A method and system for water treatment of boiler feed water pH

Abstract

The present application relates to the technical field of boiler water treatment, and specifically discloses a boiler feed water pH value water treatment method and system, which synchronously collects dynamic fluctuation data and static ion component data of water quality; performs time series decomposition and normalization on the dynamic data to form a dynamic characteristic sequence, and performs normalization on the static data to form a static characteristic vector; mines the implicit coupling relationship between the two through an iterative optimization algorithm, constructs a dynamic correlation matrix quantifying the influence of each component on water quality stability; based on the matrix and a target pH interval, simulates multiple regulation paths in a state transition network and reversely generates an operation variable correction instruction set; finally, an asynchronous analysis and signal mapping drive an actuator, and a closed-loop optimization is realized based on feedback, which realizes accurate identification and advanced regulation of buffer components, ensures that the pH value is stable and up to standard, and supports complete resource utilization of complex water sources.

Description

Technical Field

[0001] This invention relates to the field of boiler water treatment technology, and specifically to a method and system for treating the pH value of boiler feedwater. Background Technology

[0002] The coal chemical enterprise employs multi-component slurry coal gasification technology, producing methanol products through shift conversion, low-temperature methanol washing, methanol synthesis, and methanol distillation. It is equipped with a 3×160t / h circulating fluidized bed boiler system to supply driving steam to the air separation turbine. Simultaneously, a demineralized water system and a water supply and drainage network are constructed to supply qualified demineralized water to the boiler system and the waste heat boiler of the main chemical unit.

[0003] The existing technology has the following shortcomings:

[0004] When existing coal chemical enterprises use unconventional water sources such as mine drainage water with complex compositions as boiler feedwater, the specific components with buffering properties, such as boron, contained in the water are difficult to completely remove by conventional desalination processes. This results in the pH value of subsequent boiler feedwater and boiler water remaining consistently low and unable to meet standards. Consequently, they are forced to mix surface water to adjust the water quality. This not only restricts the utilization of mine water resources and the achievement of the goal of "zero discharge" of wastewater, but also brings the risk of corrosion and scaling to the long-term safe and stable operation of the boiler system. Summary of the Invention

[0005] The purpose of this invention is to provide a water treatment method and system for adjusting the pH value of boiler feedwater, so as to solve the problems mentioned above.

[0006] The objective of this invention can be achieved through the following technical solutions:

[0007] A method for treating the pH value of boiler feedwater includes the following steps:

[0008] S1: Collect two types of water quality data in real time from multiple set locations of the target water source and subsequent treatment process. The first type is dynamic component data reflecting real-time fluctuations in water quality, and the second type is static component data reflecting stable water quality characteristics.

[0009] S2: Integrate and standardize the dynamic component data according to the time series to form a dynamic feature sequence; at the same time, perform structuring and normalization processing on the static component data to form a static feature vector.

[0010] S3: Based on dynamic feature sequences and static feature vectors, a comprehensive analysis is performed through iterative optimization algorithms to uncover the implicit coupling relationship between dynamic component data and static component data; based on the implicit coupling relationship, a dynamic correlation matrix is ​​constructed that can characterize the impact of specific buffering components in the target water source on the stability of the outlet water quality of the entire treatment process;

[0011] S4: Based on the dynamic correlation matrix and combined with the preset final water quality pH target range, the real-time correction instruction set for key operational variables in the treatment process is derived in reverse.

[0012] S5: Input the real-time correction instruction set into the control system of the processing flow, dynamically adjust the operating parameters of the corresponding steps, so that the output water quality of the processing flow meets the preset pH target range requirements, and complete the treatment of boiler feedwater.

[0013] As a further aspect of the present invention: the formation of the dynamic feature sequence specifically includes:

[0014] The original time series composed of dynamic component data is decomposed to obtain multiple intrinsic components that characterize different fluctuation period characteristics.

[0015] Based on the statistical distribution of each intrinsic component within a preset historical window, each intrinsic component is adaptively scaled to unify the numerical range of each intrinsic component to a standard range.

[0016] All intrinsic components, after adaptive scaling, are recombined according to their corresponding fluctuation cycles to generate a dynamic feature sequence.

[0017] As a further aspect of the present invention: the decomposition of the original time series composed of dynamic component data specifically includes:

[0018] The original time series is subjected to a nonlinear transformation to generate an intermediate sequence with local fluctuation characteristics;

[0019] Based on the sign consistency of the rate of change of adjacent data points in the intermediate sequence, baseline components with stable monotonic trends are identified and extracted.

[0020] The baseline component is subtracted from the original time series to obtain the residual sequence. The residual sequence is then subjected to iterative filtering based on the peak spectral density to separate two independent components that represent high-frequency random fluctuations and mid-frequency periodic fluctuations, respectively.

[0021] The baseline component, high-frequency random fluctuation component, and mid-frequency periodic fluctuation component are collectively output as multiple intrinsic components.

[0022] As a further aspect of the present invention: the process of mining the implicit coupling relationship is as follows:

[0023] The static feature vectors are expanded into static matrices corresponding to the time points of the dynamic feature sequence, thus constructing a spatiotemporally aligned joint data field;

[0024] In the joint data field, a weight network reflecting the dynamic-static correlation is initialized based on the matching degree between the gradient direction of the dynamic feature sequence and the feature direction of the static matrix.

[0025] With the goal of minimizing the difference between the predicted and measured values ​​of the dynamic feature sequence, the weight network is adjusted through multiple rounds of feedback until the output stabilizes.

[0026] The structure and parameters of the stabilized weighted network are adjusted and encoded to generate a dynamic correlation matrix representing the implicit coupling relationship.

[0027] As a further aspect of the present invention: the encoding to generate a dynamic correlation matrix representing implicit coupling relationships specifically includes:

[0028] Based on the adjusted and stabilized weight network, the correlation strength coefficient between each feature component in the static feature vector and each time point of the dynamic feature sequence is calculated.

[0029] Statistical significance tests were performed on all the calculated correlation strength coefficients to screen out the set of key feature components that have stable statistical associations with changes in dynamic feature sequences;

[0030] Based on the relative contribution of each component in the key feature component set to the weight network, a stability influence factor reflecting the buffering capacity against fluctuations in effluent water quality is assigned to each component.

[0031] The key feature components and their corresponding stability influencing factors are integrated and arranged according to preset rules to construct a dynamic correlation matrix.

[0032] As a further aspect of the present invention: the reverse derivation of the real-time correction instruction set for key operational variables in the processing flow specifically includes:

[0033] By combining the dynamic correlation matrix with the real-time status parameters of the current processing flow, a spatiotemporal state transition network reflecting the evolution of water quality from the current state to the target state is constructed.

[0034] In the spatiotemporal state transition network, with the final target pH range of water quality as the endpoint, a recursive calculation method is used to forward simulate multiple state transition paths from the starting point to the endpoint.

[0035] Based on each state transition path, the adjustment operations required for key operational variables to make the water quality transfer along the path are derived in reverse, forming a single path correction scheme.

[0036] All single-path correction schemes are evaluated for convergence and their benefits are weighed. The scheme with the best overall performance is selected, and all adjustment operations in the scheme are integrated into a real-time correction instruction set.

[0037] As a further aspect of the present invention: the forward simulation of multiple state transition paths from the starting point to the ending point specifically includes:

[0038] In the spatiotemporal state transition network, a set of state transition constraints is defined to reflect the allowable range of change of key water quality indicators within adjacent time units.

[0039] Starting from the initial state representing the current water quality, based on state transition constraints and dynamic correlation matrix, the various discrete states that the water quality may reach in each subsequent set time unit are recursively calculated, and these discrete states are marked as path nodes.

[0040] All continuous state sequences that start from the initial state and eventually reach the target interval through different path nodes are identified as multiple state transition paths.

[0041] As a further aspect of the present invention: the dynamic adjustment of the operating parameters of the corresponding steps specifically includes:

[0042] The real-time correction instruction set is parsed asynchronously and in parallel, and the real-time correction instruction set is decomposed into multiple independent sub-instructions corresponding to different processing steps;

[0043] Based on the timestamp and operation parameter identifier carried by each independent sub-instruction, the independent sub-instruction is matched with a predefined operation parameter mapping table to generate control signals that can directly drive the physical execution unit;

[0044] The control signals are sent to the corresponding physical execution units in a time-division manner according to the timestamps of the control signals, so that the physical execution units can adjust the operation parameters according to the content of the control signals.

[0045] After the adjustment is completed, real-time water quality analysis data of key nodes in the processing flow are collected and compared with the expected adjustment target. The verification results are used as feedback information for the generation and optimization of subsequent correction instruction sets to form closed-loop control.

[0046] As a further aspect of the present invention: the step of decomposing the real-time correction instruction set into multiple independent sub-instructions corresponding to different processing steps specifically includes:

[0047] The real-time correction instruction set is divided into multiple consecutive instruction segments according to a preset time window;

[0048] Based on the step identifier carried in the instruction, each instruction fragment is simultaneously assigned to different parsing threads using a hash mapping method, with each thread responsible for processing the instruction of one processing step.

[0049] Within each parsing thread, operation parameters, execution timing, and control logic are extracted from instruction fragments according to predefined instruction syntax rules and encapsulated into independent sub-instructions;

[0050] Each encapsulated independent sub-instruction is injected into the dynamic scheduling queue, awaiting subsequent execution.

[0051] A water treatment system for boiler feedwater pH value, comprising:

[0052] The multi-source water quality data synchronous acquisition module is used to continuously acquire two types of water quality data in real time from multiple set locations of the target water source and subsequent treatment process. The first type is dynamic component data reflecting real-time fluctuations in water quality, and the second type is static component data reflecting stable water quality characteristics.

[0053] The multimodal water quality feature construction module integrates and standardizes dynamic component data according to time series to form dynamic feature sequences; at the same time, it performs structuring and normalization processing on static component data to form static feature vectors.

[0054] The coupling relationship mining and correlation matrix construction module, based on dynamic feature sequences and static feature vectors, performs comprehensive analysis through iterative optimization algorithms to mine the implicit coupling relationships between dynamic component data and static component data; based on the implicit coupling relationships, it constructs a dynamic correlation matrix that can characterize the impact of specific buffering components in the target water source on the stability of the outlet water quality of the entire treatment process;

[0055] The reverse derivation and real-time instruction generation module, based on the dynamic correlation matrix and combined with the preset final water quality pH target range, reverse derivation of the real-time correction instruction set for key operational variables in the treatment process.

[0056] The instruction execution and closed-loop control module will modify the control system of the instruction set input processing flow in real time, dynamically adjust the operating parameters of the corresponding steps, so that the output water quality of the processing flow meets the preset pH target range requirements, and complete the treatment of boiler feedwater.

[0057] The beneficial effects of this invention are:

[0058] (1) This invention, through in-depth correlation analysis and influence quantification of static components (especially buffering ions such as borate) in the target water source, can accurately identify and quantify the buffering effect of specific components on the final effluent pH value. On this basis, by constructing a spatiotemporal state transition network and simulating multi-path adjustment schemes, it achieves advanced, coordinated, and refined control of key operational variables such as dosing and process parameters. This enables the method to effectively overcome the problem of persistently low pH value caused by the use of complex water sources such as mine drainage water. While ensuring that the pH value of boiler feedwater is stable and meets the standards, it achieves complete resource utilization of unconventional water resources such as mine water, reduces dependence on high-quality surface water, and meets the strict environmental protection requirements of "zero discharge" of wastewater.

[0059] (2) This invention constructs a closed-loop control system covering the entire process from data perception and intelligent analysis to optimized execution. This method not only relies on real-time feedback but also achieves feedforward predictive control through data-driven modeling and multi-path forward simulation, thereby enhancing the system's robustness and adaptability in the face of water quality fluctuations and process disturbances. Continuous feedback optimization based on actual execution results enables the control strategy to evolve dynamically and continuously improve adjustment accuracy. This closed-loop intelligent control mechanism effectively avoids problems such as adjustment lag, over-reliance on human experience, and frequent oscillations in traditional methods, ensuring the long-term, highly stable automatic operation of the boiler water treatment system and reducing operation and maintenance costs and safety risks. Attached Figure Description

[0060] The invention will now be further described with reference to the accompanying drawings.

[0061] Figure 1 This is a flowchart of the method of the present invention;

[0062] Figure 2 This is a system block diagram of the present invention. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Please see Figure 1 As shown, the present invention is a water treatment method for boiler feedwater pH value, comprising the following steps:

[0065] S1: Collect two types of water quality data in real time from multiple set locations of the target water source and subsequent treatment process. The first type is dynamic component data reflecting real-time fluctuations in water quality, and the second type is static component data reflecting stable water quality characteristics.

[0066] S2: Integrate and standardize the dynamic component data according to the time series to form a dynamic feature sequence; at the same time, perform structuring and normalization processing on the static component data to form a static feature vector.

[0067] S3: Based on dynamic feature sequences and static feature vectors, a comprehensive analysis is performed through iterative optimization algorithms to uncover the implicit coupling relationship between dynamic component data and static component data; based on the implicit coupling relationship, a dynamic correlation matrix is ​​constructed that can characterize the impact of specific buffering components in the target water source on the stability of the outlet water quality of the entire treatment process;

[0068] S4: Based on the dynamic correlation matrix and combined with the preset final water quality pH target range, the real-time correction instruction set for key operational variables in the treatment process is derived in reverse.

[0069] S5: Input the real-time correction instruction set into the control system of the processing flow, dynamically adjust the operating parameters of the corresponding steps, so that the output water quality of the processing flow meets the preset pH target range requirements, and complete the treatment of boiler feedwater.

[0070] In S1, for the first type of dynamic component data reflecting real-time fluctuations in water quality, the acquisition process is as follows: online pH sensors and online conductivity sensors are installed at key locations such as the target water source inlet, the membrane treatment unit's permeate outlet, and the final desalinated water outlet. These sensors continuously measure and record the instantaneous pH and conductivity values ​​of the water body at a frequency of no less than once per minute, thereby forming a dynamic data stream that can reflect water quality fluctuations on a minute or even second level.

[0071] For the second type of static component data reflecting the stable characteristics of water quality, the collection process is as follows: At the same points mentioned above, an automatic sampling device is set up to collect water samples at a fixed frequency of once per hour. The collected water samples are immediately transported to an online ion chromatograph for analysis, accurately determining the concentrations of borate ions, silicate ions, and other characteristic anions. These ion concentration data represent the relatively stable compositional characteristics of the water quality over a timescale of several hours.

[0072] In S2, the processing objective for dynamic component data is to transform the original, continuous time series data into a sequence that clearly characterizes the fluctuations at different time scales. The specific processing steps are as follows: First, a nonlinear transformation is performed on the original time series composed of dynamic component data. This transformation involves calculating the absolute value of the difference between each data point in the original sequence and its preceding data point, resulting in a first-order difference absolute value sequence. Then, a sliding window with a window length of 30 data points is applied to this first-order difference absolute value sequence, and the standard deviation of the data within each window is calculated, generating a new intermediate sequence that emphasizes the intensity of local fluctuations. Second, a baseline component is identified based on this intermediate sequence. Specifically, the intermediate sequence is iterated, and when the signs (positive or negative) of the rate of change corresponding to more than 50 consecutive data points remain consistent, the data segment is determined to have a stable monotonic trend. Data segments corresponding to these intervals in the original time series are extracted, and a linear fitting method is used to connect the start and end points of each segment, forming a smooth baseline component representing a long-term, slow trend. Finally, this baseline component is subtracted from the original time series to obtain a residual sequence. For this residual sequence, iterative filtering based on spectral density peaks is performed to separate other components: the spectrum of the residual sequence is calculated, and the frequency corresponding to the highest energy peak in the spectrum is identified as the principal periodic frequency; a bandpass filter is constructed centered on this frequency to filter the residual sequence, and the extracted signal is the component representing the mid-frequency periodic fluctuation; this mid-frequency periodic fluctuation component is then subtracted from the residual sequence, and the remaining signal is considered as the component representing high-frequency random noise. Thus, the original time series is decomposed into three intrinsic components: the baseline component, the mid-frequency periodic fluctuation component, and the high-frequency random fluctuation component.

[0073] Next, each intrinsic component obtained above undergoes adaptive scaling. For any component to be processed, the system reads all its historical data within a preset historical window (usually 24 hours). The 5th and 95th percentiles of this historical dataset are calculated. During scaling, all current data values ​​of the component are subtracted from the 5th percentile, and then divided by the difference between the 95th and 5th percentiles. This operation ensures that the processed values ​​of the component mostly fall within the standard range of 0 to 1. All three intrinsic components are scaled independently using this method. Finally, the adaptively scaled baseline component, mid-frequency periodic fluctuation component, and high-frequency random fluctuation component are concatenated sequentially from low to high frequency to generate a complete, standardized dynamic feature sequence.

[0074] For static component data, the goal is to transform discrete ion concentration analysis results into a unified numerical vector. The specific process involves: arranging a set of ion concentration data obtained from each online ion chromatography analysis according to a fixed order of ion types (e.g., borate, silicate, chloride, sulfate). For each ion concentration data point, it is divided by the maximum concentration value statistically obtained from long-term historical data, thus normalizing the data so that the numerical value for each ion is between 0 and 1. Finally, these normalized concentration values ​​are combined in a predetermined order to form a static feature vector.

[0075] In S3, the process of uncovering implicit coupling relationships is implemented through the following steps. First, the spatiotemporal alignment of the data field is constructed. Static feature vectors represent the overall composition of water quality at a certain moment. To enable them to correspond with the dynamic feature sequence that changes over time, temporal alignment processing is required. Specifically, each static feature vector is associated with its acquisition time. For each time point in the dynamic feature sequence, the static feature vector corresponding to its timestamp or the closest static feature vector is found. If there is no corresponding static vector for a certain dynamic time point, the interpolation or nearest neighbor value of its adjacent static vectors is used as the static feature of that time point. All the static features corresponding to all time points are arranged in chronological order to form a static matrix that is time-aligned with the dynamic feature sequence. This static matrix and the dynamic feature sequence are aligned in the time dimension, together forming a spatiotemporal joint data field.

[0076] Secondly, a weighted network reflecting the dynamic and static correlation is initialized. This network can be viewed as a multi-layered connection structure. The initialization is based on the degree of matching between the change pattern of the dynamic feature sequence and the features of the static matrix. Specifically, the direction of change (i.e., gradient direction) of the dynamic feature sequence at each time point prior to the previous time point is calculated, while simultaneously analyzing the change pattern of each feature component in the static matrix at different time points. The initial matching degree is evaluated by calculating the correlation coefficient between the change pattern sequence of each static feature component and the gradient direction sequence of the dynamic feature sequence. After normalizing the calculated correlation coefficients, they are used as the initial weight values ​​connecting the static feature component to the prediction node of the dynamic feature sequence, thus completing the initialization of the weighted network.

[0077] The correlation coefficient is calculated as follows: standardize the two sequences so that their mean is zero and their standard deviation is one; calculate the sum of the products of the corresponding data points of the two standardized sequences; divide the sum by the sequence length minus one to obtain the correlation coefficient.

[0078] Next, the weight network is iteratively adjusted multiple times to optimize its performance. The goal of the adjustment is to minimize the difference between the dynamic feature sequence predicted by the weight network based on the input historical static matrix data and the actual dynamic feature sequence. This difference is measured by calculating the sum of squared differences between the predicted and actual sequences at each time point. In each adjustment round, the predicted values ​​are first calculated based on the current weights and static matrix data to obtain the sum of differences. Then, using the principle of error backpropagation, the gradient (i.e., sensitivity to change) of this sum of differences with respect to each weight value in the network is calculated. The specific adjustment amount for each weight value is the gradient value multiplied by a preset learning rate (e.g., 1 / 1000), and the adjustment direction is to reduce the sum of differences. This process of prediction, difference calculation, gradient calculation, and weight adjustment is repeated. When the change in the sum of differences is less than 1 / 100 in ten consecutive iterations, the output of the weight network is considered to have stabilized, and the adjustment process ends.

[0079] Finally, the stabilized weighted network is encoded into a dynamic association matrix. This process consists of four sub-steps. First, calculate the association strength coefficient. Based on the stable weighted network, for the i-th component in the static feature vector, its association strength coefficient with the dynamic feature sequence at time t is equal to the sum of the weight products of all paths in the network that start from this static component, pass through network connections, and ultimately affect the prediction of the dynamic value at time t. This is the association strength coefficient. Second, perform a statistical significance test. Using the bootstrap method, the time order of the dynamic feature sequence is randomly shuffled 1000 times. After each shuffle, the association strength coefficient is recalculated to obtain a random distribution. The actual calculated coefficient is compared with this random distribution. If the actual coefficient value is greater than 95% of the values ​​in the random distribution, the association is considered statistically significant, and the corresponding static feature component is included in the key feature component set. Third, assign stability influence factors. For each component in the key set, its average association strength coefficient at all time points is calculated, and this average value is divided by the sum of the average association strength coefficients of all components in the set. The resulting ratio is the relative contribution of that component. This contribution is directly defined as the stability influence factor of that component. Fourth, construct the matrix. Arrange the selected key feature components in descending order of their stability impact factors, and store the identifier of each component (e.g., "borate") and its corresponding numerical stability impact factor in pairs, ultimately forming a dynamic correlation matrix. Each row of this matrix corresponds to a different key static component, and its value directly quantifies the magnitude of that component's buffering capacity against fluctuations in the final effluent water quality.

[0080] In S4, firstly, a spatiotemporal state transition network reflecting the evolution of water quality status is constructed. The latest state of the dynamic feature sequence collected and processed at the current moment, the static feature vector, and key process parameters obtained from real-time monitoring data (such as reverse osmosis feed water pressure and mixed bed cumulative operating time) are collectively defined as the "current composite state" of water quality and process. This state is the starting point of the network. The influence factors of each static component on water quality stability quantified in the dynamic correlation matrix are transformed into "transition weights" of the influence of different component concentration changes on the overall state during state transitions in this network. The endpoint of the network is set as the set of all possible water quality states that conform to a preset pH target range (e.g., 9.0 to 10.5). Each node in the network represents a discretized "composite state" at a specific moment, and the directed connections between nodes represent the state transition possibilities that can be achieved within a unit control cycle (e.g., 5 minutes) by adjusting specific operational variables (such as dosing pump frequency and valve opening).

[0081] Secondly, in this spatiotemporal state transition network, multiple feasible paths from the current state to the target state are simulated in a forward manner. This simulation process specifically includes the following steps: First, define state transition constraints. For example, it is stipulated that within a single control cycle, the maximum allowable change in the simulated pH value shall not exceed 0.3 units, and the simulated change in the concentration of key ions shall not exceed 10% of its current value. These constraints constitute the boundary conditions for state transition. Second, starting from the starting state node characterizing the current water quality, recursively extrapolate. Based on the stability influence factors of different components in the dynamic correlation matrix and the preset state transition constraints, calculate multiple new states that may be reached in the next control cycle after applying all allowable combinations of operational variables in the current state. Each new state must satisfy all constraints and will be created as a new path node. Third, this process is recursively performed, that is, each newly generated node is used as a new starting point to continue extrapolating the possible states in its subsequent cycles. This process continues until the extrapolated states enter the preset endpoint (target pH range) state set, or reach the preset maximum number of extrapolation steps (e.g., simulating 12 control cycles for the next hour). Fourth, record all complete state sequences that start from the starting node, pass through a series of intermediate nodes, and finally reach any endpoint node. Each such sequence is then identified as a possible state transition path.

[0082] Based on each simulated state transition path, a reverse deduction is performed to generate specific operational instruction schemes. The deduction process starts from the end of the path and traverses each state node along the path in reverse. For two adjacent state nodes on the path, i.e., the preceding state (earlier time) and the subsequent state (later time), their state differences are analyzed. These differences are specifically reflected in the simulated pH value, ion concentration, and process parameter values. By querying a predefined "operating variable-state influence" mapping table, one or more key operating variable adjustments that can most effectively drive the state transition from the preceding state to the subsequent state can be found (e.g., increasing the reducing agent dosage by 2 ml per minute, or raising the reverse osmosis feed water pH setpoint by 0.1). Arranging the adjustments derived from the reverse deduction for all adjacent state nodes on the path in chronological order constitutes a complete "single path correction scheme" bound to the time axis.

[0083] Finally, all generated single-path correction schemes are comprehensively evaluated and optimized to form the final real-time correction instruction set. The evaluation is conducted from two dimensions: first, "convergence," which assesses the smoothness of the scheme's execution; its quantitative indicator is the sum of the adjustment magnitudes of all operational variables in the scheme, with a smaller sum indicating a smoother adjustment; second, "efficiency," which assesses the effectiveness and speed at which the scheme achieves its goal; its quantitative indicator is the degree of closeness between the predicted state and the ideal target state center after the scheme's execution. To this end, a comprehensive efficiency index is defined. To weigh these two aspects: ;

[0084] in, This represents the overall performance evaluation value of the proposed solution; a higher value indicates better overall performance. This represents the sum of the adjustments made to all operational variables in the scheme. It is a reference base value for the sum of adjustment magnitudes, used for normalization. The Euclidean distance between the predicted endpoint state and the ideal target state of the representative scheme. It is a reference distance baseline value used for normalization. It is a tradeoff coefficient between 0 and 1, used to adjust the emphasis on "operational smoothness" and "target proximity", and can be set to 0.5.

[0085] Calculate the overall performance index of each scheme. Then, select The solution with the highest value is selected as the optimal solution in terms of overall performance. All key operational variable adjustments included in this optimal solution, arranged according to future time series, are extracted, integrated, and formatted to generate the "real-time correction instruction set" to be issued in this control cycle. This instruction set will directly guide the precise actions of subsequent processing units.

[0086] In S5, the real-time correction instruction set is first asynchronously and in parallel parsed, breaking it down into independent control sub-instructions corresponding to different physical processing units. The specific process is as follows: The real-time correction instruction set, containing a sequence of operations over a future period (e.g., the next 30 minutes), is divided into multiple consecutive instruction segments at fixed time intervals (e.g., one segment every 5 minutes). Each instruction segment contains all the operations to be executed within that time period. Based on the processing step identifier carried in each operation command (e.g., a unique numerical code identifying a "reverse osmosis dosing pump"), the commands in each instruction segment are simultaneously distributed to different independent parsing threads for processing using a hash mapping method. Specifically, the hash mapping method involves inputting the step identifier into a preset hash function, which outputs an integer. This integer is used to assign the command to the corresponding fixed-numbered parsing thread. Each parsing thread is specifically responsible for processing all commands for a particular processing step. Within each parsing thread, key information is extracted from the received command data according to predefined instruction syntax rules. This includes the target value of the operational parameters to be adjusted (e.g., "set the dosing frequency to 50 Hz"), the absolute or relative timestamp of the command execution (e.g., "execute at 300 seconds after receiving the instruction"), and necessary control logic (e.g., "gradually change to the target value at a rate of 0.5 Hz per second"). This information is then encapsulated into a single, uniformly formatted, and complete sub-instruction. All encapsulated sub-instructions are injected into the dynamic scheduling queue corresponding to their respective target processing steps, awaiting extraction and execution by the next step.

[0087] Secondly, control signals that can directly drive physical devices are generated based on independent sub-instructions. This process is as follows: when an independent sub-instruction in a dynamic scheduling queue arrives at its execution time window, the control unit retrieves the instruction from the queue. Based on the operation parameter identifier carried in the instruction, a lookup and matching operation is performed in a predefined operation parameter mapping table. This mapping table establishes the correspondence between operation parameter identifiers and specific physical execution units (such as a specific control register address of a frequency converter), and defines the physical signal type (such as 4-20 mA analog current, 0-10 V voltage, or switching quantity). Generating the control signal involves converting the target value of the operation parameter in the instruction into a specific signal value or state that the physical execution unit can receive, according to the mapping table. For example, if the target value is a pump frequency of 50 Hz, and the mapping table specifies a corresponding analog current signal of 12 mA, then a control signal with a value of 12 mA is generated.

[0088] Next, the generated control signals are sent to the physical actuators in a time-division manner to adjust the operating parameters. The control unit, based on the timestamp attached to each control signal, strictly adheres to the timing requirements and sends it to the corresponding physical actuator, such as a frequency converter, regulating valve, or switchgear, via fieldbus or analog signal lines. Upon receiving the control signal, the physical actuator changes its output according to the signal's content, thereby adjusting the actual operating parameters, such as changing the pump speed, valve opening, or heater power.

[0089] Finally, the adjusted effects are verified and feedback is optimized to form a closed-loop control. At a predetermined time point after all adjustment operations based on the instruction set have been completed (e.g., 15 minutes after the adjustment begins), real-time water quality analysis data, mainly pH value and related ion concentration, are re-collected at key nodes of the treatment process (especially the final effluent). The actual measured pH value is compared with the predicted pH target value expected when the instruction set was generated, and the deviation is calculated. This deviation value, along with the comparison between the actual water quality data and the predicted data, is sent back to the instruction generation step as a feedback information package. When generating new real-time correction instruction sets in the future, this feedback information will be used to calibrate the prediction model and correct the parameters in the state transition network, thereby optimizing the accuracy and control effect of subsequent instructions, thus forming a continuous, self-improving closed-loop control process.

[0090] Please see Figure 2 As shown, a water treatment system for boiler feedwater pH value includes:

[0091] The multi-source water quality data synchronous acquisition module is used to continuously acquire two types of water quality data in real time from multiple set locations of the target water source and subsequent treatment process. The first type is dynamic component data reflecting real-time fluctuations in water quality, and the second type is static component data reflecting stable water quality characteristics.

[0092] The multimodal water quality feature construction module integrates and standardizes dynamic component data according to time series to form dynamic feature sequences; at the same time, it performs structuring and normalization processing on static component data to form static feature vectors.

[0093] The coupling relationship mining and correlation matrix construction module, based on dynamic feature sequences and static feature vectors, performs comprehensive analysis through iterative optimization algorithms to mine the implicit coupling relationships between dynamic component data and static component data; based on the implicit coupling relationships, it constructs a dynamic correlation matrix that can characterize the impact of specific buffering components in the target water source on the stability of the outlet water quality of the entire treatment process;

[0094] The reverse derivation and real-time instruction generation module, based on the dynamic correlation matrix and combined with the preset final water quality pH target range, reverse derivation of the real-time correction instruction set for key operational variables in the treatment process.

[0095] The instruction execution and closed-loop control module will modify the control system of the instruction set input processing flow in real time, dynamically adjust the operating parameters of the corresponding steps, so that the output water quality of the processing flow meets the preset pH target range requirements, and complete the treatment of boiler feedwater.

[0096] The working principle of this invention is as follows: First, two types of water quality data are simultaneously collected from the water source inlet and key nodes of the water treatment process: dynamic fluctuation data continuously acquired by online pH and conductivity sensors, and static component data periodically acquired through automatic sampling and ion chromatography analysis. Second, the dynamic data is decomposed into a time series, separating the intrinsic components characterizing long-term trends, periodic fluctuations, and random noise, which are then normalized and combined into a dynamic feature sequence; simultaneously, the static ion concentration data is normalized to form a static feature vector. Next, by constructing a spatiotemporally aligned data field and initializing a weight network, an iterative optimization algorithm is used to mine the implicit coupling relationship between dynamic and static data, thereby screening key static components and quantifying their impact, generating a dynamic correlation matrix. Then, a spatiotemporal state transition network is constructed in conjunction with the target pH range, forward simulating multiple possible paths from the current state to the target state, backward deducing the corresponding operational variable adjustment schemes, and selecting the optimal scheme through comprehensive evaluation to generate a real-time correction instruction set. Finally, the instruction set is asynchronously parsed and mapped, converted into specific control signals and sent to each physical execution unit in a time-sharing manner to drive the actions of equipment such as dosing and valves. Based on the deviation between the measured data and the predicted value of the adjusted water quality, feedback optimization is performed to form an adaptive closed-loop control, thereby achieving precise and stable control of the pH value of boiler feedwater and ensuring water quality meets standards while making full use of complex water sources such as mine drainage water.

[0097] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for treating the pH value of boiler feedwater, characterized in that, Includes the following steps: S1: Collect two types of water quality data in real time from multiple set locations of the target water source and subsequent treatment process. The first type is dynamic component data reflecting real-time fluctuations in water quality, and the second type is static component data reflecting stable water quality characteristics. S2: Integrate and standardize the dynamic component data according to the time series to form a dynamic feature sequence; at the same time, perform structuring and normalization processing on the static component data to form a static feature vector. S3: Based on dynamic feature sequences and static feature vectors, a comprehensive analysis is performed through iterative optimization algorithms to uncover the implicit coupling relationship between dynamic component data and static component data; based on the implicit coupling relationship, a dynamic correlation matrix is ​​constructed that can characterize the impact of specific buffering components in the target water source on the stability of the outlet water quality of the entire treatment process; The process of uncovering the hidden coupling relationship is as follows: The static feature vectors are expanded into static matrices corresponding to the time points of the dynamic feature sequence, thus constructing a spatiotemporally aligned joint data field; In the joint data field, a weight network reflecting the dynamic-static correlation is initialized based on the matching degree between the gradient direction of the dynamic feature sequence and the feature direction of the static matrix. With the goal of minimizing the difference between the predicted and measured values ​​of the dynamic feature sequence, the weight network is adjusted through multiple rounds of feedback until the output stabilizes. The structure and parameters of the stabilized weighted network are adjusted and encoded to generate a dynamic correlation matrix representing the implicit coupling relationship; The encoding process generates a dynamic correlation matrix representing the implicit coupling relationship, specifically including: Based on the adjusted and stabilized weight network, the correlation strength coefficient between each feature component in the static feature vector and each time point of the dynamic feature sequence is calculated. Statistical significance tests were performed on all the calculated correlation strength coefficients to screen out the set of key feature components that have stable statistical associations with changes in dynamic feature sequences; Based on the relative contribution of each component in the key feature component set in the weight network, a stability influence factor reflecting the buffering capacity against fluctuations in effluent water quality is assigned to each component. The key feature components and their corresponding stability influencing factors are integrated and arranged according to preset rules to construct a dynamic correlation matrix; S4: Based on the dynamic correlation matrix and the preset final water quality pH target range, a set of real-time correction instructions for key operational variables in the treatment process is derived in reverse, specifically including: By combining the dynamic correlation matrix with the real-time status parameters of the current processing flow, a spatiotemporal state transition network reflecting the evolution of water quality from the current state to the target state is constructed. The construction process of the spatiotemporal state transition network is as follows: the latest state of the dynamic feature sequence collected and processed at the current moment, the static feature vector, and the key process parameters obtained from real-time monitoring data are collectively defined as the current composite state of water quality and process. The current composite state is the starting point of the spatiotemporal state transition network. The influence factors of each static component on water quality stability quantified in the dynamic correlation matrix are transformed into the transfer weights of the influence of different component concentration changes on the overall state when the state transition is carried out in the spatiotemporal state transition network. The endpoint of the spatiotemporal state transition network is set as the set of all possible water quality states that meet the preset pH target range. Each node in the network represents a discrete composite state at a specific moment. The directed connection between nodes represents the possibility of state transition that can be achieved by adjusting specific operating variables within a unit control cycle. The specific operating variables include the dosing pump frequency and the valve opening degree. In the spatiotemporal state transition network, with the final target pH range of water quality as the endpoint, a recursive calculation method is used to forward simulate multiple state transition paths from the starting point to the endpoint. Based on each state transition path, the adjustment operations required for key operational variables to make the water quality transfer along the path are derived in reverse, forming a single path correction scheme. All single-path correction schemes are evaluated for convergence and their benefits are weighed. The scheme with the best overall performance is selected, and all adjustment operations in the scheme are integrated into a real-time correction instruction set. The forward simulation of multiple state transition paths from the starting point to the ending point specifically includes: In the spatiotemporal state transition network, a set of state transition constraints is defined to reflect the allowable range of change of key water quality indicators within adjacent time units. Starting from the initial state representing the current water quality, based on state transition constraints and dynamic correlation matrix, the various discrete states that the water quality may reach in each subsequent set time unit are recursively calculated, and these discrete states are marked as path nodes. All continuous state sequences that start from the initial state, pass through different path nodes, and finally arrive at the target interval are identified as multiple state transition paths; S5: Input the real-time correction instruction set into the control system of the processing flow, dynamically adjust the operating parameters of the corresponding steps, so that the output water quality of the processing flow meets the preset pH target range requirements, and complete the treatment of boiler feedwater.

2. The water treatment method for adjusting the pH value of boiler feedwater according to claim 1, characterized in that, The formation of the dynamic feature sequence specifically includes: The original time series composed of dynamic component data is decomposed to obtain multiple intrinsic components that characterize different fluctuation period characteristics. Based on the statistical distribution of each intrinsic component within a preset historical window, each intrinsic component is adaptively scaled to unify the numerical range of each intrinsic component to a standard range. All intrinsic components, after adaptive scaling, are recombined according to their corresponding fluctuation cycles to generate a dynamic feature sequence.

3. The water treatment method for adjusting the pH value of boiler feedwater according to claim 2, characterized in that, The decomposition of the original time series composed of dynamic component data specifically includes: The original time series is subjected to a nonlinear transformation to generate an intermediate sequence with local fluctuation characteristics; Based on the sign consistency of the rate of change of adjacent data points in the intermediate sequence, baseline components with stable monotonic trends are identified and extracted. The baseline component is subtracted from the original time series to obtain the residual sequence. The residual sequence is then subjected to iterative filtering based on the peak spectral density to separate two independent components that represent high-frequency random fluctuations and mid-frequency periodic fluctuations, respectively. The baseline component, high-frequency random fluctuation component, and mid-frequency periodic fluctuation component are collectively output as multiple intrinsic components.

4. The water treatment method for adjusting the pH value of boiler feedwater according to claim 1, characterized in that, The dynamic adjustment of the operation parameters for the corresponding steps specifically includes: The real-time correction instruction set is parsed asynchronously and in parallel, and the real-time correction instruction set is decomposed into multiple independent sub-instructions corresponding to different processing steps; Based on the timestamp and operation parameter identifier carried by each independent sub-instruction, the independent sub-instruction is matched with a predefined operation parameter mapping table to generate control signals that can directly drive the physical execution unit; The control signals are sent to the corresponding physical execution units in a time-division manner according to the timestamps of the control signals, so that the physical execution units can adjust the operation parameters according to the content of the control signals. After the adjustment is completed, real-time water quality analysis data of key nodes in the processing flow are collected and compared with the expected adjustment target. The verification results are used as feedback information for the generation and optimization of subsequent correction instruction sets to form closed-loop control.

5. The water treatment method for boiler feedwater pH value according to claim 4, characterized in that, The process of breaking down the real-time correction instruction set into multiple independent sub-instructions corresponding to different processing steps specifically includes: The real-time correction instruction set is divided into multiple consecutive instruction segments according to a preset time window; Based on the step identifier carried in the instruction, each instruction fragment is simultaneously assigned to different parsing threads using a hash mapping method, with each thread responsible for processing the instruction of one processing step. Within each parsing thread, operation parameters, execution timing, and control logic are extracted from instruction fragments according to predefined instruction syntax rules and encapsulated into independent sub-instructions; Each encapsulated individual sub-instruction is injected into the dynamic scheduling queue, awaiting subsequent execution.

6. A water treatment system for boiler feedwater pH value, characterized in that, A water treatment method for adjusting the pH value of boiler feedwater according to any one of claims 1-5, comprising: The multi-source water quality data synchronous acquisition module is used to continuously acquire two types of water quality data in real time from multiple set locations of the target water source and subsequent treatment process. The first type is dynamic component data reflecting real-time fluctuations in water quality, and the second type is static component data reflecting stable water quality characteristics. The multimodal water quality feature construction module integrates and standardizes dynamic component data according to time series to form dynamic feature sequences; at the same time, it performs structuring and normalization processing on static component data to form static feature vectors. The coupling relationship mining and correlation matrix construction module, based on dynamic feature sequences and static feature vectors, performs comprehensive analysis through iterative optimization algorithms to mine the implicit coupling relationships between dynamic component data and static component data; based on the implicit coupling relationships, it constructs a dynamic correlation matrix that can characterize the impact of specific buffering components in the target water source on the stability of the outlet water quality of the entire treatment process; The reverse derivation and real-time instruction generation module, based on the dynamic correlation matrix and combined with the preset final water quality pH target range, reverse derivation of the real-time correction instruction set for key operational variables in the treatment process. The instruction execution and closed-loop control module will modify the control system of the instruction set input processing flow in real time, dynamically adjust the operating parameters of the corresponding steps, so that the output water quality of the processing flow meets the preset pH target range requirements, and complete the treatment of boiler feedwater.

Images