Mine water evaporation-freezing crystallization process monitoring system and method

Abstract

The invention discloses a mine water evaporation-freezing crystallization process monitoring system and a mine water evaporation-freezing crystallization process monitoring method, which relate to the technical field of process monitoring, and are characterized in that when evaporation section parameters and freezing section parameters float, self-adaptive regulation and control are carried out according to a cooperative control rule, and influence weights of the evaporation section parameters and the freezing section parameters on a crystallization process are dynamically analyzed; performing real-time coupling analysis on evaporation section parameters-freezing section parameters and a crystallization process, predicting a fluctuation risk level of the crystallization process, and generating a corresponding intervention strategy based on a prediction result, the intervention strategy comprising automatic injection of a trace scale inhibitor or starting of pulse backwashing in a high risk, and adjustment of a freezing section cooling gradient in a medium risk, and crystal directional growth is guided, the stirring frequency is optimized at low risk, and crystal particles are dispersed. The monitoring system can master the mine water quality change, the equipment operation condition and the crystallized salt output condition in real time, dynamically adjust the process parameters, guarantee the stable and efficient operation of the process, and improve the resource recovery rate and the automation level at the same time.

Description

Technical Field

[0001] This invention relates to the field of process monitoring technology, specifically to a monitoring system and method for mine water evaporation-freeze crystallization process. Background Technology

[0002] In the production process of coal mines or other underground mines, mine drainage is an unavoidable process. Since mine water usually contains high concentrations of inorganic salts, suspended solids, organic matter and other impurities, direct discharge not only pollutes the environment, but also wastes water resources. In order to achieve the reduction, resource utilization and harmless treatment of mine water, evaporation-freeze crystallization technology is widely used in the field of deep treatment of mine water and salt resource recovery.

[0003] Evaporation-freeze crystallization is a process that combines low-temperature freezing with evaporation and concentration to gradually separate water from mine water and precipitate dissolved salts in crystal form. This process typically includes the following key steps:

[0004] Mine water pretreatment: removes suspended impurities and some organic pollutants to ensure the stable operation of subsequent treatment systems.

[0005] Evaporation and concentration: The water in the mine water is vaporized by heating or low-pressure evaporation devices, so that the salts are gradually enriched.

[0006] Freeze-crystallization: Using a low-temperature environment, salts in the concentrate are gradually crystallized out, achieving solid-liquid separation.

[0007] Crystallized salt recovery: Crystallized salt is separated, dehydrated, and dried for use as industrial raw material or for harmless disposal.

[0008] Because this process involves the coordinated operation of multiple devices and involves a complex process environment and variable media properties, it places high demands on real-time monitoring and intelligent control of operating status, process parameters, equipment load, and effluent quality. Traditional manual inspection and single-point instrument monitoring methods are no longer sufficient to meet the needs of modern mine water treatment for efficient, intelligent, and precise operation and management.

[0009] Based on this, this application proposes a monitoring system and method for mine water evaporation-freeze crystallization process, which can monitor changes in mine water quality, equipment operation status and crystallization salt production in real time, dynamically adjust process parameters, ensure stable and efficient operation of the process, improve resource recovery rate and automation level, reduce operating energy consumption and labor input, and promote the resource utilization of mine water and the construction of green mines. Summary of the Invention

[0010] The purpose of this invention is to provide a monitoring system and method for mine water evaporation-freeze crystallization process to address the shortcomings of the prior art.

[0011] To achieve the above object, the present application provides the following technical solutions: a mine water evaporation-freezing crystallization process monitoring method, the monitoring method comprising the following steps:

[0012] The monitoring system dynamically collects evaporation section parameters and freezing section parameters;

[0013] When the evaporation section parameters and freezing section parameters fluctuate, adaptive control is performed according to the cooperative control rule;

[0014] The influence weight of the evaporation section parameter and freezing section parameter fluctuation on the crystallization process is dynamically analyzed by the LSTM network;

[0015] The evaporation section parameter-freezing section parameter and the crystallization process are analyzed in real time, the fluctuation risk level of the crystallization process is predicted, and the corresponding intervention strategy is generated based on the prediction result.

[0016] In a preferred embodiment, the real-time coupling analysis of the evaporation section parameter-freezing section parameter and the crystallization process is performed to predict the fluctuation risk level of the crystallization process, comprising the following steps:

[0017] The key process parameters of the evaporation section and the freezing section are weighted and fused according to their influence weight on the crystallization process to form a crystallization driving feature vector;

[0018] The short-term trend window is constructed by using the crystallization driving feature vector to quantify the change trend of the crystallization process, assess the sensitivity and potential fluctuation risk, and obtain the crystallization sensitivity index CSI t ;

[0019] The risk interval is divided according to the crystallization sensitivity index to judge the risk level:

[0020] If CSI t > θ3, it is high risk; if θ2 < CSI t ≤ θ3, it is medium risk; if CSI t ≤ θ2, it is low risk.

[0021] In a preferred embodiment, the crystallization driving feature vector is formed, and the expression is:

[0022] In the formula, CDV t is the crystallization driving feature vector at time t, is the attention weight of the i th evaporation section parameter output by the LSTM at time t on the crystallization process, is the measured value of the i th evaporation section parameter at time t, is the attention weight of the j th freezing section parameter output by the LSTM at time t on the crystallization process, is the measured value of the jth freezing section parameter at time t, m is the number of evaporation section parameters, and n is the number of freezing section parameters;

[0023] The crystallization sensitivity index is obtained, and the expression is: In the formula, CSI t is the crystallization sensitivity index at time t, CDV t-k:t is the crystallization driving feature vector in the time window from t-k to t, Var(*) is the variance in the window, is the mean value in the window, and γ is the adjustment factor, and k is the sliding window length.

[0024] In a preferred embodiment, the influence weight of the evaporation section parameter and the freezing section parameter fluctuation on the crystallization process is dynamically analyzed by an LSTM network, including the following steps:

[0025] The time series is modeled by constructing a multi-layer LSTM network, and the LSTM structure includes: a forgetting gate, an input gate, a state update gate, and an output gate.

[0026] The attention weight is learned based on the LSTM attention mechanism, and the importance of each parameter at different time points to the current prediction result is dynamically evaluated.

[0027] In a preferred embodiment, the attention weight is learned based on the LSTM attention mechanism, and the importance of each parameter at different time points to the current prediction result is dynamically evaluated, including the following steps:

[0028] A time series data input matrix is constructed, each row representing a time point and each column representing a process parameter.

[0029] The hidden state vector of each time point is encoded by LSTM.

[0030] Based on the current hidden state and historical input features, the attention score value is calculated.

[0031] The attention score value is normalized to the attention weight by the Softmax function.

[0032] In a preferred embodiment, when the evaporation section parameter and the freezing section parameter fluctuate, adaptive control is performed according to the collaborative control rule, including the following steps:

[0033] Through continuous sliding window analysis of the data flow matrix, the sliding mean, standard deviation, and coefficient of variation statistical characteristics are used to identify whether the evaporation section or freezing section parameter is in a state of severe fluctuation.

[0034] The identification result is converted into a numerical type deviation index to form a fluctuation trend vector.

[0035] The cooperative control rule logic structure is: if the evaporation section temperature rises + the water inflow fluctuates, and at the same time is accompanied by the freezing section supercooling degree decreases, it is determined that the nucleation is blocked and the crystal formation is unstable, and the cooling rate needs to be adjusted and a small amount of crystal nucleus inducer is injected;

[0036] According to the matching result, an adaptive control strategy is executed, and the strategy includes:

[0037] Temperature-cold gradient linkage regulation: when the evaporation temperature suddenly rises to cause the concentration rate to accelerate, the freezing temperature is automatically adjusted to be low or the freezing rate is increased to maintain the stability of the crystal nucleation zone;

[0038] Flow-Viscosity dynamic coupling compensation: when the water inflow increases to cause the viscosity in the crystallizer to rise, the crystal nucleus generation rate is reduced or the stirring frequency is optimized to prevent crystal agglomeration.

[0039] In a preferred embodiment, by continuously sliding window analysis on the data flow matrix, the system uses sliding mean, standard deviation, and coefficient of variation statistical characteristics to identify whether the evaporation section or freezing section parameters are in a state of severe fluctuation, including the following steps:

[0040] A fixed length time window is set, each window contains a plurality of sampling points, and the entire time sequence is divided into a series of sliding windows, each window slides by a set step;

[0041] The sliding mean is calculated: In the formula, μ w is the sliding mean in the window w, x i is the parameter value of the i-th sampling point, t is the current time point, and w is the window size;

[0042] The sliding standard deviation is calculated: In the formula, σ w is the standard deviation in the window w, and the coefficient of variation is calculated: In the formula, CV w is the coefficient of variation in the window w, and ∈ is a numerical stabilization constant;

[0043] If σ w > σ 阈值 or CV w > CV 阈值 , it is identified that the parameter is in a fluctuation state.

[0044] In a preferred embodiment, the monitoring system dynamically collects evaporation section parameters and freezing section parameters, including the following steps:

[0045] In the evaporation section, the collected parameters include evaporation temperature, water inflow, concentration liquid salinity, and evaporation rate;

[0046] In the freezing section, the collected parameters include freezing temperature, supercooling degree, crystal nucleus generation rate and fluid viscosity in the crystallizer;

[0047] All collected data are filtered, denoised, interpolated, normalized and standardized to construct a unified time series structure and form an evaporation-freezing data stream matrix.

[0048] In a preferred embodiment, the expression of the evaporation-freezing data stream matrix is:

[0049] In the formula, X t is the evaporation-freezing data stream matrix of the tth time step, is the evaporation temperature of the tth time step, is the water inflow of the tth time step, is the concentration liquid salinity of the tth time step, is the evaporation rate of the tth time step, is the freezing temperature of the tth time step, is the supercooling degree of the tth time step, is the crystal nucleus generation rate of the tth time step, is the fluid viscosity of the tth time step.

[0050] The application also provides a mine water evaporation-freezing crystallization process monitoring system, which comprises a data acquisition unit, an adaptive control unit and a blockage risk prediction unit.

[0051] The data acquisition unit dynamically acquires evaporation section parameters and freezing section parameters.

[0052] The adaptive control unit performs adaptive control according to the cooperative control rules when the evaporation section parameters and the freezing section parameters fluctuate.

[0053] The blockage risk prediction unit dynamically analyzes the influence weight of the evaporation section parameter and the freezing section parameter fluctuation on the crystallization process through an LSTM network, performs real-time coupling analysis on the evaporation section parameter-freezing section parameter and the crystallization process, predicts the fluctuation risk level of the crystallization process, and generates a corresponding intervention strategy based on the prediction result.

[0054] In the above technical solution, the application provides the following technical effects and advantages:

[0055] The application dynamically collects evaporation section parameters and freezing section parameters, and when the evaporation section parameters and the freezing section parameters fluctuate, self-adaptive regulation and control are performed according to the cooperative control rule, and the influence weight of the evaporation section parameter and the freezing section parameter fluctuation on the crystallization process is dynamically analyzed through the LSTM network, the evaporation section parameter-freezing section parameter and the crystallization process are analyzed in real time, the fluctuation risk level of the crystallization process is predicted, and the corresponding intervention strategy is generated based on the prediction result, the intervention strategy includes automatically injecting a small amount of scale inhibitor or starting pulse backwashing when the risk is high, adjusting the freezing section cooling gradient when the risk is medium, guiding the directional growth of crystallization, and optimizing the stirring frequency when the risk is low. The crystallization particle is dispersed. The monitoring system can master the mine water quality change, equipment operation condition and crystalline salt output in real time, dynamically adjust the process parameters, ensure the stable and efficient operation of the process, improve the resource recovery rate and automation level, reduce the operation energy consumption and labor input, promote the mine water resource utilization and green mine construction. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed in the embodiments will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments described in the present application, and other drawings can also be obtained by those skilled in the art based on these drawings.

[0057] Figure 1 The method flowchart of the present application.

[0058] Figure 2 The system unit diagram of the present application.

[0059] Figure 3 The method timing diagram of the present application.

[0060] Figure 4 The method mind map of the present application. DETAILED DESCRIPTION

[0061] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be described clearly and completely in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0062] Embodiment 1: please refer to Figure 1 and Figures 3-4 The mine water evaporation-freezing crystallization process monitoring method described in the present embodiment includes the following steps:

[0063] The monitoring system dynamically collects evaporation section parameters and freezing section parameters. The evaporation section parameters include evaporation temperature, water inflow, concentrated liquid salinity, evaporation rate and other parameters. The freezing section parameters include freezing temperature, supercooling degree, crystal nucleus generation rate, fluid viscosity in the crystallizer and other parameters. When the evaporation section parameters and the freezing section parameters fluctuate, adaptive regulation is performed according to the collaborative control rule, and the influence weight of the fluctuation of the evaporation section parameters and the freezing section parameters on the crystallization process is dynamically analyzed through the LSTM network. Real-time coupling analysis is performed on the evaporation section parameters-freezing section parameters and the crystallization process, the fluctuation risk level of the crystallization process is predicted, and corresponding intervention strategies are generated based on the prediction results. The intervention strategies include automatically injecting a small amount of scale inhibitor or starting pulse backwashing in high-risk situations, adjusting the freezing section temperature gradient in medium-risk situations, guiding directional growth of crystallization, and optimizing stirring frequency in low-risk situations to disperse crystalline particles.

[0064] The present application dynamically collects evaporation section parameters and freezing section parameters, and when the evaporation section parameters and the freezing section parameters fluctuate, adaptive regulation is performed according to the collaborative control rule, and the influence weight of the fluctuation of the evaporation section parameters and the freezing section parameters on the crystallization process is dynamically analyzed through the LSTM network. Real-time coupling analysis is performed on the evaporation section parameters-freezing section parameters and the crystallization process, the fluctuation risk level of the crystallization process is predicted, and corresponding intervention strategies are generated based on the prediction results. The intervention strategies include automatically injecting a small amount of scale inhibitor or starting pulse backwashing in high-risk situations, adjusting the freezing section temperature gradient in medium-risk situations, guiding directional growth of crystallization, and optimizing stirring frequency in low-risk situations to disperse crystalline particles. The monitoring system can real-time master the change of mine water quality, the running condition of equipment and the output of crystalline salt, dynamically adjust the process parameters, ensure stable and efficient operation of the process, improve resource recovery rate and automation level, reduce operation energy consumption and labor input, and promote mine water resource utilization and green mine construction.

[0065] In embodiment 2, the monitoring system dynamically collects evaporation section parameters and freezing section parameters. The evaporation section parameters include evaporation temperature, water inflow, concentrated liquid salinity, evaporation rate and other parameters. The freezing section parameters include freezing temperature, supercooling degree, crystal nucleus generation rate, fluid viscosity in the crystallizer and other parameters. The following steps are included:

[0066] In the mine water evaporation-freezing crystallization process monitoring system, the first step is to dynamically collect and model the key process parameters of the evaporation section and the freezing section. This step aims to build a high-time-efficiency and high-precision data foundation. By deploying multiple types of industrial-grade intelligent sensors, the running state variables of the evaporation and freezing sections are systematically collected, providing bottom support for subsequent process regulation and risk prediction.

[0067] In the evaporation section, the collected parameters include:

[0068] Evaporation temperature: As a direct embodiment of heat input, its change directly affects the speed and degree of salt concentration in mine water, and is related to the formation of crystallization saturation conditions;

[0069] Inlet flow rate: This index affects the system load and heat exchange efficiency. Excessive or fluctuating flow rate can cause the concentration rhythm to be out of sync;

[0070] Concentrate salinity: It is an important index for determining whether the output liquid in the evaporation section is close to the crystallization critical point. Excessive salinity indicates that the crystallization zone may face the risk of supersaturation;

[0071] Evaporation rate: Reflects the water evaporation efficiency per unit time, and indirectly reflects the degree of contamination of the heat exchange surface and the change of heat transfer resistance.

[0072] In the freezing section, the collected parameters include:

[0073] Freezing temperature: Determines the starting point of crystal formation and is the core variable for regulating crystal morphology and growth rate;

[0074] Supercooling degree: The amplitude by which the system temperature is lower than the crystallization point of the solution is an important inducement for the formation of crystal nuclei;

[0075] Crystal nucleus generation rate: Characterizes the activity level of the granulation process in the crystallizer. Excessive rate may lead to increased risk of particle agglomeration and blockage;

[0076] Fluid viscosity in the crystallizer: Reflects the mass transfer characteristics of the fluid inside the freezing section and the concentration level of the crystalline material, which is an important parameter for crystal movement, distribution, and wall adhesion tendency.

[0077] In addition, all collected data will be filtered, denoised, missing interpolated, normalized, and standardized to construct a unified time series structure and form a high-resolution data flow matrix for the entire evaporation-freezing process. The system will also adaptively adjust the collection frequency and sampling window according to actual operating load and environmental changes to ensure the representativeness and real-time nature of the data.

[0078] The expression of the evaporation-freezing data flow matrix is:

[0079] In the formula, X t is the evaporation-freezing data flow matrix at the tth time step, is the evaporation temperature at the tth time step, is the inlet flow rate at the tth time step, is the concentrate salinity at the tth time step, is the evaporation rate at the tth time step, is the freezing temperature at the tth time step, is the supercooling degree at the tth time step, is the crystal nucleus generation rate at the tth time step, Viscosity of crystallizer fluid for the t-th time step.

[0080] The data stream matrix is a structure that fuses multi-dimensional features and time series, reflecting the joint state of each key process parameter at time t. It has the following characteristics: Multivariate: Fusion of process characteristics of evaporation and freezing sections; Sequentiality: preserves the time evolution relationship, suitable for model input such as LSTM and Transformer; High-resolution: each row represents a high time resolution sampling point, capable of capturing short-term mutations or trends; Spatial coupling: evaporation section parameters affect freezing section response, which can be used for parameter coupling analysis and dynamic modeling through column relationship.

[0081] Through the execution of this step, the monitoring system realizes fine perception of the complex heat-mass transfer process under high-salt mine water environment, and establishes a solid data foundation for subsequent parameter floating detection, dynamic modeling, risk identification and intelligent intervention.

[0082] When the evaporation section parameters and freezing section parameters float, adaptive control is performed according to synergistic control rules, including the following steps:

[0083] After completing the dynamic collection of multi-dimensional running parameters of the evaporation section and freezing section and constructing the high-resolution data stream matrix, the monitoring system enters the second phase, that is, adaptive regulation of floating parameters through synergistic control rules, so as to stabilize the crystallization process and reduce the risk of crystallization blockage.

[0084] Through continuous sliding window analysis of the data stream matrix, the system uses statistical characteristics such as moving average, standard deviation, and coefficient of variation, specifically:

[0085] Set a fixed length time window, each window contains a number of sampling points (such as sampling once per second, a total of 300 points). Divide the entire time series into a series of sliding windows, each window slides by a set step.

[0086] Calculate the moving average: In the formula, μ w is the moving average in the window w, x i is the parameter value (such as evaporation temperature) of the i-th sampling point, t is the current time point, w is the window size, representing the current analysis time length, the moving average is used to capture the local average level of the current parameter, identify short-term trends, and assist in determining whether there is a sustained increasing or decreasing trend.

[0087] Calculate the moving standard deviation:

[0088] In the formula, σ w μ represents the standard deviation within the window, which measures the intensity of fluctuations. w x is the moving average within this window. i Let be the parameter value at the i-th sampling point (e.g., evaporation temperature). The moving standard deviation reflects the degree of parameter dispersion within the window and is a core indicator for judging the severity of fluctuations. The larger the standard deviation, the more severe the fluctuations.

[0089] Calculate the coefficient of variation (CV): In the formula, σ w μ is the standard deviation within the window. w Let CV be the moving average within window w. w Here, represents the coefficient of variation within the window, and ∈ represents the numerical stabilization constant, enabling a unified comparison of parameters with different dimensions and magnitudes. CV w The higher the value, the more unstable the fluctuation of the parameter, which is suitable for comparing fluctuation risk across multiple parameters.

[0090] If σ w >σ 阈值 or CV w >CV 阈值 If the parameter is considered to be in a state of drastic fluctuation, then this sliding window analysis method allows the monitoring system to quickly and quantitatively identify various dynamic disturbances in the evaporation-freezing process, thereby constructing a data-driven stability assessment mechanism. This mechanism provides key criteria for judging fluctuation inputs for subsequent coordinated control, adaptive strategy generation, and crystallization risk prediction, effectively enhancing the response sensitivity and prevention capabilities of the entire system.

[0091] The system identifies whether the parameters of the evaporation or freezing section are in a state of drastic fluctuation. If the rate of change of the evaporation temperature exceeds the set rate of change threshold, it is identified as being in a state of drastic fluctuation. If the salinity of the concentrate shows a continuous upward trend, it is identified as being in a state of drastic fluctuation. If the subcooling occurs multiple peaks in a short period of time (e.g., 2 hours), it is identified as being in a state of drastic fluctuation. The identification results are converted into numerical deviation indicators to form a fluctuation trend vector for use in the next step of rule comparison.

[0092] The monitoring system incorporates an evaporation-freezing coordinated control rule library, which is derived from extensive operational data training and mechanistic model analysis. The rules contain the following logical structure: "If the temperature in the evaporation section rises while the influent flow rate fluctuates, and this is accompanied by a decrease in the supercooling of the freezing section, it is determined that nucleation is hindered and crystal formation is unstable, requiring an increase in the cooling rate and a small injection of a crystal nucleation inducer." This rule library supports fuzzy matching and parameter tolerance identification, enabling it to extract the optimal control strategy based on the joint fluctuation pattern of multiple parameters and compare it with the feedback effects of similar historical scenarios to select the optimal response path.

[0093] After completing the sliding window statistical analysis, the monitoring system expresses the degree of drastic fluctuation of each core operating parameter (such as evaporation temperature, subcooling, and concentrate salinity) in numerical form, forming a multi-dimensional fluctuation trend vector, defined as follows:

[0094] D t =[d1(t),d2(t),...,d n [(t)], where D t This indicates that the parameter deviates from the trend vector at time point t, d i (t) is the fluctuation deviation index of the i-th parameter at the current time, defined as: In the formula, is the real-time value of parameter i, is the mean of parameter i within the sliding window w, is the standard deviation of parameter i within the sliding window w, and ∈ is a small constant (usually 0.1) to avoid division by zero. This expression can be regarded as a normalized deviation measure, and the larger the value, the greater the deviation from the "steady state".

[0095] The monitoring system loads the embedded evaporation-freezing coupling control rule library, which has the following key features:

[0096] Statistical training model based on long-term operational data;

[0097] Supplemented by mechanistic models of thermodynamic and kinetic processes (such as nucleation models and solute precipitation rate models);

[0098] The comprehensive expert knowledge graph is transformed into an executable rule set.

[0099] Rule expression structure: Each rule typically contains a logical structure of "ternary condition + control strategy". For example:

[0100] IF (Evaporation section temperature rises ∧ Inlet water flow rate fluctuations ∧ Subcooling of freezing section decreases);

[0101] THEN (determine that nucleation is blocked + start the cooling rate adjustment + inject a small amount of nucleation inducer).

[0102] A fuzzy match is made between the current deviation vector and the historical typical fluctuation scenarios using a multidimensional Euclidean distance or cosine similarity algorithm, and a certain range of parameter fluctuation deviation is tolerated. If the matching score between the current deviation vector and the historical typical fluctuation scenarios is greater than the matching threshold, it is determined that there is an effective match.

[0103] After the system matches the preliminary response strategy, it further searches the historical case library (Historical-Response-Repository) and performs the following actions:

[0104] A historical feedback result scoring function is set, considering key indicators such as intervention duration, crystallization stability recovery time, and crystalline particle uniformity improvement value. The response paths with the highest scores are extracted as candidates. Strategy feedback is iteratively optimized:

[0105] Using reinforcement learning or Bayesian optimization methods, the effects of candidate strategies are predicted, and the path with the smallest intervention and the largest stability improvement is selected. The strategy parameters (such as antifouling agent injection rate and freezing section cooling rate) are dynamically adjusted to achieve fine-tuned response path control. The above steps are part of the prior art, and will not be described in detail in this application.

[0106] According to the matching result, an adaptive control strategy is executed, including but not limited to:

[0107] Temperature-cold gradient linkage control: when the evaporation temperature suddenly rises, causing the concentration rate to accelerate, the system can automatically lower the freezing temperature or increase the freezing rate to maintain the stability of the crystal nucleation zone;

[0108] Flow-Viscosity Dynamic Coupling Compensation: When the water inflow increases, causing the viscosity in the crystallizer to rise, the system can temporarily reduce the crystal nucleation rate or optimize the stirring frequency to prevent crystal agglomeration.

[0109] Through LSTM (Long Short-Term Memory) network dynamic analysis of the influence weight of evaporation section parameter and freezing section parameter fluctuation on the crystallization process, it is the key link to realize intelligent prediction and collaborative control. Through LSTM network dynamic analysis of the influence weight of evaporation section parameter and freezing section parameter fluctuation on the crystallization process, including the following steps:

[0110] A multi-layer LSTM network is constructed to model the time series. The core structure of LSTM includes: forget gate, input gate, state update gate, and output gate. These gating mechanisms ensure that the network has good perception ability for trends, mutations, time lags, and other phenomena in long time series.

[0111] The attention mechanism based on LSTM learns attention weights and dynamically evaluates the importance of each parameter to the current prediction result at different time points. A code example is shown below:

[0112] import_torch

[0113] import_torch.nn_as_nn

[0114] import_torch.nn.functional_as_F

[0115] class_LSTMWithAttention(nn.Module):

[0116] def__init__(self, input_dim, hidden_dim, output_dim, lstm_layers=2, dropout=0.2):

[0117] super(LSTMWithAttention, self).__init__()

[0118] self.hidden_dim = hidden_dim

[0119] self.lstm_layers=lstm_layers

[0120] #LSTM layer

[0121] self.lstm=nn.LSTM(input_dim, hidden_dim, num_layers=lstm_layers, batch_first=True, dropout=dropout)

[0122] #Trainable parameters of the attention mechanism

[0123] self.attn_W_h=nn.Linear(hidden_dim, hidden_dim)

[0124] self.attn_W_x=nn.Linear(input_dim, hidden_dim)

[0125] self.attn_v=nn.Linear(hidden_dim, 1, bias=False)

[0126] #Output Layer

[0127] self.fc = nn.Linear(hidden_dim, output_dim)

[0128] def attention_scores(self, h_t, x_t):

[0129] # h_t: [batch, hidden_dim]

[0130] # x_t: [batch, time_steps, input_dim]

[0131] batch_size, time_steps, input_dim = x_t.size()

[0132] h_expand = h_t.unsqueeze(1).expand(-1, time_steps, -1) # [batch, time, hidden_dim]

[0133] score = self.attn_v(torch.tanh(self.attn_W_h(h_expand) + self.attn_W_x(x_t))) # [batch, time, 1]

[0134] attn_weights = F.softmax(score, dim=1) # time dimension normalized

[0135] Return attn_weights

[0136] def forward(self, x):

[0137] # x: [batch, time_steps, input_dim]

[0138] lstm_out, (h_n, c_n) = self.lstm(x) # lstm_out: [batch, time, hidden]

[0139] h_last = lstm_out[:, -1, :] # Take the output of the last layer as reference

[0140] attn_weights = self.attention_scores(h_last, x) # [batch, time, 1]

[0141] # Attention weighted sum

[0142] context_vector = torch.sum(attn_weights * lstm_out, dim=1) # [batch, hidden_dim]

[0143] output = self.fc(context_vector) # final prediction

[0144] return_output, attn_weights.

[0145] In the above code, the model structure is explained as follows:

[0146] Input data dimensions: [batch_size, time_steps, num_features].

[0147] LSTM layer: multiple layers stacked, output time series hidden state.

[0148] Attention mechanism: for each time step, calculate the attention weight of each feature.

[0149] Output: predict a certain target in the crystallization process (such as crystal particle size, blockage risk level, etc.).

[0150] Use the model code example:

[0151] # Assuming there are 10 process parameters, time steps are 30, and the output is 1-dimensional (such as blockage risk value)

[0152] model = LSTMWithAttention(input_dim=10, hidden_dim=64, output_dim=1)

[0153] # Simulate input [batch_size, time_steps, input_dim]

[0154] x_sample = torch.randn(16, 30, 10)

[0155] # Perform forward propagation

[0156] output, attention_weights = model(x_sample)

[0157] print("Prediction result:", output.shape) # [16, 1]

[0158] print("Attention weight dimensions:", attention_weights.shape) # [16, 30, 1]

[0159] Attention weights can be extracted for visualization analysis, such as drawing heatmaps to see which parameters have the greatest impact on crystallization at what time points.

[0160] By integrating an LSTM neural network with an attention mechanism, the influence of each input parameter at different time points is dynamically learned during time series prediction, thereby assessing the contribution of each parameter to the crystallization state (such as crystal size distribution, crystallization rate, and blockage risk). The overall process includes the following sub-steps:

[0161] Construct a time series data input matrix, where each row represents a time point and each column represents a process parameter;

[0162] The hidden state vector at each time point is encoded using LSTM;

[0163] Based on the current hidden state and historical input features, the attention score is calculated, expressed as e. i,t =v T ·tanh(W h h t +W x x i,t +b), where e i,t h is the attention score of the i-th parameter at time t. t Let x be the hidden state vector of the LSTM at time t. i,t For the i-th input feature (i.e., the process parameter value) at time t, W h W x Here is the weight matrix in the attention mechanism, b is the bias term, and v T Here is a trainable parameter vector, and tanh(*) is the hyperbolic tangent function used to introduce nonlinearity;

[0164] The attention weights are normalized using the Softmax function, and the expression is: In the formula, α i,t Let be the attention weight of the i-th parameter to the prediction result at time t, and N be the total number of input parameters;

[0165] The weights are used for subsequent predictions or to explain the relative influence of each parameter on the result.

[0166] Assuming the current time t, there are three key input parameters: evaporation temperature T = 78℃, concentrate salinity S = 95g / L, and subcooling ΔT = 2.5℃.

[0167] Assumption: Hidden state vector h t = [0.5, -0.2], parameter vector x1,t = 78, x 2,t = 95, x 3,t = 2.5, weight matrix setting (simplified as linear transformation): W h = [0.2, 0.1], dimension and h t match, W x = 0.05, each input is a scalar, v = 1.0, b = 0, the attention score value calculated by substituting the formula is:

[0168] Take the first parameter (evaporation temperature) as an example: e 1,t = 1.0 tanh (0.2 0.5 + 0.1 - 0.2 + 0.05 78) = tanh (0.1 - 0.02 + 3.9) = tanh (3.98) ≈ 0.999;

[0169] Similarly:

[0170] e 2,t = tanh (0.1 - 0.02 + 0.05 95) - tanh (5.83) ≈ 0.9999;

[0171] e 3,t = tanh (0.1 - 0.02 + 0.05 2.5) = tanh (0.1 - 0.02 + 0.125) = tanh (0.205) ≈ 0.202;

[0172] After Softmax normalization:

[0173]

[0174] Real-time coupling analysis of evaporation section parameters-freezing section parameters and crystallization process, prediction of crystallization process fluctuation risk level, generation of corresponding intervention strategy based on prediction results, intervention strategy including automatic injection of trace amount of scale inhibitor or start of pulse backwashing in high risk, adjustment of freezing section cooling gradient in medium risk, guiding of crystallization directional growth, optimization of stirring frequency in low risk, dispersion of crystallization particles, including the following steps:

[0175] The key process parameters of the evaporation section and the freezing section are weighted and fused according to the influence weight of the parameters on the crystallization process, and a crystallization driving feature vector (Crystallization-Driving-Vector, abbreviated as CDV) is formed, and the expression is:

[0176] In the formula, CDV t is the crystallization driving feature vector at time t, is the attention weight of the i-th evaporation section parameter output by LSTM at time t on the crystallization process, is the measured value of the i-th evaporation section parameter (such as evaporation temperature, concentration of concentrated liquid, etc.) at time t, is the attention weight of the j-th freezing section parameter output by the LSTM at time t on the crystallization process, is the measured value of the j-th freezing section parameter (such as supercooling degree, crystal nucleus generation rate, etc.) at time t, m is the number of evaporation section parameters, and n is the number of freezing section parameters.

[0177] A short-term trend window is constructed using the time series CDV vector to quantify the change trend of the crystallization process and evaluate the sensitivity and potential fluctuation risk of the system, and the expression is:

[0178] In the formula, CSI t is the crystallization sensitivity index at time t, CDV t-k:t is the crystallization driving feature vector in the time window from t-k to t, Var(*) is the variance in the window, reflecting the fluctuation degree, is the mean value in the window, and γ is an adjustment factor for amplifying or suppressing sensitivity (usually set between 0.8-1.2), and k is the sliding window length (such as 5-10 time steps).

[0179] According to the crystallization sensitivity index, the risk interval is divided, and the risk level is judged:

[0180] If CSI t > θ3, it is high risk (High-Risk);

[0181] If θ2 < CSI t ≤ θ3, it is medium risk (Medium-Risk);

[0182] If CSI t ≤ θ2, it is low risk (Low-Risk);

[0183] Wherein, θ2 and θ3 are empirical threshold values, such as set according to historical statistics: θ2 = 0.3, θ3 = 0.6.

[0184] Based on the predicted pipe blockage risk level, the monitoring system will trigger differentiated intervention strategies in real time to realize the optimization and control of the crystallization process and prevent the occurrence of abnormal situations such as excessive aggregation of crystals and pipe blockage. The intervention strategy is divided into high risk, medium risk and low risk according to the risk level, and each level is matched with specific countermeasures. The following are the specific implementation steps:

[0185] The system receives the risk level prediction results (such as "High-Risk", "Medium-Risk" or "Low-Risk") output by the LSTM network and the CSI analysis module, maps them to control trigger codes (for example: high risk = 3, medium risk = 2, low risk = 1), and matches the corresponding intervention strategies.

[0186] The system has an intervention strategy repository built-in, and each risk level is bound to a set of refined response mechanisms, which are matched as follows:

[0187] ① High risk level (High-Risk): The nucleation rate decreases significantly, the concentration of the solution increases sharply, and the supercooling degree remains abnormally low, indicating that the crystal is oversaturated and the fluid flow in the crystallizer is decreasing.

[0188] Intervention measures: Automatically inject a small amount of scale inhibitor (such as phosphate) into the crystallizer inlet, the dosage is calculated based on flow and concentration feedback; start the pulse backwash module to periodically clean the crystallization pipeline, set the flushing frequency and time (such as 1 time every 10 minutes, 60 seconds each time); adjust the refrigeration load of the refrigeration section compressor to form a transient cooling wave to suppress the coalescence of large crystal particles.

[0189] ② Medium risk level (Medium-Risk): Evaporation rate fluctuates, crystallization trend nonlinearly fluctuates, and crystal nucleus growth rate is unstable but does not break through the critical threshold.

[0190] Intervention measures: Adjust the temperature gradient of the refrigeration section (the temperature drop speed increases from 1.5℃ / min to 2.0℃ / min) to promote the formation of small-sized particles; enable directional crystallization induction mechanism to set electromagnetic disturbance sources or directional crystal nucleus seeds in the crystallizer to make crystal growth along a controllable direction; moderately increase the stirring frequency (such as from 60 rpm to 80 rpm) to alleviate the formation of local saturation areas in the fluid.

[0191] ③ Low risk level (Low-Risk): System parameter fluctuations are smooth, crystal nucleus generation and growth rates fluctuate within normal ranges, and crystallization has no blocking trend.

[0192] Intervention measures: Optimize the stirring frequency and period to avoid crystal aggregation in local static areas (such as using a periodic high-low speed alternating stirring strategy based on flow feedback); turn on the particle dispersion control module to perform ultrasonic or mechanical oscillation treatment on the slurry in the crystallizer; continuously record the operating state and system response results under the current parameters for optimizing the subsequent fuzzy control rule base.

[0193] Strategy execution feedback and adjustment closed loop:

[0194] During the execution of the intervention strategy, the system continuously collects the following parameters for feedback closed-loop adjustment: the change in the concentrated liquid salinity within 5 minutes after the injection of the scale inhibitor; the comparison of the fluid viscosity in the freezing section before and after the pulse backwashing; the response curve of the crystal nucleus generation rate after adjusting the cooling gradient.

[0195] According to the feedback, the system dynamically evaluates the effect of the intervention strategy, and if the effect is not good (such as the salinity continues to rise, and the viscosity does not decrease significantly), the strategy upgrade or diversion (for example, switching from the medium risk to the high risk scheme) is triggered.

[0196] Embodiment 3: Please refer to Figure 2 As shown in the figure, the mine water evaporation-freezing crystallization process monitoring system described in this embodiment includes a data acquisition unit, an adaptive control unit, and a blockage risk prediction unit.

[0197] The data acquisition unit: dynamically acquires evaporation section parameters and freezing section parameters, the evaporation section parameters including evaporation temperature, water inflow, concentrated liquid salinity, evaporation rate and other parameters, the freezing section parameters including freezing temperature, supercooling degree, crystal nucleus generation rate, fluid viscosity in the crystallizer and other parameters, and the evaporation section parameters and the freezing section parameters are sent to the adaptive control unit and the blockage risk prediction unit.

[0198] The adaptive control unit: when the evaporation section parameters and the freezing section parameters fluctuate, adaptive control is performed according to the cooperative control rule;

[0199] The blockage risk prediction unit: the influence weight of the evaporation section parameter and the freezing section parameter fluctuation on the crystallization process is dynamically analyzed through the LSTM network, the evaporation section parameter-freezing section parameter and the crystallization process are analyzed in real time, the crystallization process fluctuation risk level is predicted, the corresponding intervention strategy is generated based on the prediction result, the intervention strategy includes automatically injecting a small amount of scale inhibitor or starting pulse backwashing in high risk, adjusting the cooling gradient of the freezing section to guide the directional growth of crystallization in medium risk, and optimizing the stirring frequency to disperse the crystalline particles in low risk.

[0200] It should be understood that the term "and / or" herein only describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B, which can represent three cases of A alone, A and B together, and B alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents an "or" relationship between the front and rear associated objects, but it can also represent an "and / or" relationship, which can be understood according to the context before and after.

[0201] It should be understood that in various embodiments of the present application, the size of the sequence number of the above processes does not mean the order of execution, and the execution order of the processes should be determined by their functions and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0202] Those skilled in the art can understand that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized by electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software manner depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to realize the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application. Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described system, device and unit can refer to the corresponding processes in the foregoing method embodiments, which will not be described here.

[0203] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited to this. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A monitoring method for mine water evaporation-freeze crystallization process, characterized in that: The monitoring method includes the following steps: The monitoring system dynamically collects parameters from the evaporation section and the freezing section. When the parameters of the evaporation section and the freezing section fluctuate, adaptive adjustment is performed according to the coordinated control rules; The influence weights of fluctuations in evaporation and freezing section parameters on the crystallization process are dynamically analyzed using an LSTM network. Real-time coupled analysis of evaporation section parameters, freezing section parameters, and crystallization process is performed to predict the risk level of crystallization process fluctuations, and corresponding intervention strategies are generated based on the prediction results.

2. The monitoring method for mine water evaporation-freeze crystallization process according to claim 1, characterized in that: Real-time coupled analysis of evaporation section parameters, freezing section parameters, and the crystallization process is performed to predict the risk level of crystallization process fluctuations. This includes the following steps: The key process parameters of the evaporation section and the freezing section are weighted and fused according to their influence on the crystallization process to form a crystallization-driven feature vector. By constructing a short-term trend window using crystallization-driven eigenvectors, the changing trend of the crystallization process is quantified, and the crystallization sensitivity index (CSI) is obtained. t ; Risk ranges are divided based on crystallization sensitivity indicators, and risk levels are determined accordingly. If CSI t If θ2 > θ3, then it is high risk; if θ2 > θ3, then it is high risk. <CSI t If ≤θ3, it is considered medium risk; if CSI t If the value is ≤θ2, then it is considered low risk.

3. The monitoring method for mine water evaporation-freeze crystallization process according to claim 2, characterized in that: The crystallization-driven eigenvector is formed, expressed as: In the formula, CDV t Let be the crystallization-driven eigenvector at time t. The attention weights of the i-th evaporation segment parameters output by the LSTM at time t for the crystallization process. Let be the measured value of the parameter of the i-th evaporation section at time t. The attention weight of the j-th frozen section parameter in the LSTM output at time t for the crystallization process. Let m be the measured value of the j-th freezing section parameter at time t, m be the number of evaporation section parameters, and n be the number of freezing section parameters. The crystallization sensitivity index is obtained by the following expression: In the formula, CSI t CDV is the crystallization sensitivity index at time t. t-k:t Let Var(*) be the crystallization-driven eigenvector within the time window from tk to t, and Var(*) be the variance within the window. Let γ be the mean value within the window, γ be the adjustment factor, and k be the sliding window length.

4. The monitoring method for mine water evaporation-freeze crystallization process according to claim 3, characterized in that: The influence weights of fluctuations in evaporation and freezing parameters on the crystallization process are dynamically analyzed using an LSTM network, including the following steps: The time series is modeled by constructing a multi-layer LSTM network. The LSTM structure includes: forget gate, input gate, state update gate and output gate. The attention mechanism based on LSTM learns attention weights and dynamically evaluates the importance of each parameter to the current prediction result at different time points.

5. The monitoring method for mine water evaporation-freeze crystallization process according to claim 4, characterized in that: The attention mechanism based on LSTM learns attention weights and dynamically evaluates the importance of each parameter to the current prediction result at different time points, including the following steps: Construct a time series data input matrix, where each row represents a time point and each column represents a process parameter; The hidden state vector at each time point is encoded using LSTM; Calculate the attention score based on the current hidden state and historical input features; The attention weights are normalized using the Softmax function.

6. The monitoring method for mine water evaporation-freeze crystallization process according to claim 5, characterized in that: When the parameters of the evaporation section and the freezing section fluctuate, adaptive adjustment is performed according to the coordinated control rules, including the following steps: By analyzing the data flow matrix through a continuous sliding window, and utilizing the statistical characteristics of the moving mean, standard deviation, and coefficient of variation, we can identify whether the parameters of the evaporation or freezing section are in a state of drastic fluctuation. The identification results are converted into numerical deviation indicators to form a floating trend vector; The logical structure of the collaborative control rule is as follows: if the temperature of the evaporation section rises and the influent flow rate fluctuates, and at the same time the supercooling of the freezing section decreases, it is determined that nucleation is hindered and there is a risk of unstable crystal formation. The cooling rate needs to be increased and a small amount of crystal nucleation inducer needs to be injected. Based on the matching results, an adaptive adjustment strategy is executed, which includes: Temperature-cold gradient linkage control: When the evaporation temperature rises suddenly and the concentration rate increases, the freezing temperature is automatically lowered or the freezing rate is increased to maintain the stability of the crystal nucleation zone; Flow-viscosity dynamic coupling compensation: When the influent flow rate increases, the viscosity inside the crystallizer increases, reducing the crystal nucleus generation rate or optimizing the stirring frequency to prevent crystal agglomeration.

7. The monitoring method for mine water evaporation-freeze crystallization process according to claim 6, characterized in that: By performing continuous sliding window analysis on the data flow matrix, the system utilizes statistical characteristics such as the moving mean, standard deviation, and coefficient of variation to identify whether the parameters of the evaporation or freezing section are in a state of drastic fluctuation, including the following steps: Set a fixed-length time window, with each window containing several sampling points, and divide the entire time series into a series of sliding windows, with each window sliding according to a set step size; Calculate the moving average: In the formula, μ w Let x be the moving average within window w. i Let t be the parameter value of the i-th sampling point, t be the current time point, and w be the window size; Calculate the sliding standard deviation: In the formula, σ w Calculate the coefficient of variation for the standard deviation within window w: In the formula, CV w is the coefficient of variation within window w, and ∈ is the numerical stabilization constant; If σ w >σ 阈值 or CV w >CV 阈值 The identification parameters are fluctuating.

8. The monitoring method for mine water evaporation-freeze crystallization process according to claim 7, characterized in that: The monitoring system dynamically collects parameters from the evaporation section and the freezing section, including the following steps: In the evaporation section, the parameters collected include evaporation temperature, influent flow rate, concentrate salinity, and evaporation rate. In the freezing section, the parameters collected include freezing temperature, supercooling, crystal nucleation rate, and fluid viscosity inside the crystallizer; All collected data undergoes filtering, noise reduction, missing data imputation, normalization, and standardization to construct a unified time series structure, forming an evaporation-freezing data stream matrix.

9. The monitoring method for mine water evaporation-freeze crystallization process according to claim 8, characterized in that: The expression for the evaporation-freezing data stream matrix is: In the formula, X t Let be the evaporation-freezing data stream matrix at time step t. Let be the evaporation temperature at time step t. Let be the inflow rate at time step t. Let be the salinity of the concentrate at time step t. Let be the evaporation rate at time step t. Let be the freezing temperature at time step t. Let be the supercooling at time step t. Let be the nucleus generation rate at time step t. Let be the viscosity of the crystallizer fluid at time step t.

10. A monitoring system for mine water evaporation-freeze crystallization process, used to implement the monitoring method according to any one of claims 1-9, characterized in that: It includes a data acquisition unit, an adaptive control unit, and a congestion risk prediction unit; Data acquisition unit: dynamically acquires parameters of the evaporation section and the freezing section; Adaptive control unit: When the parameters of the evaporation section and the freezing section fluctuate, it performs adaptive control according to the cooperative control rules; Blockage Risk Prediction Unit: Dynamically analyzes the influence weight of evaporation section parameters and freezing section parameters on the crystallization process through LSTM network, performs real-time coupling analysis of evaporation section parameters, freezing section parameters and crystallization process, predicts the risk level of crystallization process fluctuation, and generates corresponding intervention strategies based on the prediction results.

Images