies the bias assignment to the channel edge weights of the target state. This constraint fixes the application point of the bias assignment at the bottom layer of the main algorithm. Existing post-processing calibration simply multiplies the model output or sensor readings by empirical coefficients, which cannot change how the environmental variable channels within the model affect the target state. This scheme allows the bias assignment to directly act on the channel edge weights of the environmental variable channels pointing to the chlorophyll target state channels, thus redistributing the original target injection path. In this way, the environmentally induced bias state and the true chlorophyll state become updatable objects within the DAG, allowing subsequent S5 and S6 to continue using them along the same main solution chain, instead of taking the DAG prediction results outside for further modification.

[0056] In the channel correlation discovery unit of DAG, the original channel edge weights between environmental variable channels and chlorophyll target states are denoted as the set of channel edge weights to be retargeted. . This indicates that the temperature channel points to the chlorophyll target state channel. The right-hand side of the channel to be redirected This indicates that the pressure channel points to the chlorophyll target state channel. The right-hand side of the channel to be redirected This indicates that the platform's lifting motion status channel points to the chlorophyll target status channel. The edge weight of the channel to be redirected. The edge weight of the channel to be redirected... The value is denoted as and make Environmentally oriented basic quantities Correspondence values ​​in Consistent, Therefore, the set of edge weights for the channel to be redirected. According to the channel sequence and deviation assignment based on temperature, pressure, and platform lifting motion status. Binding.

[0057] Set environmental induced deviation status node and chlorophyll true state node Environmentally induced deviation state nodes Used to determine the true state node of chlorophyll based on sensor response offset information caused by temperature and pressure motion. Used for chlorophyll-related information based on states not assigned to environmentally induced biases. For any environmental variable channel Assign the corresponding deviation to the quantity This serves as the basis for allocating edge weights for state redirection. To ensure that the edge weights of the channels to be redirected maintain the same numerical scale before and after redirection, and that both the baseline values ​​of the biased and true edge attributes are generated from directly usable edge weight values, the following rules are adopted to determine them. and : ; In the formula, Represents environment variable channels The corresponding deviation belongs to the basic quantity of the edge; Represents environment variable channels The corresponding actual number of belonging edges; This indicates that the value is 1 if the condition inside the parentheses is true and 0 if it is false. Indicates the edge weight of the channel to be redirected The value; Indicates the effective threshold of edge weights. and They have the same numerical scale; Represents a symbolic function; This represents the absolute value operation; Represents environment variable channels The corresponding deviation attribution; This means that the value inside the parentheses is limited to the range of 0 to 1; Indicates that the environment points to the basic quantity. Corresponding environment variable channels The basic quantities related to the channel; The channel index indicates the temperature channel, pressure channel, and platform lifting motion status channel.

[0058] Here, qc is still mapped to the environment variable channel Cc, where Cc represents the specific environment variable channel and c represents the index of that channel, taking T, P, or U. This notation is used to ensure that the deviation assignment, the edge weight of the channel to be redirected, and the subsequent edge weight allocation maintain the same channel order.

[0059] Then, AE is used to represent the environment pointing to the overall basic quantity, and ac represents the basic quantity related to the channel corresponding to the environment variable channel Cc. Therefore, ac is used here to take the specific channel value, and AE is only retained as its source object. I think this can still keep the hierarchical relationship between the overall object and the single channel component clear.

[0060] Based on the deviation belonging to the basic quantity , and Set the edge weights of the channel to be redirected The target state of the corresponding edge is determined by the chlorophyll target state channel. Redirect to the environmentally induced deviation state node This forms the set of bias-attributed edges. . The temperature channel points to the environmentally induced deviation state node. The deviation belongs to the edge. The pressure channel points to the environmentally induced deviation state node. The deviation belongs to the edge. The platform's lifting motion status channel points to the environmental induced deviation status node. The deviation belonging edge. The set of deviation belonging edges. This approach retains the original environmental variable channels as the source, maintains the sampling index binding relationships within the same observation window, and only changes the target state of the channel edge weights. The improvement focuses on the reassignment of information injection direction. For a full-ocean-depth mobile platform, temperature, pressure, and platform elevation / reduction motion are not useless variables but need to be sent to the correct state; therefore, this approach does not delete environmental variable channels, nor does it treat all environmental variables as predictive features. Instead, it sends the deviation portion confirmed by the hysteresis response into the environmentally induced deviation state.

[0061] Based on the true belonging edge base quantity , and Set the edge weights of the channel to be redirected The edge weights that do not belong to the environmentally induced bias state point to the chlorophyll true state node. This forms the set of true belonging edges. . The temperature channel is pointed to the chlorophyll true state node. The true belonging edge, The pressure channel points to the chlorophyll true state node. The true belonging edge, The platform's lifting motion status channel points to the chlorophyll's actual state node. The true attribution edge. After completing the state redirection of the channel edge weight destination, the set of biased attribution edges will be... and the set of true belonging edges Combined into channel edge weights and the channel edge weight results Input S5 is used to update the environmentally induced bias state and the true chlorophyll state. This indicates that the redirected edges are not statically labeled, but continue to participate in state updates. Existing technologies, even when identifying certain environmental factors affecting readings, often remain at the level of compensation amounts or correction curves; this scheme combines the bias-attributed edge set and the true-attributed edge set into a channel edge weight result, which is then input into S5, allowing the environmentally induced bias state and the true chlorophyll state to obtain updated information separately in the DAG. This continuous reference relationship ensures that the bias assignment amount, edge weight redirection, state update, and dynamic correction belong to the same algorithm chain, rather than being a post-hoc fusion of multiple independent results.

[0062] S5. Update the environmentally induced deviation state based on the deviation-assigned edge, and update the true chlorophyll state based on the true-assigned edge to form a structured chlorophyll state; including: Based on the deviation attribution edge, the deviation injection information of the environmental variable channel to the environmental induced deviation state is determined in the DAG; the environmental induced deviation state is updated based on the deviation injection information to form the updated environmental induced deviation state. Based on the true attribution edge, determine the true injection information of environmental variable channels into the true state of chlorophyll in the DAG; The true state of chlorophyll is updated based on the actual injection information to form the updated true state of chlorophyll. Based on the updated environmentally induced bias state and the updated true chlorophyll state, a structured chlorophyll state is formed; Using structured chlorophyll state as the basis for dynamic correction of raw chlorophyll fluorescence readings, and closing S5 and S6, is key to demonstrating the continuity of the technology chain. Existing corrections may directly apply temperature or turbidity compensation to the raw chlorophyll fluorescence readings, relying on external parameters or empirical models. Here, the dynamic correction is based on structured chlorophyll state, which already incorporates the updated results of both the biased and true biased edges. Therefore, the dynamic correction amount is not calculated based on a single environmental variable, but rather generated from the state object after separating the environmentally induced biased state from the true chlorophyll state, allowing the correction process to inherit the state redirection results of the internal edge weights of the DAG.

[0063] One of the core aspects of S5: The bias-attributed edge and the true-attributed edge are used as the channel edge weights in the DAG. The bias-attributed edge is used to update the environmentally induced bias state, and the true-attributed edge is used to update the chlorophyll true state. The structured chlorophyll state is formed based on the updated environmentally induced bias state and the updated chlorophyll true state.

[0064] Step S5 is as follows: Channel edge weight results based on S4 output , For the set of edges to which the deviation belongs, The set of true belonging edges. The DAG simultaneously receives time-correlated representations of the temperature channel, pressure channel, platform lifting motion state channel, and chlorophyll target state channel, denoted as follows: , , and In the fixed example configuration, , , and All Dimensional time-dependent representation, sampling index is The channel represents the dimension index. .

[0065] According to the set of edges to which the deviation belongs In the DAG, the deviation injection information of environmental variable channels to the environmentally induced deviation state is determined. This is the underlying update method for the environmentally induced deviation state. Ordinary models may uniformly concatenate environmental variable channels and send them to the target prediction layer, and the response path and deviation source on the sampling index will be smoothed out during feature fusion. This scheme establishes adjacency aggregation connections pointing to the environmentally induced deviation state nodes through deviation attribution edges, retains the channel representation at each sampling index, and then injects them in a weighted manner according to the edge weight value of the deviation attribution edge. In this way, the contributions of temperature, pressure, and platform lifting and lowering motion states to the sensor response offset enter the same state node along the time index, forming deviation injection information that can be dynamically corrected and read. Time-related representation of connection to temperature channel ,Will Time-related representation of connection to pressure channel ,Will Time-related representation of connection to the platform lifting motion status channel .Will , and The edge weights are denoted as follows: , and DAG is determined by the set of edges to which the deviation belongs. Establish a state node pointing from the environmental variable channel to the environmental induced deviation. Adjacency aggregation join. Adjacency aggregation join retains each sampling index. The 32-dimensional channel representation at the location is used to weight and inject the information of the corresponding environmental variable channel according to the edge weight value of the deviation attribution edge, forming the deviation injection information. .

[0066] Based on the set of true belonging edges In the DAG, the true injection information of environmental variable channels into the actual state of chlorophyll is determined. , and Connected to respectively , and Furthermore, the time-correlation representation of the chlorophyll target state channel is used. This constraint avoids writing the real-state update as an environmental variable-driven process. Unlike methods that directly use temperature, pressure, and platform motion as the source of real chlorophyll changes, this scheme uses the time-correlation representation of the chlorophyll target state channel as the main state input for updating the real chlorophyll state. The true attribution edge only accepts environmental information not assigned to the environmentally induced bias state. This structure preserves the dominance of continuous chlorophyll changes in the real state while allowing necessary environmental information to enter in a controlled manner through the true attribution edge, preventing sensor response offsets from replacing real chlorophyll changes. As the main state input for updating the true state of chlorophyll. , and The edge weights are denoted as follows: , and A DAG is defined according to the set of true belonging edges. Establish a node pointing from environmental variable channels to the actual state of chlorophyll. Adjacency aggregation join, and in the same sampling index The 32-dimensional representation of the environmental variable channels and The 32-dimensional representation in the data is used for state binding to form the actual injected information. .

[0067] Let the environmental induced deviation state be denoted as The true state of chlorophyll is recorded as For the sampling index Using deviation injection information initialization Using real injection information initialization For the sampling index Using the same sampling index, the environmental variable channel time correlation representation, the chlorophyll target state channel time correlation representation, the bias-attributed edge value, and the true-attributed edge value, the state update rules are formed according to the following rules. and : ; In the formula, Indicates sampling index Channels represent dimensional indexes Update results of environmentally induced bias state at the location; Indicates sampling index Channels represent dimensional indexes Updated results of chlorophyll true state at the location; Indicates the sampling index; Channels represent dimension indexes; The state continuity coefficient represents the state of environmentally induced deviation. The state continuity coefficient represents the true state of chlorophyll; This indicates the environmental induced deviation state value at the previous sampling index; This represents the true chlorophyll state value at the previous sampling index; Indicates the channel index of the environment variable; Indicates the temperature channel; Indicates a pressure channel; This indicates the platform's lifting and lowering motion status channel; Represents environment variable channels The edge weight value corresponding to the edge to which the deviation belongs; Represents environment variable channels The edge weight corresponding to the actual belonging edge; Represents environment variable channels In the sampling index Channels represent dimensional indexes The time-related numerical value at that location; Indicates the chlorophyll target state channel at the sampling index Channels represent dimensional indexes The time-related numerical value at that location; This represents the summation of the temperature channel, pressure channel, and platform lifting motion status channel; This represents the absolute value operation; A positive constant that prevents the denominator from being zero; a constant in the denominator. Channel representing chlorophyll target state Unit injection coefficient in chlorophyll true state update.

[0068] In this invention, the constant 1 in the denominator represents the unit injection coefficient of the chlorophyll target state channel HF in the update of the chlorophyll true state. Its function is to retain the baseline weight of HF as the master state input. This constant can also be expressed as a fixed positive coefficient, and is set to 1 in a fixed example configuration to keep the calculation scale clear; other values ​​are also acceptable.

[0069] After completing the state update of all sampling indices, the updated environmental induced bias state is as follows. for Dimensions, updated true state of chlorophyll for Dimension. Based on the updated environmentally induced bias state. And the updated true state of chlorophyll To form a structured chlorophyll state , Structured chlorophyll state for The structured chlorophyll state is organized into two dimensions: the first 32 dimensions correspond to the environmentally induced bias state, and the latter 32 dimensions correspond to the true chlorophyll state. This dimensional organization transforms the abstract structured chlorophyll state into actionable data objects. Existing correction results typically only contain a single chlorophyll value, making it difficult to trace which part originates from the sensor's environmental response. This solution retains both environmentally induced bias state and true chlorophyll state representations within the structured chlorophyll state, enabling S6 to calculate dynamic correction amounts from the environmentally induced bias state and determine state consistency from the true chlorophyll state. This organization is not intended to increase the number of dimensions, but rather to allow the two states to be readable, usable, and jointly support correction within the same DAG main link. The DAG represents the structured chlorophyll state. Transmitted to S6, and the structured chlorophyll state is then transferred. This serves as the basis for dynamically correcting the original fluorescence readings of chlorophyll.

[0070] S6. Dynamically correct the raw chlorophyll fluorescence readings based on the structured chlorophyll state to obtain the corrected chlorophyll state; including: Based on the structured chlorophyll state, determine the dynamic correction amount of the environmentally induced bias state relative to the original chlorophyll fluorescence reading; Based on the dynamic correction amount, the original chlorophyll fluorescence readings are dynamically corrected to form the initial corrected chlorophyll state; Based on the actual chlorophyll state in the structured chlorophyll state, the initial corrected chlorophyll state is judged for state consistency, and a state consistency judgment result is formed. Based on the state consistency determination results, the initial corrected chlorophyll state is adjusted to obtain the corrected chlorophyll state.

[0071] Step S6 is as follows: Structured chlorophyll state obtained based on S5 ,in . For sampling index The 32-dimensional environmental induced deviation state at that location. For sampling index The 32-dimensional true state of chlorophyll at the location. Simultaneous reception of raw chlorophyll fluorescence readings from the full-depth observation sequence. , for Dimension. S6 does not redetermine the biased and true attribution edges, but rather operates within the structured chlorophyll state. Based on the separation of environmentally induced bias state and true chlorophyll state, the raw chlorophyll fluorescence readings were analyzed. Perform dynamic correction.

[0072] Based on the structured chlorophyll state First, start with each Read the environmental induced deviation state Output layer settings for correction mapping , This represents the correction mapping parameters. Correction mapping by and raw chlorophyll fluorescence readings under the same sampling index Input: Sample index Dynamic correction amount at the location This forms a dynamic correction sequence. Dynamic correction amount These are signed numerical values. Positive values ​​represent the original fluorescence reading of chlorophyll. It includes positive environmental-induced bias, and negative values ​​represent raw chlorophyll fluorescence readings. It includes negative environmental induced bias. Set the correction amplitude threshold. ,when Greater than At that time, Limited to Within the corresponding amplitude range.

[0073] The signed dynamic correction value reflects that the correction is not a fixed deduction or unidirectional compensation. Temperature decreases, pressure increases, and platform elevation movements in the full-depth profile can cause positive or negative shifts in the raw chlorophyll fluorescence readings; existing static compensation methods struggle to express the changes in the direction of deviation at different stages of the same profile. This scheme reads the correction basis from the environmentally induced deviation state in the structured chlorophyll state, forming a dynamic correction value that can be positive or negative, ensuring that the correction action aligns with the source of deviation at the sampling index. This processing truly transforms the environmentally induced deviation state of S5 into an executable correction operation of S6.

[0074] Based on dynamic correction amount Raw fluorescence readings of chlorophyll Perform dynamic correction. For each sampling index chlorophyll raw fluorescence readings Subtract dynamic correction amount To form an initial corrected chlorophyll state Initial correction of chlorophyll state for The sampling index order and pressure order are kept consistent with the full ocean depth observation sequence.

[0075] Based on the structured chlorophyll state The true state of chlorophyll in For the initial corrected chlorophyll state A state consistency determination is performed (this determination prevents dynamic correction from deviating from the true chlorophyll state based solely on the bias state. Ordinary reading compensation methods typically calculate the correction value and output it directly, with the prediction stage passively accepting the result; this scheme also utilizes the true chlorophyll state from the structured chlorophyll state to form a true state reference value, performing a consistency determination on the initial corrected chlorophyll state. This step ensures that dynamic correction is simultaneously constrained by both the environmentally induced bias state and the true chlorophyll state, forming a closed loop of "bias deduction" and "true state verification," rather than applying environmental compensation parameters in isolation to the original fluorescence reading). The output layer sets a true state mapping. , Represents the parameters of the real state mapping. Real state mapping by Input: Sample index Reference value of the actual state at the location And form a real state reference sequence. Set a state consistency threshold. .when and The absolute difference between them is no greater than At that time, the sampling index will be used. Corresponding state consistency determination result Set to consistent; when and The absolute difference between them is greater than At that time, the sampling index will be used. Corresponding state consistency determination result This is set to inconsistent. This results in a state consistency determination. .

[0076] Based on the state consistency determination result For the initial corrected chlorophyll state Make adjustments. Set the adjustment amplitude threshold. .when When consistent, Directly used as a correction for chlorophyll state .when Inconsistent, and Greater than At that time, in accordance with no more than Amplitude reduction .when Inconsistent, and Less than At that time, in accordance with no more than The amplitude increased After completing all sampling index processing, the corrected chlorophyll state is obtained. Correcting chlorophyll status for The dimension is used as the chlorophyll state input for S7 to generate chlorophyll prediction results.

[0077] S7. Generate chlorophyll prediction results based on the corrected chlorophyll state, and output sensor data results including the corrected chlorophyll state and chlorophyll prediction results.

[0078] Based on the corrected chlorophyll state, determine the chlorophyll state input used to generate chlorophyll prediction results; The chlorophyll state input is fed into the prediction output process of the DAG to generate chlorophyll prediction results; Based on the corrected chlorophyll state and chlorophyll prediction results, sensor data results are generated; The output includes sensor data results that correct for chlorophyll status and chlorophyll prediction.

[0079] In this embodiment, step S7 specifically includes:

[0080] Corrected chlorophyll state based on S6 , Indicates sampling index Corrected chlorophyll state at the location. In the fixed example configuration, chlorophyll state is corrected. for The system maintains the acquisition time sequence and pressure sequence corresponding to the full ocean depth observation sequence. S7 no longer receives raw chlorophyll fluorescence readings as prediction input, nor does it receive environmentally induced bias states as separate prediction inputs; instead, it uses the dynamically corrected chlorophyll state. This serves as the data foundation for subsequent prediction output processes.

[0081] According to the corrected chlorophyll status Determine the chlorophyll state inputs used to generate chlorophyll prediction results. In the fixed example configuration, Write directly according to the sampling index order And retain the 1D state value at each sampling index. Chlorophyll state input No additional fields for temperature, pressure, and platform motion state are added. This section explains that the prediction input constraints are derived from the previous structured separation results. Conventional exogenous variable prediction continues to input temperature, pressure, and motion state during the prediction phase, and the model may still interpret sensor response biases as future chlorophyll changes. In this scheme, these environmental information elements have already participated in state separation through temperature, pressure, motion hysteresis response feature groups, bias attribution, bias attribution edges, and true attribution edges in S2 to S5. Therefore, S7 only receives the chlorophyll state input formed by correcting the chlorophyll state. Thus, the prediction output process receives the corrected continuous chlorophyll state, rather than a multivariate input that is again mixed with environmentally induced biases. Temperature, pressure, and platform motion state have already participated in structured state separation in S2 to S5 through temperature, pressure, motion hysteresis response feature groups, bias attribution, bias attribution edges, and true attribution edges; the prediction output process only receives the chlorophyll state input formed by correcting the chlorophyll state. This ensures continuous changes in the predicted input after correction for the actual state of chlorophyll.

[0082] Input chlorophyll status The prediction output process of the DAG is accessed. The prediction output process of the DAG is located after the output layer, and it receives... Chlorophyll status input and according to the sampling index to The corrected chlorophyll state changes are read sequentially. The prediction output process sets 16 prediction sampling time points as output nodes, and the prediction sampling time index is denoted as... At each prediction sampling time, the output node is connected to the chlorophyll state input. The corresponding continuous state representation is used to output a 1D predicted value. The output nodes at 16 prediction sampling times jointly generate the chlorophyll prediction result. , Indicates the first Chlorophyll prediction results at each prediction sampling time. In a fixed example configuration, the chlorophyll prediction results... for dimension.

[0083] According to the corrected chlorophyll status Chlorophyll prediction results To generate sensor data results Sensor data results Organized in order of result type, the first 64 output positions correspond to the corrected chlorophyll state. The last 16 output positions correspond to the chlorophyll prediction results. Sensor data results for Output object. Output sensor data results. At the same time, the sampling index corresponding to the corrected chlorophyll state is retained. chlorophyll prediction results and corresponding prediction sampling time index To make sensor data results It can correspond to the correction results within the observed full-depth ocean profile and the prediction results at the subsequent prediction sampling time, respectively.

[0084] Figure 2 In the figure, (a), (b), and (c) respectively show the changes in temperature, pressure, and chlorophyll state profiles during the continuous full-ocean-depth observation process of the mobile platform. Figure 2 The temperature profile in (a) shows the pattern that seawater temperature decreases rapidly with depth and then tends to stabilize. Figure 2 The pressure profile in (b) shows an approximately linear increase with increasing depth; Figure 2 (c) Comparison of raw chlorophyll fluorescence readings and corrected chlorophyll states illustrates that the raw readings are subject to environmental response disturbances at different depth intervals, while the corrected states are closer to the trend of continuous profile changes.

[0085] Figure 3 Using the sampling sequence number as the horizontal axis, the continuous changes in the original chlorophyll fluorescence readings, corrected chlorophyll state, and true chlorophyll state are compared. The figure shows that at the beginning and end of the profile, the original fluorescence readings exhibit significant fluctuations and shifts, while the corrected chlorophyll state's trajectory is more consistent with the true chlorophyll state. This figure demonstrates the mitigating effect of dynamic correction on the environmentally induced bias in the original readings.

[0086] Figure 4 This paper demonstrates the corresponding changes in the temperature-pressure motion driving relationship and the deviation assignment during continuous sampling, and indicates the results of the hysteresis response determination. In the intervals where the temperature-pressure motion driving relationship increases, the deviation assignment usually increases synchronously, indicating that the influence of temperature, pressure, and the state of plateau motion on the raw chlorophyll fluorescence readings is not a typical exogenous correlation, but rather enters the deviation assignment calculation through the hysteresis response process.

[0087] Figure 5 This diagram illustrates the bias-attributed edges and true-attributed edges formed after the state redirection of channel edge weights. Bias-attributed edges are used to receive environmental variable channel information that should point to the environmentally induced bias state, while true-attributed edges retain valid information pointing to the true state of chlorophyll. The two types of edge weights exhibit dynamic separation at different sampling stages, reflecting the constraint effect of the bias-attributed amount on the direction of information injection within the DAG channel-related discovery units.

[0088] Figure 6The differences in corrected mean absolute error and predicted mean absolute error between the complete scheme and several ablation schemes were compared. The complete scheme had the lowest error; after removing the characteristic group of temperature and pressure motion hysteresis response, the deviation assignment, the state redirection of the channel edge weights, or the environmentally induced deviation state update, the errors all increased to varying degrees.

[0089] Figure 7 The corrected mean absolute error of each ablation scheme is displayed according to depth range. The darker the color, the higher the error. The complete scheme maintains low error across all depth ranges, while the conventional DAG and the scheme that removes channel edge weights and redirects the state shows more significant errors in shallow and mid-deep layers.

[0090] Figure 8 A three-dimensional surface was constructed using sampling sequence number, depth, and chlorophyll state. The surface represents the continuous change of chlorophyll state across the full ocean depth profile, and the color indicates the amount of deviation. The red frame represents the original chlorophyll fluorescence reading before correction, and the bottom projection represents the dynamic correction.

Claims

1. A method for dynamic correction and prediction of profile data of a seawater optical chlorophyll sensor, characterized in that, Includes the following steps: S1. Acquire raw chlorophyll fluorescence readings, temperature, pressure, and platform elevation and descent status during continuous full-ocean-depth observations of the mobile platform to form a full-ocean-depth observation sequence; S2. Determine the temperature and pressure motion driving relationship based on the full ocean depth observation sequence, and generate a temperature and pressure motion hysteresis response characteristic group by combining the hysteresis change of the original chlorophyll fluorescence readings relative to the temperature and pressure motion driving relationship. S3. Input the temperature and pressure motion hysteresis response feature group into the channel correlation discovery unit of the DAG to calculate the deviation assignment between the environmental variable channel and the chlorophyll target state; S4. Based on the deviation assignment, the state of the channel edge weight in the channel correlation discovery unit of the DAG is redirected to form the deviation assignment edge pointing to the environmentally induced deviation state and the true assignment edge pointing to the true chlorophyll state. In S4, based on the deviation assignment, the channel weights to be redirected between the environmental variable channel and the chlorophyll target state are determined in the channel correlation discovery unit of the DAG; Based on the edge weights of the channel to be redirected and the deviation attribution quantity, determine the edge weight portion of the edge weights of the channel to be redirected that belongs to the environmentally induced deviation state, and form the basic quantity of deviation attribution edge weights. Based on the basic quantity of the deviation attribution edge, the target state of the channel edge weight to be redirected is redirected from the chlorophyll target state to the environmentally induced deviation state, forming a deviation attribution edge pointing to the environmentally induced deviation state. Based on the portion of the edge weights in the channel to be redirected that do not belong to the environmentally induced deviation state, a true attribution edge base quantity pointing to the true state of chlorophyll is formed. The bias-attributed edge and the true-attributed edge are used as the channel edge weights in the DAG to update the environmentally induced bias state and the true chlorophyll state. To keep the same value scale of the channel edge weight to be redirected before and after the redirection, and to make both the bias attribution edge base quantity and the true attribution edge base quantity generated by the directly usable edge weight value, the following rules are adopted to determine and : ； In the formula, Represents environment variable channels The corresponding deviation belongs to the basic quantity of the edge; Represents environment variable channels The corresponding actual number of belonging edges; This indicates that the value is 1 if the condition inside the parentheses is true and 0 if it is false. Indicates the edge weight of the channel to be redirected The value; Indicates the effective threshold of edge weights. and They have the same numerical scale; Represents a symbolic function; This represents the absolute value operation; Represents environment variable channels The corresponding deviation attribution; This indicates that the value within the parentheses is limited to the range of 0 to 1; Indicates that the environment points to the basic quantity. Corresponding environment variable channels The basic quantities related to the channel; This indicates that the channel index retrieves the temperature channel, pressure channel, and platform lifting motion status channel; S5. Update the environmentally induced deviation state based on the deviation-assigned edge, and update the true chlorophyll state based on the true-assigned edge to form a structured chlorophyll state. S6. Dynamically correct the original chlorophyll fluorescence readings based on the structured chlorophyll state to obtain the corrected chlorophyll state. S7. Generate chlorophyll prediction results based on the corrected chlorophyll state, and output sensor data results including the corrected chlorophyll state and chlorophyll prediction results.

2. The method for dynamic correction and prediction of seawater optical chlorophyll sensor profile data according to claim 1, characterized in that, S1 collects raw chlorophyll fluorescence readings, temperature, pressure, and platform elevation and descent status during continuous full-ocean-depth observations of the mobile platform, and generates raw observation data based on the collection time. The full ocean depth continuous observation boundary was determined based on the pressure changes in the raw observation data, and the raw chlorophyll fluorescence readings, temperature, pressure, and platform rise and fall motion status within the full ocean depth continuous observation boundary were used to form the boundary observation data. Based on the acquisition time in the observation data within the boundary, the original chlorophyll fluorescence readings, temperature, pressure, and platform lifting and lowering motion status are correlated over time to form time-corresponding observation data; A full-ocean-depth observation sequence is generated based on the pressure order and acquisition time order in the time-corresponding observation data.

3. The method for dynamic correction and prediction of seawater optical chlorophyll sensor profile data according to claim 1, characterized in that, In S2, the temperature and pressure movement sequence within the same observation window is determined based on the temperature, pressure, and platform rise and fall motion status in the full ocean depth observation sequence; Based on the temperature and pressure motion change sequence, a temperature and pressure motion driving relationship is constructed to characterize the combined effects of temperature, pressure, and platform lifting and lowering motion. Based on the temperature and pressure motion driving relationship, hysteresis matching was performed on the original chlorophyll fluorescence readings to obtain the hysteresis change of the original chlorophyll fluorescence readings relative to the temperature and pressure motion driving relationship. Based on the hysteresis change, determine whether there is a corresponding relationship between the original chlorophyll fluorescence reading and the initial temperature and pressure motion driving the response, and form a hysteresis response determination result; Based on the temperature and pressure motion driving relationship and the hysteresis response determination results, a temperature and pressure motion hysteresis response feature set is generated for the channel correlation discovery unit used as input to the DAG.

4. The method for dynamic correction and prediction of seawater optical chlorophyll sensor profile data according to claim 1, characterized in that, In S3, the temperature and pressure motion hysteresis response feature group is input into the channel correlation discovery unit of the DAG, and the temperature, pressure and platform lifting motion state are channelized and mapped according to the temperature and pressure motion hysteresis response feature group to form environmental variable channels. Based on the correspondence between environmental variable channels and chlorophyll target states within the same observation window, calculate the channel-related basic quantities between environmental variable channels and chlorophyll target states; Based on the characteristic set of temperature and pressure motion hysteresis response, the source of deviation is determined for the relevant basic quantities of the channel, and the result of the deviation source determination is formed. Based on the determination of the source of deviation, the portion of the channel-related basic quantities caused by the temperature and pressure motion driving relationship is assigned and calculated to form the deviation correlation quantity; By mapping the deviation correlation quantity to the channel-related basic quantity, the deviation attribution quantity between the environmental variable channel and the chlorophyll target state is obtained; The deviation assignment is used as the basis for determining the state redirection of the channel edge weight in the channel correlation discovery unit of the DAG.

5. The method for dynamic correction and prediction of seawater optical chlorophyll sensor profile data according to claim 4, characterized in that, For each environment variable channel The deviation attribution is determined using the following formula: ； In the formula, Represents environment variable channels The corresponding deviation attribution; This indicates the temperature channel, pressure channel, and platform lifting / lowering motion status channel; Indicates that the environment points to the basic quantity. Corresponding environment variable channels The basic quantities related to the channel; This indicates that the value within the parentheses is limited to the range of 0 to 1; This indicates that the value is 1 if the condition inside the parentheses is true and 0 if it is false. Indicates the sampling index; This indicates the number of sampling times within the same observation window; Indicates sampling index The result of the delayed response determination at the location; Indicates sampling index Environment variable channel Effective driving markers in thermo-pressure motion driving relationships; Indicates the source support threshold; Indicates sampling index Lagging changes at the location; Represents a positive constant to prevent the denominator from being zero; This represents the absolute value operation; Will , and The deviation attribution is obtained by combining the channels of temperature, pressure, and platform lifting motion in sequence. , , and These correspond to the temperature channel, pressure channel, and platform lifting motion status channel, respectively.

6. The method for dynamic correction and prediction of seawater optical chlorophyll sensor profile data according to claim 1, characterized in that, Step S5: The bias-assigned edge and the true-assigned edge are used as the channel edge weights in the DAG. The bias-assigned edge is used to update the environmentally induced bias state, and the true-assigned edge is used to update the chlorophyll true state. The structured chlorophyll state is formed based on the updated environmentally induced bias state and the updated chlorophyll true state. Let the environmental induced deviation state be denoted as The true state of chlorophyll is recorded as For sampling index Using the same sampling index, the environmental variable channel time correlation representation, the chlorophyll target state channel time correlation representation, the bias-attributed edge value, and the true-attributed edge value, the state update rules are formed according to the following rules. and : ； In the formula, Indicates sampling index Channels represent dimensional indexes Update results of environmentally induced bias state at the location; Indicates sampling index Channels represent dimensional indexes Updated results of chlorophyll true state at the location; Indicates the sampling index; Channels represent dimension indexes; The state continuity coefficient represents the state of environmentally induced deviation. The state continuity coefficient represents the true state of chlorophyll; This indicates the environmental induced deviation state value at the previous sampling index; This represents the true chlorophyll state value at the previous sampling index; Indicates the channel index of the environment variable; Indicates the temperature channel; Indicates a pressure channel; This indicates the platform's lifting and lowering motion status channel; Represents environment variable channels The edge weight value corresponding to the edge to which the deviation belongs; Represents environment variable channels The edge weight corresponding to the actual belonging edge; Represents environment variable channels In the sampling index Channels represent dimensional indexes The time-related numerical value at that location; The chlorophyll target state channel is represented at the sampling index. Channels represent dimensional indexes The time-related numerical value at that location; This represents the summation of the temperature channel, pressure channel, and platform lifting motion status channel; This represents the absolute value operation; Represents a positive constant to prevent the denominator from being zero; Channel representing chlorophyll target state Unit injection coefficient in chlorophyll true state update.

7. The method for dynamic correction and prediction of seawater optical chlorophyll sensor profile data according to claim 1, characterized in that, S6 determines the dynamic correction amount of the environmentally induced bias state relative to the original chlorophyll fluorescence reading based on the structured chlorophyll state. Based on the dynamic correction amount, the original chlorophyll fluorescence readings are dynamically corrected to form the initial corrected chlorophyll state; Based on the actual chlorophyll state in the structured chlorophyll state, the initial corrected chlorophyll state is judged for state consistency, and a state consistency judgment result is formed. Based on the state consistency determination results, the initial corrected chlorophyll state is adjusted to obtain the corrected chlorophyll state.

8. The method for dynamic correction and prediction of seawater optical chlorophyll sensor profile data according to claim 1, characterized in that, S7 determines the chlorophyll state input used to generate chlorophyll prediction results based on the corrected chlorophyll state. The chlorophyll state input is fed into the prediction output process of the DAG to generate chlorophyll prediction results; Based on the corrected chlorophyll status and chlorophyll prediction results, sensor data results are generated. The output includes sensor data results that correct for chlorophyll status and chlorophyll prediction.

Images